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 (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
  30  * Use is subject to license terms.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/lov/lov_obd.c
  37  *
  38  * Author: Phil Schwan <phil@clusterfs.com>
  39  * Author: Peter Braam <braam@clusterfs.com>
  40  * Author: Mike Shaver <shaver@clusterfs.com>
  41  * Author: Nathan Rutman <nathan@clusterfs.com>
  42  */
  43
  44 #ifndef EXPORT_SYMTAB
  45 # define EXPORT_SYMTAB
  46 #endif
  47 #define DEBUG_SUBSYSTEM S_LOV
  48 #ifdef __KERNEL__
  49 #include <libcfs/libcfs.h>
  50 #else
  51 #include <liblustre.h>
  52 #endif
  53
  54 #include <obd_support.h>
  55 #include <lustre_lib.h>
  56 #include <lustre_net.h>
  57 #include <lustre/lustre_idl.h>
  58 #include <lustre_dlm.h>
  59 #include <lustre_mds.h>
  60 #include <lustre_debug.h>
  61 #include <obd_class.h>
  62 #include <obd_lov.h>
  63 #include <obd_ost.h>
  64 #include <lprocfs_status.h>
  65 #include <lustre_param.h>
  66 #include <cl_object.h>
  67 #include <lustre/ll_fiemap.h>
  68
  69 #include "lov_internal.h"
  70
  71 /* Keep a refcount of lov->tgt usage to prevent racing with addition/deletion.
  72    Any function that expects lov_tgts to remain stationary must take a ref. */
  73 static void lov_getref(struct obd_device *obd)
  74 {
  75         struct lov_obd *lov = &obd->u.lov;
  76
  77         /* nobody gets through here until lov_putref is done */
  78         cfs_mutex_down(&lov->lov_lock);
  79         cfs_atomic_inc(&lov->lov_refcount);
  80         cfs_mutex_up(&lov->lov_lock);
  81         return;
  82 }
  83
  84 static void __lov_del_obd(struct obd_device *obd, struct lov_tgt_desc *tgt);
  85
  86 static void lov_putref(struct obd_device *obd)
  87 {
  88         struct lov_obd *lov = &obd->u.lov;
  89
  90         cfs_mutex_down(&lov->lov_lock);
  91         /* ok to dec to 0 more than once -- ltd_exp's will be null */
  92         if (cfs_atomic_dec_and_test(&lov->lov_refcount) && lov->lov_death_row) {
  93                 CFS_LIST_HEAD(kill);
  94                 int i;
  95                 struct lov_tgt_desc *tgt, *n;
  96                 CDEBUG(D_CONFIG, "destroying %d lov targets\n",
  97                        lov->lov_death_row);
  98                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
  99                         tgt = lov->lov_tgts[i];
 100
 101                         if (!tgt || !tgt->ltd_reap)
 102                                 continue;
 103                         cfs_list_add(&tgt->ltd_kill, &kill);
 104                         /* XXX - right now there is a dependency on ld_tgt_count
 105                          * being the maximum tgt index for computing the
 106                          * mds_max_easize. So we can't shrink it. */
 107                         lov_ost_pool_remove(&lov->lov_packed, i);
 108                         lov->lov_tgts[i] = NULL;
 109                         lov->lov_death_row--;
 110                 }
 111                 cfs_mutex_up(&lov->lov_lock);
 112
 113                 cfs_list_for_each_entry_safe(tgt, n, &kill, ltd_kill) {
 114                         cfs_list_del(&tgt->ltd_kill);
 115                         /* Disconnect */
 116                         __lov_del_obd(obd, tgt);
 117                 }
 118         } else {
 119                 cfs_mutex_up(&lov->lov_lock);
 120         }
 121 }
 122
 123 static int lov_set_osc_active(struct obd_device *obd, struct obd_uuid *uuid,
 124                               int activate);
 125 static int lov_notify(struct obd_device *obd, struct obd_device *watched,
 126                       enum obd_notify_event ev, void *data);
 127
 128
 129 #define MAX_STRING_SIZE 128
 130 int lov_connect_obd(struct obd_device *obd, __u32 index, int activate,
