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  * Copyright (C) 2002, 2003 Cluster File Systems, Inc.
   5  * Author: Phil Schwan <phil@clusterfs.com>
   6  *         Peter Braam <braam@clusterfs.com>
   7  *         Mike Shaver <shaver@clusterfs.com>
   8  *
   9  *   This file is part of Lustre, http://www.lustre.org.
  10  *
  11  *   Lustre is free software; you can redistribute it and/or
  12  *   modify it under the terms of version 2 of the GNU General Public
  13  *   License as published by the Free Software Foundation.
  14  *
  15  *   Lustre is distributed in the hope that it will be useful,
  16  *   but WITHOUT ANY WARRANTY; without even the implied warranty of
  17  *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  18  *   GNU General Public License for more details.
  19  *
  20  *   You should have received a copy of the GNU General Public License
  21  *   along with Lustre; if not, write to the Free Software
  22  *   Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  23  */
  24
  25 #ifndef EXPORT_SYMTAB
  26 # define EXPORT_SYMTAB
  27 #endif
  28 #define DEBUG_SUBSYSTEM S_LOV
  29 #ifdef __KERNEL__
  30 #include <linux/slab.h>
  31 #include <linux/module.h>
  32 #include <linux/init.h>
  33 #include <linux/slab.h>
  34 #include <linux/pagemap.h>
  35 #include <linux/seq_file.h>
  36 #include <asm/div64.h>
  37 #else
  38 #include <liblustre.h>
  39 #endif
  40
  41 #include <linux/obd_support.h>
  42 #include <linux/lustre_lib.h>
  43 #include <linux/lustre_net.h>
  44 #include <linux/lustre_idl.h>
  45 #include <linux/lustre_dlm.h>
  46 #include <linux/lustre_mds.h>
  47 #include <linux/obd_class.h>
  48 #include <linux/obd_lov.h>
  49 #include <linux/obd_ost.h>
  50 #include <linux/lprocfs_status.h>
  51 #include <linux/lustre_audit.h>
  52 #include "lov_internal.h"
  53
  54 /* obd methods */
  55 #define MAX_STRING_SIZE 128
  56 static int lov_connect_obd(struct obd_device *obd, struct lov_tgt_desc *tgt,
  57                            int activate, struct obd_connect_data *conn_data,
  58                            unsigned long connect_flags)
  59 {
  60         struct obd_uuid lov_osc_uuid = { "LOV_OSC_UUID" };
  61         struct obd_uuid *tgt_uuid = &tgt->uuid;
  62
  63 #ifdef __KERNEL__
  64         struct proc_dir_entry *lov_proc_dir;
  65 #endif
  66         struct lov_obd *lov = &obd->u.lov;
  67         struct lustre_handle conn = {0, };
  68         struct obd_device *tgt_obd;
  69         int rc;
  70         ENTRY;
  71
  72         tgt_obd = class_find_client_obd(tgt_uuid, OBD_OSC_DEVICENAME,
  73                                         &obd->obd_uuid);
  74
  75         if (!tgt_obd) {
  76                 CERROR("Target %s not attached\n", tgt_uuid->uuid);
  77                 RETURN(-EINVAL);
  78         }
  79
  80         if (!tgt_obd->obd_set_up) {
  81                 CERROR("Target %s not set up\n", tgt_uuid->uuid);
  82                 RETURN(-EINVAL);
  83         }
  84
  85         if (activate) {
  86                 tgt_obd->obd_no_recov = 0;
  87                 ptlrpc_activate_import(tgt_obd->u.cli.cl_import);
  88         }
  89
  90         if (tgt_obd->u.cli.cl_import->imp_invalid) {
  91                 CERROR("not connecting OSC %s; administratively "
  92                        "disabled\n", tgt_uuid->uuid);
  93                 rc = obd_register_observer(tgt_obd, obd);
  94                 if (rc) {
  95                         CERROR("Target %s register_observer error %d; "
  96                                "will not be able to reactivate\n",
  97                                tgt_uuid->uuid, rc);
  98                 }
  99                 RETURN(0);
 100         }
 101
 102         rc = obd_connect(&conn, tgt_obd, &lov_osc_uuid, conn_data,
 103                          connect_flags);
 104         if (rc) {
 105                 CERROR("Target %s connect error %d\n", tgt_uuid->uuid, rc);
 106                 RETURN(rc);
 107         }
 108         tgt->ltd_exp = class_conn2export(&conn);
 109
 110         rc = obd_register_observer(tgt_obd, obd);
 111         if (rc) {
 112                 CERROR("Target %s register_observer error %d\n",
 113                        tgt_uuid->uuid, rc);
 114                 obd_disconnect(tgt->ltd_exp, 0);
 115                 tgt->ltd_exp = NULL;
 116                 RETURN(rc);
 117         }
 118
 119         lov_tgt_set_flags(lov, tgt, LTD_ACTIVE);
 120
 121 #ifdef __KERNEL__
 122         lov_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
 123         if (lov_proc_dir) {
 124                 struct obd_device *osc_obd = class_conn2obd(&conn);
 125                 struct proc_dir_entry *osc_symlink;
 126                 char name[MAX_STRING_SIZE + 1];
 127
 128                 LASSERT(osc_obd != NULL);
 129                 LASSERT(osc_obd->obd_type != NULL);
 130                 LASSERT(osc_obd->obd_type->typ_name != NULL);
 131                 name[MAX_STRING_SIZE] = '\0';
 132                 snprintf(name, MAX_STRING_SIZE, "../../../%s/%s",
 133                          osc_obd->obd_type->typ_name,
 134                          osc_obd->obd_name);
 135                 osc_symlink = proc_symlink(osc_obd->obd_name, lov_proc_dir,
 136                                            name);
 137                 if (osc_symlink == NULL) {
 138                         CERROR("could not register LOV target "
 139                                "/proc/fs/lustre/%s/%s/target_obds/%s\n",
 140                                obd->obd_type->typ_name, obd->obd_name,
 141                                osc_obd->obd_name);
 142                         lprocfs_remove(lov_proc_dir);
 143                         lov_proc_dir = NULL;
 144                 }
 145         }
 146 #endif
 147
 148         RETURN(0);
 149 }
 150
 151 static int lov_connect(struct lustre_handle *conn, struct obd_device *obd,
 152                        struct obd_uuid *cluuid, struct obd_connect_data *data,
 153                        unsigned long flags)
 154 {
 155 #ifdef __KERNEL__
 156         struct proc_dir_entry *lov_proc_dir;
 157 #endif
 158         struct lov_obd *lov = &obd->u.lov;
 159         struct lov_tgt_desc *tgt;
 160         struct obd_export *exp;
 161         int rc, rc2, i;
 162         ENTRY;
 163
 164         rc = class_connect(conn, obd, cluuid);
 165         if (rc)
 166                 RETURN(rc);
 167
 168         exp = class_conn2export(conn);
 169
 170         /* We don't want to actually do the underlying connections more than
 171          * once, so keep track. */
 172         lov->refcount++;
 173         if (lov->refcount > 1) {
 174                 class_export_put(exp);
 175                 RETURN(0);
 176         }
 177
 178 #ifdef __KERNEL__
 179         lov_proc_dir = lprocfs_register("target_obds", obd->obd_proc_entry,
 180                                         NULL, NULL);
 181         if (IS_ERR(lov_proc_dir)) {
 182                 CERROR("could not register /proc/fs/lustre/%s/%s/target_obds.",
 183                        obd->obd_type->typ_name, obd->obd_name);
 184                 lov_proc_dir = NULL;
 185         }
 186 #endif
 187
 188         /* connect_flags is the MDS number, save for use in lov_add_obd */
 189         lov->lov_connect_flags = flags;
 190         for (i = 0, tgt = lov->tgts; i < lov->desc.ld_tgt_count; i++, tgt++) {
 191                 if (obd_uuid_empty(&tgt->uuid))
 192                         continue;
 193                 rc = lov_connect_obd(obd, tgt, 0, data, flags);
 194                 if (rc)
 195                         GOTO(out_disc, rc);
 196         }
 197
 198         class_export_put(exp);
 199         RETURN (0);
 200
 201  out_disc:
 202 #ifdef __KERNEL__
 203         if (lov_proc_dir)
 204                 lprocfs_remove(lov_proc_dir);
 205 #endif
 206
 207         while (i-- > 0) {
 208                 struct obd_uuid uuid;
 209                 --tgt;
 210                 lov_tgt_clear_flags(lov, tgt, LTD_ACTIVE);
 211
 212                 /* save for CERROR below; (we know it's terminated) */
 213                 uuid = tgt->uuid;
 214                 rc2 = obd_disconnect(tgt->ltd_exp, 0);
 215                 if (rc2)
 216                         CERROR("error: LOV target %s disconnect on OST idx %d: "
 217                                "rc = %d\n", uuid.uuid, i, rc2);
 218         }
 219         class_disconnect(exp, 0);
 220         RETURN (rc);
 221 }
 222
 223 static int lov_disconnect_obd(struct obd_device *obd,
 224                               struct lov_tgt_desc *tgt,
 225                               unsigned long flags)
 226 {
 227 #ifdef __KERNEL__
 228         struct proc_dir_entry *lov_proc_dir;
 229 #endif
 230         struct obd_device *osc_obd = class_exp2obd(tgt->ltd_exp);
 231         struct lov_obd *lov = &obd->u.lov;
 232         int rc;
 233         ENTRY;
 234
 235 #ifdef __KERNEL__
 236         lov_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
 237         if (lov_proc_dir) {
 238                 struct proc_dir_entry *osc_symlink;
 239
 240                 osc_symlink = lprocfs_srch(lov_proc_dir, osc_obd->obd_name);
 241                 if (osc_symlink) {
 242                         lprocfs_remove(osc_symlink);
 243                 } else {
 244                         CERROR("/proc/fs/lustre/%s/%s/target_obds/%s missing\n",
 245                                obd->obd_type->typ_name, obd->obd_name,
 246                                osc_obd->obd_name);
 247                 }
 248         }
 249 #endif
 250         if (obd->obd_no_recov) {
 251                 /* Pass it on to our clients.
 252                  * XXX This should be an argument to disconnect,
 253                  * XXX not a back-door flag on the OBD.  Ah well.
 254                  */
 255                 if (osc_obd)
 256                         osc_obd->obd_no_recov = 1;
 257         }
 258
 259         /* XXX - shouldn't this be after the obd_disconnect() ? */
 260         obd_register_observer(tgt->ltd_exp->exp_obd, NULL);
 261
 262         rc = obd_disconnect(tgt->ltd_exp, flags);
 263         if (rc) {
 264                 if (lov_tgt_active(lov, tgt, 0)) {
 265                         lov_tgt_decref(lov, tgt);
 266                         CERROR("Target %s disconnect error %d\n",
 267                                tgt->uuid.uuid, rc);
 268                 }
 269         }
 270
 271         CDEBUG(D_CONFIG, "idx: %d flags: 0x%x active: %d\n",
 272                tgt - lov->tgts, tgt->ltd_flags,
 273                lov->desc.ld_active_tgt_count);
 274
 275         lov_tgt_clear_flags(lov, tgt, LTD_ACTIVE);
 276         tgt->ltd_exp = NULL;
 277         RETURN(0);
 278 }
 279
 280 static int lov_disconnect(struct obd_export *exp, unsigned long flags)
 281 {
 282         struct obd_device *obd = class_exp2obd(exp);
 283 #ifdef __KERNEL__
 284         struct proc_dir_entry *lov_proc_dir;
 285 #endif
 286         struct lov_obd *lov = &obd->u.lov;
 287         struct lov_tgt_desc *tgt;
 288         int rc, i;
 289         ENTRY;
 290
 291         if (!lov->tgts)
 292                 goto out_local;
 293
 294         /* Only disconnect the underlying layers on the final disconnect. */
 295         lov->refcount--;
 296         if (lov->refcount != 0)
 297                 goto out_local;
 298
 299         for (i = 0, tgt = lov->tgts; i < lov->desc.ld_tgt_count; i++, tgt++) {
 300                 if (tgt->ltd_exp)
 301                         lov_disconnect_obd(obd, tgt, flags);
 302         }
 303
 304 #ifdef __KERNEL__
 305         lov_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
 306         if (lov_proc_dir) {
 307                 lprocfs_remove(lov_proc_dir);
 308         } else {
 309                 CERROR("/proc/fs/lustre/%s/%s/target_obds missing.",
 310                        obd->obd_type->typ_name, obd->obd_name);
 311         }
 312 #endif
 313
 314  out_local:
 315         rc = class_disconnect(exp, 0);
 316         RETURN(rc);
 317 }
 318
 319 /* Error codes:
 320  *
 321  *  -EINVAL  : UUID can't be found in the LOV's target list
 322  *  -ENOTCONN: The UUID is found, but the target connection is bad (!)
 323  *  -EBADF   : The UUID is found, but the OBD is the wrong type (!)
