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) 2002-2004 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 the Lustre file system, http://www.lustre.org
  10  *   Lustre is a trademark of Cluster File Systems, Inc.
  11  *
  12  *   You may have signed or agreed to another license before downloading
  13  *   this software.  If so, you are bound by the terms and conditions
  14  *   of that agreement, and the following does not apply to you.  See the
  15  *   LICENSE file included with this distribution for more information.
  16  *
  17  *   If you did not agree to a different license, then this copy of Lustre
  18  *   is open source software; you can redistribute it and/or modify it
  19  *   under the terms of version 2 of the GNU General Public License as
  20  *   published by the Free Software Foundation.
  21  *
  22  *   In either case, Lustre is distributed in the hope that it will be
  23  *   useful, but WITHOUT ANY WARRANTY; without even the implied warranty
  24  *   of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  25  *   license text for more details.
  26  */
  27
  28 #ifndef EXPORT_SYMTAB
  29 # define EXPORT_SYMTAB
  30 #endif
  31 #define DEBUG_SUBSYSTEM S_LOV
  32 #ifdef __KERNEL__
  33 #include <libcfs/libcfs.h>
  34 #else
  35 #include <liblustre.h>
  36 #endif
  37
  38 #include <obd_support.h>
  39 #include <lustre_lib.h>
  40 #include <lustre_net.h>
  41 #include <lustre/lustre_idl.h>
  42 #include <lustre_dlm.h>
  43 #include <lustre_mds.h>
  44 #include <lustre_debug.h>
  45 #include <obd_class.h>
  46 #include <obd_lov.h>
  47 #include <obd_ost.h>
  48 #include <lprocfs_status.h>
  49 #include <lustre_param.h>
  50
  51 #include "lov_internal.h"
  52
  53
  54 /* FIXME add lov_get/putrefs around every access to lov->tgts for on-line non-
  55    quiescent ost removal */
  56 /* Keep a refcount of lov->tgt usage to prevent racing with deletion */
  57 static void lov_getref(struct obd_device *obd)
  58 {
  59         struct lov_obd *lov = &obd->u.lov;
  60
  61         /* nobody gets through here until lov_putref is done */
  62         mutex_down(&lov->lov_lock);
  63         atomic_inc(&lov->refcount);
  64         mutex_up(&lov->lov_lock);
  65         return;
  66 }
  67
  68 static void __lov_del_obd(struct obd_device *obd, struct lov_tgt_desc *tgt);
  69
  70 static void lov_putref(struct obd_device *obd)
  71 {
  72         struct lov_obd *lov = &obd->u.lov;
  73         mutex_down(&lov->lov_lock);
  74         /* ok to dec to 0 more than once -- ltd_exp's will be null */
  75         if (atomic_dec_and_test(&lov->refcount) && lov->death_row) {
  76                 struct lov_tgt_desc *tgt;
  77                 int i;
  78                 CDEBUG(D_CONFIG, "destroying %d lov targets\n", lov->death_row);
  79                 for (i = 0, tgt = lov->tgts; i < lov->desc.ld_tgt_count;
  80                       i++, tgt++) {
  81                         if (!tgt->reap)
  82                                 continue;
  83                         /* Disconnect and delete from list */
  84                         __lov_del_obd(obd, tgt);
  85                         lov->death_row--;
  86                 }
  87         }
  88         mutex_up(&lov->lov_lock);
  89 }
  90
  91 #define MAX_STRING_SIZE 128
  92 static int lov_connect_obd(struct obd_device *obd, struct lov_tgt_desc *tgt,
  93                            int activate, struct obd_connect_data *data)
  94 {
  95         struct lov_obd *lov = &obd->u.lov;
  96         struct obd_uuid *tgt_uuid = &tgt->uuid;
  97         struct obd_device *tgt_obd;
  98         struct obd_uuid lov_osc_uuid = { "LOV_OSC_UUID" };
  99         struct lustre_handle conn = {0, };
 100         struct obd_import *imp;
 101 #ifdef __KERNEL__
 102         cfs_proc_dir_entry_t *lov_proc_dir;
 103 #endif
 104         int rc;
 105         ENTRY;
 106
 107         tgt_obd = class_find_client_obd(tgt_uuid, LUSTRE_OSC_NAME,
 108                                         &obd->obd_uuid);
 109
 110         if (!tgt_obd) {
 111                 CERROR("Target %s not attached\n", obd_uuid2str(tgt_uuid));
 112                 RETURN(-EINVAL);
 113         }
 114
 115         CDEBUG(D_CONFIG, "Connect tgt %s (%s)\n", obd_uuid2str(tgt_uuid),
 116                tgt_obd->obd_name);
 117
 118         if (!tgt_obd->obd_set_up) {
 119                 CERROR("Target %s not set up\n", obd_uuid2str(tgt_uuid));
 120                 RETURN(-EINVAL);
 121         }
 122
 123         if (activate) {
 124                 tgt_obd->obd_no_recov = 0;
 125                 /* FIXME this is probably supposed to be
 126                    ptlrpc_set_import_active.  Horrible naming. */
 127                 ptlrpc_activate_import(tgt_obd->u.cli.cl_import);
 128         }
 129
 130         /*
 131          * Divine LOV knows that OBDs under it are OSCs.
 132          */
 133         imp = tgt_obd->u.cli.cl_import;
 134
 135         if (imp->imp_invalid) {
 136                 CERROR("not connecting OSC %s; administratively "
 137                        "disabled\n", obd_uuid2str(tgt_uuid));
 138                 rc = obd_register_observer(tgt_obd, obd);
 139                 if (rc) {
 140                         CERROR("Target %s register_observer error %d; "
 141                                "will not be able to reactivate\n",
 142                                obd_uuid2str(tgt_uuid), rc);
 143                 }
 144                 RETURN(0);
 145         }
 146
 147         rc = obd_connect(NULL, &conn, tgt_obd, &lov_osc_uuid, data);
 148         if (rc) {
 149                 CERROR("Target %s connect error %d\n",
 150                        obd_uuid2str(tgt_uuid), rc);
 151                 RETURN(rc);
 152         }
 153         tgt->ltd_exp = class_conn2export(&conn);
 154
 155         rc = obd_register_observer(tgt_obd, obd);
 156         if (rc) {
 157                 CERROR("Target %s register_observer error %d\n",
 158                        obd_uuid2str(tgt_uuid), rc);
 159                 obd_disconnect(tgt->ltd_exp);
 160                 tgt->ltd_exp = NULL;
 161                 RETURN(rc);
 162         }
 163
 164         tgt->active = 1;
 165         tgt->reap = 0;
 166         lov->desc.ld_active_tgt_count++;
 167
 168 #ifdef __KERNEL__
 169         lov_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
 170         if (lov_proc_dir) {
 171                 struct obd_device *osc_obd = class_conn2obd(&conn);
 172                 cfs_proc_dir_entry_t *osc_symlink;
 173                 char name[MAX_STRING_SIZE];
 174
 175                 LASSERT(osc_obd != NULL);
 176                 LASSERT(osc_obd->obd_type != NULL);
 177                 LASSERT(osc_obd->obd_type->typ_name != NULL);
 178                 snprintf(name, MAX_STRING_SIZE, "../../../%s/%s",
 179                          osc_obd->obd_type->typ_name,
 180                          osc_obd->obd_name);
 181                 osc_symlink = proc_symlink(osc_obd->obd_name, lov_proc_dir,
 182                                            name);
 183                 if (osc_symlink == NULL) {
 184                         CERROR("could not register LOV target "
 185                                "/proc/fs/lustre/%s/%s/target_obds/%s.",
 186                                obd->obd_type->typ_name, obd->obd_name,
 187                                osc_obd->obd_name);
 188                         lprocfs_remove(lov_proc_dir);
 189                         lov_proc_dir = NULL;
 190                 }
 191         }
 192 #endif
 193
 194         RETURN(0);
 195 }
 196
 197 static int lov_connect(const struct lu_context *ctx,
 198                        struct lustre_handle *conn, struct obd_device *obd,
 199                        struct obd_uuid *cluuid, struct obd_connect_data *data)
 200 {
 201         struct lov_obd *lov = &obd->u.lov;
 202         int rc;
 203         ENTRY;
 204
 205         lov->ocd.ocd_connect_flags = OBD_CONNECT_EMPTY;
 206         if (data)
 207                 lov->ocd = *data;
 208
 209         rc = class_connect(conn, obd, cluuid);
 210         if (!rc)
 211                 lov->connects++;
 212         CDEBUG(D_CONFIG, "connect #%d\n", lov->connects);
 213
 214         /* target connects are done in lov_add_target */
 215
 216         RETURN (rc);
 217 }
 218
 219 static int lov_disconnect_obd(struct obd_device *obd, struct lov_tgt_desc *tgt)
 220 {
 221         cfs_proc_dir_entry_t *lov_proc_dir;
 222         struct obd_device *osc_obd = class_exp2obd(tgt->ltd_exp);
 223         struct lov_obd *lov = &obd->u.lov;
 224         int rc;
 225         ENTRY;
 226
 227         CDEBUG(D_CONFIG, "%s: disconnecting target %s\n",
 228                obd->obd_name, osc_obd->obd_name);
 229
 230         lov_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
 231         if (lov_proc_dir) {
 232                 cfs_proc_dir_entry_t *osc_symlink;
 233
 234                 osc_symlink = lprocfs_srch(lov_proc_dir, osc_obd->obd_name);
 235                 if (osc_symlink) {
 236                         lprocfs_remove(osc_symlink);
 237                 } else {
 238                         CERROR("/proc/fs/lustre/%s/%s/target_obds/%s missing.",
 239                                obd->obd_type->typ_name, obd->obd_name,
 240                                osc_obd->obd_name);
 241                 }
 242         }
 243
 244         if (obd->obd_no_recov) {
 245                 /* Pass it on to our clients.
 246                  * XXX This should be an argument to disconnect,
 247                  * XXX not a back-door flag on the OBD.  Ah well.
 248                  */
 249                 if (osc_obd)
 250                         osc_obd->obd_no_recov = 1;
 251         }
 252
 253         obd_register_observer(osc_obd, NULL);
 254
 255         rc = obd_disconnect(tgt->ltd_exp);
 256         if (rc) {
 257                 if (tgt->active) {
 258                         CERROR("Target %s disconnect error %d\n",
 259                                tgt->uuid.uuid, rc);
 260                 }
 261                 rc = 0;
 262         }
 263
 264         if (tgt->active) {
 265                 tgt->active = 0;
 266                 lov->desc.ld_active_tgt_count--;
 267         }
 268
 269         tgt->ltd_exp = NULL;
 270         RETURN(0);
 271 }
 272
 273 static int lov_del_target(struct obd_device *obd, struct obd_uuid *uuidp,
 274                           int index, int gen);
 275
 276 static int lov_disconnect(struct obd_export *exp)
 277 {
 278         struct obd_device *obd = class_exp2obd(exp);
 279         struct lov_obd *lov = &obd->u.lov;
 280         struct lov_tgt_desc *tgt;
 281         int i, rc;
 282         ENTRY;
 283
 284         if (!lov->tgts)
 285                 goto out;
 286
 287         /* Only disconnect the underlying layers on the final disconnect. */
 288         lov->connects--;
 289         if (lov->connects != 0) {
 290                 /* why should there be more than 1 connect? */
 291                 CERROR("disconnect #%d\n", lov->connects);
 292                 goto out;
 293         }
 294
 295         /* Let's hold another reference so lov_del_obd doesn't spin through
 296            putref every time */
 297         lov_getref(obd);
 298         for (i = 0, tgt = lov->tgts; i < lov->desc.ld_tgt_count; i++, tgt++) {
 299                 if (tgt->ltd_exp) {
 300                         /* Disconnection is the last we know about an obd */
 301                         lov_del_target(obd, &tgt->uuid, i, tgt->ltd_gen);
 302                 }
 303         }
 304         lov_putref(obd);
 305
 306 out:
 307         rc = class_disconnect(exp); /* bz 9811 */
 308         RETURN(rc);
 309 }
 310
 311 /* Error codes:
 312  *
 313  *  -EINVAL  : UUID can't be found in the LOV's target list
 314  *  -ENOTCONN: The UUID is found, but the target connection is bad (!)
 315  *  -EBADF   : The UUID is found, but the OBD is the wrong type (!)
