lustre/lov/lov_obd.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * GPL HEADER START
   5  *
   6  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License version 2 only,
  10  * as published by the Free Software Foundation.
  11  *
  12  * This program is distributed in the hope that it will be useful, but
  13  * WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * General Public License version 2 for more details (a copy is included
  16  * in the LICENSE file that accompanied this code).
  17  *
  18  * You should have received a copy of the GNU General Public License
  19  * version 2 along with this program; If not, see
  20  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  21  *
  22  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  23  * CA 95054 USA or visit www.sun.com if you need additional information or
  24  * have any questions.
  25  *
  26  * GPL HEADER END
  27  */
  28 /*
  29  * Copyright  2008 Sun Microsystems, Inc. All rights reserved
  30  * Use is subject to license terms.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/lov/lov_obd.c
  37  *
  38  * Author: Phil Schwan <phil@clusterfs.com>
  39  * Author: Peter Braam <braam@clusterfs.com>
  40  * Author: Mike Shaver <shaver@clusterfs.com>
  41  * Author: Nathan Rutman <nathan@clusterfs.com>
  42  */
  43
  44 #ifndef EXPORT_SYMTAB
  45 # define EXPORT_SYMTAB
  46 #endif
  47 #define DEBUG_SUBSYSTEM S_LOV
  48 #ifdef __KERNEL__
  49 #include <libcfs/libcfs.h>
  50 #else
  51 #include <liblustre.h>
  52 #endif
  53
  54 #include <obd_support.h>
  55 #include <lustre_lib.h>
  56 #include <lustre_net.h>
  57 #include <lustre/lustre_idl.h>
  58 #include <lustre_dlm.h>
  59 #include <lustre_mds.h>
  60 #include <lustre_debug.h>
  61 #include <obd_class.h>
  62 #include <obd_lov.h>
  63 #include <obd_ost.h>
  64 #include <lprocfs_status.h>
  65 #include <lustre_param.h>
  66 #include <cl_object.h>
  67 #include <lustre/ll_fiemap.h>
  68
  69 #include "lov_internal.h"
  70
  71 /* Keep a refcount of lov->tgt usage to prevent racing with addition/deletion.
  72    Any function that expects lov_tgts to remain stationary must take a ref. */
  73 static void lov_getref(struct obd_device *obd)
  74 {
  75         struct lov_obd *lov = &obd->u.lov;
  76
  77         /* nobody gets through here until lov_putref is done */
  78         mutex_down(&lov->lov_lock);
  79         atomic_inc(&lov->lov_refcount);
  80         mutex_up(&lov->lov_lock);
  81         return;
  82 }
  83
  84 static void __lov_del_obd(struct obd_device *obd, struct lov_tgt_desc *tgt);
  85
  86 static void lov_putref(struct obd_device *obd)
  87 {
  88         struct lov_obd *lov = &obd->u.lov;
  89
  90         mutex_down(&lov->lov_lock);
  91         /* ok to dec to 0 more than once -- ltd_exp's will be null */
  92         if (atomic_dec_and_test(&lov->lov_refcount) && lov->lov_death_row) {
  93                 CFS_LIST_HEAD(kill);
  94                 int i;
  95                 struct lov_tgt_desc *tgt, *n;
  96                 CDEBUG(D_CONFIG, "destroying %d lov targets\n",
  97                        lov->lov_death_row);
  98                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
  99                         tgt = lov->lov_tgts[i];
 100
 101                         if (!tgt || !tgt->ltd_reap)
 102                                 continue;
 103                         list_add(&tgt->ltd_kill, &kill);
 104                         /* XXX - right now there is a dependency on ld_tgt_count
 105                          * being the maximum tgt index for computing the
 106                          * mds_max_easize. So we can't shrink it. */
 107                         lov_ost_pool_remove(&lov->lov_packed, i);
 108                         lov->lov_tgts[i] = NULL;
 109                         lov->lov_death_row--;
 110                 }
 111                 mutex_up(&lov->lov_lock);
 112
 113                 list_for_each_entry_safe(tgt, n, &kill, ltd_kill) {
 114                         list_del(&tgt->ltd_kill);
 115                         /* Disconnect */
 116                         __lov_del_obd(obd, tgt);
 117                 }
 118         } else {
 119                 mutex_up(&lov->lov_lock);
 120         }
 121 }
 122
 123 static int lov_set_osc_active(struct obd_device *obd, struct obd_uuid *uuid,
 124                               int activate);
 125 static int lov_notify(struct obd_device *obd, struct obd_device *watched,
 126                       enum obd_notify_event ev, void *data);
 127
 128
 129 #define MAX_STRING_SIZE 128
 130 int lov_connect_obd(struct obd_device *obd, __u32 index, int activate,
 131                     struct obd_connect_data *data)
 132 {
 133         struct lov_obd *lov = &obd->u.lov;
 134         struct obd_uuid tgt_uuid;
 135         struct obd_device *tgt_obd;
 136         static struct obd_uuid lov_osc_uuid = { "LOV_OSC_UUID" };
 137         struct obd_import *imp;
 138
 139 #ifdef __KERNEL__
 140         cfs_proc_dir_entry_t *lov_proc_dir;
 141 #endif
 142         int rc;
 143         ENTRY;
 144
 145         if (!lov->lov_tgts[index])
 146                 RETURN(-EINVAL);
 147
 148         tgt_uuid = lov->lov_tgts[index]->ltd_uuid;
 149         tgt_obd = lov->lov_tgts[index]->ltd_obd;
 150
 151         if (!tgt_obd->obd_set_up) {
 152                 CERROR("Target %s not set up\n", obd_uuid2str(&tgt_uuid));
 153                 RETURN(-EINVAL);
 154         }
 155
 156         /* override the sp_me from lov */
 157         tgt_obd->u.cli.cl_sp_me = lov->lov_sp_me;
 158
 159         if (data && (data->ocd_connect_flags & OBD_CONNECT_INDEX))
 160                 data->ocd_index = index;
 161
 162         /*
 163          * Divine LOV knows that OBDs under it are OSCs.
 164          */
 165         imp = tgt_obd->u.cli.cl_import;
 166
 167         if (activate) {
 168                 tgt_obd->obd_no_recov = 0;
 169                 /* FIXME this is probably supposed to be
 170                    ptlrpc_set_import_active.  Horrible naming. */
 171                 ptlrpc_activate_import(imp);
 172         }
 173
 174         rc = obd_register_observer(tgt_obd, obd);
 175         if (rc) {
 176                 CERROR("Target %s register_observer error %d\n",
 177                        obd_uuid2str(&tgt_uuid), rc);
 178                 RETURN(rc);
 179         }
 180
 181
 182         if (imp->imp_invalid) {
 183                 CERROR("not connecting OSC %s; administratively "
 184                        "disabled\n", obd_uuid2str(&tgt_uuid));
 185                 RETURN(0);
 186         }
 187
 188         rc = obd_connect(NULL, &lov->lov_tgts[index]->ltd_exp, tgt_obd,
 189                          &lov_osc_uuid, data, NULL);
 190         if (rc || !lov->lov_tgts[index]->ltd_exp) {
 191                 CERROR("Target %s connect error %d\n",
 192                        obd_uuid2str(&tgt_uuid), rc);
 193                 RETURN(-ENODEV);
 194         }
 195
 196         lov->lov_tgts[index]->ltd_reap = 0;
 197
 198         CDEBUG(D_CONFIG, "Connected tgt idx %d %s (%s) %sactive\n", index,
 199                obd_uuid2str(&tgt_uuid), tgt_obd->obd_name, activate ? "":"in");
 200
 201 #ifdef __KERNEL__
 202         lov_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
 203         if (lov_proc_dir) {
 204                 struct obd_device *osc_obd = lov->lov_tgts[index]->ltd_exp->exp_obd;
 205                 cfs_proc_dir_entry_t *osc_symlink;
 206                 char name[MAX_STRING_SIZE];
 207
 208                 LASSERT(osc_obd != NULL);
 209                 LASSERT(osc_obd->obd_magic == OBD_DEVICE_MAGIC);
 210                 LASSERT(osc_obd->obd_type->typ_name != NULL);
 211                 snprintf(name, MAX_STRING_SIZE, "../../../%s/%s",
 212                          osc_obd->obd_type->typ_name,
 213                          osc_obd->obd_name);
 214                 osc_symlink = lprocfs_add_symlink(osc_obd->obd_name, lov_proc_dir,
 215                                                   name);
 216                 if (osc_symlink == NULL) {
 217                         CERROR("could not register LOV target "
 218                                "/proc/fs/lustre/%s/%s/target_obds/%s.",
 219                                obd->obd_type->typ_name, obd->obd_name,
 220                                osc_obd->obd_name);
 221                         lprocfs_remove(&lov_proc_dir);
 222                 }
 223         }
 224 #endif
 225
 226         rc = qos_add_tgt(obd, index);
 227         if (rc)
 228                 CERROR("qos_add_tgt failed %d\n", rc);
 229
 230         RETURN(0);
 231 }
 232
 233 static int lov_connect(const struct lu_env *env,
 234                        struct obd_export **exp, struct obd_device *obd,
 235                        struct obd_uuid *cluuid, struct obd_connect_data *data,
 236                        void *localdata)
 237 {
 238         struct lov_obd *lov = &obd->u.lov;
 239         struct lov_tgt_desc *tgt;
 240         struct lustre_handle conn;
 241         int i, rc;
 242         ENTRY;
 243
 244         CDEBUG(D_CONFIG, "connect #%d\n", lov->lov_connects);
 245
 246         rc = class_connect(&conn, obd, cluuid);
 247         if (rc)
 248                 RETURN(rc);
 249
 250         *exp = class_conn2export(&conn);
 251
 252         /* Why should there ever be more than 1 connect? */
 253         lov->lov_connects++;
 254         LASSERT(lov->lov_connects == 1);
 255
 256         memset(&lov->lov_ocd, 0, sizeof(lov->lov_ocd));
 257         if (data)
 258                 lov->lov_ocd = *data;
 259
 260         obd_getref(obd);
 261         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
 262                 tgt = lov->lov_tgts[i];
 263                 if (!tgt || obd_uuid_empty(&tgt->ltd_uuid))
 264                         continue;
 265                 /* Flags will be lowest common denominator */
 266                 rc = lov_connect_obd(obd, i, tgt->ltd_activate, &lov->lov_ocd);
 267                 if (rc) {
 268                         CERROR("%s: lov connect tgt %d failed: %d\n",
 269                                obd->obd_name, i, rc);
 270                         continue;
 271                 }
 272                 /* connect to administrative disabled ost */
 273                 if (!lov->lov_tgts[i]->ltd_exp)
 274                         continue;
 275
 276                 rc = lov_notify(obd, lov->lov_tgts[i]->ltd_exp->exp_obd,
 277                                 OBD_NOTIFY_CONNECT, (void *)&i);
 278                 if (rc) {
 279                         CERROR("%s error sending notify %d\n",
 280                                obd->obd_name, rc);
 281                 }
 282         }
 283         obd_putref(obd);
 284
 285         RETURN(0);
 286 }
 287
 288 static int lov_disconnect_obd(struct obd_device *obd, struct lov_tgt_desc *tgt)
 289 {
 290         cfs_proc_dir_entry_t *lov_proc_dir;
 291         struct lov_obd *lov = &obd->u.lov;
 292         struct obd_device *osc_obd;
 293         int rc;
 294         ENTRY;
 295
 296         osc_obd = class_exp2obd(tgt->ltd_exp);
 297         CDEBUG(D_CONFIG, "%s: disconnecting target %s\n",
 298                obd->obd_name, osc_obd->obd_name);
 299
 300         if (tgt->ltd_active) {
 301                 tgt->ltd_active = 0;
 302                 lov->desc.ld_active_tgt_count--;
 303                 tgt->ltd_exp->exp_obd->obd_inactive = 1;
 304         }
 305
 306         lov_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
 307         if (lov_proc_dir) {
 308                 cfs_proc_dir_entry_t *osc_symlink;
 309
 310                 osc_symlink = lprocfs_srch(lov_proc_dir, osc_obd->obd_name);
 311                 if (osc_symlink) {
 312                         lprocfs_remove(&osc_symlink);
 313                 } else {
 314                         CERROR("/proc/fs/lustre/%s/%s/target_obds/%s missing.",
 315                                obd->obd_type->typ_name, obd->obd_name,
 316                                osc_obd->obd_name);
 317                 }
 318         }
 319
 320         if (osc_obd) {
 321                 /* Pass it on to our clients.
 322                  * XXX This should be an argument to disconnect,
 323                  * XXX not a back-door flag on the OBD.  Ah well.
 324                  */
 325                 osc_obd->obd_force = obd->obd_force;
 326                 osc_obd->obd_fail = obd->obd_fail;
 327                 osc_obd->obd_no_recov = obd->obd_no_recov;
 328         }
 329
 330         obd_register_observer(osc_obd, NULL);
 331
 332         rc = obd_disconnect(tgt->ltd_exp);
 333         if (rc) {
 334                 CERROR("Target %s disconnect error %d\n",
 335                        tgt->ltd_uuid.uuid, rc);
 336                 rc = 0;
 337         }
 338
 339         qos_del_tgt(obd, tgt);
 340
 341         tgt->ltd_exp = NULL;
 342         RETURN(0);
 343 }
 344
 345 static int lov_disconnect(struct obd_export *exp)
 346 {
 347         struct obd_device *obd = class_exp2obd(exp);
 348         struct lov_obd *lov = &obd->u.lov;
 349         int i, rc;
 350         ENTRY;
 351
 352         if (!lov->lov_tgts)
 353                 goto out;
 354
 355         /* Only disconnect the underlying layers on the final disconnect. */
 356         lov->lov_connects--;
 357         if (lov->lov_connects != 0) {
 358                 /* why should there be more than 1 connect? */
 359                 CERROR("disconnect #%d\n", lov->lov_connects);
 360                 goto out;
 361         }
 362
 363         /* Let's hold another reference so lov_del_obd doesn't spin through
 364            putref every time */
 365         obd_getref(obd);
 366
 367         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
 368                 if (lov->lov_tgts[i] && lov->lov_tgts[i]->ltd_exp) {
 369                         /* Disconnection is the last we know about an obd */
 370                         lov_del_target(obd, i, 0, lov->lov_tgts[i]->ltd_gen);
 371                 }
 372         }
 373         obd_putref(obd);
 374
 375 out:
 376         rc = class_disconnect(exp); /* bz 9811 */
 377         RETURN(rc);
 378 }
 379
 380 /* Error codes:
 381  *
 382  *  -EINVAL  : UUID can't be found in the LOV's target list
 383  *  -ENOTCONN: The UUID is found, but the target connection is bad (!)