 131                     struct obd_connect_data *data)
 132 {
 133         struct lov_obd *lov = &obd->u.lov;
 134         struct obd_uuid *tgt_uuid;
 135         struct obd_device *tgt_obd;
 136         static struct obd_uuid lov_osc_uuid = { "LOV_OSC_UUID" };
 137         struct obd_import *imp;
 138 #ifdef __KERNEL__
 139         cfs_proc_dir_entry_t *lov_proc_dir;
 140 #endif
 141         int rc;
 142         ENTRY;
 143
 144         if (!lov->lov_tgts[index])
 145                 RETURN(-EINVAL);
 146
 147         tgt_uuid = &lov->lov_tgts[index]->ltd_uuid;
 148         tgt_obd = lov->lov_tgts[index]->ltd_obd;
 149
 150         if (!tgt_obd->obd_set_up) {
 151                 CERROR("Target %s not set up\n", obd_uuid2str(tgt_uuid));
 152                 RETURN(-EINVAL);
 153         }
 154
 155         /* override the sp_me from lov */
 156         tgt_obd->u.cli.cl_sp_me = lov->lov_sp_me;
 157
 158         if (data && (data->ocd_connect_flags & OBD_CONNECT_INDEX))
 159                 data->ocd_index = index;
 160
 161         /*
 162          * Divine LOV knows that OBDs under it are OSCs.
 163          */
 164         imp = tgt_obd->u.cli.cl_import;
 165
 166         if (activate) {
 167                 tgt_obd->obd_no_recov = 0;
 168                 /* FIXME this is probably supposed to be
 169                    ptlrpc_set_import_active.  Horrible naming. */
 170                 ptlrpc_activate_import(imp);
 171         }
 172
 173         rc = obd_register_observer(tgt_obd, obd);
 174         if (rc) {
 175                 CERROR("Target %s register_observer error %d\n",
 176                        obd_uuid2str(tgt_uuid), rc);
 177                 RETURN(rc);
 178         }
 179
 180
 181         if (imp->imp_invalid) {
 182                 CERROR("not connecting OSC %s; administratively "
 183                        "disabled\n", obd_uuid2str(tgt_uuid));
 184                 RETURN(0);
 185         }
 186
 187         rc = obd_connect(NULL, &lov->lov_tgts[index]->ltd_exp, tgt_obd,
 188                          &lov_osc_uuid, data, NULL);
 189         if (rc || !lov->lov_tgts[index]->ltd_exp) {
 190                 CERROR("Target %s connect error %d\n",
 191                        obd_uuid2str(tgt_uuid), rc);
 192                 RETURN(-ENODEV);
 193         }
 194
 195         lov->lov_tgts[index]->ltd_reap = 0;
 196
 197         CDEBUG(D_CONFIG, "Connected tgt idx %d %s (%s) %sactive\n", index,
 198                obd_uuid2str(tgt_uuid), tgt_obd->obd_name, activate ? "":"in");
 199
 200 #ifdef __KERNEL__
 201         lov_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
 202         if (lov_proc_dir) {
 203                 struct obd_device *osc_obd = lov->lov_tgts[index]->ltd_exp->exp_obd;
 204                 cfs_proc_dir_entry_t *osc_symlink;
 205
 206                 LASSERT(osc_obd != NULL);
 207                 LASSERT(osc_obd->obd_magic == OBD_DEVICE_MAGIC);
 208                 LASSERT(osc_obd->obd_type->typ_name != NULL);
 209
 210                 osc_symlink = lprocfs_add_symlink(osc_obd->obd_name,
 211                                                   lov_proc_dir,
 212                                                   "../../../%s/%s",
 213                                                   osc_obd->obd_type->typ_name,
 214                                                   osc_obd->obd_name);
 215                 if (osc_symlink == NULL) {
 216                         CERROR("could not register LOV target "
 217                                 "/proc/fs/lustre/%s/%s/target_obds/%s.",
 218                                 obd->obd_type->typ_name, obd->obd_name,
 219                                 osc_obd->obd_name);
 220                         lprocfs_remove(&lov_proc_dir);
 221                 }
 222         }
 223 #endif
 224