 316  */
 317 static int lov_set_osc_active(struct lov_obd *lov, struct obd_uuid *uuid,
 318                               int activate)
 319 {
 320         struct lov_tgt_desc *tgt;
 321         int i, rc = 0;
 322         ENTRY;
 323
 324         CDEBUG(D_INFO, "Searching in lov %p for uuid %s (activate=%d)\n",
 325                lov, uuid->uuid, activate);
 326
 327         for (i = 0, tgt = lov->tgts; i < lov->desc.ld_tgt_count; i++, tgt++) {
 328                 if (tgt->ltd_exp == NULL)
 329                         continue;
 330
 331                 CDEBUG(D_INFO, "lov idx %d is %s conn "LPX64"\n",
 332                        i, tgt->uuid.uuid, tgt->ltd_exp->exp_handle.h_cookie);
 333                 if (strncmp(uuid->uuid, tgt->uuid.uuid, sizeof uuid->uuid) == 0)
 334                         break;
 335         }
 336
 337         if (i == lov->desc.ld_tgt_count)
 338                 GOTO(out, rc = -EINVAL);
 339
 340         if (tgt->active == activate) {
 341                 CDEBUG(D_INFO, "OSC %s already %sactive!\n", uuid->uuid,
 342                        activate ? "" : "in");
 343                 GOTO(out, rc);
 344         }
 345
 346         CDEBUG(D_INFO, "Marking OSC %s %sactive\n", uuid->uuid,
 347                activate ? "" : "in");
 348
 349         tgt->active = activate;
 350         if (activate)
 351                 lov->desc.ld_active_tgt_count++;
 352         else
 353                 lov->desc.ld_active_tgt_count--;
 354
 355         EXIT;
 356  out:
 357         return rc;
 358 }
 359
 360 static int lov_notify(struct obd_device *obd, struct obd_device *watched,
 361                       enum obd_notify_event ev, void *data)
 362 {
 363         int rc = 0;
 364         ENTRY;
 365
 366         if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE) {
 367                 struct obd_uuid *uuid;
 368
 369                 LASSERT(watched);
 370
 371                 if (strcmp(watched->obd_type->typ_name, LUSTRE_OSC_NAME)) {
 372                         CERROR("unexpected notification of %s %s!\n",
 373                                watched->obd_type->typ_name,
 374                                watched->obd_name);
 375                         RETURN(-EINVAL);
 376                 }
 377                 uuid = &watched->u.cli.cl_target_uuid;
 378
 379                 /* Set OSC as active before notifying the observer, so the
 380                  * observer can use the OSC normally.
 381                  */
 382                 lov_getref(obd);
 383                 rc = lov_set_osc_active(&obd->u.lov, uuid,
 384                                         ev == OBD_NOTIFY_ACTIVE);
 385                 lov_putref(obd);
 386                 if (rc) {
 387                         CERROR("%sactivation of %s failed: %d\n",
 388                                (ev == OBD_NOTIFY_ACTIVE) ? "" : "de",
 389                                obd_uuid2str(uuid), rc);
 390                         RETURN(rc);
 391                 }
 392         }
 393
 394         /* Pass the notification up the chain. */
 395         if (watched) {
 396                 rc = obd_notify_observer(obd, watched, ev, data);
 397         } else {
 398                 /* NULL watched means all osc's in the lov (only for syncs) */
 399                 struct lov_obd *lov = &obd->u.lov;
 400                 struct lov_tgt_desc *tgt;
 401                 struct obd_device *tgt_obd;
 402                 int i;
 403                 for (i = 0, tgt = lov->tgts; i < lov->desc.ld_tgt_count;
 404                       i++, tgt++) {
 405                         if (obd_uuid_empty(&tgt->uuid))
 406                                 continue;
 407                         tgt_obd = class_exp2obd(tgt->ltd_exp);
 408                         rc = obd_notify_observer(obd, tgt_obd, ev, data);
 409                         if (rc) {
 410                                 CERROR("%s: notify %s of %s failed %d\n",
 411                                        obd->obd_name,
 412                                        obd->obd_observer->obd_name,
 413                                        tgt_obd->obd_name, rc);
 414                                 break;
 415                         }
 416                 }
 417         }
 418
 419         RETURN(rc);
 420 }
 421
 422 static int lov_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
 423                           int index, int gen, int active)
 424 {
 425         struct lov_obd *lov = &obd->u.lov;
 426         struct lov_tgt_desc *tgt;
 427         struct obd_connect_data *ocd = NULL;
 428         __u32 bufsize, idx;
 429         int rc;
 430         ENTRY;
 431
 432         CDEBUG(D_CONFIG, "uuid:%s idx:%d gen:%d active:%d\n",
 433                uuidp->uuid, index, gen, active);
 434
 435         if (index < 0) {
 436                 CERROR("request to add OBD %s at invalid index: %d\n",
 437                        uuidp->uuid, index);
 438                 RETURN(-EINVAL);
 439         }
 440
 441         if (gen <= 0) {
 442                 CERROR("request to add OBD %s with invalid generation: %d\n",
 443                        uuidp->uuid, gen);
 444                 RETURN(-EINVAL);
 445         }
 446
 447         bufsize = sizeof(struct lov_tgt_desc) * (index + 1);
 448         if (bufsize > lov->bufsize) {
 449                 OBD_ALLOC(tgt, bufsize);
 450                 if (tgt == NULL) {
 451                         CERROR("couldn't allocate %d bytes for new table.\n",
 452                                bufsize);
 453                         RETURN(-ENOMEM);
 454                 }
 455
 456                 if (lov->tgts) {
 457                         int i;
 458                         memcpy(tgt, lov->tgts, lov->bufsize);
 459                         LASSERT(index == lov->desc.ld_tgt_count);
 460                         for (i = 0; i < index; i++) {
 461                                 INIT_LIST_HEAD(&tgt[i].qos_bavail_list);
 462                                 list_splice(&lov->tgts[i].qos_bavail_list,
 463                                             &tgt[i].qos_bavail_list);
 464                         }
 465                         OBD_FREE(lov->tgts, lov->bufsize);
 466                 }
 467
 468                 lov->tgts = tgt;
 469                 lov->bufsize = bufsize;
 470                 CDEBUG(D_CONFIG, "tgts: %p bufsize: %d\n",
 471                        lov->tgts, lov->bufsize);
 472         }
 473
 474         tgt = &lov->tgts[index];
 475         if (!obd_uuid_empty(&tgt->uuid)) {
 476                 CERROR("UUID %s already assigned at LOV target index %d\n",
 477                        obd_uuid2str(&tgt->uuid), index);
 478                 RETURN(-EEXIST);
 479         }
 480
 481         tgt->uuid = *uuidp;
 482         /* XXX - add a sanity check on the generation number. */
 483         tgt->ltd_gen = gen;
 484         tgt->index = index;
 485         INIT_LIST_HEAD(&tgt->qos_bavail_list);
 486
 487         if (index >= lov->desc.ld_tgt_count)
 488                 lov->desc.ld_tgt_count = index + 1;
 489
 490         CDEBUG(D_CONFIG, "idx=%d ltd_gen=%d ld_tgt_count=%d\n",
 491                 index, tgt->ltd_gen, lov->desc.ld_tgt_count);
 492
 493         if (tgt->ltd_exp) {
 494                 struct obd_device *osc_obd;
 495
 496                 osc_obd = class_exp2obd(tgt->ltd_exp);
 497                 if (osc_obd)
 498                         osc_obd->obd_no_recov = 0;
 499         }
 500
 501         if (lov->ocd.ocd_connect_flags != OBD_CONNECT_EMPTY) {
 502                 /* Keep the original connect flags pristine */
 503                 OBD_ALLOC(ocd, sizeof(*ocd));
 504                 if (!ocd)
 505                         RETURN(-ENOMEM);
 506                 *ocd = lov->ocd;
 507         }
 508         rc = lov_connect_obd(obd, tgt, active, ocd);
 509         if (ocd)
 510                 OBD_FREE(ocd, sizeof(*ocd));
 511         if (rc)
 512                 GOTO(out, rc);
 513
 514         idx = index;
 515         rc = lov_notify(obd, tgt->ltd_exp->exp_obd,
 516                         active ? OBD_NOTIFY_ACTIVE : OBD_NOTIFY_INACTIVE,
 517                         (void *)&idx);
 518
 519 out:
 520         if (rc) {
 521                 CERROR("add failed (%d), deleting %s\n", rc,
 522                        (char *)tgt->uuid.uuid);
 523                 lov_del_target(obd, &tgt->uuid, index, 0);
 524         }
 525         RETURN(rc);
 526 }
 527
 528 /* Schedule a target for deletion */
 529 static int lov_del_target(struct obd_device *obd, struct obd_uuid *uuidp,
 530                           int index, int gen)
 531 {
 532         struct lov_obd *lov = &obd->u.lov;
 533         struct lov_tgt_desc *tgt;
 534         int count = lov->desc.ld_tgt_count;
 535         int rc = 0;
 536         ENTRY;
 537
 538         if (index >= count) {
 539                 CERROR("LOV target index %d >= number of LOV OBDs %d.\n",
 540                        index, count);
 541                 RETURN(-EINVAL);
 542         }
 543
 544         tgt = &lov->tgts[index];
 545
 546         if (obd_uuid_empty(&tgt->uuid)) {
 547                 CERROR("LOV target at index %d is not setup.\n", index);
 548                 RETURN(-EINVAL);
 549         }
 550
 551         if (!obd_uuid_equals(uuidp, &tgt->uuid)) {
 552                 CERROR("LOV target UUID %s at index %d doesn't match %s.\n",
 553                        tgt->uuid.uuid, index, uuidp->uuid);
 554                 RETURN(-EINVAL);
 555         }
 556
 557         CDEBUG(D_CONFIG, "uuid: %s idx: %d gen: %d exp: %p active: %d\n",
 558                tgt->uuid.uuid, index, tgt->ltd_gen, tgt->ltd_exp, tgt->active);
 559
 560         lov_getref(obd);
 561         tgt->reap = 1;
 562         lov->death_row++;
 563         /* we really delete it from lov_putref */
 564         lov_putref(obd);
 565
 566         RETURN(rc);
 567 }
 568
 569 static void __lov_del_obd(struct obd_device *obd, struct lov_tgt_desc *tgt)
 570 {
 571         struct obd_device *osc_obd;
 572
 573         LASSERT(tgt->reap);
 574         osc_obd = class_exp2obd(tgt->ltd_exp);
 575
 576         CDEBUG(D_CONFIG, "Removing tgt %s : %s\n", tgt->uuid.uuid,
 577                osc_obd ? osc_obd->obd_name : "<no obd>");
 578
 579         if (tgt->ltd_exp)
 580                 lov_disconnect_obd(obd, tgt);
 581
 582         /* XXX - right now there is a dependency on ld_tgt_count being the
 583          * maximum tgt index for computing the mds_max_easize. So we can't
 584          * shrink it. */
 585
 586         /* lt_gen = 0 will mean it will not match the gen of any valid loi */
 587         memset(tgt, 0, sizeof(*tgt));
 588
 589         /* Manual cleanup - no cleanup logs to clean up the osc's.  We must
 590            do it ourselves. And we can't do it from lov_cleanup,
 591            because we just lost our only reference to it. */
 592         if (osc_obd) {
 593                 /* Use lov's force/fail flags. */
 594                 osc_obd->obd_force = obd->obd_force;
 595                 osc_obd->obd_fail = obd->obd_fail;
 596                 class_manual_cleanup(osc_obd);
 597         }
 598 }
 599
 600 static void lov_fix_desc(struct lov_desc *desc)
 601 {
 602         if (desc->ld_default_stripe_size < PTLRPC_MAX_BRW_SIZE) {
 603                 CWARN("Increasing default_stripe_size "LPU64" to %u\n",
 604                       desc->ld_default_stripe_size, PTLRPC_MAX_BRW_SIZE);
 605                 desc->ld_default_stripe_size = PTLRPC_MAX_BRW_SIZE;
 606         } else if (desc->ld_default_stripe_size & (LOV_MIN_STRIPE_SIZE - 1)) {
 607                 CWARN("default_stripe_size "LPU64" isn't a multiple of %u\n",
 608                       desc->ld_default_stripe_size, LOV_MIN_STRIPE_SIZE);
 609                 desc->ld_default_stripe_size &= ~(LOV_MIN_STRIPE_SIZE - 1);
 610                 CWARN("changing to "LPU64"\n", desc->ld_default_stripe_size);
 611        }
 612
 613         if (desc->ld_default_stripe_count == 0)
 614                 desc->ld_default_stripe_count = 1;
 615
 616         /* from lov_setstripe */
 617         if ((desc->ld_pattern != 0) &&
 618             (desc->ld_pattern != LOV_PATTERN_RAID0)) {
 619                 CDEBUG(D_IOCTL, "bad userland stripe pattern: %#x\n",
 620                        desc->ld_pattern);
 621                 desc->ld_pattern = 0;
 622         }
 623 }
