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