 225         rc = qos_add_tgt(obd, index);
 226         if (rc)
 227                 CERROR("qos_add_tgt failed %d\n", rc);
 228
 229         RETURN(0);
 230 }
 231
 232 static int lov_connect(const struct lu_env *env,
 233                        struct obd_export **exp, struct obd_device *obd,
 234                        struct obd_uuid *cluuid, struct obd_connect_data *data,
 235                        void *localdata)
 236 {
 237         struct lov_obd *lov = &obd->u.lov;
 238         struct lov_tgt_desc *tgt;
 239         struct lustre_handle conn;
 240         int i, rc;
 241         ENTRY;
 242
 243         CDEBUG(D_CONFIG, "connect #%d\n", lov->lov_connects);
 244
 245         rc = class_connect(&conn, obd, cluuid);
 246         if (rc)
 247                 RETURN(rc);
 248
 249         *exp = class_conn2export(&conn);
 250
 251         /* Why should there ever be more than 1 connect? */
 252         lov->lov_connects++;
 253         LASSERT(lov->lov_connects == 1);
 254
 255         memset(&lov->lov_ocd, 0, sizeof(lov->lov_ocd));
 256         if (data)
 257                 lov->lov_ocd = *data;
 258
 259         obd_getref(obd);
 260         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
 261                 tgt = lov->lov_tgts[i];
 262                 if (!tgt || obd_uuid_empty(&tgt->ltd_uuid))
 263                         continue;
 264                 /* Flags will be lowest common denominator */
 265                 rc = lov_connect_obd(obd, i, tgt->ltd_activate, &lov->lov_ocd);
 266                 if (rc) {
 267                         CERROR("%s: lov connect tgt %d failed: %d\n",
 268                                obd->obd_name, i, rc);
 269                         continue;
 270                 }
 271                 /* connect to administrative disabled ost */
 272                 if (!lov->lov_tgts[i]->ltd_exp)
 273                         continue;
 274
 275                 rc = lov_notify(obd, lov->lov_tgts[i]->ltd_exp->exp_obd,
 276                                 OBD_NOTIFY_CONNECT, (void *)&i);
 277                 if (rc) {
 278                         CERROR("%s error sending notify %d\n",
 279                                obd->obd_name, rc);
 280                 }
 281         }
 282         obd_putref(obd);
 283
 284         RETURN(0);
 285 }
 286
 287 static int lov_disconnect_obd(struct obd_device *obd, struct lov_tgt_desc *tgt)
 288 {
 289         cfs_proc_dir_entry_t *lov_proc_dir;
 290         struct lov_obd *lov = &obd->u.lov;
 291         struct obd_device *osc_obd;
 292         int rc;
 293         ENTRY;
 294
 295         osc_obd = class_exp2obd(tgt->ltd_exp);
 296         CDEBUG(D_CONFIG, "%s: disconnecting target %s\n",
 297                obd->obd_name, osc_obd->obd_name);
 298
 299         if (tgt->ltd_active) {
 300                 tgt->ltd_active = 0;
 301                 lov->desc.ld_active_tgt_count--;
 302                 tgt->ltd_exp->exp_obd->obd_inactive = 1;
 303         }
 304
 305         lov_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
 306         if (lov_proc_dir) {
 307                 cfs_proc_dir_entry_t *osc_symlink;
 308
 309                 osc_symlink = lprocfs_srch(lov_proc_dir, osc_obd->obd_name);
 310                 if (osc_symlink) {
 311                         lprocfs_remove(&osc_symlink);
 312                 } else {
 313                         CERROR("/proc/fs/lustre/%s/%s/target_obds/%s missing.",
 314                                obd->obd_type->typ_name, obd->obd_name,
 315                                osc_obd->obd_name);
 316                 }
 317         }
 318
 319         if (osc_obd) {
 320                 /* Pass it on to our clients.
 321                  * XXX This should be an argument to disconnect,
 322                  * XXX not a back-door flag on the OBD.  Ah well.