 624
 625 static int lov_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
 626 {
 627         struct lprocfs_static_vars lvars;
 628         struct lov_desc *desc;
 629         struct lov_obd *lov = &obd->u.lov;
 630         struct lov_tgt_desc *tgts;
 631         int count, i;
 632         ENTRY;
 633
 634         if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
 635                 CERROR("LOV setup requires a descriptor\n");
 636                 RETURN(-EINVAL);
 637         }
 638
 639         desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
 640
 641         if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
 642                 CERROR("descriptor size wrong: %d > %d\n",
 643                        (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
 644                 RETURN(-EINVAL);
 645         }
 646
 647         if (desc->ld_magic != LOV_DESC_MAGIC) {
 648                 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
 649                             CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
 650                                    obd->obd_name, desc);
 651                             lustre_swab_lov_desc(desc);
 652                 } else {
 653                         CERROR("%s: Bad lov desc magic: %#x\n",
 654                                obd->obd_name, desc->ld_magic);
 655                         RETURN(-EINVAL);
 656                 }
 657         }
 658
 659         lov_fix_desc(desc);
 660
 661         /* Because of 64-bit divide/mod operations only work with a 32-bit
 662          * divisor in a 32-bit kernel, we cannot support a stripe width
 663          * of 4GB or larger on 32-bit CPUs. */
 664         count = desc->ld_default_stripe_count;
 665         if ((count > 0 ? count : desc->ld_tgt_count) *
 666             desc->ld_default_stripe_size > ~0UL) {
 667                 CERROR("LOV: stripe width "LPU64"x%u > %lu on 32-bit system\n",
 668                        desc->ld_default_stripe_size, count, ~0UL);
 669                 RETURN(-EINVAL);
 670         }
 671
 672         /* Allocate space for target list */
 673         if (desc->ld_tgt_count)
 674                 count = desc->ld_tgt_count;
 675         lov->bufsize = sizeof(struct lov_tgt_desc) * max(count, 1);
 676         OBD_ALLOC(lov->tgts, lov->bufsize);
 677         if (lov->tgts == NULL) {
 678                 CERROR("Out of memory\n");
 679                 RETURN(-EINVAL);
 680         }
 681         for (i = 0, tgts = lov->tgts; i < max(count, 1); i++, tgts++) {
 682                 tgts->index = i;
 683                 INIT_LIST_HEAD(&tgts->qos_bavail_list);
 684         }
 685
 686         desc->ld_active_tgt_count = 0;
 687         lov->desc = *desc;
 688         sema_init(&lov->lov_lock, 1);
 689         atomic_set(&lov->refcount, 0);
 690         INIT_LIST_HEAD(&lov->qos_bavail_list);
 691
 692         lprocfs_init_vars(lov, &lvars);
 693         lprocfs_obd_setup(obd, lvars.obd_vars);
 694 #ifdef LPROCFS
 695         {
 696                 cfs_proc_dir_entry_t *entry;
 697
 698                 entry = create_proc_entry("target_obd", 0444,
 699                                           obd->obd_proc_entry);
 700                 if (entry != NULL) {
 701                         entry->proc_fops = &lov_proc_target_fops;
 702                         entry->data = obd;
 703                 }
 704         }
 705 #endif
 706
 707         RETURN(0);
 708 }
 709
 710 static int lov_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
 711 {
 712         int rc = 0;
 713         ENTRY;
 714
 715         switch (stage) {
 716         case OBD_CLEANUP_EARLY: {
 717                 struct lov_obd *lov = &obd->u.lov;
 718                 int i;
 719                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
 720                         if (!lov->tgts[i].active)
 721                                 continue;
 722                         obd_precleanup(class_exp2obd(lov->tgts[i].ltd_exp),
 723                                        OBD_CLEANUP_EARLY);
 724                 }
 725                 break;
 726         }
 727         case OBD_CLEANUP_EXPORTS:
 728                 break;
 729         case OBD_CLEANUP_SELF_EXP:
 730                 rc = obd_llog_finish(obd, 0);
 731                 if (rc != 0)
 732                         CERROR("failed to cleanup llogging subsystems\n");
 733                 break;
 734         case OBD_CLEANUP_OBD:
 735                 break;
 736         }
 737         RETURN(rc);
 738 }
 739
 740 static int lov_cleanup(struct obd_device *obd)
 741 {
 742         struct lov_obd *lov = &obd->u.lov;
 743
 744         lprocfs_obd_cleanup(obd);
 745         if (lov->tgts) {
 746                 int i;
 747                 struct lov_tgt_desc *tgt;
 748                 for (i = 0, tgt = lov->tgts;
 749                       i < lov->desc.ld_tgt_count; i++, tgt++) {
 750                         /* We should never get here - these should have
 751                            been removed in the disconnect. */
 752                         if (!obd_uuid_empty(&tgt->uuid)) {
 753                                 CERROR("lov tgt %d not cleaned!"
 754                                        " deathrow=%d, lovrc=%d\n",
 755                                        i, lov->death_row,
 756                                        atomic_read(&lov->refcount));
 757                                 lov_del_target(obd, &tgt->uuid, i, 0);
 758                         }
 759                 }
 760                 OBD_FREE(lov->tgts, lov->bufsize);
 761         }
 762         RETURN(0);
 763 }
 764
 765 static int lov_process_config(struct obd_device *obd, obd_count len, void *buf)
 766 {
 767         struct lustre_cfg *lcfg = buf;
 768         struct obd_uuid obd_uuid;
 769         int cmd;
 770         int index;
 771         int gen;
 772         int rc = 0;
 773         ENTRY;
 774
 775         switch(cmd = lcfg->lcfg_command) {
 776         case LCFG_LOV_ADD_OBD:
 777         case LCFG_LOV_ADD_INA:
 778         case LCFG_LOV_DEL_OBD: {
 779                 /* lov_modify_tgts add  0:lov_mdsA  1:ost1_UUID  2:0  3:1 */
 780                 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
 781                         GOTO(out, rc = -EINVAL);
 782
 783                 obd_str2uuid(&obd_uuid,  lustre_cfg_buf(lcfg, 1));
 784
 785                 if (sscanf(lustre_cfg_buf(lcfg, 2), "%d", &index) != 1)
 786                         GOTO(out, rc = -EINVAL);
 787                 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", &gen) != 1)
 788                         GOTO(out, rc = -EINVAL);
 789                 if (cmd == LCFG_LOV_ADD_OBD)
 790                         rc = lov_add_target(obd, &obd_uuid, index, gen, 1);
 791                 else if (cmd == LCFG_LOV_ADD_INA)
 792                         rc = lov_add_target(obd, &obd_uuid, index, gen, 0);
 793                 else
 794                         rc = lov_del_target(obd, &obd_uuid, index, gen);
 795                 GOTO(out, rc);
 796         }
 797         case LCFG_PARAM: {
 798                 int i;
 799                 struct lov_obd *lov = &obd->u.lov;
 800                 struct lov_desc *desc = &(lov->desc);
 801                 if (!desc)
 802                         GOTO(out, rc = -EINVAL);
 803                 /* see jt_obd_lov_getconfig for variable names */
 804                 /* setparam 0:lov_mdsA 1:default_stripe_size=1048576
 805                    2:default_stripe_pattern=0 3:default_stripe_offset=0 */
 806                 for (i = 1; i < lcfg->lcfg_bufcount; i++) {
 807                         char *key, *sval;
 808                         long val;
 809                         key = lustre_cfg_buf(lcfg, i);
 810                         sval = strchr(key, '=');
 811                         if (!sval || (*(sval + 1) == 0)) {
 812                                 CERROR("Can't parse param %s\n", key);
 813                                 rc = -EINVAL;
 814                                 /* continue parsing other params */
 815                                 continue;
 816                         }
 817                         *sval = 0;
 818                         val = simple_strtol(sval + 1, NULL, 0);
 819                         if (strcmp(key, PARAM_D_STRIPE_SIZE) == 0)
 820                                 desc->ld_default_stripe_size = val;
 821                         else if (strcmp(key, PARAM_D_STRIPE_COUNT) == 0)
 822                                 desc->ld_default_stripe_count = val;
 823                         else if (strcmp(key, PARAM_D_STRIPE_OFFSET) == 0)
 824                                 desc->ld_default_stripe_offset = val;
 825                         else if (strcmp(key, PARAM_D_STRIPE_PATTERN) == 0)
 826                                 desc->ld_pattern = val;
 827                         else {
 828                                 CERROR("Unknown param %s\n", key);
 829                                 rc = -EINVAL;
 830                                 /* continue parsing other params */
 831                                 continue;
 832                         }
 833                         LCONSOLE_INFO("set %s to %ld\n", key, val);
 834                 }
 835                 lov_fix_desc(desc);
 836                 GOTO(out, rc);
 837         }
 838         default: {
 839                 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
 840                 GOTO(out, rc = -EINVAL);
 841
 842         }
 843         }
 844 out:
 845         RETURN(rc);
 846 }
 847
 848 #ifndef log2
 849 #define log2(n) ffz(~(n))
 850 #endif
 851
 852 static int lov_clear_orphans(struct obd_export *export, struct obdo *src_oa,
 853                              struct lov_stripe_md **ea,
 854                              struct obd_trans_info *oti)
 855 {
 856         struct lov_obd *lov;
 857         struct obdo *tmp_oa;
 858         struct obd_uuid *ost_uuid = NULL;
 859         int rc = 0, i;
 860         ENTRY;
 861
 862         LASSERT(src_oa->o_valid & OBD_MD_FLFLAGS &&
 863                 src_oa->o_flags == OBD_FL_DELORPHAN);
 864
 865         lov = &export->exp_obd->u.lov;
 866
 867         tmp_oa = obdo_alloc();
 868         if (tmp_oa == NULL)
 869                 RETURN(-ENOMEM);
 870
 871         if (src_oa->o_valid & OBD_MD_FLINLINE) {
 872                 ost_uuid = (struct obd_uuid *)src_oa->o_inline;
 873                 CDEBUG(D_HA, "clearing orphans only for %s\n",
 874                        ost_uuid->uuid);
 875         }
 876
 877         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
 878                 struct lov_stripe_md obj_md;
 879                 struct lov_stripe_md *obj_mdp = &obj_md;
 880                 int err;
 881
 882                 /* if called for a specific target, we don't
 883                    care if it is not active. */
 884                 if (lov->tgts[i].active == 0 && ost_uuid == NULL) {
 885                         CDEBUG(D_HA, "lov idx %d inactive\n", i);
 886                         continue;
 887                 }
 888
 889                 if (ost_uuid && !obd_uuid_equals(ost_uuid, &lov->tgts[i].uuid))
 890                         continue;
 891
 892                 CDEBUG(D_CONFIG,"Clear orphans for %d:%s\n", i, ost_uuid->uuid);
 893
 894                 memcpy(tmp_oa, src_oa, sizeof(*tmp_oa));
 895
 896                 LASSERT(lov->tgts[i].ltd_exp);
 897                 /* XXX: LOV STACKING: use real "obj_mdp" sub-data */
 898                 err = obd_create(lov->tgts[i].ltd_exp, tmp_oa, &obj_mdp, oti);
 899                 if (err)
 900                         /* This export will be disabled until it is recovered,
 901                            and then orphan recovery will be completed. */
 902                         CERROR("error in orphan recovery on OST idx %d/%d: "
 903                                "rc = %d\n", i, lov->desc.ld_tgt_count, err);
 904
 905                 if (ost_uuid)
 906                         break;
 907         }
 908         obdo_free(tmp_oa);
 909         RETURN(rc);
 910 }
 911
 912 static int lov_recreate(struct obd_export *exp, struct obdo *src_oa,
 913                         struct lov_stripe_md **ea, struct obd_trans_info *oti)
 914 {
 915         struct lov_stripe_md *obj_mdp, *lsm;
 916         struct lov_obd *lov = &exp->exp_obd->u.lov;