 324  */
 325 static int lov_set_osc_active(struct lov_obd *lov, struct obd_uuid *uuid,
 326                               int activate)
 327 {
 328         struct lov_tgt_desc *tgt;
 329         int i, rc = 0;
 330         ENTRY;
 331
 332         CDEBUG(D_INFO, "Searching in lov %p for uuid %s (activate=%d)\n",
 333                lov, uuid->uuid, activate);
 334
 335         lov_tgts_lock(lov);
 336         for (i = 0, tgt = lov->tgts; i < lov->desc.ld_tgt_count; i++, tgt++) {
 337                 if (tgt->ltd_exp == NULL)
 338                         continue;
 339
 340                 CDEBUG(D_INFO, "lov idx %d is %s conn "LPX64"\n",
 341                        i, tgt->uuid.uuid, tgt->ltd_exp->exp_handle.h_cookie);
 342
 343                 if (obd_uuid_equals(uuid, &tgt->uuid))
 344                         break;
 345         }
 346
 347         if (i == lov->desc.ld_tgt_count)
 348                 GOTO(out, rc = -EINVAL);
 349
 350         if (activate) {
 351                 if (tgt->ltd_flags & LTD_ACTIVE) {
 352                         CDEBUG(D_CONFIG|D_INFO, "OSC %s already active!\n",
 353                                uuid->uuid);
 354                         GOTO(out, rc);
 355                 }
 356                 CDEBUG(D_INFO, "Marking OSC %s active\n", uuid->uuid);
 357                 tgt->ltd_flags |= LTD_ACTIVE;
 358                 lov->desc.ld_active_tgt_count++;
 359         } else {
 360                 if ((tgt->ltd_flags & LTD_ACTIVE) == 0) {
 361                         CDEBUG(D_CONFIG|D_INFO, "OSC %s already inactive!\n",
 362                                uuid->uuid);
 363                         GOTO(out, rc);
 364                 }
 365                 CDEBUG(D_INFO, "Marking OSC %s inactive\n", uuid->uuid);
 366                 tgt->ltd_flags &= ~LTD_ACTIVE;
 367                 lov->desc.ld_active_tgt_count--;
 368         }
 369
 370         EXIT;
 371  out:
 372         lov_tgts_unlock(lov);
 373         return rc;
 374 }
 375
 376 static int lov_notify(struct obd_device *obd, struct obd_device *watched,
 377                       int active, void *data)
 378 {
 379         struct obd_uuid *uuid;
 380         int rc;
 381         ENTRY;
 382
 383         if (strcmp(watched->obd_type->typ_name, OBD_OSC_DEVICENAME)) {
 384                 CERROR("unexpected notification of %s %s!\n",
 385                        watched->obd_type->typ_name,
 386                        watched->obd_name);
 387                 return -EINVAL;
 388         }
 389
 390         uuid = &watched->u.cli.cl_import->imp_target_uuid;
 391
 392         /* Set OSC as active before notifying the observer, so the
 393          * observer can use the OSC normally.
 394          */
 395         rc = lov_set_osc_active(&obd->u.lov, uuid, active);
 396         if (rc) {
 397                 CERROR("%sactivation of %s failed: %d\n",
 398                        active ? "" : "de", uuid->uuid, rc);
 399                 RETURN(rc);
 400         }
 401
 402         if (obd->obd_observer)
 403                 /* Pass the notification up the chain. */
 404                 rc = obd_notify(obd->obd_observer, watched, active, data);
 405
 406         RETURN(rc);
 407 }
 408
 409 int lov_attach(struct obd_device *dev, obd_count len, void *data)
 410 {
 411         struct lprocfs_static_vars lvars;
 412         int rc;
 413
 414         lprocfs_init_vars(lov, &lvars);
 415         rc = lprocfs_obd_attach(dev, lvars.obd_vars);
 416         if (rc == 0) {
 417 #ifdef __KERNEL__
 418                 struct proc_dir_entry *entry;
 419
 420                 entry = create_proc_entry("target_obd_status", 0444,
 421                                           dev->obd_proc_entry);
 422                 if (entry == NULL) {
 423                         rc = -ENOMEM;
 424                 } else {
 425                         entry->proc_fops = &lov_proc_target_fops;
 426                         entry->data = dev;
 427                 }
 428 #endif
 429         }
 430         return rc;
 431 }
 432
 433 int lov_detach(struct obd_device *dev)
 434 {
 435         return lprocfs_obd_detach(dev);
 436 }
 437
 438 static int lov_setup(struct obd_device *obd, obd_count len, void *buf)
 439 {
 440         struct lov_obd *lov = &obd->u.lov;
 441         struct lustre_cfg *lcfg = buf;
 442         struct lov_desc *desc;
 443         int count;
 444         ENTRY;
 445
 446         if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
 447                 CERROR("LOV setup requires a descriptor\n");
 448                 RETURN(-EINVAL);
 449         }
 450
 451         desc = (struct lov_desc *)lustre_cfg_string(lcfg, 1);
 452         if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
 453                 CERROR("descriptor size wrong: %d > %d\n",
 454                        (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
 455                 RETURN(-EINVAL);
 456         }
 457
 458         /* Because of 64-bit divide/mod operations only work with a 32-bit
 459          * divisor in a 32-bit kernel, we cannot support a stripe width
 460          * of 4GB or larger on 32-bit CPUs.
 461          */
 462
 463         count = desc->ld_default_stripe_count;
 464         if (count && (count * desc->ld_default_stripe_size) > ~0UL) {
 465                 CERROR("LOV: stripe width "LPU64"x%u > %lu on 32-bit system\n",
 466                        desc->ld_default_stripe_size, count, ~0UL);
 467                 RETURN(-EINVAL);
 468         }
 469         if (desc->ld_tgt_count > 0) {
 470                 lov->bufsize= sizeof(struct lov_tgt_desc) * desc->ld_tgt_count;
 471         } else {
 472                 lov->bufsize = sizeof(struct lov_tgt_desc) * LOV_MAX_TGT_COUNT;
 473         }
 474         OBD_ALLOC(lov->tgts, lov->bufsize);
 475         if (lov->tgts == NULL) {
 476                 lov->bufsize = 0;
 477                 CERROR("couldn't allocate %d bytes for target table.\n",
 478                        lov->bufsize);
 479                 RETURN(-EINVAL);
 480         }
 481
 482         desc->ld_tgt_count = 0;
 483         desc->ld_active_tgt_count = 0;
 484         lov->desc = *desc;
 485         spin_lock_init(&lov->lov_lock);
 486         sema_init(&lov->lov_llog_sem, 1);
 487         init_waitqueue_head(&lov->lov_tgt_waitq);
 488
 489         RETURN(0);
 490 }
 491
 492 static int lov_cleanup(struct obd_device *obd, int flags)
 493 {
 494         struct lov_obd *lov = &obd->u.lov;
 495
 496         OBD_FREE(lov->tgts, lov->bufsize);
 497         RETURN(0);
 498 }
 499
 500 static int
 501 lov_add_obd(struct obd_device *obd, struct obd_uuid *uuidp, int index, int gen)
 502 {
 503         struct lov_obd *lov = &obd->u.lov;
 504         struct lov_tgt_desc *tgt;
 505         int rc;
 506         ENTRY;
 507
 508         CDEBUG(D_CONFIG, "uuid: %s idx: %d gen: %d\n",
 509                uuidp->uuid, index, gen);
 510
 511         if ((index < 0) || (index >= LOV_MAX_TGT_COUNT)) {
 512                 CERROR("request to add OBD %s at invalid index: %d\n",
 513                        uuidp->uuid, index);
 514                 RETURN(-EINVAL);
 515         }
 516
 517         if (gen <= 0) {
 518                 CERROR("request to add OBD %s with invalid generation: %d\n",
 519                        uuidp->uuid, gen);
 520                 RETURN(-EINVAL);
 521         }
 522
 523         tgt = lov->tgts + index;
 524
 525         lov_tgts_lock(lov);
 526
 527         if (!obd_uuid_empty(&tgt->uuid)) {
 528                 lov_tgts_unlock(lov);
 529                 CERROR("OBD already assigned at LOV target index %d\n",
 530                        index);
 531                 RETURN(-EEXIST);
 532         }
 533
 534         tgt->uuid = *uuidp;
 535         /* XXX - add a sanity check on the generation number. */
 536         tgt->ltd_gen = gen;
 537         tgt->ltd_flags = 0;
 538
 539         if (index >= lov->desc.ld_tgt_count)
 540                 lov->desc.ld_tgt_count = index + 1;
 541
 542         lov_tgts_unlock(lov);
 543
 544         CDEBUG(D_CONFIG, "idx: %d ltd_gen: %d ld_tgt_count: %d\n",
 545                 index, tgt->ltd_gen, lov->desc.ld_tgt_count);
 546
 547         if (lov->refcount == 0)
 548                 RETURN(0);
 549
 550         /* We don't need to lock the tgt entry here because the
 551          * config events are single threaded. Therefore only the
 552          * current thread can be changing the tgts tbl. */
 553         if (tgt->ltd_exp) {
 554                 struct obd_device *osc_obd;
 555
 556                 osc_obd = class_exp2obd(tgt->ltd_exp);
 557                 if (osc_obd)
 558                         osc_obd->obd_no_recov = 0;
 559         }
 560
 561         rc = lov_connect_obd(obd, tgt, 0, NULL, lov->lov_connect_flags);
 562         if (rc)
 563                 GOTO(out, rc);
 564
 565         if (obd->obd_observer) {
 566                 /* tell the mds_lov about the new target */
 567                 rc = obd_notify(obd->obd_observer, tgt->ltd_exp->exp_obd, 1,
 568                                 (void *)index);
 569         }
 570
 571         GOTO(out, rc);
 572  out:
 573         if (rc && tgt->ltd_exp != NULL)
 574                 lov_disconnect_obd(obd, tgt, 0);
 575         return rc;
 576 }
 577
 578 static int
 579 lov_del_obd(struct obd_device *obd, struct obd_uuid *uuidp, int index, int gen)
 580 {
 581         struct lov_obd *lov = &obd->u.lov;
 582         struct lov_tgt_desc *tgt;
 583         struct l_wait_info lwi = { 0 };
 584         int count = lov->desc.ld_tgt_count;
 585         int rc = 0;
 586         ENTRY;
 587
 588         CDEBUG(D_CONFIG, "uuid: %s idx: %d gen: %d\n",
 589                uuidp->uuid, index, gen);
 590
 591         if (index >= count) {
 592                 CERROR("LOV target index %d >= number of LOV OBDs %d.\n",
 593                        index, count);
 594                 RETURN(-EINVAL);
 595         }
 596
 597         tgt = lov->tgts + index;
 598
 599         if (obd_uuid_empty(&tgt->uuid)) {
 600                 CERROR("LOV target at index %d is not setup.\n", index);
 601                 RETURN(-EINVAL);
 602         }
 603
 604         if (!obd_uuid_equals(uuidp, &tgt->uuid)) {
 605                 CERROR("LOV target UUID %s at index %d doesn't match %s.\n",
 606                        tgt->uuid.uuid, index, uuidp->uuid);
 607                 RETURN(-EINVAL);
 608         }
 609
 610         lov_tgt_set_flags(lov, tgt, LTD_DEL_PENDING);
 611
 612         if (tgt->ltd_exp) {
 613                 struct obd_device *osc_obd;
 614
 615                 rc = obd_cancel_unused(tgt->ltd_exp, NULL,
 616                                        LDLM_FL_CONFIG_CHANGE, NULL);
 617                 if (rc != 0)
 618                         CWARN("obd_cancel_unused(osc): %d\n", rc);
 619
 620                 osc_obd = class_exp2obd(tgt->ltd_exp);
 621                 if (osc_obd) {
 622                         osc_obd->obd_no_recov = 1;
 623                         rc = obd_llog_finish(osc_obd, &osc_obd->obd_llogs, 1);
 624                         if (rc)
 625                                 CERROR("osc_llog_finish error: %d\n", rc);
 626                 }
 627
 628                 lov_disconnect_obd(obd, tgt, 0);
 629         }
 630
 631         CDEBUG(D_CONFIG, "Sleeping until LOV target index %d UUID %s "
 632                "is quiesced; ltd_refcount: %d ld_active_tgt_count: %d.\n",
 633                index, uuidp->uuid, tgt->ltd_refcount,
 634                lov->desc.ld_active_tgt_count);
 635
 636         /* XXX - should we set a timeout ? */
 637         lwi = LWI_TIMEOUT_INTR(0, NULL, NULL, NULL);
 638         rc = l_wait_event(lov->lov_tgt_waitq, (tgt->ltd_refcount == 0), &lwi);
 639         if (rc) {
 640                 lov_tgt_clear_flags(lov, tgt, LTD_DEL_PENDING);
 641                 CERROR("LOV target delete UUID %s index %d aborted: %d.\n",
 642                        uuidp->uuid, index, rc);
 643                 RETURN(rc);
 644         }
 645
 646         CDEBUG(D_CONFIG, "LOV target index %d UUID %s is quiesced; "
 647                "ltd_refcount: %d ld_active_tgt_count: %d.\n",
 648                index, uuidp->uuid, tgt->ltd_refcount,
 649                lov->desc.ld_active_tgt_count);
 650
 651         /* XXX - right now there is a dependency on ld_tgt_count being the
 652          * maximum tgt index for computing the mds_max_easize. So we can't
 653          * shrink it. */
 654
 655         /* lt_gen = 0 will mean it will not match the gen of any valid loi */
 656         memset(tgt, 0, sizeof(*tgt));
 657         RETURN(rc);
 658 }
 659
 660 static int lov_process_config(struct obd_device *obd, obd_count len, void *buf)