 323                  */
 324                 osc_obd->obd_force = obd->obd_force;
 325                 osc_obd->obd_fail = obd->obd_fail;
 326                 osc_obd->obd_no_recov = obd->obd_no_recov;
 327         }
 328
 329         obd_register_observer(osc_obd, NULL);
 330
 331         rc = obd_disconnect(tgt->ltd_exp);
 332         if (rc) {
 333                 CERROR("Target %s disconnect error %d\n",
 334                        tgt->ltd_uuid.uuid, rc);
 335                 rc = 0;
 336         }
 337
 338         qos_del_tgt(obd, tgt);
 339
 340         tgt->ltd_exp = NULL;
 341         RETURN(0);
 342 }
 343
 344 static int lov_disconnect(struct obd_export *exp)
 345 {
 346         struct obd_device *obd = class_exp2obd(exp);
 347         struct lov_obd *lov = &obd->u.lov;
 348         int i, rc;
 349         ENTRY;
 350
 351         if (!lov->lov_tgts)
 352                 goto out;
 353
 354         /* Only disconnect the underlying layers on the final disconnect. */
 355         lov->lov_connects--;
 356         if (lov->lov_connects != 0) {
 357                 /* why should there be more than 1 connect? */
 358                 CERROR("disconnect #%d\n", lov->lov_connects);
 359                 goto out;
 360         }
 361
 362         /* Let's hold another reference so lov_del_obd doesn't spin through
 363            putref every time */
 364         obd_getref(obd);
 365
 366         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
 367                 if (lov->lov_tgts[i] && lov->lov_tgts[i]->ltd_exp) {
 368                         /* Disconnection is the last we know about an obd */
 369                         lov_del_target(obd, i, 0, lov->lov_tgts[i]->ltd_gen);
 370                 }
 371         }
 372         obd_putref(obd);
 373
 374 out:
 375         rc = class_disconnect(exp); /* bz 9811 */
 376         RETURN(rc);
 377 }
 378
 379 /* Error codes:
 380  *
 381  *  -EINVAL  : UUID can't be found in the LOV's target list
 382  *  -ENOTCONN: The UUID is found, but the target connection is bad (!)
 383  *  -EBADF   : The UUID is found, but the OBD is the wrong type (!)
 384  *  any >= 0 : is log target index
 385  */
 386 static int lov_set_osc_active(struct obd_device *obd, struct obd_uuid *uuid,
 387                               int activate)
 388 {
 389         struct lov_obd *lov = &obd->u.lov;
 390         struct lov_tgt_desc *tgt;
 391         int index;
 392         ENTRY;
 393
 394         CDEBUG(D_INFO, "Searching in lov %p for uuid %s (activate=%d)\n",
 395                lov, uuid->uuid, activate);
 396
 397         obd_getref(obd);
 398         for (index = 0; index < lov->desc.ld_tgt_count; index++) {
 399                 tgt = lov->lov_tgts[index];
 400                 if (!tgt || !tgt->ltd_exp)
 401                         continue;
 402
 403                 CDEBUG(D_INFO, "lov idx %d is %s conn "LPX64"\n",
 404                        index, obd_uuid2str(&tgt->ltd_uuid),
 405                        tgt->ltd_exp->exp_handle.h_cookie);
 406                 if (obd_uuid_equals(uuid, &tgt->ltd_uuid))
 407                         break;
 408         }
 409
 410         if (index == lov->desc.ld_tgt_count)
 411                 GOTO(out, index = -EINVAL);
 412
 413         if (lov->lov_tgts[index]->ltd_active == activate) {
 414                 CDEBUG(D_INFO, "OSC %s already %sactive!\n", uuid->uuid,
 415                        activate ? "" : "in");
 416                 GOTO(out, index);
 417         }
 418
 419         CDEBUG(D_CONFIG, "Marking OSC %s %sactive\n", obd_uuid2str(uuid),
 420                activate ? "" : "in");
 421
 422         lov->lov_tgts[index]->ltd_active = activate;
 423
 424         if (activate) {
 425                 lov->desc.ld_active_tgt_count++;
 426                 lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 0;
 427         } else {
 428                 lov->desc.ld_active_tgt_count--;
 429                 lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 1;
 430         }
 431         /* remove any old qos penalty */
 432         lov->lov_tgts[index]->ltd_qos.ltq_penalty = 0;
 433
 434  out:
 435         obd_putref(obd);
 436         RETURN(index);
 437 }
 438
 439 static int lov_notify(struct obd_device *obd, struct obd_device *watched,
 440                       enum obd_notify_event ev, void *data)
 441 {
 442         int rc = 0;
 443         ENTRY;
 444
 445         if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE) {
 446                 struct obd_uuid *uuid;
 447
 448                 LASSERT(watched);
 449
 450                 if (strcmp(watched->obd_type->typ_name, LUSTRE_OSC_NAME)) {
 451                         CERROR("unexpected notification of %s %s!\n",
 452                                watched->obd_type->typ_name,
 453                                watched->obd_name);
 454                         RETURN(-EINVAL);
 455                 }
 456                 uuid = &watched->u.cli.cl_target_uuid;
 457
 458                 /* Set OSC as active before notifying the observer, so the
 459                  * observer can use the OSC normally.