 384  *  -EBADF   : The UUID is found, but the OBD is the wrong type (!)
 385  *  any >= 0 : is log target index
 386  */
 387 static int lov_set_osc_active(struct obd_device *obd, struct obd_uuid *uuid,
 388                               int activate)
 389 {
 390         struct lov_obd *lov = &obd->u.lov;
 391         struct lov_tgt_desc *tgt;
 392         int index;
 393         ENTRY;
 394
 395         CDEBUG(D_INFO, "Searching in lov %p for uuid %s (activate=%d)\n",
 396                lov, uuid->uuid, activate);
 397
 398         obd_getref(obd);
 399         for (index = 0; index < lov->desc.ld_tgt_count; index++) {
 400                 tgt = lov->lov_tgts[index];
 401                 if (!tgt || !tgt->ltd_exp)
 402                         continue;
 403
 404                 CDEBUG(D_INFO, "lov idx %d is %s conn "LPX64"\n",
 405                        index, obd_uuid2str(&tgt->ltd_uuid),
 406                        tgt->ltd_exp->exp_handle.h_cookie);
 407                 if (obd_uuid_equals(uuid, &tgt->ltd_uuid))
 408                         break;
 409         }
 410
 411         if (index == lov->desc.ld_tgt_count)
 412                 GOTO(out, index = -EINVAL);
 413
 414         if (lov->lov_tgts[index]->ltd_active == activate) {
 415                 CDEBUG(D_INFO, "OSC %s already %sactive!\n", uuid->uuid,
 416                        activate ? "" : "in");
 417                 GOTO(out, index);
 418         }
 419
 420         CDEBUG(D_CONFIG, "Marking OSC %s %sactive\n", obd_uuid2str(uuid),
 421                activate ? "" : "in");
 422
 423         lov->lov_tgts[index]->ltd_active = activate;
 424
 425         if (activate) {
 426                 lov->desc.ld_active_tgt_count++;
 427                 lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 0;
 428         } else {
 429                 lov->desc.ld_active_tgt_count--;
 430                 lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 1;
 431         }
 432         /* remove any old qos penalty */
 433         lov->lov_tgts[index]->ltd_qos.ltq_penalty = 0;
 434
 435  out:
 436         obd_putref(obd);
 437         RETURN(index);
 438 }
 439
 440 static int lov_notify(struct obd_device *obd, struct obd_device *watched,
 441                       enum obd_notify_event ev, void *data)
 442 {
 443         int rc = 0;
 444         ENTRY;
 445
 446         if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE) {
 447                 struct obd_uuid *uuid;
 448
 449                 LASSERT(watched);
 450
 451                 if (strcmp(watched->obd_type->typ_name, LUSTRE_OSC_NAME)) {
 452                         CERROR("unexpected notification of %s %s!\n",
 453                                watched->obd_type->typ_name,
 454                                watched->obd_name);
 455                         RETURN(-EINVAL);
 456                 }
 457                 uuid = &watched->u.cli.cl_target_uuid;
 458
 459                 /* Set OSC as active before notifying the observer, so the
 460                  * observer can use the OSC normally.
 461                  */
 462                 rc = lov_set_osc_active(obd, uuid, ev == OBD_NOTIFY_ACTIVE);
 463                 if (rc < 0) {
 464                         CERROR("%sactivation of %s failed: %d\n",
 465                                (ev == OBD_NOTIFY_ACTIVE) ? "" : "de",
 466                                obd_uuid2str(uuid), rc);
 467                         RETURN(rc);
 468                 }
 469                 /* active event should be pass lov target index as data */
 470                 data = &rc;
 471         }
 472
 473         /* Pass the notification up the chain. */
 474         if (watched) {
 475                 rc = obd_notify_observer(obd, watched, ev, data);
 476         } else {
 477                 /* NULL watched means all osc's in the lov (only for syncs) */
 478                 /* sync event should be send lov idx as data */
 479                 struct lov_obd *lov = &obd->u.lov;
 480                 struct obd_device *tgt_obd;
 481                 int i;
 482                 obd_getref(obd);
 483                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
 484                         /* don't send sync event if target not
 485                          * connected/activated */
 486                         if (!lov->lov_tgts[i] ||
 487                             !lov->lov_tgts[i]->ltd_active)
 488                                  continue;
 489
 490                         if ((ev == OBD_NOTIFY_SYNC) ||
 491                             (ev == OBD_NOTIFY_SYNC_NONBLOCK))
 492                                 data = &i;
 493
 494                         tgt_obd = class_exp2obd(lov->lov_tgts[i]->ltd_exp);
 495
 496                         rc = obd_notify_observer(obd, tgt_obd, ev, data);
 497                         if (rc) {
 498                                 CERROR("%s: notify %s of %s failed %d\n",
 499                                        obd->obd_name,
 500                                        obd->obd_observer->obd_name,
 501                                        tgt_obd->obd_name, rc);
 502                                 break;
 503                         }
 504                 }
 505                 obd_putref(obd);
 506         }
 507
 508         RETURN(rc);
 509 }
 510
 511 int lov_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
 512                    __u32 index, int gen, int active)
 513 {
 514         struct lov_obd *lov = &obd->u.lov;
 515         struct lov_tgt_desc *tgt;
 516         struct obd_device *tgt_obd;
 517         int rc;
 518         ENTRY;
 519
 520         CDEBUG(D_CONFIG, "uuid:%s idx:%d gen:%d active:%d\n",
 521                uuidp->uuid, index, gen, active);
 522
 523         if (gen <= 0) {
 524                 CERROR("request to add OBD %s with invalid generation: %d\n",
 525                        uuidp->uuid, gen);
 526                 RETURN(-EINVAL);
 527         }
 528
 529         tgt_obd = class_find_client_obd(uuidp, LUSTRE_OSC_NAME,
 530                                         &obd->obd_uuid);
 531         if (tgt_obd == NULL)
 532                 RETURN(-EINVAL);
 533
 534         mutex_down(&lov->lov_lock);
 535
 536         if ((index < lov->lov_tgt_size) && (lov->lov_tgts[index] != NULL)) {
 537                 tgt = lov->lov_tgts[index];
 538                 CERROR("UUID %s already assigned at LOV target index %d\n",
 539                        obd_uuid2str(&tgt->ltd_uuid), index);
 540                 mutex_up(&lov->lov_lock);
 541                 RETURN(-EEXIST);
 542         }
 543
 544         if (index >= lov->lov_tgt_size) {
 545                 /* We need to reallocate the lov target array. */
 546                 struct lov_tgt_desc **newtgts, **old = NULL;
 547                 __u32 newsize, oldsize = 0;
 548
 549                 newsize = max(lov->lov_tgt_size, (__u32)2);
 550                 while (newsize < index + 1)
 551                         newsize = newsize << 1;
 552                 OBD_ALLOC(newtgts, sizeof(*newtgts) * newsize);
 553                 if (newtgts == NULL) {
 554                         mutex_up(&lov->lov_lock);
 555                         RETURN(-ENOMEM);
 556                 }
 557
 558                 if (lov->lov_tgt_size) {
 559                         memcpy(newtgts, lov->lov_tgts, sizeof(*newtgts) *
 560                                lov->lov_tgt_size);
 561                         old = lov->lov_tgts;
 562                         oldsize = lov->lov_tgt_size;
 563                 }
 564
 565                 lov->lov_tgts = newtgts;
 566                 lov->lov_tgt_size = newsize;
 567 #ifdef __KERNEL__
 568                 smp_rmb();
 569 #endif
 570                 if (old)
 571                         OBD_FREE(old, sizeof(*old) * oldsize);
 572
 573                 CDEBUG(D_CONFIG, "tgts: %p size: %d\n",
 574                        lov->lov_tgts, lov->lov_tgt_size);
 575         }
 576
 577         OBD_ALLOC_PTR(tgt);
 578         if (!tgt) {
 579                 mutex_up(&lov->lov_lock);
 580                 RETURN(-ENOMEM);
 581         }
 582
 583         rc = lov_ost_pool_add(&lov->lov_packed, index, lov->lov_tgt_size);
 584         if (rc) {
 585                 mutex_up(&lov->lov_lock);
 586                 OBD_FREE_PTR(tgt);
 587                 RETURN(rc);
 588         }
 589
 590         memset(tgt, 0, sizeof(*tgt));
 591         tgt->ltd_uuid = *uuidp;
 592         tgt->ltd_obd = tgt_obd;
 593         /* XXX - add a sanity check on the generation number. */
 594         tgt->ltd_gen = gen;
 595         tgt->ltd_index = index;
 596         tgt->ltd_activate = active;
 597         lov->lov_tgts[index] = tgt;
 598         if (index >= lov->desc.ld_tgt_count)
 599                 lov->desc.ld_tgt_count = index + 1;
 600
 601         mutex_up(&lov->lov_lock);
 602
 603         CDEBUG(D_CONFIG, "idx=%d ltd_gen=%d ld_tgt_count=%d\n",
 604                 index, tgt->ltd_gen, lov->desc.ld_tgt_count);
 605
 606         if (lov->lov_connects == 0) {
 607                 /* lov_connect hasn't been called yet. We'll do the
 608                    lov_connect_obd on this target when that fn first runs,
 609                    because we don't know the connect flags yet. */
 610                 RETURN(0);
 611         }
 612
 613         obd_getref(obd);
 614
 615         rc = lov_connect_obd(obd, index, active, &lov->lov_ocd);
 616         if (rc)
 617                 GOTO(out, rc);
 618
 619         /* connect to administrative disabled ost */
 620         if (!tgt->ltd_exp)
 621                 GOTO(out, rc = 0);
 622
 623         rc = lov_notify(obd, tgt->ltd_exp->exp_obd,
 624                         active ? OBD_NOTIFY_CONNECT : OBD_NOTIFY_INACTIVE,
 625                         (void *)&index);
 626
 627 out:
 628         if (rc) {
 629                 CERROR("add failed (%d), deleting %s\n", rc,
 630                        obd_uuid2str(&tgt->ltd_uuid));
 631                 lov_del_target(obd, index, 0, 0);
 632         }
 633         obd_putref(obd);
 634         RETURN(rc);
 635 }
 636
 637 /* Schedule a target for deletion */
 638 int lov_del_target(struct obd_device *obd, __u32 index,
 639                    struct obd_uuid *uuidp, int gen)
 640 {
 641         struct lov_obd *lov = &obd->u.lov;
 642         int count = lov->desc.ld_tgt_count;
 643         int rc = 0;
 644         ENTRY;
 645
 646         if (index >= count) {
 647                 CERROR("LOV target index %d >= number of LOV OBDs %d.\n",
 648                        index, count);
 649                 RETURN(-EINVAL);
 650         }
 651
 652         obd_getref(obd);
 653
 654         if (!lov->lov_tgts[index]) {
 655                 CERROR("LOV target at index %d is not setup.\n", index);
 656                 GOTO(out, rc = -EINVAL);
 657         }
 658
 659         if (uuidp && !obd_uuid_equals(uuidp, &lov->lov_tgts[index]->ltd_uuid)) {
 660                 CERROR("LOV target UUID %s at index %d doesn't match %s.\n",
 661                        lov_uuid2str(lov, index), index,
 662                        obd_uuid2str(uuidp));
 663                 GOTO(out, rc = -EINVAL);
 664         }
 665
 666         CDEBUG(D_CONFIG, "uuid: %s idx: %d gen: %d exp: %p active: %d\n",
 667                lov_uuid2str(lov, index), index,
 668                lov->lov_tgts[index]->ltd_gen, lov->lov_tgts[index]->ltd_exp,
 669                lov->lov_tgts[index]->ltd_active);
 670
 671         lov->lov_tgts[index]->ltd_reap = 1;
 672         lov->lov_death_row++;
 673         /* we really delete it from obd_putref */
 674 out:
 675         obd_putref(obd);
 676
 677         RETURN(rc);
 678 }
 679
 680 static void __lov_del_obd(struct obd_device *obd, struct lov_tgt_desc *tgt)
 681 {
 682         struct obd_device *osc_obd;
 683
 684         LASSERT(tgt);
 685         LASSERT(tgt->ltd_reap);
 686
 687         osc_obd = class_exp2obd(tgt->ltd_exp);
 688
 689         CDEBUG(D_CONFIG, "Removing tgt %s : %s\n",
 690                tgt->ltd_uuid.uuid,
 691                osc_obd ? osc_obd->obd_name : "<no obd>");
 692
 693         if (tgt->ltd_exp)
 694                 lov_disconnect_obd(obd, tgt);
 695
 696         OBD_FREE_PTR(tgt);
 697
 698         /* Manual cleanup - no cleanup logs to clean up the osc's.  We must
 699            do it ourselves. And we can't do it from lov_cleanup,
 700            because we just lost our only reference to it. */
 701         if (osc_obd)
 702                 class_manual_cleanup(osc_obd);
 703 }
 704
 705 void lov_fix_desc_stripe_size(__u64 *val)
 706 {
 707         if (*val < PTLRPC_MAX_BRW_SIZE) {
 708                 LCONSOLE_WARN("Increasing default stripe size to min %u\n",
 709                               PTLRPC_MAX_BRW_SIZE);
 710                 *val = PTLRPC_MAX_BRW_SIZE;
 711         } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) {
 712                 *val &= ~(LOV_MIN_STRIPE_SIZE - 1);
 713                 LCONSOLE_WARN("Changing default stripe size to "LPU64" (a "
 714                               "multiple of %u)\n",
 715                               *val, LOV_MIN_STRIPE_SIZE);
 716         }
 717 }
 718
 719 void lov_fix_desc_stripe_count(__u32 *val)
 720 {
 721         if (*val == 0)
 722                 *val = 1;
 723 }
 724
 725 void lov_fix_desc_pattern(__u32 *val)
 726 {
 727         /* from lov_setstripe */
 728         if ((*val != 0) && (*val != LOV_PATTERN_RAID0)) {
 729                 LCONSOLE_WARN("Unknown stripe pattern: %#x\n", *val);
 730                 *val = 0;
 731         }
 732 }
 733
 734 void lov_fix_desc_qos_maxage(__u32 *val)
 735 {
 736         /* fix qos_maxage */
 737         if (*val == 0)
 738                 *val = QOS_DEFAULT_MAXAGE;
 739 }
 740
 741 void lov_fix_desc(struct lov_desc *desc)
 742 {
 743         lov_fix_desc_stripe_size(&desc->ld_default_stripe_size);
 744         lov_fix_desc_stripe_count(&desc->ld_default_stripe_count);
 745         lov_fix_desc_pattern(&desc->ld_pattern);
 746         lov_fix_desc_qos_maxage(&desc->ld_qos_maxage);
 747 }
 748
 749 int lov_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
 750 {
 751         struct lprocfs_static_vars lvars = { 0 };
 752         struct lov_desc *desc;
 753         struct lov_obd *lov = &obd->u.lov;
 754         int count;
 755         int rc;
 756         ENTRY;
 757