 917         unsigned ost_idx;
 918         int rc, i;
 919         ENTRY;
 920
 921         LASSERT(src_oa->o_valid & OBD_MD_FLFLAGS &&
 922                 src_oa->o_flags & OBD_FL_RECREATE_OBJS);
 923
 924         OBD_ALLOC(obj_mdp, sizeof(*obj_mdp));
 925         if (obj_mdp == NULL)
 926                 RETURN(-ENOMEM);
 927
 928         ost_idx = src_oa->o_nlink;
 929         lsm = *ea;
 930         if (lsm == NULL)
 931                 GOTO(out, rc = -EINVAL);
 932         if (ost_idx >= lov->desc.ld_tgt_count)
 933                 GOTO(out, rc = -EINVAL);
 934
 935         for (i = 0; i < lsm->lsm_stripe_count; i++) {
 936                 if (lsm->lsm_oinfo[i].loi_ost_idx == ost_idx) {
 937                         if (lsm->lsm_oinfo[i].loi_id != src_oa->o_id)
 938                                 GOTO(out, rc = -EINVAL);
 939                         break;
 940                 }
 941         }
 942         if (i == lsm->lsm_stripe_count)
 943                 GOTO(out, rc = -EINVAL);
 944
 945         rc = obd_create(lov->tgts[ost_idx].ltd_exp, src_oa, &obj_mdp, oti);
 946 out:
 947         OBD_FREE(obj_mdp, sizeof(*obj_mdp));
 948         RETURN(rc);
 949 }
 950
 951 /* the LOV expects oa->o_id to be set to the LOV object id */
 952 static int lov_create(struct obd_export *exp, struct obdo *src_oa,
 953                       struct lov_stripe_md **ea, struct obd_trans_info *oti)
 954 {
 955         struct lov_obd *lov;
 956         struct lov_request_set *set = NULL;
 957         struct obd_statfs osfs;
 958         unsigned long maxage;
 959         struct lov_request *req;
 960         int rc = 0;
 961         ENTRY;
 962
 963         LASSERT(ea != NULL);
 964         if (exp == NULL)
 965                 RETURN(-EINVAL);
 966
 967         if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
 968             src_oa->o_flags == OBD_FL_DELORPHAN) {
 969                 rc = lov_clear_orphans(exp, src_oa, ea, oti);
 970                 RETURN(rc);
 971         }
 972
 973         lov = &exp->exp_obd->u.lov;
 974         if (!lov->desc.ld_active_tgt_count)
 975                 RETURN(-EIO);
 976
 977         /* Recreate a specific object id at the given OST index */
 978         if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
 979             (src_oa->o_flags & OBD_FL_RECREATE_OBJS)) {
 980                  rc = lov_recreate(exp, src_oa, ea, oti);
 981                  RETURN(rc);
 982         }
 983
 984         maxage = cfs_time_shift(-lov->desc.ld_qos_maxage);
 985         obd_statfs(exp->exp_obd, &osfs, maxage);
 986
 987         rc = lov_prep_create_set(exp, ea, src_oa, oti, &set);
 988         if (rc)
 989                 RETURN(rc);
 990
 991         list_for_each_entry(req, &set->set_list, rq_link) {
 992                 /* XXX: LOV STACKING: use real "obj_mdp" sub-data */
 993                 rc = obd_create(lov->tgts[req->rq_idx].ltd_exp,
 994                                 req->rq_oa, &req->rq_md, oti);
 995                 lov_update_create_set(set, req, rc);
 996         }
 997         rc = lov_fini_create_set(set, ea);
 998         RETURN(rc);
 999 }
1000
1001 #define ASSERT_LSM_MAGIC(lsmp)                                                  \
1002 do {                                                                            \
1003         LASSERT((lsmp) != NULL);                                                \
1004         LASSERTF(((lsmp)->lsm_magic == LOV_MAGIC ||                             \
1005                  (lsmp)->lsm_magic == LOV_MAGIC_JOIN), "%p->lsm_magic=%x\n",    \
1006                  (lsmp), (lsmp)->lsm_magic);                                    \
1007 } while (0)
1008
1009 static int lov_destroy(struct obd_export *exp, struct obdo *oa,
1010                        struct lov_stripe_md *lsm, struct obd_trans_info *oti,
1011                        struct obd_export *md_exp)
1012 {
1013         struct lov_request_set *set;
1014         struct lov_request *req;
1015         struct list_head *pos;
1016         struct lov_obd *lov;
1017         int rc = 0;
1018         ENTRY;
1019
1020         ASSERT_LSM_MAGIC(lsm);
1021
1022         if (!exp || !exp->exp_obd)
1023                 RETURN(-ENODEV);
1024
1025         lov = &exp->exp_obd->u.lov;
1026         rc = lov_prep_destroy_set(exp, oa, lsm, oti, &set);
1027         if (rc)
1028                 RETURN(rc);
1029
1030         list_for_each (pos, &set->set_list) {
1031                 int err;
1032                 req = list_entry(pos, struct lov_request, rq_link);
1033
1034                 /* XXX update the cookie position */
1035                 oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1036                 rc = obd_destroy(lov->tgts[req->rq_idx].ltd_exp, req->rq_oa,
1037                                  NULL, oti, NULL);
1038                 err = lov_update_common_set(set, req, rc);
1039                 if (rc) {
1040                         CERROR("error: destroying objid "LPX64" subobj "
1041                                LPX64" on OST idx %d: rc = %d\n",
1042                                set->set_oa->o_id, req->rq_oa->o_id,
1043                                req->rq_idx, rc);
1044                         if (!rc)
1045                                 rc = err;
1046                 }
1047         }
1048         lov_fini_destroy_set(set);
1049         if (rc == 0) {
1050                 LASSERT(lsm_op_find(lsm->lsm_magic) != NULL);
1051                 rc = lsm_op_find(lsm->lsm_magic)->lsm_destroy(lsm, oa, md_exp);
1052         }
1053         RETURN(rc);
1054 }
1055
1056 static int lov_getattr(struct obd_export *exp, struct obdo *oa,
1057                        struct lov_stripe_md *lsm)
1058 {
1059         struct lov_request_set *set;
1060         struct lov_request *req;
1061         struct list_head *pos;
1062         struct lov_obd *lov;
1063         int err = 0, rc = 0;
1064         ENTRY;
1065
1066         ASSERT_LSM_MAGIC(lsm);
1067
1068         if (!exp || !exp->exp_obd)
1069                 RETURN(-ENODEV);
1070
1071         lov = &exp->exp_obd->u.lov;
1072
1073         rc = lov_prep_getattr_set(exp, oa, lsm, &set);
1074         if (rc)
1075                 RETURN(rc);
1076
1077         list_for_each (pos, &set->set_list) {
1078                 req = list_entry(pos, struct lov_request, rq_link);
1079
1080                 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
1081                        "%u\n", oa->o_id, req->rq_stripe, req->rq_oa->o_id,
1082                        req->rq_idx);
1083
1084                 rc = obd_getattr(lov->tgts[req->rq_idx].ltd_exp,
1085                                  req->rq_oa, NULL);
1086                 err = lov_update_common_set(set, req, rc);
1087                 if (err) {
1088                         CERROR("error: getattr objid "LPX64" subobj "
1089                                LPX64" on OST idx %d: rc = %d\n",
1090                                set->set_oa->o_id, req->rq_oa->o_id,
1091                                req->rq_idx, err);
1092                         break;
1093                 }
1094         }
1095
1096         rc = lov_fini_getattr_set(set);
1097         if (err)
1098                 rc = err;
1099         RETURN(rc);
1100 }
1101
1102 static int lov_getattr_interpret(struct ptlrpc_request_set *rqset, void *data,
1103                                  int rc)
1104 {
1105         struct lov_request_set *lovset = (struct lov_request_set *)data;
1106         ENTRY;
1107
1108         /* don't do attribute merge if this aysnc op failed */
1109         if (rc) {
1110                 lovset->set_completes = 0;
1111                 lov_fini_getattr_set(lovset);
1112         } else {
1113                 rc = lov_fini_getattr_set(lovset);
1114         }
1115         RETURN (rc);
1116 }
1117
1118 static int lov_getattr_async(struct obd_export *exp, struct obdo *oa,
1119                               struct lov_stripe_md *lsm,
1120                               struct ptlrpc_request_set *rqset)
1121 {
1122         struct lov_request_set *lovset;
1123         struct lov_obd *lov;
1124         struct list_head *pos;
1125         struct lov_request *req;
1126         int rc = 0;
1127         ENTRY;
1128
1129         ASSERT_LSM_MAGIC(lsm);
1130
1131         if (!exp || !exp->exp_obd)
1132                 RETURN(-ENODEV);
1133
1134         lov = &exp->exp_obd->u.lov;
1135
1136         rc = lov_prep_getattr_set(exp, oa, lsm, &lovset);
1137         if (rc)
1138                 RETURN(rc);
1139
1140         CDEBUG(D_INFO, "objid "LPX64": %ux%u byte stripes\n",
1141                lsm->lsm_object_id, lsm->lsm_stripe_count, lsm->lsm_stripe_size);
1142
1143         list_for_each (pos, &lovset->set_list) {
1144                 req = list_entry(pos, struct lov_request, rq_link);
1145
1146                 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
1147                        "%u\n", oa->o_id, req->rq_stripe, req->rq_oa->o_id,
1148                        req->rq_idx);
1149                 rc = obd_getattr_async(lov->tgts[req->rq_idx].ltd_exp,
1150                                        req->rq_oa, NULL, rqset);
1151                 if (rc) {
1152                         CERROR("error: getattr objid "LPX64" subobj "
1153                                LPX64" on OST idx %d: rc = %d\n",
1154                                lovset->set_oa->o_id, req->rq_oa->o_id,
1155                                req->rq_idx, rc);
1156                         GOTO(out, rc);
1157                 }
1158                 lov_update_common_set(lovset, req, rc);
1159         }
1160
1161         LASSERT(rc == 0);
1162         LASSERT (rqset->set_interpret == NULL);
1163         rqset->set_interpret = lov_getattr_interpret;
1164         rqset->set_arg = (void *)lovset;
1165         RETURN(rc);
1166 out:
1167         LASSERT(rc);
1168         lov_fini_getattr_set(lovset);
1169         RETURN(rc);
1170 }
1171
1172 static int lov_setattr(struct obd_export *exp, struct obdo *src_oa,
1173                        struct lov_stripe_md *lsm, struct obd_trans_info *oti)
1174 {
1175         struct lov_request_set *set;
1176         struct lov_obd *lov;
1177         struct list_head *pos;
1178         struct lov_request *req;
1179         int err = 0, rc = 0;
1180         ENTRY;
1181
1182         ASSERT_LSM_MAGIC(lsm);
1183
1184         if (!exp || !exp->exp_obd)
1185                 RETURN(-ENODEV);
1186
1187         /* for now, we only expect the following updates here */
1188         LASSERT(!(src_oa->o_valid & ~(OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLMODE |
1189                                       OBD_MD_FLATIME | OBD_MD_FLMTIME | OBD_MD_FLCTIME |
1190                                       OBD_MD_FLFLAGS | OBD_MD_FLSIZE | OBD_MD_FLGROUP |
1191                                       OBD_MD_FLUID | OBD_MD_FLGID | OBD_MD_FLINLINE |
1192                                       OBD_MD_FLFID | OBD_MD_FLGENER)));
1193         lov = &exp->exp_obd->u.lov;
1194         rc = lov_prep_setattr_set(exp, src_oa, lsm, oti, &set);
1195         if (rc)
1196                 RETURN(rc);
1197
1198         list_for_each (pos, &set->set_list) {
1199                 req = list_entry(pos, struct lov_request, rq_link);
1200
1201                 rc = obd_setattr(lov->tgts[req->rq_idx].ltd_exp, req->rq_oa,
1202                                  NULL, NULL);
1203                 err = lov_update_setattr_set(set, req, rc);
1204                 if (err) {
1205                         CERROR("error: setattr objid "LPX64" subobj "
1206                                LPX64" on OST idx %d: rc = %d\n",
1207                                set->set_oa->o_id, req->rq_oa->o_id,
1208                                req->rq_idx, err);
1209                         if (!rc)
1210                                 rc = err;
1211                 }
1212         }
1213         err = lov_fini_setattr_set(set);
1214         if (!rc)
1215                 rc = err;
1216         RETURN(rc);
1217 }
1218
1219 static int lov_setattr_async(struct obd_export *exp, struct obdo *src_oa,
1220                        struct lov_stripe_md *lsm, struct obd_trans_info *oti)
1221 {
1222         struct lov_obd *lov;
1223         struct lov_oinfo *loi = NULL;
1224         int rc = 0, err;
1225         obd_id objid = src_oa->o_id;
1226         int i;
1227         ENTRY;
1228
1229         ASSERT_LSM_MAGIC(lsm);