 460                  */
 461                 rc = lov_set_osc_active(obd, uuid, ev == OBD_NOTIFY_ACTIVE);
 462                 if (rc < 0) {
 463                         CERROR("%sactivation of %s failed: %d\n",
 464                                (ev == OBD_NOTIFY_ACTIVE) ? "" : "de",
 465                                obd_uuid2str(uuid), rc);
 466                         RETURN(rc);
 467                 }
 468                 /* active event should be pass lov target index as data */
 469                 data = &rc;
 470         }
 471
 472         /* Pass the notification up the chain. */
 473         if (watched) {
 474                 rc = obd_notify_observer(obd, watched, ev, data);
 475         } else {
 476                 /* NULL watched means all osc's in the lov (only for syncs) */
 477                 /* sync event should be send lov idx as data */
 478                 struct lov_obd *lov = &obd->u.lov;
 479                 struct obd_device *tgt_obd;
 480                 int i;
 481                 obd_getref(obd);
 482                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
 483                         /* don't send sync event if target not
 484                          * connected/activated */
 485                         if (!lov->lov_tgts[i] ||
 486                             !lov->lov_tgts[i]->ltd_active)
 487                                  continue;
 488
 489                         if ((ev == OBD_NOTIFY_SYNC) ||
 490                             (ev == OBD_NOTIFY_SYNC_NONBLOCK))
 491                                 data = &i;
 492
 493                         tgt_obd = class_exp2obd(lov->lov_tgts[i]->ltd_exp);
 494
 495                         rc = obd_notify_observer(obd, tgt_obd, ev, data);
 496                         if (rc) {
 497                                 CERROR("%s: notify %s of %s failed %d\n",
 498                                        obd->obd_name,
 499                                        obd->obd_observer->obd_name,
 500                                        tgt_obd->obd_name, rc);
 501                                 break;
 502                         }
 503                 }
 504                 obd_putref(obd);
 505         }
 506
 507         RETURN(rc);
 508 }
 509
 510 int lov_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
 511                    __u32 index, int gen, int active)
 512 {
 513         struct lov_obd *lov = &obd->u.lov;
 514         struct lov_tgt_desc *tgt;
 515         struct obd_device *tgt_obd;
 516         int rc;
 517         ENTRY;
 518
 519         CDEBUG(D_CONFIG, "uuid:%s idx:%d gen:%d active:%d\n",
 520                uuidp->uuid, index, gen, active);
 521
 522         if (gen <= 0) {
 523                 CERROR("request to add OBD %s with invalid generation: %d\n",
 524                        uuidp->uuid, gen);
 525                 RETURN(-EINVAL);
 526         }
 527
 528         tgt_obd = class_find_client_obd(uuidp, LUSTRE_OSC_NAME,
 529                                         &obd->obd_uuid);
 530         if (tgt_obd == NULL)
 531                 RETURN(-EINVAL);
 532
 533         cfs_mutex_down(&lov->lov_lock);
 534
 535         if ((index < lov->lov_tgt_size) && (lov->lov_tgts[index] != NULL)) {
 536                 tgt = lov->lov_tgts[index];
 537                 CERROR("UUID %s already assigned at LOV target index %d\n",
 538                        obd_uuid2str(&tgt->ltd_uuid), index);