 758         if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
 759                 CERROR("LOV setup requires a descriptor\n");
 760                 RETURN(-EINVAL);
 761         }
 762
 763         desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
 764
 765         if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
 766                 CERROR("descriptor size wrong: %d > %d\n",
 767                        (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
 768                 RETURN(-EINVAL);
 769         }
 770
 771         if (desc->ld_magic != LOV_DESC_MAGIC) {
 772                 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
 773                             CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
 774                                    obd->obd_name, desc);
 775                             lustre_swab_lov_desc(desc);
 776                 } else {
 777                         CERROR("%s: Bad lov desc magic: %#x\n",
 778                                obd->obd_name, desc->ld_magic);
 779                         RETURN(-EINVAL);
 780                 }
 781         }
 782
 783         lov_fix_desc(desc);
 784
 785         /* Because of 64-bit divide/mod operations only work with a 32-bit
 786          * divisor in a 32-bit kernel, we cannot support a stripe width
 787          * of 4GB or larger on 32-bit CPUs. */
 788         count = desc->ld_default_stripe_count;
 789         if ((count > 0 ? count : desc->ld_tgt_count) *
 790             desc->ld_default_stripe_size > 0xffffffff) {
 791                 CERROR("LOV: stripe width "LPU64"x%u > 4294967295 bytes\n",
 792                        desc->ld_default_stripe_size, count);
 793                 RETURN(-EINVAL);
 794         }
 795
 796         desc->ld_active_tgt_count = 0;
 797         lov->desc = *desc;
 798         lov->lov_tgt_size = 0;
 799
 800         sema_init(&lov->lov_lock, 1);
 801         atomic_set(&lov->lov_refcount, 0);
 802         CFS_INIT_LIST_HEAD(&lov->lov_qos.lq_oss_list);
 803         init_rwsem(&lov->lov_qos.lq_rw_sem);
 804         lov->lov_sp_me = LUSTRE_SP_CLI;
 805         lov->lov_qos.lq_dirty = 1;
 806         lov->lov_qos.lq_rr.lqr_dirty = 1;
 807         lov->lov_qos.lq_reset = 1;
 808         /* Default priority is toward free space balance */
 809         lov->lov_qos.lq_prio_free = 232;
 810         /* Default threshold for rr (roughly 17%) */
 811         lov->lov_qos.lq_threshold_rr = 43;
 812         /* Init statfs fields */
 813         OBD_ALLOC_PTR(lov->lov_qos.lq_statfs_data);
 814         if (NULL == lov->lov_qos.lq_statfs_data)
 815                 RETURN(-ENOMEM);
 816         cfs_waitq_init(&lov->lov_qos.lq_statfs_waitq);
 817
 818         lov->lov_pools_hash_body = lustre_hash_init("POOLS",
 819                                                     HASH_POOLS_CUR_BITS,
 820                                                     HASH_POOLS_MAX_BITS,
 821                                                     &pool_hash_operations, 0);
 822         CFS_INIT_LIST_HEAD(&lov->lov_pool_list);
 823         lov->lov_pool_count = 0;
 824         rc = lov_ost_pool_init(&lov->lov_packed, 0);
 825         if (rc)
 826                 RETURN(rc);
 827         rc = lov_ost_pool_init(&lov->lov_qos.lq_rr.lqr_pool, 0);
 828         if (rc) {
 829                 lov_ost_pool_free(&lov->lov_packed);
 830                 RETURN(rc);
 831         }
 832
 833         lprocfs_lov_init_vars(&lvars);
 834         lprocfs_obd_setup(obd, lvars.obd_vars);
 835 #ifdef LPROCFS
 836         {
 837                 int rc;
 838
 839                 rc = lprocfs_seq_create(obd->obd_proc_entry, "target_obd",
 840                                         0444, &lov_proc_target_fops, obd);
 841                 if (rc)
 842                         CWARN("Error adding the target_obd file\n");
 843         }
 844 #endif
 845         lov->lov_pool_proc_entry = lprocfs_register("pools",
 846                                                     obd->obd_proc_entry,
 847                                                     NULL, NULL);
 848
 849         RETURN(0);
 850 }
 851
 852 static int lov_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
 853 {
 854         int rc = 0;
 855         struct lov_obd *lov = &obd->u.lov;
 856
 857         ENTRY;
 858
 859         switch (stage) {
 860         case OBD_CLEANUP_EARLY: {
 861                 int i;
 862                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
 863                         if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_active)
 864                                 continue;
 865                         obd_precleanup(class_exp2obd(lov->lov_tgts[i]->ltd_exp),
 866                                        OBD_CLEANUP_EARLY);
 867                 }
 868                 break;
 869         }
 870         case OBD_CLEANUP_EXPORTS:
 871                 rc = obd_llog_finish(obd, 0);
 872                 if (rc != 0)
 873                         CERROR("failed to cleanup llogging subsystems\n");
 874                 break;
 875         }
 876         RETURN(rc);
 877 }
 878
 879 static int lov_cleanup(struct obd_device *obd)
 880 {
 881         struct lov_obd *lov = &obd->u.lov;
 882         struct list_head *pos, *tmp;
 883         struct pool_desc *pool;
 884
 885         list_for_each_safe(pos, tmp, &lov->lov_pool_list) {
 886                 pool = list_entry(pos, struct pool_desc, pool_list);
 887                 /* free pool structs */
 888                 CDEBUG(D_INFO, "delete pool %p\n", pool);
 889                 lov_pool_del(obd, pool->pool_name);
 890         }
 891         lustre_hash_exit(lov->lov_pools_hash_body);
 892         lov_ost_pool_free(&(lov->lov_qos.lq_rr.lqr_pool));
 893         lov_ost_pool_free(&lov->lov_packed);
 894
 895         if (lov->lov_tgts) {
 896                 int i;
 897                 obd_getref(obd);
 898                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
 899                         if (!lov->lov_tgts[i])
 900                                 continue;
 901
 902                         /* Inactive targets may never have connected */
 903                         if (lov->lov_tgts[i]->ltd_active ||
 904                             atomic_read(&lov->lov_refcount))
 905                             /* We should never get here - these
 906                                should have been removed in the
 907                              disconnect. */
 908                                 CERROR("lov tgt %d not cleaned!"
 909                                        " deathrow=%d, lovrc=%d\n",
 910                                        i, lov->lov_death_row,
 911                                        atomic_read(&lov->lov_refcount));
 912                         lov_del_target(obd, i, 0, 0);
 913                 }
 914                 obd_putref(obd);
 915                 OBD_FREE(lov->lov_tgts, sizeof(*lov->lov_tgts) *
 916                          lov->lov_tgt_size);
 917                 lov->lov_tgt_size = 0;
 918         }
 919
 920         /* clear pools parent proc entry only after all pools is killed */
 921         lprocfs_obd_cleanup(obd);
 922
 923         OBD_FREE_PTR(lov->lov_qos.lq_statfs_data);
 924         RETURN(0);
 925 }
 926
 927 int lov_process_config_base(struct obd_device *obd, struct lustre_cfg *lcfg,
 928                             __u32 *indexp, int *genp)
 929 {
 930         struct obd_uuid obd_uuid;
 931         int cmd;
 932         int rc = 0;
 933         ENTRY;
 934
 935         switch(cmd = lcfg->lcfg_command) {
 936         case LCFG_LOV_ADD_OBD:
 937         case LCFG_LOV_ADD_INA:
 938         case LCFG_LOV_DEL_OBD: {
 939                 __u32 index;
 940                 int gen;
 941                 /* lov_modify_tgts add  0:lov_mdsA  1:ost1_UUID  2:0  3:1 */
 942                 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
 943                         GOTO(out, rc = -EINVAL);
 944
 945                 obd_str2uuid(&obd_uuid,  lustre_cfg_buf(lcfg, 1));
 946
 947                 if (sscanf(lustre_cfg_buf(lcfg, 2), "%d", indexp) != 1)
 948                         GOTO(out, rc = -EINVAL);
 949                 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", genp) != 1)
 950                         GOTO(out, rc = -EINVAL);
 951                 index = *indexp;
 952                 gen = *genp;
 953                 if (cmd == LCFG_LOV_ADD_OBD)
 954                         rc = lov_add_target(obd, &obd_uuid, index, gen, 1);
 955                 else if (cmd == LCFG_LOV_ADD_INA)
 956                         rc = lov_add_target(obd, &obd_uuid, index, gen, 0);
 957                 else
 958                         rc = lov_del_target(obd, index, &obd_uuid, gen);
 959                 GOTO(out, rc);
 960         }
 961         case LCFG_PARAM: {
 962                 struct lprocfs_static_vars lvars = { 0 };
 963                 struct lov_desc *desc = &(obd->u.lov.desc);
 964
 965                 if (!desc)
 966                         GOTO(out, rc = -EINVAL);
 967
 968                 lprocfs_lov_init_vars(&lvars);
 969
 970                 rc = class_process_proc_param(PARAM_LOV, lvars.obd_vars,
 971                                               lcfg, obd);
 972                 if (rc > 0)
 973                         rc = 0;
 974                 GOTO(out, rc);
 975         }
 976         case LCFG_POOL_NEW:
 977         case LCFG_POOL_ADD:
 978         case LCFG_POOL_DEL:
 979         case LCFG_POOL_REM:
 980                 GOTO(out, rc);
 981
 982         default: {
 983                 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
 984                 GOTO(out, rc = -EINVAL);
 985
 986         }
 987         }
 988 out:
 989         RETURN(rc);
 990 }
 991
 992 #ifndef log2
 993 #define log2(n) ffz(~(n))
 994 #endif
 995
 996 static int lov_clear_orphans(struct obd_export *export, struct obdo *src_oa,
 997                              struct lov_stripe_md **ea,
 998                              struct obd_trans_info *oti)
 999 {
1000         struct lov_obd *lov;
1001         struct obdo *tmp_oa;
1002         struct obd_uuid *ost_uuid = NULL;
1003         int rc = 0, i;
1004         ENTRY;
1005
1006         LASSERT(src_oa->o_valid & OBD_MD_FLFLAGS &&
1007                 src_oa->o_flags == OBD_FL_DELORPHAN);
1008
1009         lov = &export->exp_obd->u.lov;
1010
1011         OBDO_ALLOC(tmp_oa);
1012         if (tmp_oa == NULL)
1013                 RETURN(-ENOMEM);
1014
1015         if (oti->oti_ost_uuid) {
1016                 ost_uuid = oti->oti_ost_uuid;
1017                 CDEBUG(D_HA, "clearing orphans only for %s\n",
1018                        ost_uuid->uuid);
1019         }
1020
1021         obd_getref(export->exp_obd);
1022         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
1023                 struct lov_stripe_md obj_md;
1024                 struct lov_stripe_md *obj_mdp = &obj_md;
1025                 struct lov_tgt_desc *tgt;
1026                 int err;
1027
1028                 tgt = lov->lov_tgts[i];
1029                 if (!tgt)
1030                         continue;
1031
1032                 /* if called for a specific target, we don't
1033                    care if it is not active. */
1034                 if (!lov->lov_tgts[i]->ltd_active && ost_uuid == NULL) {
1035                         CDEBUG(D_HA, "lov idx %d inactive\n", i);
1036                         continue;
1037                 }
1038
1039                 if (ost_uuid && !obd_uuid_equals(ost_uuid, &tgt->ltd_uuid))
1040                         continue;
1041
1042                 CDEBUG(D_CONFIG,"Clear orphans for %d:%s\n", i,
1043                        obd_uuid2str(ost_uuid));
1044
1045                 memcpy(tmp_oa, src_oa, sizeof(*tmp_oa));
1046
1047                 LASSERT(lov->lov_tgts[i]->ltd_exp);
1048                 /* XXX: LOV STACKING: use real "obj_mdp" sub-data */
1049                 err = obd_create(lov->lov_tgts[i]->ltd_exp,
1050                                  tmp_oa, &obj_mdp, oti);
1051                 if (err) {
1052                         /* This export will be disabled until it is recovered,
1053                            and then orphan recovery will be completed. */
1054                         CERROR("error in orphan recovery on OST idx %d/%d: "
1055                                "rc = %d\n", i, lov->desc.ld_tgt_count, err);
1056                         rc = err;
1057                 }
1058
1059                 if (ost_uuid)
1060                         break;
1061         }
1062         obd_putref(export->exp_obd);
1063
1064         OBDO_FREE(tmp_oa);
1065         RETURN(rc);
1066 }
1067
1068 static int lov_recreate(struct obd_export *exp, struct obdo *src_oa,
1069                         struct lov_stripe_md **ea, struct obd_trans_info *oti)
1070 {
1071         struct lov_stripe_md *obj_mdp, *lsm;
1072         struct lov_obd *lov = &exp->exp_obd->u.lov;
1073         unsigned ost_idx;
1074         int rc, i;
1075         ENTRY;
1076
1077         LASSERT(src_oa->o_valid & OBD_MD_FLFLAGS &&
1078                 src_oa->o_flags & OBD_FL_RECREATE_OBJS);
1079
1080         OBD_ALLOC(obj_mdp, sizeof(*obj_mdp));
1081         if (obj_mdp == NULL)
1082                 RETURN(-ENOMEM);
1083
1084         ost_idx = src_oa->o_nlink;
1085         lsm = *ea;
1086         if (lsm == NULL)
1087                 GOTO(out, rc = -EINVAL);
1088         if (ost_idx >= lov->desc.ld_tgt_count ||
1089             !lov->lov_tgts[ost_idx])
1090                 GOTO(out, rc = -EINVAL);
1091
1092         for (i = 0; i < lsm->lsm_stripe_count; i++) {
1093                 if (lsm->lsm_oinfo[i]->loi_ost_idx == ost_idx) {
1094                         if (lsm->lsm_oinfo[i]->loi_id != src_oa->o_id)
1095                                 GOTO(out, rc = -EINVAL);
1096                         break;
1097                 }
1098         }
1099         if (i == lsm->lsm_stripe_count)
1100                 GOTO(out, rc = -EINVAL);
1101
1102         rc = obd_create(lov->lov_tgts[ost_idx]->ltd_exp, src_oa, &obj_mdp, oti);
1103 out:
1104         OBD_FREE(obj_mdp, sizeof(*obj_mdp));
1105         RETURN(rc);
1106 }
1107
1108 /* the LOV expects oa->o_id to be set to the LOV object id */
1109 static int lov_create(struct obd_export *exp, struct obdo *src_oa,
1110                       struct lov_stripe_md **ea, struct obd_trans_info *oti)
1111 {