 661 {
 662         struct lustre_cfg *lcfg = buf;
 663         struct obd_uuid obd_uuid;
 664         int cmd;
 665         int index;
 666         int gen;
 667         int rc = 0;
 668         ENTRY;
 669
 670         switch(cmd = lcfg->lcfg_command) {
 671         case LCFG_LOV_ADD_OBD:
 672         case LCFG_LOV_DEL_OBD: {
 673                 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
 674                         GOTO(out, rc = -EINVAL);
 675
 676                 obd_str2uuid(&obd_uuid, lustre_cfg_string(lcfg, 1));
 677
 678                 if (sscanf(lustre_cfg_buf(lcfg, 2), "%d", &index) != 1)
 679                         GOTO(out, rc = -EINVAL);
 680                 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", &gen) != 1)
 681                         GOTO(out, rc = -EINVAL);
 682                 if (cmd == LCFG_LOV_ADD_OBD)
 683                         rc = lov_add_obd(obd, &obd_uuid, index, gen);
 684                 else
 685                         rc = lov_del_obd(obd, &obd_uuid, index, gen);
 686                 GOTO(out, rc);
 687         }
 688         default: {
 689                 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
 690                 GOTO(out, rc = -EINVAL);
 691
 692         }
 693         }
 694 out:
 695         RETURN(rc);
 696 }
 697
 698 #ifndef log2
 699 #define log2(n) ffz(~(n))
 700 #endif
 701
 702 static int lov_clear_orphans(struct obd_export *export,
 703                              struct obdo *src_oa,
 704                              struct lov_stripe_md **ea,
 705                              struct obd_trans_info *oti)
 706 {
 707         struct lov_obd *lov;
 708         struct obdo *tmp_oa;
 709         struct obd_uuid *ost_uuid = NULL;
 710         struct lov_tgt_desc *tgt;
 711         int rc = 0, i;
 712         ENTRY;
 713
 714         LASSERT(src_oa->o_valid & OBD_MD_FLFLAGS &&
 715                 src_oa->o_flags == OBD_FL_DELORPHAN);
 716
 717         lov = &export->exp_obd->u.lov;
 718
 719         tmp_oa = obdo_alloc();
 720         if (tmp_oa == NULL)
 721                 RETURN(-ENOMEM);
 722
 723         if (src_oa->o_valid & OBD_MD_FLINLINE) {
 724                 ost_uuid = (struct obd_uuid *)src_oa->o_inline;
 725                 CDEBUG(D_HA, "clearing orphans only for %s\n",
 726                        ost_uuid->uuid);
 727         }
 728
 729         for (i = 0, tgt = lov->tgts; i < lov->desc.ld_tgt_count; i++, tgt++) {
 730                 struct lov_stripe_md obj_md;
 731                 struct lov_stripe_md *obj_mdp = &obj_md;
 732                 int active = 0;
 733                 int err;
 734
 735                 /* if called for a specific target, we don't
 736                    care if it is not active. */
 737                 if (lov_tgt_active(lov, tgt, 0)) {
 738                         active = 1;
 739                 } else if (ost_uuid == NULL) {
 740                         CDEBUG(D_HA, "lov idx %d inactive\n", i);
 741                         continue;
 742                 }
 743
 744                 if (ost_uuid && !obd_uuid_equals(ost_uuid, &tgt->uuid)) {
 745                         if (active)
 746                                 lov_tgt_decref(lov, tgt);
 747                         continue;
 748                 }
 749
 750                 /*
 751                  * setting up objid OSS objects should be destroyed starting
 752                  * from it.
 753                  */
 754                 memcpy(tmp_oa, src_oa, sizeof(*tmp_oa));
 755                 tmp_oa->o_valid |= OBD_MD_FLID;
 756                 tmp_oa->o_id = oti->oti_objid[i];
 757
 758                 /* XXX: LOV STACKING: use real "obj_mdp" sub-data */
 759                 err = obd_create(tgt->ltd_exp, tmp_oa, NULL, 0, &obj_mdp, oti);
 760                 if (err)
 761                         /* This export will be disabled until it is recovered,
 762                            and then orphan recovery will be completed. */
 763                         CERROR("error in orphan recovery on OST idx %d/%d: "
 764                                "rc = %d\n", i, lov->desc.ld_tgt_count, err);
 765
 766                 if (active)
 767                         lov_tgt_decref(lov, tgt);
 768
 769                 if (ost_uuid)
 770                         break;
 771         }
 772
 773         obdo_free(tmp_oa);
 774         RETURN(rc);
 775 }
 776
 777 /* the LOV expects oa->o_id to be set to the LOV object id */
 778 static int
 779 lov_create(struct obd_export *exp, struct obdo *src_oa,
 780            void *acl, int acl_size, struct lov_stripe_md **ea,
 781            struct obd_trans_info *oti)
 782 {
 783         struct lov_request_set *set = NULL;
 784         struct list_head *pos;
 785         struct lov_obd *lov;
 786         int rc = 0;
 787         ENTRY;
 788
 789         LASSERT(ea != NULL);
 790         if (exp == NULL)
 791                 RETURN(-EINVAL);
 792
 793         if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
 794             src_oa->o_flags == OBD_FL_DELORPHAN) {
 795                 rc = lov_clear_orphans(exp, src_oa, ea, oti);
 796                 RETURN(rc);
 797         }
 798
 799         LASSERT(oti->oti_flags & OBD_MODE_CROW);
 800
 801         /* main creation loop */
 802         rc = lov_prep_create_set(exp, ea, src_oa, oti, &set);
 803         if (rc)
 804                 RETURN(rc);
 805
 806         lov = &exp->exp_obd->u.lov;
 807
 808         list_for_each (pos, &set->set_list) {
 809                 struct lov_request *req =
 810                         list_entry(pos, struct lov_request, rq_link);
 811
 812                 /* XXX: LOV STACKING: use real "obj_mdp" sub-data */
 813                 rc = obd_create(lov->tgts[req->rq_idx].ltd_exp,
 814                                 req->rq_oa, NULL, 0, &req->rq_md, oti);
 815                 lov_update_create_set(set, req, rc);
 816         }
 817         rc = lov_fini_create_set(set, ea);
 818         RETURN(rc);
 819 }
 820
 821 #define lsm_bad_magic(LSMP)                                     \
 822 ({                                                              \
 823         struct lov_stripe_md *_lsm__ = (LSMP);                  \
 824         int _ret__ = 0;                                         \
 825         if (!_lsm__) {                                          \
 826                 CERROR("LOV requires striping ea\n");           \
 827                 _ret__ = 1;                                     \
 828         } else if (_lsm__->lsm_magic != LOV_MAGIC) {            \
 829                 CERROR("LOV striping magic bad %#x != %#x\n",   \
 830                        _lsm__->lsm_magic, LOV_MAGIC);           \
 831                 _ret__ = 1;                                     \
 832         }                                                       \
 833         _ret__;                                                 \
 834 })
 835
 836 static int lov_destroy(struct obd_export *exp, struct obdo *oa,
 837                        struct lov_stripe_md *lsm, struct obd_trans_info *oti)
 838 {
 839         struct lov_request_set *set;
 840         struct lov_request *req;
 841         struct list_head *pos;
 842         struct lov_obd *lov;
 843         int rc = 0;
 844         ENTRY;
 845
 846         if (lsm_bad_magic(lsm))
 847                 RETURN(-EINVAL);
 848
 849         if (!exp || !exp->exp_obd)
 850                 RETURN(-ENODEV);
 851
 852         lov = &exp->exp_obd->u.lov;
 853         rc = lov_prep_destroy_set(exp, oa, lsm, oti, &set);
 854         if (rc)
 855                 RETURN(rc);
 856
 857         list_for_each (pos, &set->set_list) {
 858                 int err;
 859                 req = list_entry(pos, struct lov_request, rq_link);
 860
 861                 /* XXX update the cookie position */
 862                 oti->oti_logcookies = set->set_cookies + req->rq_stripe;
 863                 rc = obd_destroy(lov->tgts[req->rq_idx].ltd_exp, req->rq_oa,
 864                                  NULL, oti);
 865                 err = lov_update_common_set(set, req, rc);
 866                 if (rc) {
 867                         CERROR("error: destroying objid "LPX64" subobj "
 868                                LPX64" on OST idx %d: rc = %d\n",
 869                                set->set_oa->o_id, req->rq_oa->o_id,
 870                                req->rq_idx, rc);
 871                         if (!rc)
 872                                 rc = err;
 873                 }
 874         }
 875         lov_fini_destroy_set(set);
 876         RETURN(rc);
 877 }
 878
 879 static int lov_getattr(struct obd_export *exp, struct obdo *oa,
 880                        struct lov_stripe_md *lsm)
 881 {
 882         struct lov_request_set *set;
 883         struct lov_request *req;
 884         struct list_head *pos;
 885         struct lov_obd *lov;
 886         int err = 0, rc = 0;
 887         ENTRY;
 888
 889         if (lsm_bad_magic(lsm))
 890                 RETURN(-EINVAL);
 891
 892         if (!exp || !exp->exp_obd)
 893                 RETURN(-ENODEV);
 894
 895         lov = &exp->exp_obd->u.lov;
 896
 897         rc = lov_prep_getattr_set(exp, oa, lsm, &set);
 898         if (rc)
 899                 RETURN(rc);
 900
 901         list_for_each (pos, &set->set_list) {
 902                 req = list_entry(pos, struct lov_request, rq_link);
 903
 904                 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
 905                        "%u\n", oa->o_id, req->rq_stripe, req->rq_oa->o_id,
 906                        req->rq_idx);
 907
 908                 rc = obd_getattr(lov->tgts[req->rq_idx].ltd_exp,
 909                                  req->rq_oa, NULL);
 910                 err = lov_update_common_set(set, req, rc);
 911                 if (err) {
 912                         CERROR("error: getattr objid "LPX64" subobj "
 913                                LPX64" on OST idx %d: rc = %d\n",
 914                                set->set_oa->o_id, req->rq_oa->o_id,
 915                                req->rq_idx, err);
 916                         break;
 917                 }
 918         }
 919
 920         rc = lov_fini_getattr_set(set);
 921         if (err)
 922                 rc = err;
 923         RETURN(rc);
 924 }
 925
 926 static int lov_getattr_interpret(struct ptlrpc_request_set *rqset, void *data,
 927                                  int rc)
 928 {
 929         struct lov_request_set *lovset = (struct lov_request_set *)data;
 930         ENTRY;
 931
 932         /* don't do attribute merge if this aysnc op failed */
 933         if (rc) {
 934                 lovset->set_completes = 0;
 935                 lov_fini_getattr_set(lovset);
 936         } else {
 937                 rc = lov_fini_getattr_set(lovset);
 938         }
 939         RETURN (rc);
 940 }
 941
 942 static int lov_getattr_async(struct obd_export *exp, struct obdo *oa,
 943                               struct lov_stripe_md *lsm,
 944                               struct ptlrpc_request_set *rqset)
 945 {
 946         struct lov_request_set *lovset;
 947         struct lov_obd *lov;
 948         struct list_head *pos;
 949         struct lov_request *req;
 950         int rc = 0;
 951         ENTRY;
 952
 953         if (lsm_bad_magic(lsm))
 954                 RETURN(-EINVAL);
 955
 956         if (!exp || !exp->exp_obd)
 957                 RETURN(-ENODEV);
 958
 959         lov = &exp->exp_obd->u.lov;
 960
 961         rc = lov_prep_getattr_set(exp, oa, lsm, &lovset);
 962         if (rc)
 963                 RETURN(rc);
 964
 965         CDEBUG(D_INFO, "objid "LPX64": %ux%u byte stripes\n",
 966                lsm->lsm_object_id, lsm->lsm_stripe_count, lsm->lsm_stripe_size);
 967
 968         list_for_each (pos, &lovset->set_list) {
 969                 req = list_entry(pos, struct lov_request, rq_link);
 970
 971                 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
 972                        "%u\n", oa->o_id, req->rq_stripe, req->rq_oa->o_id,
 973                        req->rq_idx);
 974                 rc = obd_getattr_async(lov->tgts[req->rq_idx].ltd_exp,
 975                                        req->rq_oa, NULL, rqset);
 976                 if (rc) {
 977                         CERROR("error: getattr objid "LPX64" subobj "
 978                                LPX64" on OST idx %d: rc = %d\n",
 979                                lovset->set_oa->o_id, req->rq_oa->o_id,
 980                                req->rq_idx, rc);
 981                         GOTO(out, rc);
 982                 }
 983                 lov_update_common_set(lovset, req, rc);