1230         LASSERT(oti);
1231         if (src_oa->o_valid & OBD_MD_FLCOOKIE)
1232                 LASSERT(oti->oti_logcookies);
1233
1234         if (!exp || !exp->exp_obd)
1235                 RETURN(-ENODEV);
1236
1237         LASSERT(!(src_oa->o_valid &  ~(OBD_MD_FLID | OBD_MD_FLUID |
1238                                        OBD_MD_FLGID| OBD_MD_FLCOOKIE |
1239                                        OBD_MD_FLFID | OBD_MD_FLGENER)));
1240         lov = &exp->exp_obd->u.lov;
1241
1242         loi = lsm->lsm_oinfo;
1243         for (i = 0; i < lsm->lsm_stripe_count; i++, loi++) {
1244                 if (lov->tgts[loi->loi_ost_idx].active == 0) {
1245                         CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
1246                         goto next;
1247                 }
1248
1249                 src_oa->o_id = loi->loi_id;
1250                 src_oa->o_stripe_idx = i;
1251
1252                 /* do chown/chgrp on OST asynchronously */
1253                 err = obd_setattr_async(lov->tgts[loi->loi_ost_idx].ltd_exp,
1254                                         src_oa, NULL, oti);
1255                 if (err) {
1256                         CERROR("error: setattr objid "LPX64" subobj "
1257                                LPX64" on OST idx %d: rc = %d\n",
1258                                objid, src_oa->o_id, i, err);
1259                         if (!rc)
1260                                 rc = err;
1261                 }
1262         next:
1263                 if (src_oa->o_valid & OBD_MD_FLCOOKIE)
1264                         oti->oti_logcookies++;
1265         }
1266
1267         RETURN(rc);
1268 }
1269
1270 /* FIXME: maybe we'll just make one node the authoritative attribute node, then
1271  * we can send this 'punch' to just the authoritative node and the nodes
1272  * that the punch will affect. */
1273 static int lov_punch(struct obd_export *exp, struct obdo *oa,
1274                      struct lov_stripe_md *lsm,
1275                      obd_off start, obd_off end, struct obd_trans_info *oti)
1276 {
1277         struct lov_request_set *set;
1278         struct lov_obd *lov;
1279         struct list_head *pos;
1280         struct lov_request *req;
1281         int err = 0, rc = 0;
1282         ENTRY;
1283
1284         ASSERT_LSM_MAGIC(lsm);
1285
1286         if (!exp || !exp->exp_obd)
1287                 RETURN(-ENODEV);
1288
1289         lov = &exp->exp_obd->u.lov;
1290         rc = lov_prep_punch_set(exp, oa, lsm, start, end, oti, &set);
1291         if (rc)
1292                 RETURN(rc);
1293
1294         list_for_each (pos, &set->set_list) {
1295                 req = list_entry(pos, struct lov_request, rq_link);
1296
1297                 rc = obd_punch(lov->tgts[req->rq_idx].ltd_exp, req->rq_oa,
1298                                NULL, req->rq_extent.start,
1299                                req->rq_extent.end, NULL);
1300                 err = lov_update_punch_set(set, req, rc);
1301                 if (err) {
1302                         CERROR("error: punch objid "LPX64" subobj "LPX64
1303                                " on OST idx %d: rc = %d\n", set->set_oa->o_id,
1304                                req->rq_oa->o_id, req->rq_idx, rc);
1305                         if (!rc)
1306                                 rc = err;
1307                 }
1308         }
1309         err = lov_fini_punch_set(set);
1310         if (!rc)
1311                 rc = err;
1312         RETURN(rc);
1313 }
1314
1315 static int lov_sync(struct obd_export *exp, struct obdo *oa,
1316                     struct lov_stripe_md *lsm, obd_off start, obd_off end)
1317 {
1318         struct lov_request_set *set;
1319         struct lov_obd *lov;
1320         struct list_head *pos;
1321         struct lov_request *req;
1322         int err = 0, rc = 0;
1323         ENTRY;
1324
1325         ASSERT_LSM_MAGIC(lsm);
1326
1327         if (!exp->exp_obd)
1328                 RETURN(-ENODEV);
1329
1330         lov = &exp->exp_obd->u.lov;
1331         rc = lov_prep_sync_set(exp, oa, lsm, start, end, &set);
1332         if (rc)
1333                 RETURN(rc);
1334
1335         list_for_each (pos, &set->set_list) {
1336                 req = list_entry(pos, struct lov_request, rq_link);
1337
1338                 rc = obd_sync(lov->tgts[req->rq_idx].ltd_exp, req->rq_oa,
1339                               NULL, req->rq_extent.start, req->rq_extent.end);
1340                 err = lov_update_common_set(set, req, rc);
1341                 if (err) {
1342                         CERROR("error: fsync objid "LPX64" subobj "LPX64
1343                                " on OST idx %d: rc = %d\n", set->set_oa->o_id,
1344                                req->rq_oa->o_id, req->rq_idx, rc);
1345                         if (!rc)
1346                                 rc = err;
1347                 }
1348         }
1349         err = lov_fini_sync_set(set);
1350         if (!rc)
1351                 rc = err;
1352         RETURN(rc);
1353 }
1354
1355 static int lov_brw_check(struct lov_obd *lov, struct obdo *oa,
1356                          struct lov_stripe_md *lsm,
1357                          obd_count oa_bufs, struct brw_page *pga)
1358 {
1359         int i, rc = 0;
1360
1361         /* The caller just wants to know if there's a chance that this
1362          * I/O can succeed */
1363         for (i = 0; i < oa_bufs; i++) {
1364                 int stripe = lov_stripe_number(lsm, pga[i].off);
1365                 int ost = lsm->lsm_oinfo[stripe].loi_ost_idx;
1366                 obd_off start, end;
1367
1368                 if (!lov_stripe_intersects(lsm, i, pga[i].off,
1369                                            pga[i].off + pga[i].count,
1370                                            &start, &end))
1371                         continue;
1372
1373                 if (lov->tgts[ost].active == 0) {
1374                         CDEBUG(D_HA, "lov idx %d inactive\n", ost);
1375                         return -EIO;
1376                 }
1377                 rc = obd_brw(OBD_BRW_CHECK, lov->tgts[ost].ltd_exp, oa,
1378                              NULL, 1, &pga[i], NULL);
1379                 if (rc)
1380                         break;
1381         }
1382         return rc;
1383 }
1384
1385 static int lov_brw(int cmd, struct obd_export *exp, struct obdo *src_oa,
1386                    struct lov_stripe_md *lsm, obd_count oa_bufs,
1387                    struct brw_page *pga, struct obd_trans_info *oti)
1388 {
1389         struct lov_request_set *set;
1390         struct lov_request *req;
1391         struct list_head *pos;
1392         struct lov_obd *lov = &exp->exp_obd->u.lov;
1393         int err, rc = 0;
1394         ENTRY;
1395
1396         ASSERT_LSM_MAGIC(lsm);
1397
1398         if (cmd == OBD_BRW_CHECK) {
1399                 rc = lov_brw_check(lov, src_oa, lsm, oa_bufs, pga);
1400                 RETURN(rc);
1401         }
1402
1403         rc = lov_prep_brw_set(exp, src_oa, lsm, oa_bufs, pga, oti, &set);
1404         if (rc)
1405                 RETURN(rc);
1406
1407         list_for_each (pos, &set->set_list) {
1408                 struct obd_export *sub_exp;
1409                 struct brw_page *sub_pga;
1410                 req = list_entry(pos, struct lov_request, rq_link);
1411
1412                 sub_exp = lov->tgts[req->rq_idx].ltd_exp;
1413                 sub_pga = set->set_pga + req->rq_pgaidx;
1414                 rc = obd_brw(cmd, sub_exp, req->rq_oa, req->rq_md,
1415                              req->rq_oabufs, sub_pga, oti);
1416                 if (rc)
1417                         break;
1418                 lov_update_common_set(set, req, rc);
1419         }
1420
1421         err = lov_fini_brw_set(set);
1422         if (!rc)
1423                 rc = err;
1424         RETURN(rc);
1425 }
1426
1427 static int lov_brw_interpret(struct ptlrpc_request_set *reqset, void *data,
1428                              int rc)
1429 {
1430         struct lov_request_set *lovset = (struct lov_request_set *)data;
1431         ENTRY;
1432
1433         if (rc) {
1434                 lovset->set_completes = 0;
1435                 lov_fini_brw_set(lovset);
1436         } else {
1437                 rc = lov_fini_brw_set(lovset);
1438         }
1439
1440         RETURN(rc);
1441 }
1442
1443 static int lov_brw_async(int cmd, struct obd_export *exp, struct obdo *oa,
1444                          struct lov_stripe_md *lsm, obd_count oa_bufs,
1445                          struct brw_page *pga, struct ptlrpc_request_set *set,
1446                          struct obd_trans_info *oti)
1447 {
1448         struct lov_request_set *lovset;
1449         struct lov_request *req;
1450         struct list_head *pos;
1451         struct lov_obd *lov = &exp->exp_obd->u.lov;
1452         int rc = 0;
1453         ENTRY;
1454
1455         ASSERT_LSM_MAGIC(lsm);
1456
1457         if (cmd == OBD_BRW_CHECK) {
1458                 rc = lov_brw_check(lov, oa, lsm, oa_bufs, pga);
1459                 RETURN(rc);
1460         }
1461
1462         rc = lov_prep_brw_set(exp, oa, lsm, oa_bufs, pga, oti, &lovset);
1463         if (rc)
1464                 RETURN(rc);
1465
1466         list_for_each (pos, &lovset->set_list) {
1467                 struct obd_export *sub_exp;
1468                 struct brw_page *sub_pga;
1469                 req = list_entry(pos, struct lov_request, rq_link);
1470
1471                 sub_exp = lov->tgts[req->rq_idx].ltd_exp;
1472                 sub_pga = lovset->set_pga + req->rq_pgaidx;
1473                 rc = obd_brw_async(cmd, sub_exp, req->rq_oa, req->rq_md,
1474                                    req->rq_oabufs, sub_pga, set, oti);
1475                 if (rc)
1476                         GOTO(out, rc);
1477                 lov_update_common_set(lovset, req, rc);
1478         }
1479         LASSERT(rc == 0);
1480         LASSERT(set->set_interpret == NULL);
1481         set->set_interpret = (set_interpreter_func)lov_brw_interpret;
1482         set->set_arg = (void *)lovset;
1483
1484         RETURN(rc);
1485 out:
1486         lov_fini_brw_set(lovset);
1487         RETURN(rc);
1488 }
1489
1490 static int lov_ap_make_ready(void *data, int cmd)
1491 {
1492         struct lov_async_page *lap = LAP_FROM_COOKIE(data);
1493
1494         return lap->lap_caller_ops->ap_make_ready(lap->lap_caller_data, cmd);
1495 }
1496
1497 static int lov_ap_refresh_count(void *data, int cmd)
1498 {
1499         struct lov_async_page *lap = LAP_FROM_COOKIE(data);
1500
1501         return lap->lap_caller_ops->ap_refresh_count(lap->lap_caller_data,
1502                                                      cmd);
1503 }
1504
1505 static void lov_ap_fill_obdo(void *data, int cmd, struct obdo *oa)
1506 {
1507         struct lov_async_page *lap = LAP_FROM_COOKIE(data);
1508
1509         lap->lap_caller_ops->ap_fill_obdo(lap->lap_caller_data, cmd, oa);
1510         /* XXX woah, shouldn't we be altering more here?  size? */
1511         oa->o_id = lap->lap_loi_id;
1512         oa->o_stripe_idx = lap->lap_stripe;
1513 }
1514
1515 static void lov_ap_completion(void *data, int cmd, struct obdo *oa, int rc)
1516 {
1517         struct lov_async_page *lap = LAP_FROM_COOKIE(data);
1518
1519         /* in a raid1 regime this would down a count of many ios
1520          * in flight, onl calling the caller_ops completion when all
1521          * the raid1 ios are complete */
1522         lap->lap_caller_ops->ap_completion(lap->lap_caller_data, cmd, oa, rc);
1523 }
1524
1525 static struct obd_async_page_ops lov_async_page_ops = {
1526         .ap_make_ready =        lov_ap_make_ready,
1527         .ap_refresh_count =     lov_ap_refresh_count,
1528         .ap_fill_obdo =         lov_ap_fill_obdo,
1529         .ap_completion =        lov_ap_completion,
1530 };
1531
1532 int lov_prep_async_page(struct obd_export *exp, struct lov_stripe_md *lsm,
1533                            struct lov_oinfo *loi, cfs_page_t *page,
1534                            obd_off offset, struct obd_async_page_ops *ops,
1535                            void *data, void **res)
1536 {
1537         struct lov_obd *lov = &exp->exp_obd->u.lov;
1538         struct lov_async_page *lap;
1539         int rc = 0;
1540         ENTRY;
1541
1542         if (!page) {
1543                 int i = 0;
1544                 /* Find an existing osc so we can get it's stupid sizeof(*oap).