1112         struct lov_obd *lov;
1113         struct obd_info oinfo;
1114         struct lov_request_set *set = NULL;
1115         struct lov_request *req;
1116         int rc = 0;
1117         ENTRY;
1118
1119         LASSERT(ea != NULL);
1120         if (exp == NULL)
1121                 RETURN(-EINVAL);
1122
1123         if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
1124             src_oa->o_flags == OBD_FL_DELORPHAN) {
1125                 rc = lov_clear_orphans(exp, src_oa, ea, oti);
1126                 RETURN(rc);
1127         }
1128
1129         lov = &exp->exp_obd->u.lov;
1130         if (!lov->desc.ld_active_tgt_count)
1131                 RETURN(-EIO);
1132
1133         obd_getref(exp->exp_obd);
1134         /* Recreate a specific object id at the given OST index */
1135         if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
1136             (src_oa->o_flags & OBD_FL_RECREATE_OBJS)) {
1137                  rc = lov_recreate(exp, src_oa, ea, oti);
1138                  GOTO(out, rc);
1139         }
1140
1141         /* issue statfs rpcs if the osfs data is older than qos_maxage - 1s,
1142          * later in alloc_qos(), we will wait for those rpcs to complete if
1143          * the osfs age is older than 2 * qos_maxage */
1144         qos_statfs_update(exp->exp_obd,
1145                           cfs_time_shift_64(-lov->desc.ld_qos_maxage) + HZ, 0);
1146
1147         rc = lov_prep_create_set(exp, &oinfo, ea, src_oa, oti, &set);
1148         if (rc)
1149                 GOTO(out, rc);
1150
1151         list_for_each_entry(req, &set->set_list, rq_link) {
1152                 /* XXX: LOV STACKING: use real "obj_mdp" sub-data */
1153                 rc = obd_create(lov->lov_tgts[req->rq_idx]->ltd_exp,
1154                                 req->rq_oi.oi_oa, &req->rq_oi.oi_md, oti);
1155                 lov_update_create_set(set, req, rc);
1156         }
1157         rc = lov_fini_create_set(set, ea);
1158 out:
1159         obd_putref(exp->exp_obd);
1160         RETURN(rc);
1161 }
1162
1163 #define ASSERT_LSM_MAGIC(lsmp)                                                  \
1164 do {                                                                            \
1165         LASSERT((lsmp) != NULL);                                                \
1166         LASSERTF(((lsmp)->lsm_magic == LOV_MAGIC_V1 ||                          \
1167                  (lsmp)->lsm_magic == LOV_MAGIC_V3 ||                           \
1168                  (lsmp)->lsm_magic == LOV_MAGIC_JOIN), "%p->lsm_magic=%x\n",    \
1169                  (lsmp), (lsmp)->lsm_magic);                                    \
1170 } while (0)
1171
1172 static int lov_destroy(struct obd_export *exp, struct obdo *oa,
1173                        struct lov_stripe_md *lsm, struct obd_trans_info *oti,
1174                        struct obd_export *md_exp, void *capa)
1175 {
1176         struct lov_request_set *set;
1177         struct obd_info oinfo;
1178         struct lov_request *req;
1179         struct list_head *pos;
1180         struct lov_obd *lov;
1181         int rc = 0, err = 0;
1182         ENTRY;
1183
1184         ASSERT_LSM_MAGIC(lsm);
1185
1186         if (!exp || !exp->exp_obd)
1187                 RETURN(-ENODEV);
1188
1189         if (oa->o_valid & OBD_MD_FLCOOKIE) {
1190                 LASSERT(oti);
1191                 LASSERT(oti->oti_logcookies);
1192         }
1193
1194         lov = &exp->exp_obd->u.lov;
1195         obd_getref(exp->exp_obd);
1196         rc = lov_prep_destroy_set(exp, &oinfo, oa, lsm, oti, &set);
1197         if (rc)
1198                 GOTO(out, rc);
1199
1200         list_for_each (pos, &set->set_list) {
1201                 req = list_entry(pos, struct lov_request, rq_link);
1202
1203                 if (oa->o_valid & OBD_MD_FLCOOKIE)
1204                         oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1205
1206                 err = obd_destroy(lov->lov_tgts[req->rq_idx]->ltd_exp,
1207                                   req->rq_oi.oi_oa, NULL, oti, NULL, capa);
1208                 err = lov_update_common_set(set, req, err);
1209                 if (err) {
1210                         CERROR("error: destroying objid "LPX64" subobj "
1211                                LPX64" on OST idx %d: rc = %d\n",
1212                                oa->o_id, req->rq_oi.oi_oa->o_id,
1213                                req->rq_idx, err);
1214                         if (!rc)
1215                                 rc = err;
1216                 }
1217         }
1218
1219         if (rc == 0) {
1220                 LASSERT(lsm_op_find(lsm->lsm_magic) != NULL);
1221                 rc = lsm_op_find(lsm->lsm_magic)->lsm_destroy(lsm, oa, md_exp);
1222         }
1223         err = lov_fini_destroy_set(set);
1224 out:
1225         obd_putref(exp->exp_obd);
1226         RETURN(rc ? rc : err);
1227 }
1228
1229 static int lov_getattr(struct obd_export *exp, struct obd_info *oinfo)
1230 {
1231         struct lov_request_set *set;
1232         struct lov_request *req;
1233         struct list_head *pos;
1234         struct lov_obd *lov;
1235         int err = 0, rc = 0;
1236         ENTRY;
1237
1238         LASSERT(oinfo);
1239         ASSERT_LSM_MAGIC(oinfo->oi_md);
1240
1241         if (!exp || !exp->exp_obd)
1242                 RETURN(-ENODEV);
1243
1244         lov = &exp->exp_obd->u.lov;
1245
1246         rc = lov_prep_getattr_set(exp, oinfo, &set);
1247         if (rc)
1248                 RETURN(rc);
1249
1250         list_for_each (pos, &set->set_list) {
1251                 req = list_entry(pos, struct lov_request, rq_link);
1252
1253                 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
1254                        "%u\n", oinfo->oi_oa->o_id, req->rq_stripe,
1255                        req->rq_oi.oi_oa->o_id, req->rq_idx);
1256
1257                 rc = obd_getattr(lov->lov_tgts[req->rq_idx]->ltd_exp,
1258                                  &req->rq_oi);
1259                 err = lov_update_common_set(set, req, rc);
1260                 if (err) {
1261                         CERROR("error: getattr objid "LPX64" subobj "
1262                                LPX64" on OST idx %d: rc = %d\n",
1263                                oinfo->oi_oa->o_id, req->rq_oi.oi_oa->o_id,
1264                                req->rq_idx, err);
1265                         break;
1266                 }
1267         }
1268
1269         rc = lov_fini_getattr_set(set);
1270         if (err)
1271                 rc = err;
1272         RETURN(rc);
1273 }
1274
1275 static int lov_getattr_interpret(struct ptlrpc_request_set *rqset,
1276                                  void *data, int rc)
1277 {
1278         struct lov_request_set *lovset = (struct lov_request_set *)data;
1279         int err;
1280         ENTRY;
1281
1282         /* don't do attribute merge if this aysnc op failed */
1283         if (rc)
1284                 lovset->set_completes = 0;
1285         err = lov_fini_getattr_set(lovset);
1286         RETURN(rc ? rc : err);
1287 }
1288
1289 static int lov_getattr_async(struct obd_export *exp, struct obd_info *oinfo,
1290                               struct ptlrpc_request_set *rqset)
1291 {
1292         struct lov_request_set *lovset;
1293         struct lov_obd *lov;
1294         struct list_head *pos;
1295         struct lov_request *req;
1296         int rc = 0, err;
1297         ENTRY;
1298
1299         LASSERT(oinfo);
1300         ASSERT_LSM_MAGIC(oinfo->oi_md);
1301
1302         if (!exp || !exp->exp_obd)
1303                 RETURN(-ENODEV);
1304
1305         lov = &exp->exp_obd->u.lov;
1306
1307         rc = lov_prep_getattr_set(exp, oinfo, &lovset);
1308         if (rc)
1309                 RETURN(rc);
1310
1311         CDEBUG(D_INFO, "objid "LPX64": %ux%u byte stripes\n",
1312                oinfo->oi_md->lsm_object_id, oinfo->oi_md->lsm_stripe_count,
1313                oinfo->oi_md->lsm_stripe_size);
1314
1315         list_for_each (pos, &lovset->set_list) {
1316                 req = list_entry(pos, struct lov_request, rq_link);
1317
1318                 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
1319                        "%u\n", oinfo->oi_oa->o_id, req->rq_stripe,
1320                        req->rq_oi.oi_oa->o_id, req->rq_idx);
1321                 rc = obd_getattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1322                                        &req->rq_oi, rqset);
1323                 if (rc) {
1324                         CERROR("error: getattr objid "LPX64" subobj "
1325                                LPX64" on OST idx %d: rc = %d\n",
1326                                oinfo->oi_oa->o_id, req->rq_oi.oi_oa->o_id,
1327                                req->rq_idx, rc);
1328                         GOTO(out, rc);
1329                 }
1330         }
1331
1332         if (!list_empty(&rqset->set_requests)) {
1333                 LASSERT(rc == 0);
1334                 LASSERT (rqset->set_interpret == NULL);
1335                 rqset->set_interpret = lov_getattr_interpret;
1336                 rqset->set_arg = (void *)lovset;
1337                 RETURN(rc);
1338         }
1339 out:
1340         if (rc)
1341                 lovset->set_completes = 0;
1342         err = lov_fini_getattr_set(lovset);
1343         RETURN(rc ? rc : err);
1344 }
1345
1346 static int lov_setattr(struct obd_export *exp, struct obd_info *oinfo,
1347                        struct obd_trans_info *oti)
1348 {
1349         struct lov_request_set *set;
1350         struct lov_obd *lov;
1351         struct list_head *pos;
1352         struct lov_request *req;
1353         int err = 0, rc = 0;
1354         ENTRY;
1355
1356         LASSERT(oinfo);
1357         ASSERT_LSM_MAGIC(oinfo->oi_md);
1358
1359         if (!exp || !exp->exp_obd)
1360                 RETURN(-ENODEV);
1361
1362         /* for now, we only expect the following updates here */
1363         LASSERT(!(oinfo->oi_oa->o_valid & ~(OBD_MD_FLID | OBD_MD_FLTYPE |
1364                                             OBD_MD_FLMODE | OBD_MD_FLATIME |
1365                                             OBD_MD_FLMTIME | OBD_MD_FLCTIME |
1366                                             OBD_MD_FLFLAGS | OBD_MD_FLSIZE |
1367                                             OBD_MD_FLGROUP | OBD_MD_FLUID |
1368                                             OBD_MD_FLGID | OBD_MD_FLFID |
1369                                             OBD_MD_FLGENER)));
1370         lov = &exp->exp_obd->u.lov;
1371         rc = lov_prep_setattr_set(exp, oinfo, oti, &set);
1372         if (rc)
1373                 RETURN(rc);
1374
1375         list_for_each (pos, &set->set_list) {
1376                 req = list_entry(pos, struct lov_request, rq_link);
1377
1378                 rc = obd_setattr(lov->lov_tgts[req->rq_idx]->ltd_exp,
1379                                  &req->rq_oi, NULL);
1380                 err = lov_update_setattr_set(set, req, rc);
1381                 if (err) {
1382                         CERROR("error: setattr objid "LPX64" subobj "
1383                                LPX64" on OST idx %d: rc = %d\n",
1384                                set->set_oi->oi_oa->o_id,
1385                                req->rq_oi.oi_oa->o_id, req->rq_idx, err);
1386                         if (!rc)
1387                                 rc = err;
1388                 }
1389         }
1390         err = lov_fini_setattr_set(set);
1391         if (!rc)
1392                 rc = err;
1393         RETURN(rc);
1394 }
1395
1396 static int lov_setattr_interpret(struct ptlrpc_request_set *rqset,
1397                                  void *data, int rc)
1398 {
1399         struct lov_request_set *lovset = (struct lov_request_set *)data;
1400         int err;
1401         ENTRY;
1402
1403         if (rc)
1404                 lovset->set_completes = 0;
1405         err = lov_fini_setattr_set(lovset);
1406         RETURN(rc ? rc : err);
1407 }
1408
1409 /* If @oti is given, the request goes from MDS and responses from OSTs are not
1410    needed. Otherwise, a client is waiting for responses. */
1411 static int lov_setattr_async(struct obd_export *exp, struct obd_info *oinfo,
1412                              struct obd_trans_info *oti,
1413                              struct ptlrpc_request_set *rqset)
1414 {
1415         struct lov_request_set *set;
1416         struct lov_request *req;
1417         struct list_head *pos;
1418         struct lov_obd *lov;
1419         int rc = 0;
1420         ENTRY;
1421
1422         LASSERT(oinfo);
1423         ASSERT_LSM_MAGIC(oinfo->oi_md);
1424         if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE) {
1425                 LASSERT(oti);
1426                 LASSERT(oti->oti_logcookies);
1427         }
1428
1429         if (!exp || !exp->exp_obd)
1430                 RETURN(-ENODEV);
1431
1432         lov = &exp->exp_obd->u.lov;
1433         rc = lov_prep_setattr_set(exp, oinfo, oti, &set);
1434         if (rc)
1435                 RETURN(rc);
1436
1437         CDEBUG(D_INFO, "objid "LPX64": %ux%u byte stripes\n",
1438                oinfo->oi_md->lsm_object_id, oinfo->oi_md->lsm_stripe_count,
1439                oinfo->oi_md->lsm_stripe_size);
1440
1441         list_for_each (pos, &set->set_list) {
1442                 req = list_entry(pos, struct lov_request, rq_link);
1443
1444                 if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE)
1445                         oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1446
1447                 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
1448                        "%u\n", oinfo->oi_oa->o_id, req->rq_stripe,
1449                        req->rq_oi.oi_oa->o_id, req->rq_idx);
1450
1451                 rc = obd_setattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1452                                        &req->rq_oi, oti, rqset);
1453                 if (rc) {
1454                         CERROR("error: setattr objid "LPX64" subobj "
1455                                LPX64" on OST idx %d: rc = %d\n",
1456                                set->set_oi->oi_oa->o_id,
1457                                req->rq_oi.oi_oa->o_id,
1458                                req->rq_idx, rc);
1459                         break;
1460                 }
1461         }
1462
1463         /* If we are not waiting for responses on async requests, return. */