 984         }
 985
 986         LASSERT(rc == 0);
 987         LASSERT (rqset->set_interpret == NULL);
 988         rqset->set_interpret = lov_getattr_interpret;
 989         rqset->set_arg = (void *)lovset;
 990         RETURN(rc);
 991 out:
 992         LASSERT(rc);
 993         lov_fini_getattr_set(lovset);
 994         RETURN(rc);
 995 }
 996
 997 static int lov_setattr(struct obd_export *exp, struct obdo *src_oa,
 998                        struct lov_stripe_md *lsm, struct obd_trans_info *oti)
 999 {
1000         struct lov_request_set *set;
1001         struct lov_obd *lov;
1002         struct list_head *pos;
1003         struct lov_request *req;
1004         int err = 0, rc = 0;
1005         ENTRY;
1006
1007         if (lsm_bad_magic(lsm))
1008                 RETURN(-EINVAL);
1009
1010         if (!exp || !exp->exp_obd)
1011                 RETURN(-ENODEV);
1012
1013         LASSERT(!(src_oa->o_valid & ~(OBD_MD_FLID|OBD_MD_FLTYPE | OBD_MD_FLMODE|
1014                                       OBD_MD_FLATIME | OBD_MD_FLMTIME |
1015                                       OBD_MD_FLCTIME | OBD_MD_FLFLAGS |
1016                                       OBD_MD_FLSIZE | OBD_MD_FLGROUP |
1017                                       OBD_MD_FLUID | OBD_MD_FLGID |
1018                                       OBD_MD_FLINLINE | OBD_MD_FLIFID)));
1019
1020         LASSERT(!(src_oa->o_valid & OBD_MD_FLGROUP) || src_oa->o_gr > 0);
1021
1022         lov = &exp->exp_obd->u.lov;
1023         rc = lov_prep_setattr_set(exp, src_oa, lsm, NULL, &set);
1024         if (rc)
1025                 RETURN(rc);
1026
1027         list_for_each (pos, &set->set_list) {
1028                 req = list_entry(pos, struct lov_request, rq_link);
1029
1030                 rc = obd_setattr(lov->tgts[req->rq_idx].ltd_exp, req->rq_oa,
1031                                  NULL, NULL);
1032                 err = lov_update_common_set(set, req, rc);
1033                 if (err) {
1034                         CERROR("error: setattr objid "LPX64" subobj "
1035                                LPX64" on OST idx %d: rc = %d\n",
1036                                set->set_oa->o_id, req->rq_oa->o_id,
1037                                req->rq_idx, err);
1038                         if (!rc)
1039                                 rc = err;
1040                 }
1041         }
1042         err = lov_fini_setattr_set(set);
1043         if (!rc)
1044                 rc = err;
1045         RETURN(rc);
1046 }
1047
1048 static int lov_revalidate_policy(struct lov_obd *lov, struct lov_stripe_md *lsm)
1049 {
1050         static int next_idx = 0;
1051         struct lov_tgt_desc *tgt;
1052         int i, count;
1053         ENTRY;
1054
1055         /* XXX - we should do something clever and take lsm
1056          * into account but just do round robin for now. */
1057
1058         /* last_idx must always be less that count because
1059          * ld_tgt_count currently cannot shrink. */
1060         count = lov->desc.ld_tgt_count;
1061
1062         for (i = next_idx, tgt = lov->tgts + i; i < count; i++, tgt++) {
1063                 if (lov_tgt_active(lov, tgt, 0)) {
1064                         next_idx = (i + 1) % count;
1065                         RETURN(i);
1066                 }
1067         }
1068
1069         for (i = 0, tgt = lov->tgts; i < next_idx; i++, tgt++) {
1070                 if (lov_tgt_active(lov, tgt, 0)) {
1071                         next_idx = (i + 1) % count;
1072                         RETURN(i);
1073                 }
1074         }
1075
1076         RETURN(-EIO);
1077 }
1078
1079 static int lov_revalidate_md(struct obd_export *exp, struct obdo *src_oa,
1080                              struct lov_stripe_md *ea,
1081                              struct obd_trans_info *oti)
1082 {
1083         struct obd_export *osc_exp;
1084         struct lov_obd *lov = &exp->exp_obd->u.lov;
1085         struct lov_stripe_md *lsm = ea;
1086         struct lov_stripe_md obj_md;
1087         struct lov_stripe_md *obj_mdp = &obj_md;
1088         struct lov_oinfo *loi;
1089         struct obdo *tmp_oa;
1090         int ost_idx, updates = 0, i;
1091         ENTRY;
1092
1093         tmp_oa = obdo_alloc();
1094         if (tmp_oa == NULL)
1095                 RETURN(-ENOMEM);
1096
1097         loi = lsm->lsm_oinfo;
1098         for (i = 0; i < lsm->lsm_stripe_count; i++, loi++) {
1099                 struct lov_tgt_desc *tgt;
1100                 int rc;
1101
1102                 if (!lov_tgt_changed(lov, loi))
1103                         continue;
1104
1105                 /* returns an active tgt. tgt->ltd_refcount is incremented. */
1106                 ost_idx = lov_revalidate_policy(lov, lsm);
1107                 if (ost_idx < 0) {
1108                         /* FIXME: punt for now. */
1109                         CERROR("lov_revalidate_policy failed; no active "
1110                                "OSCs?\n");
1111                         continue;
1112                 }
1113
1114                 /* create a new object */
1115                 memcpy(tmp_oa, src_oa, sizeof(*tmp_oa));
1116                 /* XXX: LOV STACKING: use real "obj_mdp" sub-data */
1117                 tgt = lov->tgts + ost_idx;
1118                 osc_exp = tgt->ltd_exp;
1119                 rc = obd_create(osc_exp, tmp_oa, NULL, 0, &obj_mdp, oti);
1120                 if (rc) {
1121                         lov_tgt_decref(lov, tgt);
1122                         CERROR("error creating new subobj at idx %d; "
1123                                "rc = %d\n", ost_idx, rc);
1124                         continue;
1125                 }
1126
1127                 CDEBUG(D_INODE,
1128                        "replacing idx %d gen %d objid "LPX64" subobj "LPX64" "
1129                        "with idx %d gen %d objid "LPX64" subobj "LPX64".\n",
1130                        loi->loi_ost_idx, loi->loi_ost_gen,
1131                        loi->loi_id, loi->loi_gr,
1132                        ost_idx, tgt->ltd_gen, tmp_oa->o_id, tmp_oa->o_gr);
1133
1134                 if (oti->oti_objid)
1135                         oti->oti_objid[ost_idx] = tmp_oa->o_id;
1136                 loi->loi_id = tmp_oa->o_id;
1137                 loi->loi_gr = tmp_oa->o_gr;
1138                 loi->loi_ost_idx = ost_idx;
1139                 loi->loi_ost_gen = tgt->ltd_gen;
1140                 lov_tgt_decref(lov, tgt);
1141                 updates = 1;
1142         }
1143
1144         /* If we got an error revalidating an entry there's no need to
1145          * cleanup up objects we allocated here because the bad entry
1146          * still points to a deleted OST. */
1147
1148         obdo_free(tmp_oa);
1149         RETURN(updates);
1150 }
1151
1152 /* FIXME: maybe we'll just make one node the authoritative attribute node, then
1153  * we can send this 'punch' to just the authoritative node and the nodes
1154  * that the punch will affect. */
1155 static int lov_punch(struct obd_export *exp, struct obdo *oa,
1156                      struct lov_stripe_md *lsm, obd_off start,
1157                      obd_off end, struct obd_trans_info *oti,
1158                      struct lustre_capa *capa)
1159 {
1160         struct lov_request_set *set;
1161         struct lov_obd *lov;
1162         struct list_head *pos;
1163         struct lov_request *req;
1164         int err = 0, rc = 0;
1165         ENTRY;
1166
1167         if (lsm_bad_magic(lsm))
1168                 RETURN(-EINVAL);
1169
1170         if (!exp || !exp->exp_obd)
1171                 RETURN(-ENODEV);
1172
1173         lov = &exp->exp_obd->u.lov;
1174         rc = lov_prep_punch_set(exp, oa, lsm, start, end, oti, &set);
1175         if (rc)
1176                 RETURN(rc);
1177
1178         list_for_each (pos, &set->set_list) {
1179                 req = list_entry(pos, struct lov_request, rq_link);
1180
1181                 rc = obd_punch(lov->tgts[req->rq_idx].ltd_exp, req->rq_oa,
1182                                NULL, req->rq_extent.start,
1183                                req->rq_extent.end, NULL, capa);
1184                 err = lov_update_punch_set(set, req, rc);
1185                 if (err) {
1186                         CERROR("error: punch objid "LPX64" subobj "LPX64
1187                                " on OST idx %d: rc = %d\n", set->set_oa->o_id,
1188                                req->rq_oa->o_id, req->rq_idx, rc);
1189                         if (!rc)
1190                                 rc = err;
1191                 }
1192         }
1193         err = lov_fini_punch_set(set);
1194         if (!rc)
1195                 rc = err;
1196         RETURN(rc);
1197 }
1198
1199 static int lov_sync(struct obd_export *exp, struct obdo *oa,
1200                     struct lov_stripe_md *lsm, obd_off start, obd_off end)
1201 {
1202         struct lov_request_set *set;
1203         struct lov_obd *lov;
1204         struct list_head *pos;
1205         struct lov_request *req;
1206         int err = 0, rc = 0;
1207         ENTRY;
1208
1209         if (lsm_bad_magic(lsm))
1210                 RETURN(-EINVAL);
1211
1212         if (!exp->exp_obd)
1213                 RETURN(-ENODEV);
1214
1215         lov = &exp->exp_obd->u.lov;
1216         rc = lov_prep_sync_set(exp, oa, lsm, start, end, &set);
1217         if (rc)
1218                 RETURN(rc);
1219
1220         list_for_each (pos, &set->set_list) {
1221                 req = list_entry(pos, struct lov_request, rq_link);
1222
1223                 rc = obd_sync(lov->tgts[req->rq_idx].ltd_exp, req->rq_oa,
1224                               NULL, req->rq_extent.start, req->rq_extent.end);
1225                 err = lov_update_common_set(set, req, rc);
1226                 if (err) {
1227                         CERROR("error: fsync objid "LPX64" subobj "LPX64
1228                                " on OST idx %d: rc = %d\n", set->set_oa->o_id,
1229                                req->rq_oa->o_id, req->rq_idx, rc);
1230                         if (!rc)
1231                                 rc = err;
1232                 }
1233         }
1234         err = lov_fini_sync_set(set);
1235         if (!rc)
1236                 rc = err;
1237         RETURN(rc);
1238 }
1239
1240 static int lov_brw_check(struct lov_obd *lov, struct obdo *oa,
1241                          struct lov_stripe_md *lsm,
1242                          obd_count oa_bufs, struct brw_page *pga)
1243 {
1244         int i, rc = 0;
1245         ENTRY;
1246
1247         /* The caller just wants to know if there's a chance that this
1248          * I/O can succeed */
1249         for (i = 0; i < oa_bufs; i++) {
1250                 struct lov_oinfo *loi;
1251                 struct lov_tgt_desc *tgt;
1252                 int stripe;
1253                 obd_off start, end;
1254
1255                 if (!lov_stripe_intersects(lsm, i, pga[i].disk_offset,
1256                                            pga[i].disk_offset + pga[i].count,
1257                                            &start, &end))
1258                         continue;
1259
1260                 stripe = lov_stripe_number(lsm, pga[i].disk_offset);
1261                 loi = lsm->lsm_oinfo + stripe;
1262                 tgt = lov->tgts + loi->loi_ost_idx;
1263
1264                 if (!lov_tgt_ready(lov, tgt, loi->loi_ost_gen)) {
1265                         CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
1266                         RETURN(-EIO);
1267                 }
1268                 rc = obd_brw(OBD_BRW_CHECK, tgt->ltd_exp, oa,
1269                              NULL, 1, &pga[i], NULL);
1270                 lov_tgt_decref(lov, tgt);
1271                 if (rc)
1272                         break;
1273         }
1274         RETURN(rc);
1275 }
1276
1277 static int lov_brw(int cmd, struct obd_export *exp, struct obdo *src_oa,
1278                    struct lov_stripe_md *lsm, obd_count oa_bufs,
1279                    struct brw_page *pga, struct obd_trans_info *oti)
1280 {
1281         struct lov_request_set *set;
1282         struct lov_request *req;
1283         struct list_head *pos;
1284         struct lov_obd *lov = &exp->exp_obd->u.lov;
1285         int err, rc = 0;
1286         ENTRY;
1287
1288         if (lsm_bad_magic(lsm))
1289                 RETURN(-EINVAL);
1290
1291         if (cmd == OBD_BRW_CHECK) {
1292                 rc = lov_brw_check(lov, src_oa, lsm, oa_bufs, pga);
1293                 RETURN(rc);
1294         }
1295
1296         rc = lov_prep_brw_set(exp, src_oa, lsm, oa_bufs, pga, oti, &set);
1297         if (rc)
1298                 RETURN(rc);
1299
1300         list_for_each (pos, &set->set_list) {
1301                 struct obd_export *sub_exp;