 539                 cfs_mutex_up(&lov->lov_lock);
 540                 RETURN(-EEXIST);
 541         }
 542
 543         if (index >= lov->lov_tgt_size) {
 544                 /* We need to reallocate the lov target array. */
 545                 struct lov_tgt_desc **newtgts, **old = NULL;
 546                 __u32 newsize, oldsize = 0;
 547
 548                 newsize = max(lov->lov_tgt_size, (__u32)2);
 549                 while (newsize < index + 1)
 550                         newsize = newsize << 1;
 551                 OBD_ALLOC(newtgts, sizeof(*newtgts) * newsize);
 552                 if (newtgts == NULL) {
 553                         cfs_mutex_up(&lov->lov_lock);
 554                         RETURN(-ENOMEM);
 555                 }
 556
 557                 if (lov->lov_tgt_size) {
 558                         memcpy(newtgts, lov->lov_tgts, sizeof(*newtgts) *
 559                                lov->lov_tgt_size);
 560                         old = lov->lov_tgts;
 561                         oldsize = lov->lov_tgt_size;
 562                 }
 563
 564                 lov->lov_tgts = newtgts;
 565                 lov->lov_tgt_size = newsize;
 566 #ifdef __KERNEL__
 567                 smp_rmb();
 568 #endif
 569                 if (old)
 570                         OBD_FREE(old, sizeof(*old) * oldsize);
 571
 572                 CDEBUG(D_CONFIG, "tgts: %p size: %d\n",
 573                        lov->lov_tgts, lov->lov_tgt_size);
 574         }
 575
 576         OBD_ALLOC_PTR(tgt);
 577         if (!tgt) {
 578                 cfs_mutex_up(&lov->lov_lock);
 579                 RETURN(-ENOMEM);
 580         }
 581
 582         rc = lov_ost_pool_add(&lov->lov_packed, index, lov->lov_tgt_size);
 583         if (rc) {
 584                 cfs_mutex_up(&lov->lov_lock);
 585                 OBD_FREE_PTR(tgt);
 586                 RETURN(rc);
 587         }
 588
 589         memset(tgt, 0, sizeof(*tgt));
 590         tgt->ltd_uuid = *uuidp;
 591         tgt->ltd_obd = tgt_obd;
 592         /* XXX - add a sanity check on the generation number. */
 593         tgt->ltd_gen = gen;
 594         tgt->ltd_index = index;
 595         tgt->ltd_activate = active;
 596         lov->lov_tgts[index] = tgt;
 597         if (index >= lov->desc.ld_tgt_count)
 598                 lov->desc.ld_tgt_count = index + 1;
 599
 600         cfs_mutex_up(&lov->lov_lock);
 601
 602         CDEBUG(D_CONFIG, "idx=%d ltd_gen=%d ld_tgt_count=%d\n",
 603                 index, tgt->ltd_gen, lov->desc.ld_tgt_count);
 604
 605         if (lov->lov_connects == 0) {
 606                 /* lov_connect hasn't been called yet. We'll do the
 607                    lov_connect_obd on this target when that fn first runs,
 608                    because we don't know the connect flags yet. */
 609                 RETURN(0);
 610         }
 611
 612         obd_getref(obd);
 613
 614         rc = lov_connect_obd(obd, index, active, &lov->lov_ocd);
 615         if (rc)
 616                 GOTO(out, rc);
 617
 618         /* connect to administrative disabled ost */
 619         if (!tgt->ltd_exp)
 620                 GOTO(out, rc = 0);
 621
 622         rc = lov_notify(obd, tgt->ltd_exp->exp_obd,
 623                         active ? OBD_NOTIFY_CONNECT : OBD_NOTIFY_INACTIVE,
 624                         (void *)&index);
 625
 626 out:
 627         if (rc) {
 628                 CERROR("add failed (%d), deleting %s\n", rc,