1464         if (rc || !rqset || list_empty(&rqset->set_requests)) {
1465                 int err;
1466                 if (rc)
1467                         set->set_completes = 0;
1468                 err = lov_fini_setattr_set(set);
1469                 RETURN(rc ? rc : err);
1470         }
1471
1472         LASSERT(rqset->set_interpret == NULL);
1473         rqset->set_interpret = lov_setattr_interpret;
1474         rqset->set_arg = (void *)set;
1475
1476         RETURN(0);
1477 }
1478
1479 static int lov_punch_interpret(struct ptlrpc_request_set *rqset,
1480                                void *data, int rc)
1481 {
1482         struct lov_request_set *lovset = (struct lov_request_set *)data;
1483         int err;
1484         ENTRY;
1485
1486         if (rc)
1487                 lovset->set_completes = 0;
1488         err = lov_fini_punch_set(lovset);
1489         RETURN(rc ? rc : err);
1490 }
1491
1492 /* FIXME: maybe we'll just make one node the authoritative attribute node, then
1493  * we can send this 'punch' to just the authoritative node and the nodes
1494  * that the punch will affect. */
1495 static int lov_punch(struct obd_export *exp, struct obd_info *oinfo,
1496                      struct obd_trans_info *oti,
1497                      struct ptlrpc_request_set *rqset)
1498 {
1499         struct lov_request_set *set;
1500         struct lov_obd *lov;
1501         struct list_head *pos;
1502         struct lov_request *req;
1503         int rc = 0;
1504         ENTRY;
1505
1506         LASSERT(oinfo);
1507         ASSERT_LSM_MAGIC(oinfo->oi_md);
1508
1509         if (!exp || !exp->exp_obd)
1510                 RETURN(-ENODEV);
1511
1512         lov = &exp->exp_obd->u.lov;
1513         rc = lov_prep_punch_set(exp, oinfo, oti, &set);
1514         if (rc)
1515                 RETURN(rc);
1516
1517         list_for_each (pos, &set->set_list) {
1518                 req = list_entry(pos, struct lov_request, rq_link);
1519
1520                 rc = obd_punch(lov->lov_tgts[req->rq_idx]->ltd_exp,
1521                                &req->rq_oi, NULL, rqset);
1522                 if (rc) {
1523                         CERROR("error: punch objid "LPX64" subobj "LPX64
1524                                " on OST idx %d: rc = %d\n",
1525                                set->set_oi->oi_oa->o_id,
1526                                req->rq_oi.oi_oa->o_id, req->rq_idx, rc);
1527                         break;
1528                 }
1529         }
1530
1531         if (rc || list_empty(&rqset->set_requests)) {
1532                 int err;
1533                 err = lov_fini_punch_set(set);
1534                 RETURN(rc ? rc : err);
1535         }
1536
1537         LASSERT(rqset->set_interpret == NULL);
1538         rqset->set_interpret = lov_punch_interpret;
1539         rqset->set_arg = (void *)set;
1540
1541         RETURN(0);
1542 }
1543
1544 static int lov_sync(struct obd_export *exp, struct obdo *oa,
1545                     struct lov_stripe_md *lsm, obd_off start, obd_off end,
1546                     void *capa)
1547 {
1548         struct lov_request_set *set;
1549         struct obd_info oinfo;
1550         struct lov_obd *lov;
1551         struct list_head *pos;
1552         struct lov_request *req;
1553         int err = 0, rc = 0;
1554         ENTRY;
1555
1556         ASSERT_LSM_MAGIC(lsm);
1557
1558         if (!exp->exp_obd)
1559                 RETURN(-ENODEV);
1560
1561         lov = &exp->exp_obd->u.lov;
1562         rc = lov_prep_sync_set(exp, &oinfo, oa, lsm, start, end, &set);
1563         if (rc)
1564                 RETURN(rc);
1565
1566         list_for_each (pos, &set->set_list) {
1567                 req = list_entry(pos, struct lov_request, rq_link);
1568
1569                 rc = obd_sync(lov->lov_tgts[req->rq_idx]->ltd_exp,
1570                               req->rq_oi.oi_oa, NULL,
1571                               req->rq_oi.oi_policy.l_extent.start,
1572                               req->rq_oi.oi_policy.l_extent.end, capa);
1573                 err = lov_update_common_set(set, req, rc);
1574                 if (err) {
1575                         CERROR("error: fsync objid "LPX64" subobj "LPX64
1576                                " on OST idx %d: rc = %d\n",
1577                                set->set_oi->oi_oa->o_id,
1578                                req->rq_oi.oi_oa->o_id, req->rq_idx, rc);
1579                         if (!rc)
1580                                 rc = err;
1581                 }
1582         }
1583         err = lov_fini_sync_set(set);
1584         if (!rc)
1585                 rc = err;
1586         RETURN(rc);
1587 }
1588
1589 static int lov_brw_check(struct lov_obd *lov, struct obd_info *lov_oinfo,
1590                          obd_count oa_bufs, struct brw_page *pga)
1591 {
1592         struct obd_info oinfo = { { { 0 } } };
1593         int i, rc = 0;
1594
1595         oinfo.oi_oa = lov_oinfo->oi_oa;
1596
1597         /* The caller just wants to know if there's a chance that this
1598          * I/O can succeed */
1599         for (i = 0; i < oa_bufs; i++) {
1600                 int stripe = lov_stripe_number(lov_oinfo->oi_md, pga[i].off);
1601                 int ost = lov_oinfo->oi_md->lsm_oinfo[stripe]->loi_ost_idx;
1602                 obd_off start, end;
1603
1604                 if (!lov_stripe_intersects(lov_oinfo->oi_md, i, pga[i].off,
1605                                            pga[i].off + pga[i].count - 1,
1606                                            &start, &end))
1607                         continue;
1608
1609                 if (!lov->lov_tgts[ost] || !lov->lov_tgts[ost]->ltd_active) {
1610                         CDEBUG(D_HA, "lov idx %d inactive\n", ost);
1611                         return -EIO;
1612                 }
1613
1614                 rc = obd_brw(OBD_BRW_CHECK, lov->lov_tgts[ost]->ltd_exp, &oinfo,
1615                              1, &pga[i], NULL);
1616                 if (rc)
1617                         break;
1618         }
1619         return rc;
1620 }
1621
1622 static int lov_brw(int cmd, struct obd_export *exp, struct obd_info *oinfo,
1623                    obd_count oa_bufs, struct brw_page *pga,
1624                    struct obd_trans_info *oti)
1625 {
1626         struct lov_request_set *set;
1627         struct lov_request *req;
1628         struct list_head *pos;
1629         struct lov_obd *lov = &exp->exp_obd->u.lov;
1630         int err, rc = 0;
1631         ENTRY;
1632
1633         ASSERT_LSM_MAGIC(oinfo->oi_md);
1634
1635         if (cmd == OBD_BRW_CHECK) {
1636                 rc = lov_brw_check(lov, oinfo, oa_bufs, pga);
1637                 RETURN(rc);
1638         }
1639
1640         rc = lov_prep_brw_set(exp, oinfo, oa_bufs, pga, oti, &set);
1641         if (rc)
1642                 RETURN(rc);
1643
1644         list_for_each (pos, &set->set_list) {
1645                 struct obd_export *sub_exp;
1646                 struct brw_page *sub_pga;
1647                 req = list_entry(pos, struct lov_request, rq_link);
1648
1649                 sub_exp = lov->lov_tgts[req->rq_idx]->ltd_exp;
1650                 sub_pga = set->set_pga + req->rq_pgaidx;
1651                 rc = obd_brw(cmd, sub_exp, &req->rq_oi, req->rq_oabufs,
1652                              sub_pga, oti);
1653                 if (rc)
1654                         break;
1655                 lov_update_common_set(set, req, rc);
1656         }
1657
1658         err = lov_fini_brw_set(set);
1659         if (!rc)
1660                 rc = err;
1661         RETURN(rc);
1662 }
1663
1664 static int lov_enqueue_interpret(struct ptlrpc_request_set *rqset,
1665                                  void *data, int rc)
1666 {
1667         struct lov_request_set *lovset = (struct lov_request_set *)data;
1668         ENTRY;
1669         rc = lov_fini_enqueue_set(lovset, lovset->set_ei->ei_mode, rc, rqset);
1670         RETURN(rc);
1671 }
1672
1673 static int lov_enqueue(struct obd_export *exp, struct obd_info *oinfo,
1674                        struct ldlm_enqueue_info *einfo,
1675                        struct ptlrpc_request_set *rqset)
1676 {
1677         ldlm_mode_t mode = einfo->ei_mode;
1678         struct lov_request_set *set;
1679         struct lov_request *req;
1680         struct list_head *pos;
1681         struct lov_obd *lov;
1682         ldlm_error_t rc;
1683         ENTRY;
1684
1685         LASSERT(oinfo);
1686         ASSERT_LSM_MAGIC(oinfo->oi_md);
1687         LASSERT(mode == (mode & -mode));
1688
1689         /* we should never be asked to replay a lock this way. */
1690         LASSERT((oinfo->oi_flags & LDLM_FL_REPLAY) == 0);
1691
1692         if (!exp || !exp->exp_obd)
1693                 RETURN(-ENODEV);
1694
1695         lov = &exp->exp_obd->u.lov;
1696         rc = lov_prep_enqueue_set(exp, oinfo, einfo, &set);
1697         if (rc)
1698                 RETURN(rc);
1699
1700         list_for_each (pos, &set->set_list) {
1701                 req = list_entry(pos, struct lov_request, rq_link);
1702
1703                 rc = obd_enqueue(lov->lov_tgts[req->rq_idx]->ltd_exp,
1704                                  &req->rq_oi, einfo, rqset);
1705                 if (rc != ELDLM_OK)
1706                         GOTO(out, rc);
1707         }
1708
1709         if (rqset && !list_empty(&rqset->set_requests)) {
1710                 LASSERT(rc == 0);
1711                 LASSERT(rqset->set_interpret == NULL);
1712                 rqset->set_interpret = lov_enqueue_interpret;
1713                 rqset->set_arg = (void *)set;
1714                 RETURN(rc);
1715         }
1716 out:
1717         rc = lov_fini_enqueue_set(set, mode, rc, rqset);
1718         RETURN(rc);
1719 }
1720
1721 static int lov_change_cbdata(struct obd_export *exp,
1722                              struct lov_stripe_md *lsm, ldlm_iterator_t it,
1723                              void *data)
1724 {
1725         struct lov_obd *lov;
1726         int rc = 0, i;
1727         ENTRY;
1728
1729         ASSERT_LSM_MAGIC(lsm);
1730
1731         if (!exp || !exp->exp_obd)
1732                 RETURN(-ENODEV);
1733
1734         LASSERT_MDS_GROUP(lsm->lsm_object_gr);
1735
1736         lov = &exp->exp_obd->u.lov;
1737         for (i = 0; i < lsm->lsm_stripe_count; i++) {
1738                 struct lov_stripe_md submd;
1739                 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
1740
1741                 if (!lov->lov_tgts[loi->loi_ost_idx]) {
1742                         CDEBUG(D_HA, "lov idx %d NULL \n", loi->loi_ost_idx);
1743                         continue;
1744                 }
1745
1746                 submd.lsm_object_id = loi->loi_id;
1747                 submd.lsm_object_gr = lsm->lsm_object_gr;
1748                 submd.lsm_stripe_count = 0;
1749                 rc = obd_change_cbdata(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1750                                        &submd, it, data);
1751         }
1752         RETURN(rc);
1753 }
1754
1755 static int lov_cancel(struct obd_export *exp, struct lov_stripe_md *lsm,
1756                       __u32 mode, struct lustre_handle *lockh)
1757 {
1758         struct lov_request_set *set;
1759         struct obd_info oinfo;
1760         struct lov_request *req;
1761         struct list_head *pos;
1762         struct lov_obd *lov;
1763         struct lustre_handle *lov_lockhp;
1764         int err = 0, rc = 0;
1765         ENTRY;
1766
1767         ASSERT_LSM_MAGIC(lsm);
1768
1769         if (!exp || !exp->exp_obd)
1770                 RETURN(-ENODEV);
1771
1772         LASSERT_MDS_GROUP(lsm->lsm_object_gr);
1773         LASSERT(lockh);
1774         lov = &exp->exp_obd->u.lov;
1775         rc = lov_prep_cancel_set(exp, &oinfo, lsm, mode, lockh, &set);
1776         if (rc)
1777                 RETURN(rc);
1778
1779         list_for_each (pos, &set->set_list) {
1780                 req = list_entry(pos, struct lov_request, rq_link);
1781                 lov_lockhp = set->set_lockh->llh_handles + req->rq_stripe;
1782
1783                 rc = obd_cancel(lov->lov_tgts[req->rq_idx]->ltd_exp,
1784                                 req->rq_oi.oi_md, mode, lov_lockhp);
1785                 rc = lov_update_common_set(set, req, rc);
1786                 if (rc) {
1787                         CERROR("error: cancel objid "LPX64" subobj "
1788                                LPX64" on OST idx %d: rc = %d\n",
1789                                lsm->lsm_object_id,
1790                                req->rq_oi.oi_md->lsm_object_id,
1791                                req->rq_idx, rc);
1792                         err = rc;
1793                 }
1794
1795         }
1796         lov_fini_cancel_set(set);
1797         RETURN(err);
1798 }
1799
1800 static int lov_cancel_unused(struct obd_export *exp,
1801                              struct lov_stripe_md *lsm,
1802                              int flags, void *opaque)
1803 {
1804         struct lov_obd *lov;
1805         int rc = 0, i;
1806         ENTRY;
1807
1808         if (!exp || !exp->exp_obd)
1809                 RETURN(-ENODEV);
1810
1811         lov = &exp->exp_obd->u.lov;
1812         if (lsm == NULL) {
1813                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
1814                         int err;
1815                         if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
1816                                 continue;
1817
1818                         err = obd_cancel_unused(lov->lov_tgts[i]->ltd_exp, NULL,
1819                                                 flags, opaque);
1820                         if (!rc)
1821                                 rc = err;
1822                 }
1823                 RETURN(rc);
1824         }
1825
1826         ASSERT_LSM_MAGIC(lsm);
1827
1828         LASSERT_MDS_GROUP(lsm->lsm_object_gr);
1829         for (i = 0; i < lsm->lsm_stripe_count; i++) {
1830                 struct lov_stripe_md submd;
1831                 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
1832                 int err;
1833
1834                 if (!lov->lov_tgts[loi->loi_ost_idx]) {
1835                         CDEBUG(D_HA, "lov idx %d NULL\n", loi->loi_ost_idx);
1836                         continue;
1837                 }
1838
1839                 if (!lov->lov_tgts[loi->loi_ost_idx]->ltd_active)
1840                         CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
1841