1302                 struct brw_page *sub_pga;
1303                 req = list_entry(pos, struct lov_request, rq_link);
1304
1305                 sub_exp = lov->tgts[req->rq_idx].ltd_exp;
1306                 sub_pga = set->set_pga + req->rq_pgaidx;
1307                 rc = obd_brw(cmd, sub_exp, req->rq_oa, req->rq_md,
1308                              req->rq_oabufs, sub_pga, oti);
1309                 if (rc)
1310                         break;
1311                 lov_update_common_set(set, req, rc);
1312         }
1313
1314         err = lov_fini_brw_set(set);
1315         if (!rc)
1316                 rc = err;
1317         RETURN(rc);
1318 }
1319
1320 static int lov_brw_interpret(struct ptlrpc_request_set *reqset, void *data,
1321                              int rc)
1322 {
1323         struct lov_request_set *lovset = (struct lov_request_set *)data;
1324         ENTRY;
1325
1326         if (rc) {
1327                 lovset->set_completes = 0;
1328                 lov_fini_brw_set(lovset);
1329         } else {
1330                 rc = lov_fini_brw_set(lovset);
1331         }
1332
1333         RETURN(rc);
1334 }
1335
1336 static int lov_brw_async(int cmd, struct obd_export *exp, struct obdo *oa,
1337                          struct lov_stripe_md *lsm, obd_count oa_bufs,
1338                          struct brw_page *pga, struct ptlrpc_request_set *set,
1339                          struct obd_trans_info *oti)
1340 {
1341         struct lov_request_set *lovset;
1342         struct lov_request *req;
1343         struct list_head *pos;
1344         struct lov_obd *lov = &exp->exp_obd->u.lov;
1345         int rc = 0;
1346         ENTRY;
1347
1348         if (lsm_bad_magic(lsm))
1349                 RETURN(-EINVAL);
1350
1351         if (cmd == OBD_BRW_CHECK) {
1352                 rc = lov_brw_check(lov, oa, lsm, oa_bufs, pga);
1353                 RETURN(rc);
1354         }
1355
1356         rc = lov_prep_brw_set(exp, oa, lsm, oa_bufs, pga, oti, &lovset);
1357         if (rc)
1358                 RETURN(rc);
1359
1360         list_for_each (pos, &lovset->set_list) {
1361                 struct obd_export *sub_exp;
1362                 struct brw_page *sub_pga;
1363                 req = list_entry(pos, struct lov_request, rq_link);
1364
1365                 sub_exp = lov->tgts[req->rq_idx].ltd_exp;
1366                 sub_pga = lovset->set_pga + req->rq_pgaidx;
1367                 rc = obd_brw_async(cmd, sub_exp, req->rq_oa, req->rq_md,
1368                                    req->rq_oabufs, sub_pga, set, oti);
1369                 if (rc)
1370                         GOTO(out, rc);
1371                 lov_update_common_set(lovset, req, rc);
1372         }
1373         LASSERT(rc == 0);
1374         LASSERT(set->set_interpret == NULL);
1375         set->set_interpret = (set_interpreter_func)lov_brw_interpret;
1376         set->set_arg = (void *)lovset;
1377
1378         RETURN(rc);
1379 out:
1380         lov_fini_brw_set(lovset);
1381         RETURN(rc);
1382 }
1383
1384 static int lov_ap_make_ready(void *data, int cmd)
1385 {
1386         struct lov_async_page *lap = LAP_FROM_COOKIE(data);
1387
1388         return lap->lap_caller_ops->ap_make_ready(lap->lap_caller_data, cmd);
1389 }
1390 static int lov_ap_refresh_count(void *data, int cmd)
1391 {
1392         struct lov_async_page *lap = LAP_FROM_COOKIE(data);
1393
1394         return lap->lap_caller_ops->ap_refresh_count(lap->lap_caller_data,
1395                                                      cmd);
1396 }
1397 static void lov_ap_fill_obdo(void *data, int cmd, struct obdo *oa)
1398 {
1399         struct lov_async_page *lap = LAP_FROM_COOKIE(data);
1400
1401         lap->lap_caller_ops->ap_fill_obdo(lap->lap_caller_data, cmd, oa);
1402         /* XXX woah, shouldn't we be altering more here?  size? */
1403         oa->o_id = lap->lap_loi_id;
1404 }
1405
1406 static void lov_ap_completion(void *data, int cmd, struct obdo *oa, int rc)
1407 {
1408         struct lov_async_page *lap = LAP_FROM_COOKIE(data);
1409
1410         /* in a raid1 regime this would down a count of many ios
1411          * in flight, onl calling the caller_ops completion when all
1412          * the raid1 ios are complete */
1413         lap->lap_caller_ops->ap_completion(lap->lap_caller_data, cmd, oa, rc);
1414 }
1415
1416 static struct obd_async_page_ops lov_async_page_ops = {
1417         .ap_make_ready =        lov_ap_make_ready,
1418         .ap_refresh_count =     lov_ap_refresh_count,
1419         .ap_fill_obdo =         lov_ap_fill_obdo,
1420         .ap_completion =        lov_ap_completion,
1421 };
1422
1423 static int lov_prep_async_page(struct obd_export *exp,
1424                                struct lov_stripe_md *lsm,
1425                                struct lov_oinfo *loi, struct page *page,
1426                                obd_off offset, struct obd_async_page_ops *ops,
1427                                void *data, void **res)
1428 {
1429         struct lov_obd *lov = &exp->exp_obd->u.lov;
1430         struct lov_tgt_desc *tgt;
1431         struct lov_async_page *lap;
1432         int rc, stripe;
1433         ENTRY;
1434
1435         if (lsm_bad_magic(lsm))
1436                 RETURN(-EINVAL);
1437         LASSERT(loi == NULL);
1438
1439         stripe = lov_stripe_number(lsm, offset);
1440         loi = &lsm->lsm_oinfo[stripe];
1441
1442         OBD_ALLOC(lap, sizeof(*lap));
1443         if (lap == NULL)
1444                 RETURN(-ENOMEM);
1445
1446         lap->lap_magic = LAP_MAGIC;
1447         lap->lap_caller_ops = ops;
1448         lap->lap_caller_data = data;
1449
1450         /* FIXME handle multiple oscs after landing b_raid1 */
1451         lap->lap_stripe = stripe;
1452         switch (lsm->lsm_pattern) {
1453                 case LOV_PATTERN_RAID0:
1454                         lov_stripe_offset(lsm, offset, lap->lap_stripe,
1455                                           &lap->lap_sub_offset);
1456                         break;
1457                 case LOV_PATTERN_CMOBD:
1458                         lap->lap_sub_offset = offset;
1459                         break;
1460                 default:
1461                         LBUG();
1462         }
1463
1464         /* so the callback doesn't need the lsm */
1465         lap->lap_loi_id = loi->loi_id;
1466
1467         tgt = lov->tgts + loi->loi_ost_idx;
1468
1469         if (lov_tgt_ready(lov, tgt, loi->loi_ost_gen)) {
1470                 rc = obd_prep_async_page(tgt->ltd_exp, lsm, loi, page,
1471                                          lap->lap_sub_offset,
1472                                          &lov_async_page_ops, lap,
1473                                          &lap->lap_sub_cookie);
1474         } else {
1475                 rc = -EIO;
1476         }
1477
1478         if (rc) {
1479                 OBD_FREE(lap, sizeof(*lap));
1480                 RETURN(rc);
1481         }
1482         CDEBUG(D_CACHE, "lap %p page %p cookie %p off "LPU64"\n", lap, page,
1483                lap->lap_sub_cookie, offset);
1484         *res = lap;
1485         RETURN(0);
1486 }
1487
1488 static int lov_queue_async_io(struct obd_export *exp,
1489                               struct lov_stripe_md *lsm,
1490                               struct lov_oinfo *loi, void *cookie,
1491                               int cmd, obd_off off, int count,
1492                               obd_flags brw_flags, obd_flags async_flags)
1493 {
1494         struct lov_obd *lov = &exp->exp_obd->u.lov;
1495         struct lov_tgt_desc *tgt;
1496         struct lov_async_page *lap;
1497         int rc;
1498
1499         LASSERT(loi == NULL);
1500
1501         if (lsm_bad_magic(lsm))
1502                 RETURN(-EINVAL);
1503
1504         lap = LAP_FROM_COOKIE(cookie);
1505         loi = &lsm->lsm_oinfo[lap->lap_stripe];
1506         tgt = lov->tgts + loi->loi_ost_idx;
1507
1508         if (!lov_tgt_ready(lov, tgt, loi->loi_ost_gen) &&
1509             !lov_tgt_pending(lov, tgt, loi->loi_ost_gen))
1510                  RETURN(-EIO);
1511
1512         rc = obd_queue_async_io(tgt->ltd_exp, lsm, loi, lap->lap_sub_cookie,
1513                                 cmd, off, count, brw_flags, async_flags);
1514
1515         lov_tgt_decref(lov, tgt);
1516         RETURN(rc);
1517 }
1518
1519 static int lov_set_async_flags(struct obd_export *exp,
1520                                struct lov_stripe_md *lsm,
1521                                struct lov_oinfo *loi, void *cookie,
1522                                obd_flags async_flags)
1523 {
1524         struct lov_obd *lov = &exp->exp_obd->u.lov;
1525         struct lov_tgt_desc *tgt;
1526         struct lov_async_page *lap;
1527         int rc;
1528
1529         LASSERT(loi == NULL);
1530
1531         if (lsm_bad_magic(lsm))
1532                 RETURN(-EINVAL);
1533
1534         lap = LAP_FROM_COOKIE(cookie);
1535         loi = &lsm->lsm_oinfo[lap->lap_stripe];
1536         tgt = lov->tgts + loi->loi_ost_idx;
1537
1538         if (!lov_tgt_ready(lov, tgt, loi->loi_ost_gen) &&
1539             !lov_tgt_pending(lov, tgt, loi->loi_ost_gen))
1540                  RETURN(-EIO);
1541
1542         rc = obd_set_async_flags(tgt->ltd_exp, lsm, loi, lap->lap_sub_cookie,
1543                                  async_flags);
1544
1545         lov_tgt_decref(lov, tgt);
1546         RETURN(rc);
1547 }
1548
1549 static int lov_queue_group_io(struct obd_export *exp,
1550                               struct lov_stripe_md *lsm,
1551                               struct lov_oinfo *loi,
1552                               struct obd_io_group *oig, void *cookie,
1553                               int cmd, obd_off off, int count,
1554                               obd_flags brw_flags, obd_flags async_flags)
1555 {
1556         struct lov_obd *lov = &exp->exp_obd->u.lov;
1557         struct lov_tgt_desc *tgt;
1558         struct lov_async_page *lap;
1559         int rc;
1560
1561         LASSERT(loi == NULL);
1562
1563         if (lsm_bad_magic(lsm))
1564                 RETURN(-EINVAL);
1565
1566         lap = LAP_FROM_COOKIE(cookie);
1567         loi = &lsm->lsm_oinfo[lap->lap_stripe];
1568         tgt = lov->tgts + loi->loi_ost_idx;
1569
1570         if (!lov_tgt_ready(lov, tgt, loi->loi_ost_gen) &&
1571             !lov_tgt_pending(lov, tgt, loi->loi_ost_gen))
1572                  RETURN(-EIO);
1573
1574         rc = obd_queue_group_io(tgt->ltd_exp, lsm, loi, oig,
1575                                 lap->lap_sub_cookie, cmd, off, count,
1576                                 brw_flags, async_flags);
1577
1578         lov_tgt_decref(lov, tgt);
1579         RETURN(rc);
1580 }
1581
1582 /* this isn't exactly optimal.  we may have queued sync io in oscs on
1583  * all stripes, but we don't record that fact at queue time.  so we
1584  * trigger sync io on all stripes. */
1585 static int lov_trigger_group_io(struct obd_export *exp,
1586                                 struct lov_stripe_md *lsm,
1587                                 struct lov_oinfo *loi,
1588                                 struct obd_io_group *oig)
1589 {
1590         struct lov_obd *lov = &exp->exp_obd->u.lov;
1591         int rc = 0, i, err;
1592
1593         LASSERT(loi == NULL);
1594
1595         if (lsm_bad_magic(lsm))
1596                 RETURN(-EINVAL);
1597
1598         loi = lsm->lsm_oinfo;
1599         for (i = 0; i < lsm->lsm_stripe_count; i++, loi++) {
1600                 struct lov_tgt_desc *tgt = lov->tgts + loi->loi_ost_idx;
1601
1602                 if (!lov_tgt_ready(lov, tgt, loi->loi_ost_gen)) {
1603                         CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
1604                         continue;
1605                 }
1606
1607                 err = obd_trigger_group_io(tgt->ltd_exp, lsm, loi, oig);
1608
1609                 lov_tgt_decref(lov, tgt);
1610
1611                 if (rc == 0 && err != 0)
1612                         rc = err;
1613         };
1614         RETURN(rc);
1615 }
1616
1617 static int lov_teardown_async_page(struct obd_export *exp,
1618                                    struct lov_stripe_md *lsm,
1619                                    struct lov_oinfo *loi, void *cookie)
1620 {
1621         struct lov_obd *lov = &exp->exp_obd->u.lov;
1622         struct lov_tgt_desc *tgt;
1623         struct lov_async_page *lap;
1624         int rc;
1625         ENTRY;
1626
1627         LASSERT(loi == NULL);
1628
1629         if (lsm_bad_magic(lsm))
1630                 RETURN(-EINVAL);
1631
1632         lap = LAP_FROM_COOKIE(cookie);
1633
1634         loi = &lsm->lsm_oinfo[lap->lap_stripe];
1635         tgt = lov->tgts + loi->loi_ost_idx;
1636