1545                    Only because of this layering limitation will a client
1546                    mount with no osts fail */
1547                 while (!lov->tgts[i].ltd_exp) {
1548                         i++;
1549                         if (i >= lov->desc.ld_tgt_count)
1550                                 RETURN(-ENOTBLK);
1551                 }
1552                 rc = size_round(sizeof(*lap)) +
1553                         obd_prep_async_page(lov->tgts[i].ltd_exp, NULL, NULL,
1554                                             NULL, 0, NULL, NULL, NULL);
1555                 RETURN(rc);
1556         }
1557         ASSERT_LSM_MAGIC(lsm);
1558         LASSERT(loi == NULL);
1559
1560         lap = *res;
1561         lap->lap_magic = LAP_MAGIC;
1562         lap->lap_caller_ops = ops;
1563         lap->lap_caller_data = data;
1564
1565         /* for now only raid 0 which passes through */
1566         lap->lap_stripe = lov_stripe_number(lsm, offset);
1567         lov_stripe_offset(lsm, offset, lap->lap_stripe, &lap->lap_sub_offset);
1568         loi = &lsm->lsm_oinfo[lap->lap_stripe];
1569
1570         /* so the callback doesn't need the lsm */
1571         lap->lap_loi_id = loi->loi_id;
1572
1573         lap->lap_sub_cookie = (void *)lap + size_round(sizeof(*lap));
1574
1575         rc = obd_prep_async_page(lov->tgts[loi->loi_ost_idx].ltd_exp,
1576                                  lsm, loi, page, lap->lap_sub_offset,
1577                                  &lov_async_page_ops, lap,
1578                                  &lap->lap_sub_cookie);
1579         if (rc)
1580                 RETURN(rc);
1581         CDEBUG(D_CACHE, "lap %p page %p cookie %p off "LPU64"\n", lap, page,
1582                lap->lap_sub_cookie, offset);
1583         RETURN(0);
1584 }
1585
1586 static int lov_queue_async_io(struct obd_export *exp,
1587                               struct lov_stripe_md *lsm,
1588                               struct lov_oinfo *loi, void *cookie,
1589                               int cmd, obd_off off, int count,
1590                               obd_flag brw_flags, obd_flag async_flags)
1591 {
1592         struct lov_obd *lov = &exp->exp_obd->u.lov;
1593         struct lov_async_page *lap;
1594         int rc;
1595
1596         LASSERT(loi == NULL);
1597
1598         ASSERT_LSM_MAGIC(lsm);
1599
1600         lap = LAP_FROM_COOKIE(cookie);
1601
1602         loi = &lsm->lsm_oinfo[lap->lap_stripe];
1603
1604         rc = obd_queue_async_io(lov->tgts[loi->loi_ost_idx].ltd_exp, lsm,
1605                                 loi, lap->lap_sub_cookie, cmd, off, count,
1606                                 brw_flags, async_flags);
1607         RETURN(rc);
1608 }
1609
1610 static int lov_set_async_flags(struct obd_export *exp,
1611                                struct lov_stripe_md *lsm,
1612                                struct lov_oinfo *loi, void *cookie,
1613                                obd_flag async_flags)
1614 {
1615         struct lov_obd *lov = &exp->exp_obd->u.lov;
1616         struct lov_async_page *lap;
1617         int rc;
1618
1619         LASSERT(loi == NULL);
1620
1621         ASSERT_LSM_MAGIC(lsm);
1622
1623         lap = LAP_FROM_COOKIE(cookie);
1624
1625         loi = &lsm->lsm_oinfo[lap->lap_stripe];
1626
1627         rc = obd_set_async_flags(lov->tgts[loi->loi_ost_idx].ltd_exp,
1628                                  lsm, loi, lap->lap_sub_cookie, async_flags);
1629         RETURN(rc);
1630 }
1631
1632 static int lov_queue_group_io(struct obd_export *exp,
1633                               struct lov_stripe_md *lsm,
1634                               struct lov_oinfo *loi,
1635                               struct obd_io_group *oig, void *cookie,
1636                               int cmd, obd_off off, int count,
1637                               obd_flag brw_flags, obd_flag async_flags)
1638 {
1639         struct lov_obd *lov = &exp->exp_obd->u.lov;
1640         struct lov_async_page *lap;
1641         int rc;
1642
1643         LASSERT(loi == NULL);
1644
1645         ASSERT_LSM_MAGIC(lsm);
1646
1647         lap = LAP_FROM_COOKIE(cookie);
1648
1649         loi = &lsm->lsm_oinfo[lap->lap_stripe];
1650
1651         rc = obd_queue_group_io(lov->tgts[loi->loi_ost_idx].ltd_exp, lsm, loi,
1652                                 oig, lap->lap_sub_cookie, cmd, off, count,
1653                                 brw_flags, async_flags);
1654         RETURN(rc);
1655 }
1656
1657 /* this isn't exactly optimal.  we may have queued sync io in oscs on
1658  * all stripes, but we don't record that fact at queue time.  so we
1659  * trigger sync io on all stripes. */
1660 static int lov_trigger_group_io(struct obd_export *exp,
1661                                 struct lov_stripe_md *lsm,
1662                                 struct lov_oinfo *loi,
1663                                 struct obd_io_group *oig)
1664 {
1665         struct lov_obd *lov = &exp->exp_obd->u.lov;
1666         int rc = 0, i, err;
1667
1668         LASSERT(loi == NULL);
1669
1670         ASSERT_LSM_MAGIC(lsm);
1671
1672         loi = lsm->lsm_oinfo;
1673         for (i = 0; i < lsm->lsm_stripe_count; i++, loi++) {
1674                 if (lov->tgts[loi->loi_ost_idx].active == 0) {
1675                         CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
1676                         continue;
1677                 }
1678
1679                 err = obd_trigger_group_io(lov->tgts[loi->loi_ost_idx].ltd_exp,
1680                                            lsm, loi, oig);
1681                 if (rc == 0 && err != 0)
1682                         rc = err;
1683         };
1684         RETURN(rc);
1685 }
1686
1687 static int lov_teardown_async_page(struct obd_export *exp,
1688                                    struct lov_stripe_md *lsm,
1689                                    struct lov_oinfo *loi, void *cookie)
1690 {
1691         struct lov_obd *lov = &exp->exp_obd->u.lov;
1692         struct lov_async_page *lap;
1693         int rc;
1694
1695         LASSERT(loi == NULL);
1696
1697         ASSERT_LSM_MAGIC(lsm);
1698
1699         lap = LAP_FROM_COOKIE(cookie);
1700
1701         loi = &lsm->lsm_oinfo[lap->lap_stripe];
1702
1703         rc = obd_teardown_async_page(lov->tgts[loi->loi_ost_idx].ltd_exp,
1704                                      lsm, loi, lap->lap_sub_cookie);
1705         if (rc) {
1706                 CERROR("unable to teardown sub cookie %p: %d\n",
1707                        lap->lap_sub_cookie, rc);
1708                 RETURN(rc);
1709         }
1710         RETURN(rc);
1711 }
1712
1713 static int lov_enqueue(struct obd_export *exp, struct lov_stripe_md *lsm,
1714                        __u32 type, ldlm_policy_data_t *policy, __u32 mode,
1715                        int *flags, void *bl_cb, void *cp_cb, void *gl_cb,
1716                        void *data,__u32 lvb_len, void *lvb_swabber,
1717                        struct lustre_handle *lockh)
1718 {
1719         struct lov_request_set *set;
1720         struct lov_request *req;
1721         struct list_head *pos;
1722         struct lustre_handle *lov_lockhp;
1723         struct lov_obd *lov;
1724         ldlm_error_t rc;
1725         int save_flags = *flags;
1726         ENTRY;
1727
1728         ASSERT_LSM_MAGIC(lsm);
1729
1730         /* we should never be asked to replay a lock this way. */
1731         LASSERT((*flags & LDLM_FL_REPLAY) == 0);
1732
1733         if (!exp || !exp->exp_obd)
1734                 RETURN(-ENODEV);
1735
1736         lov = &exp->exp_obd->u.lov;
1737         rc = lov_prep_enqueue_set(exp, lsm, policy, mode, lockh, &set);
1738         if (rc)
1739                 RETURN(rc);
1740
1741         list_for_each (pos, &set->set_list) {
1742                 ldlm_policy_data_t sub_policy;
1743                 req = list_entry(pos, struct lov_request, rq_link);
1744                 lov_lockhp = set->set_lockh->llh_handles + req->rq_stripe;
1745                 LASSERT(lov_lockhp);
1746
1747                 *flags = save_flags;
1748                 sub_policy.l_extent = req->rq_extent;
1749
1750                 rc = obd_enqueue(lov->tgts[req->rq_idx].ltd_exp, req->rq_md,
1751                                  type, &sub_policy, mode, flags, bl_cb,
1752                                  cp_cb, gl_cb, data, lvb_len, lvb_swabber,
1753                                  lov_lockhp);
1754                 rc = lov_update_enqueue_set(set, req, rc, save_flags);
1755                 if (rc != ELDLM_OK)
1756                         break;
1757         }
1758
1759         lov_fini_enqueue_set(set, mode);
1760         RETURN(rc);
1761 }
1762
1763 static int lov_match(struct obd_export *exp, struct lov_stripe_md *lsm,
1764                      __u32 type, ldlm_policy_data_t *policy, __u32 mode,
1765                      int *flags, void *data, struct lustre_handle *lockh)
1766 {
1767         struct lov_request_set *set;
1768         struct lov_request *req;
1769         struct list_head *pos;
1770         struct lov_obd *lov = &exp->exp_obd->u.lov;
1771         struct lustre_handle *lov_lockhp;
1772         int lov_flags, rc = 0;
1773         ENTRY;
1774
1775         ASSERT_LSM_MAGIC(lsm);
1776
1777         if (!exp || !exp->exp_obd)
1778                 RETURN(-ENODEV);
1779
1780         lov = &exp->exp_obd->u.lov;
1781         rc = lov_prep_match_set(exp, lsm, policy, mode, lockh, &set);
1782         if (rc)
1783                 RETURN(rc);
1784
1785         list_for_each (pos, &set->set_list) {
1786                 ldlm_policy_data_t sub_policy;
1787                 req = list_entry(pos, struct lov_request, rq_link);
1788                 lov_lockhp = set->set_lockh->llh_handles + req->rq_stripe;
1789                 LASSERT(lov_lockhp);
1790
1791                 lov_flags = *flags;
1792                 sub_policy.l_extent = req->rq_extent;
1793
1794                 rc = obd_match(lov->tgts[req->rq_idx].ltd_exp, req->rq_md,
1795                                type, &sub_policy, mode, &lov_flags, data,
1796                                lov_lockhp);
1797                 rc = lov_update_match_set(set, req, rc);
1798                 if (rc != 1)
1799                         break;
1800         }
1801         lov_fini_match_set(set, mode, *flags);
1802         RETURN(rc);
1803 }
1804
1805 static int lov_change_cbdata(struct obd_export *exp,
1806                              struct lov_stripe_md *lsm, ldlm_iterator_t it,
1807                              void *data)
1808 {
1809         struct lov_obd *lov;
1810         struct lov_oinfo *loi;
1811         int rc = 0, i;
1812         ENTRY;
1813
1814         ASSERT_LSM_MAGIC(lsm);
1815
1816         if (!exp || !exp->exp_obd)
1817                 RETURN(-ENODEV);
1818
1819         lov = &exp->exp_obd->u.lov;
1820         for (i = 0,loi = lsm->lsm_oinfo; i < lsm->lsm_stripe_count; i++,loi++) {
1821                 struct lov_stripe_md submd;
1822
1823                 submd.lsm_object_id = loi->loi_id;
1824                 submd.lsm_stripe_count = 0;
1825                 rc = obd_change_cbdata(lov->tgts[loi->loi_ost_idx].ltd_exp,
1826                                        &submd, it, data);
1827         }
1828         RETURN(rc);
1829 }
1830
1831 static int lov_cancel(struct obd_export *exp, struct lov_stripe_md *lsm,
1832                       __u32 mode, struct lustre_handle *lockh)
1833 {
1834         struct lov_request_set *set;
1835         struct lov_request *req;
1836         struct list_head *pos;
1837         struct lov_obd *lov = &exp->exp_obd->u.lov;
1838         struct lustre_handle *lov_lockhp;
1839         int err = 0, rc = 0;
1840         ENTRY;
1841
1842         ASSERT_LSM_MAGIC(lsm);
1843
1844         if (!exp || !exp->exp_obd)
1845                 RETURN(-ENODEV);
1846
1847         LASSERT(lockh);
1848         lov = &exp->exp_obd->u.lov;
1849         rc = lov_prep_cancel_set(exp, lsm, mode, lockh, &set);
1850         if (rc)
1851                 RETURN(rc);
1852
1853         list_for_each (pos, &set->set_list) {
1854                 req = list_entry(pos, struct lov_request, rq_link);
1855                 lov_lockhp = set->set_lockh->llh_handles + req->rq_stripe;
1856
1857                 rc = obd_cancel(lov->tgts[req->rq_idx].ltd_exp, req->rq_md,
1858                                 mode, lov_lockhp);
1859                 rc = lov_update_common_set(set, req, rc);
1860                 if (rc) {
1861                         CERROR("error: cancel objid "LPX64" subobj "
1862                                LPX64" on OST idx %d: rc = %d\n",
1863                                lsm->lsm_object_id,
1864                                req->rq_md->lsm_object_id, req->rq_idx, rc);
1865                         err = rc;
1866                 }
1867
1868         }
1869         lov_fini_cancel_set(set);
1870         RETURN(err);
1871 }
1872
1873 static int lov_cancel_unused(struct obd_export *exp,
1874                              struct lov_stripe_md *lsm,
1875                              int flags, void *opaque)
1876 {
1877         struct lov_oinfo *loi;
1878         struct lov_obd *lov;
1879         int rc = 0, i;
1880         ENTRY;
1881
1882         lov = &exp->exp_obd->u.lov;
1883         if (lsm == NULL) {