1842                 submd.lsm_object_id = loi->loi_id;
1843                 submd.lsm_object_gr = lsm->lsm_object_gr;
1844                 submd.lsm_stripe_count = 0;
1845                 err = obd_cancel_unused(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1846                                         &submd, flags, opaque);
1847                 if (err && lov->lov_tgts[loi->loi_ost_idx]->ltd_active) {
1848                         CERROR("error: cancel unused objid "LPX64" subobj "LPX64
1849                                " on OST idx %d: rc = %d\n", lsm->lsm_object_id,
1850                                loi->loi_id, loi->loi_ost_idx, err);
1851                         if (!rc)
1852                                 rc = err;
1853                 }
1854         }
1855         RETURN(rc);
1856 }
1857
1858 int lov_statfs_interpret(struct ptlrpc_request_set *rqset, void *data, int rc)
1859 {
1860         struct lov_request_set *lovset = (struct lov_request_set *)data;
1861         int err;
1862         ENTRY;
1863
1864         if (rc)
1865                 lovset->set_completes = 0;
1866
1867         err = lov_fini_statfs_set(lovset);
1868         RETURN(rc ? rc : err);
1869 }
1870
1871 static int lov_statfs_async(struct obd_device *obd, struct obd_info *oinfo,
1872                             __u64 max_age, struct ptlrpc_request_set *rqset)
1873 {
1874         struct lov_request_set *set;
1875         struct lov_request *req;
1876         struct list_head *pos;
1877         struct lov_obd *lov;
1878         int rc = 0;
1879         ENTRY;
1880
1881         LASSERT(oinfo != NULL);
1882         LASSERT(oinfo->oi_osfs != NULL);
1883
1884         lov = &obd->u.lov;
1885         rc = lov_prep_statfs_set(obd, oinfo, &set);
1886         if (rc)
1887                 RETURN(rc);
1888
1889         list_for_each (pos, &set->set_list) {
1890                 struct obd_device *osc_obd;
1891
1892                 req = list_entry(pos, struct lov_request, rq_link);
1893
1894                 osc_obd = class_exp2obd(lov->lov_tgts[req->rq_idx]->ltd_exp);
1895                 rc = obd_statfs_async(osc_obd, &req->rq_oi, max_age, rqset);
1896                 if (rc)
1897                         break;
1898         }
1899
1900         if (rc || list_empty(&rqset->set_requests)) {
1901                 int err;
1902                 if (rc)
1903                         set->set_completes = 0;
1904                 err = lov_fini_statfs_set(set);
1905                 RETURN(rc ? rc : err);
1906         }
1907
1908         LASSERT(rqset->set_interpret == NULL);
1909         rqset->set_interpret = lov_statfs_interpret;
1910         rqset->set_arg = (void *)set;
1911         RETURN(0);
1912 }
1913
1914 static int lov_statfs(struct obd_device *obd, struct obd_statfs *osfs,
1915                       __u64 max_age, __u32 flags)
1916 {
1917         struct ptlrpc_request_set *set = NULL;
1918         struct obd_info oinfo = { { { 0 } } };
1919         int rc = 0;
1920         ENTRY;
1921
1922
1923         /* for obdclass we forbid using obd_statfs_rqset, but prefer using async
1924          * statfs requests */
1925         set = ptlrpc_prep_set();
1926         if (set == NULL)
1927                 RETURN(-ENOMEM);
1928
1929         oinfo.oi_osfs = osfs;
1930         oinfo.oi_flags = flags;
1931         rc = lov_statfs_async(obd, &oinfo, max_age, set);
1932         if (rc == 0)
1933                 rc = ptlrpc_set_wait(set);
1934         ptlrpc_set_destroy(set);
1935
1936         RETURN(rc);
1937 }
1938
1939 static int lov_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1940                          void *karg, void *uarg)
1941 {
1942         struct obd_device *obddev = class_exp2obd(exp);
1943         struct lov_obd *lov = &obddev->u.lov;
1944         int i = 0, rc = 0, count = lov->desc.ld_tgt_count;
1945         struct obd_uuid *uuidp;
1946         ENTRY;
1947
1948         switch (cmd) {
1949         case IOC_OBD_STATFS: {
1950                 struct obd_ioctl_data *data = karg;
1951                 struct obd_device *osc_obd;
1952                 struct obd_statfs stat_buf = {0};
1953                 __u32 index;
1954
1955                 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
1956                 LASSERT(data->ioc_plen1 == sizeof(struct obd_statfs));
1957
1958                 if ((index >= count))
1959                         RETURN(-ENODEV);
1960
1961                 if (!lov->lov_tgts[index])
1962                         /* Try again with the next index */
1963                         RETURN(-EAGAIN);
1964                 if (!lov->lov_tgts[index]->ltd_active)
1965                         RETURN(-ENODATA);
1966
1967                 osc_obd = class_exp2obd(lov->lov_tgts[index]->ltd_exp);
1968                 if (!osc_obd)
1969                         RETURN(-EINVAL);
1970
1971                 /* got statfs data */
1972                 rc = obd_statfs(osc_obd, &stat_buf,
1973                                 cfs_time_current_64() - HZ, 0);
1974                 if (rc)
1975                         RETURN(rc);
1976                 if (copy_to_user(data->ioc_pbuf1, &stat_buf, data->ioc_plen1))
1977                         RETURN(-EFAULT);
1978                 /* copy UUID */
1979                 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(osc_obd),
1980                                  data->ioc_plen2))
1981                         RETURN(-EFAULT);
1982                 break;
1983         }
1984         case OBD_IOC_LOV_GET_CONFIG: {
1985                 struct obd_ioctl_data *data;
1986                 struct lov_desc *desc;
1987                 char *buf = NULL;
1988                 __u32 *genp;
1989
1990                 len = 0;
1991                 if (obd_ioctl_getdata(&buf, &len, (void *)uarg))
1992                         RETURN(-EINVAL);
1993
1994                 data = (struct obd_ioctl_data *)buf;
1995
1996                 if (sizeof(*desc) > data->ioc_inllen1) {
1997                         obd_ioctl_freedata(buf, len);
1998                         RETURN(-EINVAL);
1999                 }
2000
2001                 if (sizeof(uuidp->uuid) * count > data->ioc_inllen2) {
2002                         obd_ioctl_freedata(buf, len);
2003                         RETURN(-EINVAL);
2004                 }
2005
2006                 if (sizeof(__u32) * count > data->ioc_inllen3) {
2007                         obd_ioctl_freedata(buf, len);
2008                         RETURN(-EINVAL);
2009                 }
2010
2011                 desc = (struct lov_desc *)data->ioc_inlbuf1;
2012                 memcpy(desc, &(lov->desc), sizeof(*desc));
2013
2014                 uuidp = (struct obd_uuid *)data->ioc_inlbuf2;
2015                 genp = (__u32 *)data->ioc_inlbuf3;
2016                 /* the uuid will be empty for deleted OSTs */
2017                 for (i = 0; i < count; i++, uuidp++, genp++) {
2018                         if (!lov->lov_tgts[i])
2019                                 continue;
2020                         *uuidp = lov->lov_tgts[i]->ltd_uuid;
2021                         *genp = lov->lov_tgts[i]->ltd_gen;
2022                 }
2023
2024                 if (copy_to_user((void *)uarg, buf, len))
2025                         rc = -EFAULT;
2026                 obd_ioctl_freedata(buf, len);
2027                 break;
2028         }
2029         case LL_IOC_LOV_SETSTRIPE:
2030                 rc = lov_setstripe(exp, len, karg, uarg);
2031                 break;
2032         case LL_IOC_LOV_GETSTRIPE:
2033                 rc = lov_getstripe(exp, karg, uarg);
2034                 break;
2035         case LL_IOC_LOV_SETEA:
2036                 rc = lov_setea(exp, karg, uarg);
2037                 break;
2038         case OBD_IOC_QUOTACTL: {
2039                 struct if_quotactl *qctl = karg;
2040                 struct lov_tgt_desc *tgt = NULL;
2041                 struct obd_quotactl *oqctl;
2042
2043                 if (qctl->qc_valid == QC_OSTIDX) {
2044                         if (qctl->qc_idx < 0 || count <= qctl->qc_idx)
2045                                 RETURN(-EINVAL);
2046
2047                         tgt = lov->lov_tgts[qctl->qc_idx];
2048                         if (!tgt || !tgt->ltd_exp)
2049                                 RETURN(-EINVAL);
2050                 } else if (qctl->qc_valid == QC_UUID) {
2051                         for (i = 0; i < count; i++) {
2052                                 tgt = lov->lov_tgts[i];
2053                                 if (!tgt ||
2054                                     !obd_uuid_equals(&tgt->ltd_uuid,
2055                                                      &qctl->obd_uuid))
2056                                         continue;
2057
2058                                 if (tgt->ltd_exp == NULL)
2059                                         RETURN(-EINVAL);
2060
2061                                 break;
2062                         }
2063                 } else {
2064                         RETURN(-EINVAL);
2065                 }
2066
2067                 if (i >= count)
2068                         RETURN(-EAGAIN);
2069
2070                 LASSERT(tgt && tgt->ltd_exp);
2071                 OBD_ALLOC_PTR(oqctl);
2072                 if (!oqctl)
2073                         RETURN(-ENOMEM);
2074
2075                 QCTL_COPY(oqctl, qctl);
2076                 rc = obd_quotactl(tgt->ltd_exp, oqctl);
2077                 if (rc == 0) {
2078                         QCTL_COPY(qctl, oqctl);
2079                         qctl->qc_valid = QC_OSTIDX;
2080                         qctl->obd_uuid = tgt->ltd_uuid;
2081                 }
2082                 OBD_FREE_PTR(oqctl);
2083                 break;
2084         }
2085         default: {
2086                 int set = 0;
2087
2088                 if (count == 0)
2089                         RETURN(-ENOTTY);
2090
2091                 for (i = 0; i < count; i++) {
2092                         int err;
2093
2094                         /* OST was disconnected */
2095                         if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
2096                                 continue;
2097
2098                         err = obd_iocontrol(cmd, lov->lov_tgts[i]->ltd_exp,
2099                                             len, karg, uarg);
2100                         if (err == -ENODATA && cmd == OBD_IOC_POLL_QUOTACHECK) {
2101                                 RETURN(err);
2102                         } else if (err) {
2103                                 if (lov->lov_tgts[i]->ltd_active) {
2104                                         CDEBUG(err == -ENOTTY ?
2105                                                D_IOCTL : D_WARNING,
2106                                                "iocontrol OSC %s on OST "
2107                                                "idx %d cmd %x: err = %d\n",
2108                                                lov_uuid2str(lov, i),
2109                                                i, cmd, err);
2110                                         if (!rc)
2111                                                 rc = err;
2112                                 }
2113                         } else {
2114                                 set = 1;
2115                         }
2116                 }
2117                 if (!set && !rc)
2118                         rc = -EIO;
2119         }
2120         }
2121
2122         RETURN(rc);
2123 }
2124
2125 #define FIEMAP_BUFFER_SIZE 4096
2126
2127 /**
2128  * Non-zero fe_logical indicates that this is a continuation FIEMAP
2129  * call. The local end offset and the device are sent in the first
2130  * fm_extent. This function calculates the stripe number from the index.
2131  * This function returns a stripe_no on which mapping is to be restarted.
2132  *
2133  * This function returns fm_end_offset which is the in-OST offset at which
2134  * mapping should be restarted. If fm_end_offset=0 is returned then caller
2135  * will re-calculate proper offset in next stripe.
2136  * Note that the first extent is passed to lov_get_info via the value field.
2137  *
2138  * \param fiemap fiemap request header
2139  * \param lsm striping information for the file
2140  * \param fm_start logical start of mapping
2141  * \param fm_end logical end of mapping
2142  * \param start_stripe starting stripe will be returned in this
2143  */
2144 obd_size fiemap_calc_fm_end_offset(struct ll_user_fiemap *fiemap,
2145                                    struct lov_stripe_md *lsm, obd_size fm_start,
2146                                    obd_size fm_end, int *start_stripe)
2147 {
2148         obd_size local_end = fiemap->fm_extents[0].fe_logical;
2149         obd_off lun_start, lun_end;
2150         obd_size fm_end_offset;
2151         int stripe_no = -1, i;
2152
2153         if (fiemap->fm_extent_count == 0 ||
2154             fiemap->fm_extents[0].fe_logical == 0)
2155                 return 0;
2156
2157         /* Find out stripe_no from ost_index saved in the fe_device */
2158         for (i = 0; i < lsm->lsm_stripe_count; i++) {
2159                 if (lsm->lsm_oinfo[i]->loi_ost_idx ==
2160                                         fiemap->fm_extents[0].fe_device) {
2161                         stripe_no = i;
2162                         break;
2163                 }
2164         }
2165
2166         /* If we have finished mapping on previous device, shift logical
2167          * offset to start of next device */
2168         if ((lov_stripe_intersects(lsm, stripe_no, fm_start, fm_end,
2169                                    &lun_start, &lun_end)) != 0 &&
2170                                    local_end < lun_end) {
2171                 fm_end_offset = local_end;
2172                 *start_stripe = stripe_no;
2173         } else {
2174                 /* This is a special value to indicate that caller should
2175                  * calculate offset in next stripe. */
2176                 fm_end_offset = 0;
2177                 *start_stripe = (stripe_no + 1) % lsm->lsm_stripe_count;
2178         }
2179
2180         return fm_end_offset;
2181 }
2182
2183 /**
2184  * We calculate on which OST the mapping will end. If the length of mapping
2185  * is greater than (stripe_size * stripe_count) then the last_stripe will
2186  * will be one just before start_stripe. Else we check if the mapping
2187  * intersects each OST and find last_stripe.