1637         rc = obd_teardown_async_page(tgt->ltd_exp, lsm, loi,
1638                                      lap->lap_sub_cookie);
1639         lov_tgt_decref(lov, tgt);
1640
1641         if (rc) {
1642                 CERROR("unable to teardown sub cookie %p: %d\n",
1643                        lap->lap_sub_cookie, rc);
1644                 RETURN(rc);
1645         }
1646         OBD_FREE(lap, sizeof(*lap));
1647         RETURN(rc);
1648 }
1649
1650 static int lov_enqueue(struct obd_export *exp, struct lov_stripe_md *lsm,
1651                        __u32 type, ldlm_policy_data_t *policy, __u32 mode,
1652                        int *flags, void *bl_cb, void *cp_cb, void *gl_cb,
1653                        void *data,__u32 lvb_len, void *lvb_swabber,
1654                        struct lustre_handle *lockh)
1655 {
1656         struct lov_request_set *set;
1657         struct lov_request *req;
1658         struct list_head *pos;
1659         struct lustre_handle *lov_lockhp;
1660         struct lov_obd *lov;
1661         ldlm_error_t rc;
1662         int save_flags = *flags;
1663         ENTRY;
1664
1665         if (lsm_bad_magic(lsm))
1666                 RETURN(-EINVAL);
1667
1668         /* we should never be asked to replay a lock this way. */
1669         LASSERT((*flags & LDLM_FL_REPLAY) == 0);
1670
1671         if (!exp || !exp->exp_obd)
1672                 RETURN(-ENODEV);
1673
1674         lov = &exp->exp_obd->u.lov;
1675         rc = lov_prep_enqueue_set(exp, lsm, policy, mode, lockh, &set);
1676         if (rc)
1677                 RETURN(rc);
1678
1679         list_for_each (pos, &set->set_list) {
1680                 ldlm_policy_data_t sub_policy;
1681                 req = list_entry(pos, struct lov_request, rq_link);
1682                 lov_lockhp = set->set_lockh->llh_handles + req->rq_stripe;
1683                 LASSERT(lov_lockhp);
1684
1685                 *flags = save_flags;
1686                 sub_policy.l_extent.start = req->rq_extent.start;
1687                 sub_policy.l_extent.end = req->rq_extent.end;
1688
1689                 rc = obd_enqueue(lov->tgts[req->rq_idx].ltd_exp, req->rq_md,
1690                                  type, &sub_policy, mode, flags, bl_cb,
1691                                  cp_cb, gl_cb, data, lvb_len, lvb_swabber,
1692                                  lov_lockhp);
1693                 rc = lov_update_enqueue_set(set, req, rc, save_flags);
1694                 if (rc != ELDLM_OK)
1695                         break;
1696         }
1697
1698         lov_fini_enqueue_set(set, mode);
1699         RETURN(rc);
1700 }
1701
1702 static int lov_match(struct obd_export *exp, struct lov_stripe_md *lsm,
1703                      __u32 type, ldlm_policy_data_t *policy, __u32 mode,
1704                      int *flags, void *data, struct lustre_handle *lockh)
1705 {
1706         struct lov_request_set *set;
1707         struct lov_request *req;
1708         struct list_head *pos;
1709         struct lov_obd *lov = &exp->exp_obd->u.lov;
1710         struct lustre_handle *lov_lockhp;
1711         int lov_flags, rc = 0;
1712         ENTRY;
1713
1714         if (lsm_bad_magic(lsm))
1715                 RETURN(-EINVAL);
1716
1717         if (!exp || !exp->exp_obd)
1718                 RETURN(-ENODEV);
1719
1720         lov = &exp->exp_obd->u.lov;
1721         rc = lov_prep_match_set(exp, lsm, policy, mode, lockh, &set);
1722         if (rc)
1723                 RETURN(rc);
1724
1725         list_for_each (pos, &set->set_list) {
1726                 ldlm_policy_data_t sub_policy;
1727                 req = list_entry(pos, struct lov_request, rq_link);
1728                 lov_lockhp = set->set_lockh->llh_handles + req->rq_stripe;
1729                 LASSERT(lov_lockhp);
1730
1731                 sub_policy.l_extent.start = req->rq_extent.start;
1732                 sub_policy.l_extent.end = req->rq_extent.end;
1733                 lov_flags = *flags;
1734
1735                 rc = obd_match(lov->tgts[req->rq_idx].ltd_exp, req->rq_md,
1736                                type, &sub_policy, mode, &lov_flags, data,
1737                                lov_lockhp);
1738                 rc = lov_update_match_set(set, req, rc);
1739                 if (rc != 1)
1740                         break;
1741         }
1742         lov_fini_match_set(set, mode, *flags);
1743         RETURN(rc);
1744 }
1745
1746 static int lov_change_cbdata(struct obd_export *exp,
1747                              struct lov_stripe_md *lsm, ldlm_iterator_t it,
1748                              void *data)
1749 {
1750         struct lov_obd *lov;
1751         struct lov_oinfo *loi;
1752         int rc = 0, i;
1753         ENTRY;
1754
1755         if (lsm_bad_magic(lsm))
1756                 RETURN(-EINVAL);
1757
1758         if (!exp || !exp->exp_obd)
1759                 RETURN(-ENODEV);
1760
1761         LASSERT(lsm->lsm_object_gr > 0);
1762
1763         lov = &exp->exp_obd->u.lov;
1764         for (i = 0,loi = lsm->lsm_oinfo; i < lsm->lsm_stripe_count; i++,loi++) {
1765                 struct lov_tgt_desc *tgt = lov->tgts + loi->loi_ost_idx;
1766                 struct lov_stripe_md submd;
1767
1768                 submd.lsm_object_id = loi->loi_id;
1769                 submd.lsm_object_gr = lsm->lsm_object_gr;
1770                 submd.lsm_stripe_count = 0;
1771                 rc = obd_change_cbdata(tgt->ltd_exp, &submd, it, data);
1772                 lov_tgt_decref(lov, tgt);
1773         }
1774         RETURN(rc);
1775 }
1776
1777 static int lov_cancel(struct obd_export *exp, struct lov_stripe_md *lsm,
1778                       __u32 mode, struct lustre_handle *lockh)
1779 {
1780         struct lov_request_set *set;
1781         struct lov_request *req;
1782         struct list_head *pos;
1783         struct lov_obd *lov = &exp->exp_obd->u.lov;
1784         struct lustre_handle *lov_lockhp;
1785         int err = 0, rc = 0;
1786         ENTRY;
1787
1788         if (lsm_bad_magic(lsm))
1789                 RETURN(-EINVAL);
1790
1791         if (!exp || !exp->exp_obd)
1792                 RETURN(-ENODEV);
1793
1794         LASSERT(lsm->lsm_object_gr > 0);
1795
1796         LASSERT(lockh);
1797         lov = &exp->exp_obd->u.lov;
1798         rc = lov_prep_cancel_set(exp, lsm, mode, lockh, &set);
1799         if (rc)
1800                 RETURN(rc);
1801
1802         list_for_each (pos, &set->set_list) {
1803                 req = list_entry(pos, struct lov_request, rq_link);
1804                 lov_lockhp = set->set_lockh->llh_handles + req->rq_stripe;
1805
1806                 rc = obd_cancel(lov->tgts[req->rq_idx].ltd_exp, req->rq_md,
1807                                 mode, lov_lockhp);
1808                 rc = lov_update_common_set(set, req, rc);
1809                 if (rc) {
1810                         CERROR("error: cancel objid "LPX64" subobj "
1811                                LPX64" on OST idx %d: rc = %d\n",
1812                                lsm->lsm_object_id,
1813                                req->rq_md->lsm_object_id, req->rq_idx, rc);
1814                         err = rc;
1815                 }
1816
1817         }
1818         lov_fini_cancel_set(set);
1819         RETURN(err);
1820 }
1821
1822 static int lov_cancel_unused(struct obd_export *exp,
1823                              struct lov_stripe_md *lsm,
1824                              int flags, void *opaque)
1825 {
1826         struct lov_obd *lov;
1827         struct lov_oinfo *loi;
1828         struct lov_tgt_desc *tgt;
1829         int rc = 0, i;
1830         ENTRY;
1831
1832         lov = &exp->exp_obd->u.lov;
1833         if (lsm == NULL) {
1834                 tgt = lov->tgts;
1835                 for (i = 0; i < lov->desc.ld_tgt_count; i++, tgt++) {
1836                         int err;
1837
1838                         if (lov_tgt_ready(lov, tgt, 0)) {
1839                                 err = obd_cancel_unused(tgt->ltd_exp, NULL,
1840                                                         flags, opaque);
1841                                 lov_tgt_decref(lov, tgt);
1842                         } else {
1843                                 if (obd_uuid_empty(&tgt->uuid))
1844                                         continue;
1845                                 CERROR("error: cancel unused "
1846                                        "OST idx %d: OST inactive.\n", i);
1847                                 err = -EIO;
1848                         }
1849                         if (!rc)
1850                                 rc = err;
1851                 }
1852                 RETURN(rc);
1853         }
1854
1855         if (lsm_bad_magic(lsm))
1856                 RETURN(-EINVAL);
1857
1858         if (!exp || !exp->exp_obd)
1859                 RETURN(-ENODEV);
1860
1861         LASSERT(lsm->lsm_object_gr > 0);
1862
1863         for (i = 0,loi = lsm->lsm_oinfo; i < lsm->lsm_stripe_count; i++,loi++) {
1864                 struct lov_stripe_md submd;
1865                 int err;
1866
1867                 submd.lsm_object_id = loi->loi_id;
1868                 submd.lsm_object_gr = lsm->lsm_object_gr;
1869                 submd.lsm_stripe_count = 0;
1870
1871                 tgt = lov->tgts + loi->loi_ost_idx;
1872                 if (lov_tgt_ready(lov, tgt, loi->loi_ost_gen)) {
1873                         err = obd_cancel_unused(tgt->ltd_exp, &submd, flags,
1874                                                 opaque);
1875                         lov_tgt_decref(lov, tgt);
1876                         if (err && lov_tgt_ready(lov, tgt, loi->loi_ost_gen)) {
1877                                 lov_tgt_decref(lov, tgt);
1878                                 CERROR("error: cancel unused objid "LPX64" "
1879                                        "subobj "LPX64" OST idx %d: rc = %d\n",
1880                                        lsm->lsm_object_id, loi->loi_id,
1881                                        loi->loi_ost_idx, err);
1882                         }
1883                 } else {
1884                         err = -EIO;
1885                         CERROR("error: cancel unused objid "LPX64" "
1886                                "subobj "LPX64" on OST idx %d: OST inactive.\n",
1887                                lsm->lsm_object_id,
1888                                loi->loi_id, loi->loi_ost_idx);
1889                 }
1890
1891                 if (!rc)
1892                         rc = err;
1893         }
1894         RETURN(rc);
1895 }
1896
1897 #define LOV_U64_MAX ((__u64)~0ULL)
1898 #define LOV_SUM_MAX(tot, add)                                           \
1899         do {                                                            \
1900                 if ((tot) + (add) < (tot))                              \
1901                         (tot) = LOV_U64_MAX;                            \
1902                 else                                                    \
1903                         (tot) += (add);                                 \
1904         } while(0)
1905
1906 static int lov_statfs(struct obd_device *obd, struct obd_statfs *osfs,
1907                       unsigned long max_age)
1908 {
1909         struct lov_obd *lov = &obd->u.lov;
1910         struct lov_tgt_desc *tgt = lov->tgts;
1911         struct obd_statfs lov_sfs;
1912         int set = 0;
1913         int rc = 0;
1914         int i;
1915         ENTRY;
1916
1917
1918         /* We only get block data from the OBD */
1919         for (i = 0; i < lov->desc.ld_tgt_count; i++, tgt++) {
1920                 int err;
1921
1922                 if (!lov_tgt_active(lov, tgt, 0)) {
1923                         CDEBUG(D_HA, "lov idx %d inactive\n", i);
1924                         continue;
1925                 }
1926
1927                 err = obd_statfs(class_exp2obd(tgt->ltd_exp), &lov_sfs,max_age);
1928                 lov_tgt_decref(lov, tgt);
1929                 if (err) {
1930                         if (lov_tgt_active(lov, tgt, 0)) {
1931                                 lov_tgt_decref(lov, tgt);
1932                                 if (!rc)
1933                                         rc = err;
1934                         }
1935                         continue;
1936                 }
1937
1938                 if (!set) {
1939                         memcpy(osfs, &lov_sfs, sizeof(lov_sfs));
1940                         set = 1;
1941                 } else {
1942                         osfs->os_bfree += lov_sfs.os_bfree;
1943                         osfs->os_bavail += lov_sfs.os_bavail;
1944                         osfs->os_blocks += lov_sfs.os_blocks;
1945                         /* XXX not sure about this one - depends on policy.