1884                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
1885                         int err;
1886                         if (!lov->tgts[i].ltd_exp)
1887                                 continue;
1888
1889                         err = obd_cancel_unused(lov->tgts[i].ltd_exp, NULL,
1890                                                 flags, opaque);
1891                         if (!rc)
1892                                 rc = err;
1893                 }
1894                 RETURN(rc);
1895         }
1896
1897         ASSERT_LSM_MAGIC(lsm);
1898
1899         if (!exp || !exp->exp_obd)
1900                 RETURN(-ENODEV);
1901
1902         for (i = 0,loi = lsm->lsm_oinfo; i < lsm->lsm_stripe_count; i++,loi++) {
1903                 struct lov_stripe_md submd;
1904                 int err;
1905
1906                 if (lov->tgts[loi->loi_ost_idx].active == 0)
1907                         CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
1908
1909                 submd.lsm_object_id = loi->loi_id;
1910                 submd.lsm_stripe_count = 0;
1911                 err = obd_cancel_unused(lov->tgts[loi->loi_ost_idx].ltd_exp,
1912                                         &submd, flags, opaque);
1913                 if (err && lov->tgts[loi->loi_ost_idx].active) {
1914                         CERROR("error: cancel unused objid "LPX64" subobj "LPX64
1915                                " on OST idx %d: rc = %d\n", lsm->lsm_object_id,
1916                                loi->loi_id, loi->loi_ost_idx, err);
1917                         if (!rc)
1918                                 rc = err;
1919                 }
1920         }
1921         RETURN(rc);
1922 }
1923
1924 static int lov_join_lru(struct obd_export *exp,
1925                         struct lov_stripe_md *lsm, int join)
1926 {
1927         struct lov_obd *lov;
1928         struct lov_oinfo *loi;
1929         int i, count = 0;
1930         ENTRY;
1931
1932         ASSERT_LSM_MAGIC(lsm);
1933         if (!exp || !exp->exp_obd)
1934                 RETURN(-ENODEV);
1935
1936         lov = &exp->exp_obd->u.lov;
1937         for (i = 0,loi = lsm->lsm_oinfo; i < lsm->lsm_stripe_count; i++,loi++) {
1938                 struct lov_stripe_md submd;
1939                 int rc = 0;
1940
1941                 if (lov->tgts[loi->loi_ost_idx].active == 0)
1942                         CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
1943
1944                 submd.lsm_object_id = loi->loi_id;
1945                 submd.lsm_stripe_count = 0;
1946                 rc = obd_join_lru(lov->tgts[loi->loi_ost_idx].ltd_exp,
1947                                   &submd, join);
1948                 if (rc < 0) {
1949                         CERROR("join lru failed. objid: "LPX64" subobj: "LPX64
1950                                " ostidx: %d rc: %d\n", lsm->lsm_object_id,
1951                                loi->loi_id, loi->loi_ost_idx, rc);
1952                         return rc;
1953                 } else {
1954                         count += rc;
1955                 }
1956         }
1957         RETURN(count);
1958 }
1959
1960 #define LOV_U64_MAX ((__u64)~0ULL)
1961 #define LOV_SUM_MAX(tot, add)                                           \
1962         do {                                                            \
1963                 if ((tot) + (add) < (tot))                              \
1964                         (tot) = LOV_U64_MAX;                            \
1965                 else                                                    \
1966                         (tot) += (add);                                 \
1967         } while(0)
1968
1969 static int lov_statfs(struct obd_device *obd, struct obd_statfs *osfs,
1970                       cfs_time_t max_age)
1971 {
1972         struct lov_obd *lov = &obd->u.lov;
1973         struct obd_statfs lov_sfs;
1974         int set = 0;
1975         int rc = 0;
1976         int i;
1977         ENTRY;
1978
1979
1980         /* We only get block data from the OBD */
1981         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
1982                 int err;
1983                 if (!lov->tgts[i].active) {
1984                         CDEBUG(D_HA, "lov idx %d inactive\n", i);
1985                         continue;
1986                 }
1987
1988                 err = obd_statfs(class_exp2obd(lov->tgts[i].ltd_exp), &lov_sfs,
1989                                  max_age);
1990                 if (err) {
1991                         if (lov->tgts[i].active && !rc)
1992                                 rc = err;
1993                         continue;
1994                 }
1995                 qos_update(lov, i, &lov_sfs);
1996
1997                 if (!set) {
1998                         memcpy(osfs, &lov_sfs, sizeof(lov_sfs));
1999                         set = 1;
2000                 } else {
2001 #ifdef MIN_DF
2002                         /* Sandia requested that df (and so, statfs) only
2003                            returned minimal available space on
2004                            a single OST, so people would be able to
2005                            write this much data guaranteed. */
2006                         if (osfs->os_bavail > lov_sfs.os_bavail) {
2007                                 /* Presumably if new bavail is smaller,
2008                                    new bfree is bigger as well */
2009                                 osfs->os_bfree = lov_sfs.os_bfree;
2010                                 osfs->os_bavail = lov_sfs.os_bavail;
2011                         }
2012 #else
2013                         osfs->os_bfree += lov_sfs.os_bfree;
2014                         osfs->os_bavail += lov_sfs.os_bavail;
2015 #endif
2016                         osfs->os_blocks += lov_sfs.os_blocks;
2017                         /* XXX not sure about this one - depends on policy.
2018                          *   - could be minimum if we always stripe on all OBDs
2019                          *     (but that would be wrong for any other policy,
2020                          *     if one of the OBDs has no more objects left)
2021                          *   - could be sum if we stripe whole objects
2022                          *   - could be average, just to give a nice number
2023                          *
2024                          * To give a "reasonable" (if not wholly accurate)
2025                          * number, we divide the total number of free objects
2026                          * by expected stripe count (watch out for overflow).
2027                          */
2028                         LOV_SUM_MAX(osfs->os_files, lov_sfs.os_files);
2029                         LOV_SUM_MAX(osfs->os_ffree, lov_sfs.os_ffree);
2030                 }
2031         }
2032
2033         if (set) {
2034                 __u32 expected_stripes = lov_get_stripecnt(lov, 0);
2035
2036                 if (osfs->os_files != LOV_U64_MAX)
2037                         do_div(osfs->os_files, expected_stripes);
2038                 if (osfs->os_ffree != LOV_U64_MAX)
2039                         do_div(osfs->os_ffree, expected_stripes);
2040         } else if (!rc)
2041                 rc = -EIO;
2042
2043         RETURN(rc);
2044 }
2045
2046 static int lov_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
2047                          void *karg, void *uarg)
2048 {
2049         struct obd_device *obddev = class_exp2obd(exp);
2050         struct lov_obd *lov = &obddev->u.lov;
2051         int i, rc, count = lov->desc.ld_tgt_count;
2052         struct obd_uuid *uuidp;
2053         ENTRY;
2054
2055         switch (cmd) {
2056         case OBD_IOC_LOV_GET_CONFIG: {
2057                 struct obd_ioctl_data *data = karg;
2058                 struct lov_tgt_desc *tgtdesc;
2059                 struct lov_desc *desc;
2060                 char *buf = NULL;
2061                 __u32 *genp;
2062
2063                 buf = NULL;
2064                 len = 0;
2065                 if (obd_ioctl_getdata(&buf, &len, (void *)uarg))
2066                         RETURN(-EINVAL);
2067
2068                 data = (struct obd_ioctl_data *)buf;
2069
2070                 if (sizeof(*desc) > data->ioc_inllen1) {
2071                         obd_ioctl_freedata(buf, len);
2072                         RETURN(-EINVAL);
2073                 }
2074
2075                 if (sizeof(uuidp->uuid) * count > data->ioc_inllen2) {
2076                         obd_ioctl_freedata(buf, len);
2077                         RETURN(-EINVAL);
2078                 }
2079
2080                 if (sizeof(__u32) * count > data->ioc_inllen3) {
2081                         obd_ioctl_freedata(buf, len);
2082                         RETURN(-EINVAL);
2083                 }
2084
2085                 desc = (struct lov_desc *)data->ioc_inlbuf1;
2086                 memcpy(desc, &(lov->desc), sizeof(*desc));
2087
2088                 uuidp = (struct obd_uuid *)data->ioc_inlbuf2;
2089                 genp = (__u32 *)data->ioc_inlbuf3;
2090                 tgtdesc = lov->tgts;
2091                 /* the uuid will be empty for deleted OSTs */
2092                 for (i = 0; i < count; i++, uuidp++, genp++, tgtdesc++) {
2093                         obd_str2uuid(uuidp, tgtdesc->uuid.uuid);
2094                         *genp = tgtdesc->ltd_gen;
2095                 }
2096
2097                 rc = copy_to_user((void *)uarg, buf, len);
2098                 if (rc)
2099                         rc = -EFAULT;
2100                 obd_ioctl_freedata(buf, len);
2101                 break;
2102         }
2103         case LL_IOC_LOV_SETSTRIPE:
2104                 rc = lov_setstripe(exp, karg, uarg);
2105                 break;
2106         case LL_IOC_LOV_GETSTRIPE:
2107                 rc = lov_getstripe(exp, karg, uarg);
2108                 break;
2109         case LL_IOC_LOV_SETEA:
2110                 rc = lov_setea(exp, karg, uarg);
2111                 break;
2112         default: {
2113                 int set = 0;
2114
2115                 if (count == 0)
2116                         RETURN(-ENOTTY);
2117
2118                 rc = 0;
2119                 for (i = 0; i < count; i++) {
2120                         int err;
2121
2122                         /* OST was disconnected */
2123                         if (!lov->tgts[i].ltd_exp)
2124                                 continue;
2125
2126                         err = obd_iocontrol(cmd, lov->tgts[i].ltd_exp,
2127                                             len, karg, uarg);
2128                         if (err == -ENODATA && cmd == OBD_IOC_POLL_QUOTACHECK) {
2129                                 RETURN(err);
2130                         } else if (err) {
2131                                 if (lov->tgts[i].active) {
2132                                         CERROR("error: iocontrol OSC %s on OST "
2133                                                "idx %d cmd %x: err = %d\n",
2134                                                lov->tgts[i].uuid.uuid, i,
2135                                                cmd, err);
2136                                         if (!rc)
2137                                                 rc = err;
2138                                 }
2139                         } else
2140                                 set = 1;
2141                 }
2142                 if (!set && !rc)
2143                         rc = -EIO;
2144         }
2145         }
2146
2147         RETURN(rc);
2148 }
2149
2150 static int lov_get_info(struct obd_export *exp, __u32 keylen,
2151                         void *key, __u32 *vallen, void *val)
2152 {
2153         struct obd_device *obddev = class_exp2obd(exp);
2154         struct lov_obd *lov = &obddev->u.lov;
2155         int i, rc;
2156         ENTRY;
2157
2158         if (!vallen || !val)
2159                 RETURN(-EFAULT);
2160
2161         lov_getref(obddev);
2162
2163         if (keylen > strlen("lock_to_stripe") &&
2164             strcmp(key, "lock_to_stripe") == 0) {
2165                 struct {
2166                         char name[16];
2167                         struct ldlm_lock *lock;
2168                         struct lov_stripe_md *lsm;
2169                 } *data = key;
2170                 struct ldlm_res_id *res_id = &data->lock->l_resource->lr_name;
2171                 struct lov_oinfo *loi;
2172                 __u32 *stripe = val;
2173
2174                 if (*vallen < sizeof(*stripe))
2175                         GOTO(out, rc = -EFAULT);
2176                 *vallen = sizeof(*stripe);
2177
2178                 /* XXX This is another one of those bits that will need to
2179                  * change if we ever actually support nested LOVs.  It uses
2180                  * the lock's export to find out which stripe it is. */
2181                 /* XXX - it's assumed all the locks for deleted OSTs have
2182                  * been cancelled. Also, the export for deleted OSTs will
2183                  * be NULL and won't match the lock's export. */
2184                 for (i = 0, loi = data->lsm->lsm_oinfo;
2185                      i < data->lsm->lsm_stripe_count;
2186                      i++, loi++) {
2187                          if (lov->tgts[loi->loi_ost_idx].ltd_exp ==