2188  * This function returns the last_stripe and also sets the stripe_count
2189  * over which the mapping is spread
2190  *
2191  * \param lsm striping information for the file
2192  * \param fm_start logical start of mapping
2193  * \param fm_end logical end of mapping
2194  * \param start_stripe starting stripe of the mapping
2195  * \param stripe_count the number of stripes across which to map is returned
2196  *
2197  * \retval last_stripe return the last stripe of the mapping
2198  */
2199 int fiemap_calc_last_stripe(struct lov_stripe_md *lsm, obd_size fm_start,
2200                             obd_size fm_end, int start_stripe,
2201                             int *stripe_count)
2202 {
2203         int last_stripe;
2204         obd_off obd_start, obd_end;
2205         int i, j;
2206
2207         if (fm_end - fm_start > lsm->lsm_stripe_size * lsm->lsm_stripe_count) {
2208                 last_stripe = (start_stripe < 1 ? lsm->lsm_stripe_count - 1 :
2209                                                               start_stripe - 1);
2210                 *stripe_count = lsm->lsm_stripe_count;
2211         } else {
2212                 for (j = 0, i = start_stripe; j < lsm->lsm_stripe_count;
2213                      i = (i + 1) % lsm->lsm_stripe_count, j++) {
2214                         if ((lov_stripe_intersects(lsm, i, fm_start, fm_end,
2215                                                    &obd_start, &obd_end)) == 0)
2216                                 break;
2217                 }
2218                 *stripe_count = j;
2219                 last_stripe = (start_stripe + j - 1) %lsm->lsm_stripe_count;
2220         }
2221
2222         return last_stripe;
2223 }
2224
2225 /**
2226  * Set fe_device and copy extents from local buffer into main return buffer.
2227  *
2228  * \param fiemap fiemap request header
2229  * \param lcl_fm_ext array of local fiemap extents to be copied
2230  * \param ost_index OST index to be written into the fm_device field for each
2231                     extent
2232  * \param ext_count number of extents to be copied
2233  * \param current_extent where to start copying in main extent array
2234  */
2235 void fiemap_prepare_and_copy_exts(struct ll_user_fiemap *fiemap,
2236                                   struct ll_fiemap_extent *lcl_fm_ext,
2237                                   int ost_index, unsigned int ext_count,
2238                                   int current_extent)
2239 {
2240         char *to;
2241         int ext;
2242
2243         for (ext = 0; ext < ext_count; ext++) {
2244                 lcl_fm_ext[ext].fe_device = ost_index;
2245                 lcl_fm_ext[ext].fe_flags |= FIEMAP_EXTENT_NET;
2246         }
2247
2248         /* Copy fm_extent's from fm_local to return buffer */
2249         to = (char *)fiemap + fiemap_count_to_size(current_extent);
2250         memcpy(to, lcl_fm_ext, ext_count * sizeof(struct ll_fiemap_extent));
2251 }
2252
2253 /**
2254  * Break down the FIEMAP request and send appropriate calls to individual OSTs.
2255  * This also handles the restarting of FIEMAP calls in case mapping overflows
2256  * the available number of extents in single call.
2257  */
2258 static int lov_fiemap(struct lov_obd *lov, __u32 keylen, void *key,
2259                       __u32 *vallen, void *val, struct lov_stripe_md *lsm)
2260 {
2261         struct ll_fiemap_info_key *fm_key = key;
2262         struct ll_user_fiemap *fiemap = val;
2263         struct ll_user_fiemap *fm_local = NULL;
2264         struct ll_fiemap_extent *lcl_fm_ext;
2265         int count_local;
2266         unsigned int get_num_extents = 0;
2267         int ost_index = 0, actual_start_stripe, start_stripe;
2268         obd_size fm_start, fm_end, fm_length, fm_end_offset = 0;
2269         obd_size curr_loc;
2270         int current_extent = 0, rc = 0, i;
2271         int ost_eof = 0; /* EOF for object */
2272         int ost_done = 0; /* done with required mapping for this OST? */
2273         int last_stripe;
2274         int cur_stripe = 0, cur_stripe_wrap = 0, stripe_count;
2275         unsigned int buffer_size = FIEMAP_BUFFER_SIZE;
2276
2277         if (lsm == NULL)
2278                 GOTO(out, rc = 0);
2279
2280         if (fiemap_count_to_size(fm_key->fiemap.fm_extent_count) < buffer_size)
2281                 buffer_size = fiemap_count_to_size(fm_key->fiemap.fm_extent_count);
2282
2283         OBD_ALLOC(fm_local, buffer_size);
2284         if (fm_local == NULL)
2285                 GOTO(out, rc = -ENOMEM);
2286         lcl_fm_ext = &fm_local->fm_extents[0];
2287
2288         count_local = fiemap_size_to_count(buffer_size);
2289
2290         memcpy(fiemap, &fm_key->fiemap, sizeof(*fiemap));
2291         fm_start = fiemap->fm_start;
2292         fm_length = fiemap->fm_length;
2293         /* Calculate start stripe, last stripe and length of mapping */
2294         actual_start_stripe = start_stripe = lov_stripe_number(lsm, fm_start);
2295         fm_end = (fm_length == ~0ULL ? fm_key->oa.o_size :
2296                                                 fm_start + fm_length - 1);
2297         /* If fm_length != ~0ULL but fm_start+fm_length-1 exceeds file size */
2298         if (fm_end > fm_key->oa.o_size)
2299                 fm_end = fm_key->oa.o_size;
2300
2301         last_stripe = fiemap_calc_last_stripe(lsm, fm_start, fm_end,
2302                                             actual_start_stripe, &stripe_count);
2303
2304         fm_end_offset = fiemap_calc_fm_end_offset(fiemap, lsm, fm_start, fm_end,
2305                                                   &start_stripe);
2306
2307         if (fiemap->fm_extent_count == 0) {
2308                 get_num_extents = 1;
2309                 count_local = 0;
2310         }
2311
2312         /* Check each stripe */
2313         for (cur_stripe = start_stripe, i = 0; i < stripe_count;
2314              i++, cur_stripe = (cur_stripe + 1) % lsm->lsm_stripe_count) {
2315                 obd_size req_fm_len; /* Stores length of required mapping */
2316                 obd_size len_mapped_single_call;
2317                 obd_off lun_start, lun_end, obd_object_end;
2318                 unsigned int ext_count;
2319
2320                 cur_stripe_wrap = cur_stripe;
2321
2322                 /* Find out range of mapping on this stripe */
2323                 if ((lov_stripe_intersects(lsm, cur_stripe, fm_start, fm_end,
2324                                            &lun_start, &obd_object_end)) == 0)
2325                         continue;
2326
2327                 /* If this is a continuation FIEMAP call and we are on
2328                  * starting stripe then lun_start needs to be set to
2329                  * fm_end_offset */
2330                 if (fm_end_offset != 0 && cur_stripe == start_stripe)
2331                         lun_start = fm_end_offset;
2332
2333                 if (fm_length != ~0ULL) {
2334                         /* Handle fm_start + fm_length overflow */
2335                         if (fm_start + fm_length < fm_start)
2336                                 fm_length = ~0ULL - fm_start;
2337                         lun_end = lov_size_to_stripe(lsm, fm_start + fm_length,
2338                                                      cur_stripe);
2339                 } else {
2340                         lun_end = ~0ULL;
2341                 }
2342
2343                 if (lun_start == lun_end)
2344                         continue;
2345
2346                 req_fm_len = obd_object_end - lun_start;
2347                 fm_local->fm_length = 0;
2348                 len_mapped_single_call = 0;
2349
2350                 /* If the output buffer is very large and the objects have many
2351                  * extents we may need to loop on a single OST repeatedly */
2352                 ost_eof = 0;
2353                 ost_done = 0;
2354                 do {
2355                         if (get_num_extents == 0) {
2356                                 /* Don't get too many extents. */
2357                                 if (current_extent + count_local >
2358                                     fiemap->fm_extent_count)
2359                                         count_local = fiemap->fm_extent_count -
2360                                                                  current_extent;
2361                         }
2362
2363                         lun_start += len_mapped_single_call;
2364                         fm_local->fm_length = req_fm_len - len_mapped_single_call;
2365                         req_fm_len = fm_local->fm_length;
2366                         fm_local->fm_extent_count = count_local;
2367                         fm_local->fm_mapped_extents = 0;
2368                         fm_local->fm_flags = fiemap->fm_flags;
2369
2370                         fm_key->oa.o_id = lsm->lsm_oinfo[cur_stripe]->loi_id;
2371                         ost_index = lsm->lsm_oinfo[cur_stripe]->loi_ost_idx;
2372
2373                         if (ost_index < 0 || ost_index >=lov->desc.ld_tgt_count)
2374                                 GOTO(out, rc = -EINVAL);
2375
2376                         /* If OST is inactive, return extent with UNKNOWN flag */
2377                         if (!lov->lov_tgts[ost_index]->ltd_active) {
2378                                 fm_local->fm_flags |= FIEMAP_EXTENT_LAST;
2379                                 fm_local->fm_mapped_extents = 1;
2380
2381                                 lcl_fm_ext[0].fe_logical = lun_start;
2382                                 lcl_fm_ext[0].fe_length = obd_object_end -
2383                                                                       lun_start;
2384                                 lcl_fm_ext[0].fe_flags |= FIEMAP_EXTENT_UNKNOWN;
2385
2386                                 goto inactive_tgt;
2387                         }
2388
2389                         fm_local->fm_start = lun_start;
2390                         fm_local->fm_flags &= ~FIEMAP_FLAG_DEVICE_ORDER;
2391                         memcpy(&fm_key->fiemap, fm_local, sizeof(*fm_local));
2392                         *vallen=fiemap_count_to_size(fm_local->fm_extent_count);
2393                         rc = obd_get_info(lov->lov_tgts[ost_index]->ltd_exp,
2394                                           keylen, key, vallen, fm_local, lsm);
2395                         if (rc != 0)
2396                                 GOTO(out, rc);
2397
2398 inactive_tgt:
2399                         ext_count = fm_local->fm_mapped_extents;
2400                         if (ext_count == 0) {
2401                                 ost_done = 1;
2402                                 /* If last stripe has hole at the end,
2403                                  * then we need to return */
2404                                 if (cur_stripe_wrap == last_stripe) {
2405                                         fiemap->fm_mapped_extents = 0;
2406                                         goto finish;
2407                                 }
2408                                 break;
2409                         }
2410
2411                         /* If we just need num of extents then go to next device */
2412                         if (get_num_extents) {
2413                                 current_extent += ext_count;
2414                                 break;
2415                         }
2416
2417                         len_mapped_single_call = lcl_fm_ext[ext_count-1].fe_logical -
2418                                   lun_start + lcl_fm_ext[ext_count - 1].fe_length;
2419
2420                         /* Have we finished mapping on this device? */
2421                         if (req_fm_len <= len_mapped_single_call)
2422                                 ost_done = 1;
2423
2424                         /* Clear the EXTENT_LAST flag which can be present on
2425                          * last extent */
2426                         if (lcl_fm_ext[ext_count-1].fe_flags & FIEMAP_EXTENT_LAST)
2427                                 lcl_fm_ext[ext_count - 1].fe_flags &=
2428                                                             ~FIEMAP_EXTENT_LAST;
2429
2430                         curr_loc = lov_stripe_size(lsm,
2431                                            lcl_fm_ext[ext_count - 1].fe_logical+
2432                                            lcl_fm_ext[ext_count - 1].fe_length,
2433                                            cur_stripe);
2434                         if (curr_loc >= fm_key->oa.o_size)
2435                                 ost_eof = 1;
2436
2437                         fiemap_prepare_and_copy_exts(fiemap, lcl_fm_ext,
2438                                                      ost_index, ext_count,
2439                                                      current_extent);
2440
2441                         current_extent += ext_count;
2442
2443                         /* Ran out of available extents? */
2444                         if (current_extent >= fiemap->fm_extent_count)
2445                                 goto finish;
2446                 } while (ost_done == 0 && ost_eof == 0);
2447
2448                 if (cur_stripe_wrap == last_stripe)
2449                         goto finish;
2450         }
2451
2452 finish:
2453         /* Indicate that we are returning device offsets unless file just has
2454          * single stripe */
2455         if (lsm->lsm_stripe_count > 1)
2456                 fiemap->fm_flags |= FIEMAP_FLAG_DEVICE_ORDER;
2457
2458         if (get_num_extents)
2459                 goto skip_last_device_calc;
2460
2461         /* Check if we have reached the last stripe and whether mapping for that
2462          * stripe is done. */
2463         if (cur_stripe_wrap == last_stripe) {
2464                 if (ost_done || ost_eof)
2465                         fiemap->fm_extents[current_extent - 1].fe_flags |=
2466                                                              FIEMAP_EXTENT_LAST;
2467         }
2468
2469 skip_last_device_calc:
2470         fiemap->fm_mapped_extents = current_extent;
2471
2472 out:
2473         OBD_FREE(fm_local, buffer_size);
2474         return rc;
2475 }
2476
2477 static int lov_get_info(struct obd_export *exp, __u32 keylen,
2478                         void *key, __u32 *vallen, void *val,
2479                         struct lov_stripe_md *lsm)
2480 {
2481         struct obd_device *obddev = class_exp2obd(exp);
2482         struct lov_obd *lov = &obddev->u.lov;
2483         int i, rc;
2484         ENTRY;
2485
2486         if (!vallen || !val)
2487                 RETURN(-EFAULT);
2488
2489         obd_getref(obddev);
2490