1946                          *   - could be minimum if we always stripe on all OBDs
1947                          *     (but that would be wrong for any other policy,
1948                          *     if one of the OBDs has no more objects left)
1949                          *   - could be sum if we stripe whole objects
1950                          *   - could be average, just to give a nice number
1951                          *
1952                          * To give a "reasonable" (if not wholly accurate)
1953                          * number, we divide the total number of free objects
1954                          * by expected stripe count (watch out for overflow).
1955                          */
1956                         LOV_SUM_MAX(osfs->os_files, lov_sfs.os_files);
1957                         LOV_SUM_MAX(osfs->os_ffree, lov_sfs.os_ffree);
1958                 }
1959         }
1960
1961         if (set) {
1962                 __u32 expected_stripes = lov->desc.ld_default_stripe_count ?
1963                                          lov->desc.ld_default_stripe_count :
1964                                          lov->desc.ld_active_tgt_count;
1965
1966                 if ((lov->desc.ld_default_stripe_count == 0) ||
1967                     (lov->desc.ld_default_stripe_count >
1968                      lov->desc.ld_active_tgt_count)) {
1969                         expected_stripes = lov->desc.ld_active_tgt_count;
1970                 } else {
1971                         expected_stripes = lov->desc.ld_default_stripe_count;
1972                 }
1973
1974                 if (osfs->os_files != LOV_U64_MAX)
1975                         do_div(osfs->os_files, expected_stripes);
1976                 if (osfs->os_ffree != LOV_U64_MAX)
1977                         do_div(osfs->os_ffree, expected_stripes);
1978         } else if (!rc)
1979                 rc = -EIO;
1980
1981         RETURN(rc);
1982 }
1983
1984 static int lov_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1985                          void *karg, void *uarg)
1986 {
1987         struct obd_device *obddev = class_exp2obd(exp);
1988         struct lov_obd *lov = &obddev->u.lov;
1989         struct lov_tgt_desc *tgt;
1990         int i, rc, count = lov->desc.ld_tgt_count;
1991         struct obd_uuid *uuidp;
1992         ENTRY;
1993
1994         switch (cmd) {
1995         case OBD_IOC_LOV_GET_CONFIG: {
1996                 struct obd_ioctl_data *data = karg;
1997                 struct lov_desc *desc;
1998                 char *buf = NULL;
1999                 __u32 *genp;
2000
2001                 buf = NULL;
2002                 len = 0;
2003                 if (obd_ioctl_getdata(&buf, &len, (void *)uarg))
2004                         RETURN(-EINVAL);
2005
2006                 data = (struct obd_ioctl_data *)buf;
2007
2008                 if (sizeof(*desc) > data->ioc_inllen1) {
2009                         obd_ioctl_freedata(buf, len);
2010                         RETURN(-EINVAL);
2011                 }
2012
2013                 if (sizeof(uuidp->uuid) * count > data->ioc_inllen2) {
2014                         obd_ioctl_freedata(buf, len);
2015                         RETURN(-EINVAL);
2016                 }
2017
2018                 if (sizeof(__u32) * count > data->ioc_inllen3) {
2019                         obd_ioctl_freedata(buf, len);
2020                         RETURN(-EINVAL);
2021                 }
2022
2023                 desc = (struct lov_desc *)data->ioc_inlbuf1;
2024                 memcpy(desc, &(lov->desc), sizeof(*desc));
2025
2026                 uuidp = (struct obd_uuid *)data->ioc_inlbuf2;
2027                 genp = (__u32 *)data->ioc_inlbuf3;
2028                 /* the uuid will be empty for deleted OSTs */
2029                 lov_tgts_lock(lov);
2030                 tgt = lov->tgts;
2031                 for (i = 0; i < count; i++, uuidp++, genp++, tgt++) {
2032                         obd_str2uuid(uuidp, (char *)tgt->uuid.uuid);
2033                         *genp = tgt->ltd_gen;
2034                 }
2035                 lov_tgts_unlock(lov);
2036
2037                 rc = copy_to_user((void *)uarg, buf, len);
2038                 if (rc)
2039                         rc = -EFAULT;
2040                 obd_ioctl_freedata(buf, len);
2041                 break;
2042         }
2043         case LL_IOC_LOV_SETSTRIPE:
2044                 rc = lov_setstripe(exp, karg, uarg);
2045                 break;
2046         case LL_IOC_LOV_GETSTRIPE:
2047                 rc = lov_getstripe(exp, karg, uarg);
2048                 break;
2049         case LL_IOC_LOV_SETEA:
2050                 rc = lov_setea(exp, karg, uarg);
2051                 break;
2052         default: {
2053                 int set = 0;
2054                 if (count == 0)
2055                         RETURN(-ENOTTY);
2056                 rc = 0;
2057                 for (i = 0, tgt = lov->tgts; i < count; i++, tgt++) {
2058                         int err;
2059
2060                         if (!lov_tgt_active(lov, tgt, 0)) {
2061                                 CERROR("error: iocontrol failed for OSC %s on "
2062                                        "OST idx %d: OST inactive.\n",
2063                                        tgt->uuid.uuid, i);
2064                                 if (!rc)
2065                                         rc = -EIO;
2066                                 continue;
2067                         }
2068
2069                         err = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
2070                         lov_tgt_decref(lov, tgt);
2071                         if (err) {
2072                                 if (lov_tgt_ready(lov, tgt, 0)) {
2073                                         lov_tgt_decref(lov, tgt);
2074                                         CERROR("error: iocontrol OSC %s on OST "
2075                                                "idx %d cmd %x: err = %d\n",
2076                                                lov->tgts[i].uuid.uuid, i,
2077                                                cmd, err);
2078                                         if (!rc)
2079                                                 rc = err;
2080                                 }
2081                         } else
2082                                 set = 1;
2083                 }
2084                 if (!set && !rc)
2085                         rc = -EIO;
2086         }
2087         }
2088
2089         RETURN(rc);
2090 }
2091
2092 static int lov_get_info(struct obd_export *exp, __u32 keylen,
2093                         void *key, __u32 *vallen, void *val)
2094 {
2095         struct obd_device *obddev = class_exp2obd(exp);
2096         struct lov_obd *lov = &obddev->u.lov;
2097         struct lov_tgt_desc *tgt;
2098         int i;
2099         ENTRY;
2100
2101         if (!vallen || !val)
2102                 RETURN(-EFAULT);
2103
2104         if (keylen > strlen("lock_to_stripe") &&
2105             strcmp(key, "lock_to_stripe") == 0) {
2106                 struct {
2107                         char name[16];
2108                         struct ldlm_lock *lock;
2109                         struct lov_stripe_md *lsm;
2110                 } *data = key;
2111                 struct lov_oinfo *loi;
2112                 struct ldlm_res_id *res_id = &data->lock->l_resource->lr_name;
2113                 __u32 *stripe = val;
2114
2115                 if (*vallen < sizeof(*stripe))
2116                         RETURN(-EFAULT);
2117                 *vallen = sizeof(*stripe);
2118
2119                 /* XXX This is another one of those bits that will need to
2120                  * change if we ever actually support nested LOVs.  It uses
2121                  * the lock's export to find out which stripe it is. */
2122                 /* XXX - it's assumed all the locks for deleted OSTs have
2123                  * been cancelled. Also, the export for deleted OSTs will
2124                  * be NULL and won't match the lock's export. */
2125                 lov_tgts_lock(lov);
2126                 for (i = 0, loi = data->lsm->lsm_oinfo;
2127                      i < data->lsm->lsm_stripe_count;
2128                      i++, loi++) {
2129                         if (lov->tgts[loi->loi_ost_idx].ltd_exp ==
2130                                         data->lock->l_conn_export &&
2131                             loi->loi_id == res_id->name[0] &&
2132                             loi->loi_gr == res_id->name[2]) {
2133                                 lov_tgts_unlock(lov);
2134                                 *stripe = i;
2135                                 RETURN(0);
2136                         }
2137                 }
2138                 lov_tgts_unlock(lov);
2139                 LDLM_ERROR(data->lock, "lock on inode without such object");
2140                 dump_lsm(D_ERROR, data->lsm);
2141                 portals_debug_dumpstack(NULL);
2142                 RETURN(-ENXIO);
2143         } else if (keylen >= strlen("size_to_stripe") &&
2144                    strcmp(key, "size_to_stripe") == 0) {
2145                 struct {
2146                         int stripe_number;
2147                         __u64 size;
2148                         struct lov_stripe_md *lsm;
2149                 } *data = val;
2150
2151                 if (*vallen < sizeof(*data))
2152                         RETURN(-EFAULT);
2153
2154                 data->size = lov_size_to_stripe(data->lsm, data->size,
2155                                                 data->stripe_number);
2156                 RETURN(0);
2157         } else if (keylen >= strlen("last_id") && strcmp(key, "last_id") == 0) {
2158                 __u32 size = sizeof(obd_id);
2159                 obd_id *ids = val;
2160                 int rc = 0;
2161
2162                 tgt = lov->tgts;
2163                 for (i = 0; i < lov->desc.ld_tgt_count; i++, tgt++) {
2164                         if (!lov_tgt_active(lov, tgt, 0))
2165                                 continue;
2166                         rc = obd_get_info(tgt->ltd_exp, keylen, key, &size,
2167                                           ids + i);
2168                         lov_tgt_decref(lov, tgt);
2169                         if (rc != 0)
2170                                 RETURN(rc);
2171                 }
2172                 RETURN(0);
2173         } else if (keylen >= strlen("lovdesc") && strcmp(key, "lovdesc") == 0) {
2174                 struct lov_desc *desc_ret = val;
2175                 *desc_ret = lov->desc;
2176
2177                 RETURN(0);
2178         }
2179
2180         RETURN(-EINVAL);
2181 }
2182
2183 static int lov_set_info(struct obd_export *exp, obd_count keylen,
2184                         void *key, obd_count vallen, void *val)
2185 {
2186         struct obd_device *obddev = class_exp2obd(exp);
2187         struct lov_obd *lov = &obddev->u.lov;
2188         struct lov_tgt_desc *tgt;
2189         int i, rc = 0, err;
2190         ENTRY;
2191
2192 #define KEY_IS(str) \
2193         (keylen == strlen(str) && memcmp(key, str, keylen) == 0)
2194
2195         if (KEY_IS("async")) {
2196                 struct lov_desc *desc = &lov->desc;
2197
2198                 if (vallen != sizeof(int))
2199                         RETURN(-EINVAL);
2200                 lov->async = *((int*) val);
2201
2202                 lov_tgts_lock(lov);
2203                 tgt = lov->tgts;
2204                 for (i = 0; i < desc->ld_tgt_count; i++, tgt++) {
2205                         struct obd_uuid *tgt_uuid = &tgt->uuid;
2206                         struct obd_device *tgt_obd;
2207
2208                         tgt_obd = class_find_client_obd(tgt_uuid,
2209                                                         OBD_OSC_DEVICENAME,
2210                                                         &obddev->obd_uuid);
2211                         if (!tgt_obd) {
2212                                 CERROR("Target %s not attached\n",
2213                                         tgt_uuid->uuid);
2214                                 if (!rc)
2215                                         rc = -EINVAL;
2216                                 continue;
2217                         }
2218
2219                         err = obd_set_info(tgt_obd->obd_self_export,
2220                                            keylen, key, vallen, val);
2221                         if (err) {
2222                                 CERROR("Failed to set async on target %s\n",
2223                                         tgt_obd->obd_name);
2224                                 if (!rc)
2225                                         rc = err;
2226                         }
2227                 }
2228                 lov_tgts_unlock(lov);
2229                 RETURN(rc);
2230         }
2231
2232         if (KEY_IS("mds_conn")) {
2233                 if (vallen != sizeof(__u32))
2234                         RETURN(-EINVAL);
2235         } else if (KEY_IS("unlinked") || KEY_IS("unrecovery")) {
2236                 if (vallen != 0)
2237                         RETURN(-EINVAL);
2238         } else if (KEY_IS("sec") || KEY_IS("sec_flags")) {
2239                 struct lov_tgt_desc *tgt;
2240                 struct obd_export *exp;
2241                 int rc = 0, err, i;
2242
2243                 lov_tgts_lock(lov);
2244                 tgt = lov->tgts;
2245                 for (i = 0; i < lov->desc.ld_tgt_count; i++, tgt++) {
2246                         exp = tgt->ltd_exp;
2247                         /* during setup time the connections to osc might
2248                          * haven't been established.