2188                              data->lock->l_conn_export &&
2189                              loi->loi_id == res_id->name[0] &&
2190                              loi->loi_gr == res_id->name[2]) {
2191                                 *stripe = i;
2192                                 GOTO(out, rc = 0);
2193                         }
2194                 }
2195                 LDLM_ERROR(data->lock, "lock on inode without such object");
2196                 dump_lsm(D_ERROR, data->lsm);
2197                 GOTO(out, rc = -ENXIO);
2198         } else if (keylen >= strlen("last_id") && strcmp(key, "last_id") == 0) {
2199                 obd_id *ids = val;
2200                 int size = sizeof(obd_id);
2201                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2202                         if (!lov->tgts[i].active)
2203                                 continue;
2204                         rc = obd_get_info(lov->tgts[i].ltd_exp,
2205                                           keylen, key, &size, &(ids[i]));
2206                         if (rc != 0)
2207                                 GOTO(out, rc);
2208                 }
2209                 GOTO(out, rc = 0);
2210         } else if (keylen >= strlen(KEY_LOVDESC) &&
2211                    strcmp(key, KEY_LOVDESC) == 0) {
2212                 struct lov_desc *desc_ret = val;
2213                 *desc_ret = lov->desc;
2214
2215                 GOTO(out, rc = 0);
2216         }
2217
2218         rc = -EINVAL;
2219 out:
2220         lov_putref(obddev);
2221         RETURN(rc);
2222 }
2223
2224 static int lov_set_info_async(struct obd_export *exp, obd_count keylen,
2225                               void *key, obd_count vallen, void *val,
2226                               struct ptlrpc_request_set *set)
2227 {
2228         struct obd_device *obddev = class_exp2obd(exp);
2229         struct lov_obd *lov = &obddev->u.lov;
2230         int i, rc = 0, err;
2231         int no_set = !set;
2232         ENTRY;
2233
2234         if (no_set) {
2235                 set = ptlrpc_prep_set();
2236                 if (!set)
2237                         RETURN(-ENOMEM);
2238         }
2239
2240         if (KEY_IS(KEY_NEXT_ID)) {
2241                 if (vallen > lov->desc.ld_tgt_count)
2242                         RETURN(-EINVAL);
2243                 vallen = sizeof(obd_id);
2244         }
2245
2246         lov_getref(obddev);
2247
2248         if (KEY_IS(KEY_NEXT_ID) || KEY_IS("checksum")) {
2249                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2250                         /* OST was disconnected */
2251                         if (!lov->tgts[i].ltd_exp)
2252                                 continue;
2253
2254                         /* hit all OSCs, even inactive ones */
2255                         err = obd_set_info_async(lov->tgts[i].ltd_exp, keylen,
2256                                                  key, vallen,
2257                                                  ((obd_id*)val) + i, set);
2258                         if (!rc)
2259                                 rc = err;
2260                 }
2261                 GOTO(out, rc);
2262         }
2263
2264         if (KEY_IS("evict_by_nid")) {
2265                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2266                         /* OST was disconnected or is inactive */
2267                         if (!lov->tgts[i].ltd_exp || !lov->tgts[i].active)
2268                                 continue;
2269
2270                         err = obd_set_info_async(lov->tgts[i].ltd_exp, keylen,
2271                                                  key, vallen, val, set);
2272                         if (!rc)
2273                                 rc = err;
2274                 }
2275                 GOTO(out, rc);
2276         }
2277
2278         if (KEY_IS(KEY_MDS_CONN) || KEY_IS("unlinked")) {
2279                 if (vallen != 0)
2280                         GOTO(out, rc = -EINVAL);
2281         } else {
2282                 GOTO(out, rc = -EINVAL);
2283         }
2284
2285         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2286                 if (val && !obd_uuid_equals(val, &lov->tgts[i].uuid))
2287                         continue;
2288
2289                 /* OST was disconnected */
2290                 if (!lov->tgts[i].ltd_exp)
2291                         continue;
2292
2293                 if (!val && !lov->tgts[i].active)
2294                         continue;
2295
2296                 err = obd_set_info_async(lov->tgts[i].ltd_exp,
2297                                          keylen, key, vallen, val, set);
2298                 if (!rc)
2299                         rc = err;
2300         }
2301 out:
2302         lov_putref(obddev);
2303         if (no_set) {
2304                 err = ptlrpc_set_wait(set);
2305                 if (!rc)
2306                         rc = err;
2307                 ptlrpc_set_destroy(set);
2308         }
2309         RETURN(rc);
2310 }
2311
2312 static int lov_checkmd(struct obd_export *exp, struct obd_export *md_exp,
2313                        struct lov_stripe_md *lsm)
2314 {
2315         int rc;
2316         ENTRY;
2317
2318         if (!lsm)
2319                 RETURN(0);
2320         LASSERT(md_exp);
2321         LASSERT(lsm_op_find(lsm->lsm_magic) != NULL);
2322         rc = lsm_op_find(lsm->lsm_magic)->lsm_revalidate(lsm, md_exp->exp_obd);
2323
2324         RETURN(rc);
2325 }
2326
2327 int lov_test_and_clear_async_rc(struct lov_stripe_md *lsm)
2328 {
2329         struct lov_oinfo *loi;
2330         int i, rc = 0;
2331         ENTRY;
2332
2333         for (i = 0, loi = lsm->lsm_oinfo; i < lsm->lsm_stripe_count;
2334              i++, loi++) {
2335                 if (loi->loi_ar.ar_rc && !rc)
2336                         rc = loi->loi_ar.ar_rc;
2337                 loi->loi_ar.ar_rc = 0;
2338         }
2339         RETURN(rc);
2340 }
2341 EXPORT_SYMBOL(lov_test_and_clear_async_rc);
2342
2343 #if 0
2344 struct lov_multi_wait {
2345         struct ldlm_lock *lock;
2346         wait_queue_t      wait;
2347         int               completed;
2348         int               generation;
2349 };
2350
2351 int lov_complete_many(struct obd_export *exp, struct lov_stripe_md *lsm,
2352                       struct lustre_handle *lockh)
2353 {
2354         struct lov_lock_handles *lov_lockh = NULL;
2355         struct lustre_handle *lov_lockhp;
2356         struct lov_obd *lov;
2357         struct lov_oinfo *loi;
2358         struct lov_multi_wait *queues;
2359         int rc = 0, i;
2360         ENTRY;
2361
2362         ASSERT_LSM_MAGIC(lsm);
2363
2364         if (!exp || !exp->exp_obd)
2365                 RETURN(-ENODEV);
2366
2367         LASSERT(lockh != NULL);
2368         if (lsm->lsm_stripe_count > 1) {
2369                 lov_lockh = lov_handle2llh(lockh);
2370                 if (lov_lockh == NULL) {
2371                         CERROR("LOV: invalid lov lock handle %p\n", lockh);
2372                         RETURN(-EINVAL);
2373                 }
2374
2375                 lov_lockhp = lov_lockh->llh_handles;
2376         } else {
2377                 lov_lockhp = lockh;
2378         }
2379
2380         OBD_ALLOC(queues, lsm->lsm_stripe_count * sizeof(*queues));
2381         if (queues == NULL)
2382                 GOTO(out, rc = -ENOMEM);
2383
2384         lov = &exp->exp_obd->u.lov;
2385         for (i = 0, loi = lsm->lsm_oinfo; i < lsm->lsm_stripe_count;
2386              i++, loi++, lov_lockhp++) {
2387                 struct ldlm_lock *lock;
2388                 struct obd_device *obd;
2389                 unsigned long irqflags;
2390
2391                 lock = ldlm_handle2lock(lov_lockhp);
2392                 if (lock == NULL) {
2393                         CDEBUG(D_HA, "lov idx %d subobj "LPX64" no lock?\n",
2394                                loi->loi_ost_idx, loi->loi_id);
2395                         queues[i].completed = 1;
2396                         continue;
2397                 }
2398
2399                 queues[i].lock = lock;
2400                 init_waitqueue_entry(&(queues[i].wait), current);
2401                 add_wait_queue(lock->l_waitq, &(queues[i].wait));
2402
2403                 obd = class_exp2obd(lock->l_conn_export);
2404                 if (obd != NULL)
2405                         imp = obd->u.cli.cl_import;
2406                 if (imp != NULL) {
2407                         spin_lock_irqsave(&imp->imp_lock, irqflags);
2408                         queues[i].generation = imp->imp_generation;
2409                         spin_unlock_irqrestore(&imp->imp_lock, irqflags);
2410                 }
2411         }
2412
2413         lwi = LWI_TIMEOUT_INTR(obd_timeout * HZ, ldlm_expired_completion_wait,
2414                                interrupted_completion_wait, &lwd);
2415         rc = l_wait_event_added(check_multi_complete(queues, lsm), &lwi);
2416
2417         for (i = 0; i < lsm->lsm_stripe_count; i++)
2418                 remove_wait_queue(lock->l_waitq, &(queues[i].wait));
2419
2420         if (rc == -EINTR || rc == -ETIMEDOUT) {
2421
2422
2423         }
2424
2425  out:
2426         if (lov_lockh != NULL)
2427                 lov_llh_put(lov_lockh);
2428         RETURN(rc);
2429 }
2430 #endif
2431
2432 void lov_stripe_lock(struct lov_stripe_md *md)
2433 {
2434         LASSERT(md->lsm_lock_owner != cfs_current());
2435         spin_lock(&md->lsm_lock);
2436         LASSERT(md->lsm_lock_owner == NULL);
2437         md->lsm_lock_owner = cfs_current();
2438 }
2439 EXPORT_SYMBOL(lov_stripe_lock);
2440
2441 void lov_stripe_unlock(struct lov_stripe_md *md)
2442 {
2443         LASSERT(md->lsm_lock_owner == cfs_current());
2444         md->lsm_lock_owner = NULL;
2445         spin_unlock(&md->lsm_lock);
2446 }
2447 EXPORT_SYMBOL(lov_stripe_unlock);
2448
2449 struct obd_ops lov_obd_ops = {
2450         .o_owner               = THIS_MODULE,
2451         .o_setup               = lov_setup,
2452         .o_precleanup          = lov_precleanup,
2453         .o_cleanup             = lov_cleanup,
2454         .o_process_config      = lov_process_config,
2455         .o_connect             = lov_connect,
2456         .o_disconnect          = lov_disconnect,
2457         .o_statfs              = lov_statfs,
2458         .o_packmd              = lov_packmd,
2459         .o_unpackmd            = lov_unpackmd,
2460         .o_checkmd             = lov_checkmd,
2461         .o_create              = lov_create,
2462         .o_destroy             = lov_destroy,
2463         .o_getattr             = lov_getattr,
2464         .o_getattr_async       = lov_getattr_async,
2465         .o_setattr             = lov_setattr,
2466         .o_setattr_async       = lov_setattr_async,
2467         .o_brw                 = lov_brw,
2468         .o_brw_async           = lov_brw_async,
2469         .o_prep_async_page     = lov_prep_async_page,
2470         .o_queue_async_io      = lov_queue_async_io,
2471         .o_set_async_flags     = lov_set_async_flags,
2472         .o_queue_group_io      = lov_queue_group_io,
2473         .o_trigger_group_io    = lov_trigger_group_io,
2474         .o_teardown_async_page = lov_teardown_async_page,
2475         .o_merge_lvb           = lov_merge_lvb,
2476         .o_adjust_kms          = lov_adjust_kms,
2477         .o_punch               = lov_punch,
2478         .o_sync                = lov_sync,
2479         .o_enqueue             = lov_enqueue,
2480         .o_match               = lov_match,
2481         .o_change_cbdata       = lov_change_cbdata,
2482         .o_cancel              = lov_cancel,
2483         .o_cancel_unused       = lov_cancel_unused,
2484         .o_join_lru            = lov_join_lru,
2485         .o_iocontrol           = lov_iocontrol,
2486         .o_get_info            = lov_get_info,
2487         .o_set_info_async      = lov_set_info_async,
2488         .o_llog_init           = lov_llog_init,
2489         .o_llog_finish         = lov_llog_finish,
2490         .o_notify              = lov_notify,
2491 };
2492
2493 static quota_interface_t *quota_interface;
2494 extern quota_interface_t lov_quota_interface;
2495
2496 int __init lov_init(void)
2497 {
2498         struct lprocfs_static_vars lvars;
2499         int rc;
2500         ENTRY;
2501
2502         lprocfs_init_vars(lov, &lvars);
2503
2504         quota_interface = PORTAL_SYMBOL_GET(lov_quota_interface);
2505         init_obd_quota_ops(quota_interface, &lov_obd_ops);
2506
2507         rc = class_register_type(&lov_obd_ops, NULL, lvars.module_vars,
2508                                  LUSTRE_LOV_NAME, NULL);
2509         if (rc && quota_interface)
2510                 PORTAL_SYMBOL_PUT(osc_quota_interface);
2511
2512         RETURN(rc);
2513 }
2514
2515 #ifdef __KERNEL__
2516 static void /*__exit*/ lov_exit(void)
2517 {
2518         if (quota_interface)
2519                 PORTAL_SYMBOL_PUT(lov_quota_interface);
2520
2521         class_unregister_type(LUSTRE_LOV_NAME);
2522 }
2523
2524 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
2525 MODULE_DESCRIPTION("Lustre Logical Object Volume OBD driver");
2526 MODULE_LICENSE("GPL");
2527
2528 cfs_module(lov, "1.0.0", lov_init, lov_exit);
2529 #endif