2491         if (KEY_IS(KEY_LOCK_TO_STRIPE)) {
2492                 struct {
2493                         char name[16];
2494                         struct ldlm_lock *lock;
2495                 } *data = key;
2496                 struct ldlm_res_id *res_id = &data->lock->l_resource->lr_name;
2497                 struct lov_oinfo *loi;
2498                 __u32 *stripe = val;
2499
2500                 if (*vallen < sizeof(*stripe))
2501                         GOTO(out, rc = -EFAULT);
2502                 *vallen = sizeof(*stripe);
2503
2504                 /* XXX This is another one of those bits that will need to
2505                  * change if we ever actually support nested LOVs.  It uses
2506                  * the lock's export to find out which stripe it is. */
2507                 /* XXX - it's assumed all the locks for deleted OSTs have
2508                  * been cancelled. Also, the export for deleted OSTs will
2509                  * be NULL and won't match the lock's export. */
2510                 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2511                         loi = lsm->lsm_oinfo[i];
2512                         if (!lov->lov_tgts[loi->loi_ost_idx])
2513                                 continue;
2514                         if (lov->lov_tgts[loi->loi_ost_idx]->ltd_exp ==
2515                             data->lock->l_conn_export &&
2516                             osc_res_name_eq(loi->loi_id, loi->loi_gr, res_id)) {
2517                                 *stripe = i;
2518                                 GOTO(out, rc = 0);
2519                         }
2520                 }
2521                 LDLM_ERROR(data->lock, "lock on inode without such object");
2522                 dump_lsm(D_ERROR, lsm);
2523                 GOTO(out, rc = -ENXIO);
2524         } else if (KEY_IS(KEY_LAST_ID)) {
2525                 struct obd_id_info *info = val;
2526                 __u32 size = sizeof(obd_id);
2527                 struct lov_tgt_desc *tgt;
2528
2529                 LASSERT(*vallen == sizeof(struct obd_id_info));
2530                 tgt = lov->lov_tgts[info->idx];
2531
2532                 if (!tgt || !tgt->ltd_active)
2533                         GOTO(out, rc = -ESRCH);
2534
2535                 rc = obd_get_info(tgt->ltd_exp, keylen, key, &size, info->data, NULL);
2536                 GOTO(out, rc = 0);
2537         } else if (KEY_IS(KEY_LOVDESC)) {
2538                 struct lov_desc *desc_ret = val;
2539                 *desc_ret = lov->desc;
2540
2541                 GOTO(out, rc = 0);
2542         } else if (KEY_IS(KEY_FIEMAP)) {
2543                 rc = lov_fiemap(lov, keylen, key, vallen, val, lsm);
2544                 GOTO(out, rc);
2545         }
2546
2547         rc = -EINVAL;
2548
2549 out:
2550         obd_putref(obddev);
2551         RETURN(rc);
2552 }
2553
2554 static int lov_set_info_async(struct obd_export *exp, obd_count keylen,
2555                               void *key, obd_count vallen, void *val,
2556                               struct ptlrpc_request_set *set)
2557 {
2558         struct obd_device *obddev = class_exp2obd(exp);
2559         struct lov_obd *lov = &obddev->u.lov;
2560         obd_count count;
2561         int i, rc = 0, err;
2562         struct lov_tgt_desc *tgt;
2563         unsigned incr, check_uuid,
2564                  do_inactive, no_set;
2565         unsigned next_id = 0,  mds_con = 0, capa = 0;
2566         ENTRY;
2567
2568         incr = check_uuid = do_inactive = no_set = 0;
2569         if (set == NULL) {
2570                 no_set = 1;
2571                 set = ptlrpc_prep_set();
2572                 if (!set)
2573                         RETURN(-ENOMEM);
2574         }
2575
2576         obd_getref(obddev);
2577         count = lov->desc.ld_tgt_count;
2578
2579         if (KEY_IS(KEY_NEXT_ID)) {
2580                 count = vallen / sizeof(struct obd_id_info);
2581                 vallen = sizeof(obd_id);
2582                 incr = sizeof(struct obd_id_info);
2583                 do_inactive = 1;
2584                 next_id = 1;
2585         } else if (KEY_IS(KEY_CHECKSUM)) {
2586                 do_inactive = 1;
2587         } else if (KEY_IS(KEY_UNLINKED)) {
2588                 check_uuid = val ? 1 : 0;
2589         } else if (KEY_IS(KEY_EVICT_BY_NID)) {
2590                 /* use defaults:  do_inactive = incr = 0; */
2591         } else if (KEY_IS(KEY_MDS_CONN)) {
2592                 mds_con = 1;
2593         } else if (KEY_IS(KEY_CAPA_KEY)) {
2594                 capa = 1;
2595         }
2596
2597         for (i = 0; i < count; i++, val = (char *)val + incr) {
2598                 if (next_id) {
2599                         tgt = lov->lov_tgts[((struct obd_id_info*)val)->idx];
2600                 } else {
2601                         tgt = lov->lov_tgts[i];
2602                 }
2603                 /* OST was disconnected */
2604                 if (!tgt || !tgt->ltd_exp)
2605                         continue;
2606
2607                 /* OST is inactive and we don't want inactive OSCs */
2608                 if (!tgt->ltd_active && !do_inactive)
2609                         continue;
2610
2611                 if (mds_con) {
2612                         struct mds_group_info *mgi;
2613
2614                         LASSERT(vallen == sizeof(*mgi));
2615                         mgi = (struct mds_group_info *)val;
2616
2617                         /* Only want a specific OSC */
2618                         if (mgi->uuid && !obd_uuid_equals(mgi->uuid,
2619                                                 &tgt->ltd_uuid))
2620                                 continue;
2621
2622                         err = obd_set_info_async(tgt->ltd_exp,
2623                                          keylen, key, sizeof(int),
2624                                          &mgi->group, set);
2625                 } else if (next_id) {
2626                         err = obd_set_info_async(tgt->ltd_exp,
2627                                          keylen, key, vallen,
2628                                          ((struct obd_id_info*)val)->data, set);
2629                 } else if (capa) {
2630                         struct mds_capa_info *info = (struct mds_capa_info*)val;
2631
2632                         LASSERT(vallen == sizeof(*info));
2633
2634                          /* Only want a specific OSC */
2635                         if (info->uuid &&
2636                             !obd_uuid_equals(info->uuid, &tgt->ltd_uuid))
2637                                 continue;
2638
2639                         err = obd_set_info_async(tgt->ltd_exp, keylen, key,
2640                                                  sizeof(*info->capa),
2641                                                  info->capa, set);
2642
2643                 } else {
2644                         /* Only want a specific OSC */
2645                         if (check_uuid &&
2646                             !obd_uuid_equals(val, &tgt->ltd_uuid))
2647                                 continue;
2648
2649                         err = obd_set_info_async(tgt->ltd_exp,
2650                                          keylen, key, vallen, val, set);
2651                 }
2652
2653                 if (!rc)
2654                         rc = err;
2655         }
2656
2657         obd_putref(obddev);
2658         if (no_set) {
2659                 err = ptlrpc_set_wait(set);
2660                 if (!rc)
2661                         rc = err;
2662                 ptlrpc_set_destroy(set);
2663         }
2664         RETURN(rc);
2665 }
2666
2667 static int lov_checkmd(struct obd_export *exp, struct obd_export *md_exp,
2668                        struct lov_stripe_md *lsm)
2669 {
2670         int rc;
2671         ENTRY;
2672
2673         if (!lsm)
2674                 RETURN(0);
2675         LASSERT(md_exp);
2676         LASSERT(lsm_op_find(lsm->lsm_magic) != NULL);
2677         rc = lsm_op_find(lsm->lsm_magic)->lsm_revalidate(lsm, md_exp->exp_obd);
2678
2679         RETURN(rc);
2680 }
2681
2682 int lov_test_and_clear_async_rc(struct lov_stripe_md *lsm)
2683 {
2684         int i, rc = 0;
2685         ENTRY;
2686
2687         for (i = 0; i < lsm->lsm_stripe_count; i++) {
2688                 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
2689                 if (loi->loi_ar.ar_rc && !rc)
2690                         rc = loi->loi_ar.ar_rc;
2691                 loi->loi_ar.ar_rc = 0;
2692         }
2693         RETURN(rc);
2694 }
2695 EXPORT_SYMBOL(lov_test_and_clear_async_rc);
2696
2697
2698 static int lov_extent_calc(struct obd_export *exp, struct lov_stripe_md *lsm,
2699                            int cmd, __u64 *offset)
2700 {
2701         __u32 ssize = lsm->lsm_stripe_size;
2702         __u64 start;
2703
2704         start = *offset;
2705         do_div(start, ssize);
2706         start = start * ssize;
2707
2708         CDEBUG(D_DLMTRACE, "offset "LPU64", stripe %u, start "LPU64
2709                            ", end "LPU64"\n", *offset, ssize, start,
2710                            start + ssize - 1);
2711         if (cmd == OBD_CALC_STRIPE_END) {
2712                 *offset = start + ssize - 1;
2713         } else if (cmd == OBD_CALC_STRIPE_START) {
2714                 *offset = start;
2715         } else {
2716                 LBUG();
2717         }
2718
2719         RETURN(0);
2720 }
2721
2722 void lov_stripe_lock(struct lov_stripe_md *md)
2723 {
2724         LASSERT(md->lsm_lock_owner != cfs_curproc_pid());
2725         spin_lock(&md->lsm_lock);
2726         LASSERT(md->lsm_lock_owner == 0);
2727         md->lsm_lock_owner = cfs_curproc_pid();
2728 }
2729 EXPORT_SYMBOL(lov_stripe_lock);
2730
2731 void lov_stripe_unlock(struct lov_stripe_md *md)
2732 {
2733         LASSERT(md->lsm_lock_owner == cfs_curproc_pid());
2734         md->lsm_lock_owner = 0;
2735         spin_unlock(&md->lsm_lock);
2736 }
2737 EXPORT_SYMBOL(lov_stripe_unlock);
2738
2739
2740 struct obd_ops lov_obd_ops = {
2741         .o_owner               = THIS_MODULE,
2742         .o_setup               = lov_setup,
2743         .o_precleanup          = lov_precleanup,
2744         .o_cleanup             = lov_cleanup,
2745         //.o_process_config      = lov_process_config,
2746         .o_connect             = lov_connect,
2747         .o_disconnect          = lov_disconnect,
2748         .o_statfs              = lov_statfs,
2749         .o_statfs_async        = lov_statfs_async,
2750         .o_packmd              = lov_packmd,
2751         .o_unpackmd            = lov_unpackmd,
2752         .o_checkmd             = lov_checkmd,
2753         .o_create              = lov_create,
2754         .o_destroy             = lov_destroy,
2755         .o_getattr             = lov_getattr,
2756         .o_getattr_async       = lov_getattr_async,
2757         .o_setattr             = lov_setattr,
2758         .o_setattr_async       = lov_setattr_async,
2759         .o_brw                 = lov_brw,
2760         .o_merge_lvb           = lov_merge_lvb,
2761         .o_adjust_kms          = lov_adjust_kms,
2762         .o_punch               = lov_punch,
2763         .o_sync                = lov_sync,
2764         .o_enqueue             = lov_enqueue,
2765         .o_change_cbdata       = lov_change_cbdata,
2766         .o_cancel              = lov_cancel,
2767         .o_cancel_unused       = lov_cancel_unused,
2768         .o_iocontrol           = lov_iocontrol,
2769         .o_get_info            = lov_get_info,
2770         .o_set_info_async      = lov_set_info_async,
2771         .o_extent_calc         = lov_extent_calc,
2772         .o_llog_init           = lov_llog_init,
2773         .o_llog_finish         = lov_llog_finish,
2774         .o_notify              = lov_notify,
2775         .o_pool_new            = lov_pool_new,
2776         .o_pool_rem            = lov_pool_remove,
2777         .o_pool_add            = lov_pool_add,
2778         .o_pool_del            = lov_pool_del,
2779         .o_getref              = lov_getref,
2780         .o_putref              = lov_putref,
2781 };
2782
2783 static quota_interface_t *quota_interface;
2784 extern quota_interface_t lov_quota_interface;
2785
2786 cfs_mem_cache_t *lov_oinfo_slab;
2787
2788 extern struct lu_kmem_descr lov_caches[];
2789
2790 int __init lov_init(void)
2791 {
2792         struct lprocfs_static_vars lvars = { 0 };
2793         int rc, rc2;
2794         ENTRY;
2795
2796         /* print an address of _any_ initialized kernel symbol from this
2797          * module, to allow debugging with gdb that doesn't support data
2798          * symbols from modules.*/
2799         CDEBUG(D_CONSOLE, "Lustre LOV module (%p).\n", &lov_caches);
2800
2801         rc = lu_kmem_init(lov_caches);
2802         if (rc)
2803                 return rc;
2804
2805         lov_oinfo_slab = cfs_mem_cache_create("lov_oinfo",
2806                                               sizeof(struct lov_oinfo),
2807                                               0, SLAB_HWCACHE_ALIGN);
2808         if (lov_oinfo_slab == NULL) {
2809                 lu_kmem_fini(lov_caches);
2810                 return -ENOMEM;
2811         }
2812         lprocfs_lov_init_vars(&lvars);
2813
2814         request_module("lquota");
2815         quota_interface = PORTAL_SYMBOL_GET(lov_quota_interface);
2816         init_obd_quota_ops(quota_interface, &lov_obd_ops);
2817
2818         rc = class_register_type(&lov_obd_ops, NULL, lvars.module_vars,
2819                                  LUSTRE_LOV_NAME, &lov_device_type);
2820
2821         if (rc) {
2822                 if (quota_interface)
2823                         PORTAL_SYMBOL_PUT(lov_quota_interface);
2824                 rc2 = cfs_mem_cache_destroy(lov_oinfo_slab);
2825                 LASSERT(rc2 == 0);
2826                 lu_kmem_fini(lov_caches);
2827         }
2828
2829         RETURN(rc);
2830 }
2831
2832 #ifdef __KERNEL__
2833 static void /*__exit*/ lov_exit(void)
2834 {
2835         int rc;
2836
2837         lu_device_type_fini(&lov_device_type);
2838         lu_kmem_fini(lov_caches);
2839
2840         if (quota_interface)
2841                 PORTAL_SYMBOL_PUT(lov_quota_interface);
2842
2843         class_unregister_type(LUSTRE_LOV_NAME);
2844         rc = cfs_mem_cache_destroy(lov_oinfo_slab);
2845         LASSERT(rc == 0);
2846 }
2847
2848 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2849 MODULE_DESCRIPTION("Lustre Logical Object Volume OBD driver");
2850 MODULE_LICENSE("GPL");
2851
2852 cfs_module(lov, LUSTRE_VERSION_STRING, lov_init, lov_exit);
2853 #endif