2249                          */
2250                         if (exp == NULL) {
2251                                 struct obd_device *tgt_obd;
2252
2253                                 tgt_obd = class_find_client_obd(&tgt->uuid,
2254                                                                 OBD_OSC_DEVICENAME,
2255                                                                 &obddev->obd_uuid);
2256                                 if (!tgt_obd) {
2257                                         CERROR("can't set security flavor, "
2258                                                "device %s not attached?\n",
2259                                                 tgt->uuid.uuid);
2260                                         rc = -EINVAL;
2261                                         continue;
2262                                 }
2263                                 exp = tgt_obd->obd_self_export;
2264                         }
2265
2266                         err = obd_set_info(exp, keylen, key, vallen, val);
2267                         if (!rc)
2268                                 rc = err;
2269                 }
2270                 lov_tgts_unlock(lov);
2271
2272                 RETURN(rc);
2273         } else if (KEY_IS("auditlog")) {
2274                 ptl_nid_t nid = 0;
2275                 struct audit_msg * msg = val;
2276                 int i = 0, len = sizeof(nid);
2277                 struct obd_export * ost_exp;
2278                 //try to find client nid
2279                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2280                         if (!lov->tgts[i].ltd_exp)
2281                                 continue;
2282                         rc = obd_get_info(lov->tgts[i].ltd_exp, 10,
2283                                           "client_nid", &len, &nid);
2284                         if (!rc)
2285                                 break;
2286                 }
2287
2288                 i = (__u32)nid % lov->desc.ld_tgt_count;
2289
2290                 CDEBUG(D_INFO, "Select OST %i for nid %.8x\n",i, (__u32)nid);
2291
2292                 ost_exp = lov->tgts[i].ltd_exp;
2293                 if (!ost_exp) {
2294                         CERROR("can't send auditlog data,"
2295                                "device %s is not attached?\n",
2296                                lov->tgts[i].uuid.uuid);
2297                         RETURN(-EFAULT);
2298                 }
2299                 msg->nid = nid;
2300                 rc = obd_set_info(ost_exp, keylen, key, vallen, val);
2301
2302                 RETURN(rc);
2303         } else if (KEY_IS("audit_obj")) {
2304                 struct audit_lov_msg * msg = val;
2305                 int i = 0, index;
2306                 struct obd_export * ost_exp;
2307                 struct lov_oinfo *loi;
2308                 struct obdo *oa = NULL;
2309
2310                 oa = obdo_alloc();
2311                 if (oa == NULL)
2312                         RETURN(-ENOMEM);
2313
2314                 //set audit for all objects of file
2315                 CDEBUG(D_INFO, "Set audit 0x%x for objects\n", (__u32)msg->mask);
2316                 //use o_fid to store audit mask
2317                 oa->o_fid = msg->mask;
2318                 oa->o_uid = msg->uid;
2319                 oa->o_gid = msg->gid;
2320                 //iterate over all objects and set audit for each
2321                 for (i = 0, loi = msg->lsm->lsm_oinfo;
2322                      i < msg->lsm->lsm_stripe_count; i++, loi++) {
2323                         index = loi->loi_ost_idx;
2324                         ost_exp = lov->tgts[index].ltd_exp;
2325                         if (!ost_exp) {
2326                                 CERROR("can't send audit_obj data,"
2327                                        "device %s is not attached?\n",
2328                                        lov->tgts[i].uuid.uuid);
2329                                 continue;
2330                         }
2331                         oa->o_id = loi->loi_id;
2332                         oa->o_gr = loi->loi_gr;
2333
2334                         obd_set_info(ost_exp, keylen, key, sizeof(*oa), oa);
2335                 }
2336
2337                 obdo_free(oa);
2338                 RETURN(0);
2339         } else if (KEY_IS("audit")) {
2340                 int i = 0;
2341                 struct obd_export * ost_exp;
2342
2343                 //set audit for whole fs on OSS
2344                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2345                         ost_exp = lov->tgts[i].ltd_exp;
2346                         if (!ost_exp) {
2347                                 CERROR("can't send audit_fs data,"
2348                                        "device %s is not attached?\n",
2349                                        lov->tgts[i].uuid.uuid);
2350                                 continue;
2351                         }
2352                         obd_set_info(ost_exp, keylen, key, vallen, val);
2353                 }
2354
2355                 RETURN(rc);
2356         } else if (KEY_IS("flush_cred") || KEY_IS("crypto_cb") ||
2357                    KEY_IS("capa_key")) {
2358                 struct lov_tgt_desc *tgt;
2359                 int rc = 0, i;
2360
2361                 tgt = lov->tgts;
2362                 for (i = 0; i < lov->desc.ld_tgt_count; i++, tgt++) {
2363                         if (!lov_tgt_active(lov, tgt, 0))
2364                                 continue;
2365                         rc = obd_set_info(tgt->ltd_exp,
2366                                           keylen, key, vallen, val);
2367                         lov_tgt_decref(lov, tgt);
2368                         if (rc)
2369                                 RETURN(rc);
2370                 }
2371
2372                 RETURN(0);
2373         } else if (KEY_IS("setext")) {
2374                 struct lov_tgt_desc *tgt;
2375                 int rc = 0, i;
2376
2377                 for (i = 0, tgt = lov->tgts; i < lov->desc.ld_tgt_count;
2378                      i++, tgt++) {
2379                         if (!tgt->ltd_exp)
2380                                 continue;
2381                         rc = obd_set_info(tgt->ltd_exp,
2382                                           keylen, key, vallen, val);
2383                         if (rc)
2384                                 RETURN(rc);
2385                 }
2386
2387                 RETURN(0);
2388         } else {
2389                 RETURN(-EINVAL);
2390         }
2391
2392         tgt = lov->tgts;
2393         for (i = 0; i < lov->desc.ld_tgt_count; i++, tgt++) {
2394                 int active = 0;
2395
2396                 if (val) {
2397                         if (!obd_uuid_equals(val, &tgt->uuid))
2398                                 continue;
2399                 } else if (!lov_tgt_active(lov, tgt, 0)) {
2400                         continue;
2401                 } else {
2402                         active = 1;
2403                 }
2404
2405                 err = obd_set_info(tgt->ltd_exp, keylen, key, vallen, val);
2406                 if (!rc)
2407                         rc = err;
2408                 if (active)
2409                         lov_tgt_decref(lov, tgt);
2410         }
2411         RETURN(rc);
2412 #undef KEY_IS
2413 }
2414
2415 #if 0
2416 struct lov_multi_wait {
2417         struct ldlm_lock *lock;
2418         wait_queue_t      wait;
2419         int               completed;
2420         int               generation;
2421 };
2422
2423 int lov_complete_many(struct obd_export *exp, struct lov_stripe_md *lsm,
2424                       struct lustre_handle *lockh)
2425 {
2426         struct lov_lock_handles *lov_lockh = NULL;
2427         struct lustre_handle *lov_lockhp;
2428         struct lov_obd *lov;
2429         struct lov_oinfo *loi;
2430         struct lov_multi_wait *queues;
2431         int rc = 0, i;
2432         ENTRY;
2433
2434         if (lsm_bad_magic(lsm))
2435                 RETURN(-EINVAL);
2436
2437         if (!exp || !exp->exp_obd)
2438                 RETURN(-ENODEV);
2439
2440         LASSERT(lockh != NULL);
2441         if (lsm->lsm_stripe_count > 1) {
2442                 lov_lockh = lov_handle2llh(lockh);
2443                 if (lov_lockh == NULL) {
2444                         CERROR("LOV: invalid lov lock handle %p\n", lockh);
2445                         RETURN(-EINVAL);
2446                 }
2447
2448                 lov_lockhp = lov_lockh->llh_handles;
2449         } else {
2450                 lov_lockhp = lockh;
2451         }
2452
2453         OBD_ALLOC(queues, lsm->lsm_stripe_count * sizeof(*queues));
2454         if (queues == NULL)
2455                 GOTO(out, rc = -ENOMEM);
2456
2457         lov = &exp->exp_obd->u.lov;
2458         for (i = 0, loi = lsm->lsm_oinfo; i < lsm->lsm_stripe_count;
2459              i++, loi++, lov_lockhp++) {
2460                 struct ldlm_lock *lock;
2461                 struct obd_device *obd;
2462                 unsigned long irqflags;
2463
2464                 lock = ldlm_handle2lock(lov_lockhp);
2465                 if (lock == NULL) {
2466                         CDEBUG(D_HA, "lov idx %d subobj "LPX64" no lock?\n",
2467                                loi->loi_ost_idx, loi->loi_id);
2468                         queues[i].completed = 1;
2469                         continue;
2470                 }
2471
2472                 queues[i].lock = lock;
2473                 init_waitqueue_entry(&(queues[i].wait), current);
2474                 add_wait_queue(lock->l_waitq, &(queues[i].wait));
2475
2476                 obd = class_exp2obd(lock->l_conn_export);
2477                 if (obd != NULL)
2478                         imp = obd->u.cli.cl_import;
2479                 if (imp != NULL) {
2480                         spin_lock_irqsave(&imp->imp_lock, irqflags);
2481                         queues[i].generation = imp->imp_generation;
2482                         spin_unlock_irqrestore(&imp->imp_lock, irqflags);
2483                 }
2484         }
2485
2486         lwi = LWI_TIMEOUT_INTR(obd_timeout * HZ, ldlm_expired_completion_wait,
2487                                interrupted_completion_wait, &lwd);
2488         rc = l_wait_event_added(check_multi_complete(queues, lsm), &lwi);
2489
2490         for (i = 0; i < lsm->lsm_stripe_count; i++)
2491                 remove_wait_queue(lock->l_waitq, &(queues[i].wait));
2492
2493         if (rc == -EINTR || rc == -ETIMEDOUT) {
2494
2495
2496         }
2497
2498  out:
2499         if (lov_lockh != NULL)
2500                 lov_llh_put(lov_lockh);
2501         RETURN(rc);
2502 }
2503 #endif
2504
2505 struct obd_ops lov_obd_ops = {
2506         .o_owner               = THIS_MODULE,
2507         .o_attach              = lov_attach,
2508         .o_detach              = lov_detach,
2509         .o_setup               = lov_setup,
2510         .o_cleanup             = lov_cleanup,
2511         .o_process_config      = lov_process_config,
2512         .o_connect             = lov_connect,
2513         .o_disconnect          = lov_disconnect,
2514         .o_statfs              = lov_statfs,
2515         .o_packmd              = lov_packmd,
2516         .o_unpackmd            = lov_unpackmd,
2517         .o_revalidate_md       = lov_revalidate_md,
2518         .o_create              = lov_create,
2519         .o_destroy             = lov_destroy,
2520         .o_getattr             = lov_getattr,
2521         .o_getattr_async       = lov_getattr_async,
2522         .o_setattr             = lov_setattr,
2523         .o_brw                 = lov_brw,
2524         .o_brw_async           = lov_brw_async,
2525         .o_prep_async_page     = lov_prep_async_page,
2526         .o_queue_async_io      = lov_queue_async_io,
2527         .o_set_async_flags     = lov_set_async_flags,
2528         .o_queue_group_io      = lov_queue_group_io,
2529         .o_trigger_group_io    = lov_trigger_group_io,
2530         .o_teardown_async_page = lov_teardown_async_page,
2531         .o_adjust_kms          = lov_adjust_kms,
2532         .o_punch               = lov_punch,
2533         .o_sync                = lov_sync,
2534         .o_enqueue             = lov_enqueue,
2535         .o_match               = lov_match,
2536         .o_change_cbdata       = lov_change_cbdata,
2537         .o_cancel              = lov_cancel,
2538         .o_cancel_unused       = lov_cancel_unused,
2539         .o_iocontrol           = lov_iocontrol,
2540         .o_get_info            = lov_get_info,
2541         .o_set_info            = lov_set_info,
2542         .o_llog_init           = lov_llog_init,
2543         .o_llog_finish         = lov_llog_finish,
2544         .o_notify              = lov_notify,
2545 };
2546
2547 int __init lov_init(void)
2548 {
2549         struct lprocfs_static_vars lvars;
2550         int rc;
2551         ENTRY;
2552
2553         lprocfs_init_vars(lov, &lvars);
2554         rc = class_register_type(&lov_obd_ops, NULL, lvars.module_vars,
2555                                  OBD_LOV_DEVICENAME);
2556         RETURN(rc);
2557 }
2558
2559 #ifdef __KERNEL__
2560 static void /*__exit*/ lov_exit(void)
2561 {
2562         class_unregister_type(OBD_LOV_DEVICENAME);
2563 }
2564
2565 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
2566 MODULE_DESCRIPTION("Lustre Logical Object Volume OBD driver");
2567 MODULE_LICENSE("GPL");
2568
2569 module_init(lov_init);
2570 module_exit(lov_exit);
2571 #endif