 629                        obd_uuid2str(&tgt->ltd_uuid));
 630                 lov_del_target(obd, index, 0, 0);
 631         }
 632         obd_putref(obd);
 633         RETURN(rc);
 634 }
 635
 636 /* Schedule a target for deletion */
 637 int lov_del_target(struct obd_device *obd, __u32 index,
 638                    struct obd_uuid *uuidp, int gen)
 639 {
 640         struct lov_obd *lov = &obd->u.lov;
 641         int count = lov->desc.ld_tgt_count;
 642         int rc = 0;
 643         ENTRY;
 644
 645         if (index >= count) {
 646                 CERROR("LOV target index %d >= number of LOV OBDs %d.\n",
 647                        index, count);
 648                 RETURN(-EINVAL);
 649         }
 650
 651         obd_getref(obd);
 652
 653         if (!lov->lov_tgts[index]) {
 654                 CERROR("LOV target at index %d is not setup.\n", index);
 655                 GOTO(out, rc = -EINVAL);
 656         }
 657
 658         if (uuidp && !obd_uuid_equals(uuidp, &lov->lov_tgts[index]->ltd_uuid)) {
 659                 CERROR("LOV target UUID %s at index %d doesn't match %s.\n",
 660                        lov_uuid2str(lov, index), index,
 661                        obd_uuid2str(uuidp));
 662                 GOTO(out, rc = -EINVAL);
 663         }
 664
 665         CDEBUG(D_CONFIG, "uuid: %s idx: %d gen: %d exp: %p active: %d\n",
 666                lov_uuid2str(lov, index), index,
 667                lov->lov_tgts[index]->ltd_gen, lov->lov_tgts[index]->ltd_exp,
 668                lov->lov_tgts[index]->ltd_active);
 669
 670         lov->lov_tgts[index]->ltd_reap = 1;
 671         lov->lov_death_row++;
 672         /* we really delete it from obd_putref */
 673 out:
 674         obd_putref(obd);
 675
 676         RETURN(rc);
 677 }
 678
 679 static void __lov_del_obd(struct obd_device *obd, struct lov_tgt_desc *tgt)
 680 {
 681         struct obd_device *osc_obd;
 682
 683         LASSERT(tgt);
 684         LASSERT(tgt->ltd_reap);
 685
 686         osc_obd = class_exp2obd(tgt->ltd_exp);
 687
 688         CDEBUG(D_CONFIG, "Removing tgt %s : %s\n",
 689                tgt->ltd_uuid.uuid,
 690                osc_obd ? osc_obd->obd_name : "<no obd>");
 691
 692         if (tgt->ltd_exp)
 693                 lov_disconnect_obd(obd, tgt);
 694
 695         OBD_FREE_PTR(tgt);
 696
 697         /* Manual cleanup - no cleanup logs to clean up the osc's.  We must
 698            do it ourselves. And we can't do it from lov_cleanup,
 699            because we just lost our only reference to it. */
 700         if (osc_obd)
 701                 class_manual_cleanup(osc_obd);
 702 }
 703
 704 void lov_fix_desc_stripe_size(__u64 *val)
 705 {
 706         if (*val < PTLRPC_MAX_BRW_SIZE) {
 707                 LCONSOLE_WARN("Increasing default stripe size to min %u\n",
 708                               PTLRPC_MAX_BRW_SIZE);
 709                 *val = PTLRPC_MAX_BRW_SIZE;
 710         } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) {
 711                 *val &= ~(LOV_MIN_STRIPE_SIZE - 1);
 712                 LCONSOLE_WARN("Changing default stripe size to "LPU64" (a "
 713                               "multiple of %u)\n",
 714                               *val, LOV_MIN_STRIPE_SIZE);
 715         }
 716 }
 717
 718 void lov_fix_desc_stripe_count(__u32 *val)
 719 {
 720         if (*val == 0)
 721                 *val = 1;
 722 }
 723
 724 void lov_fix_desc_pattern(__u32 *val)
 725 {
 726         /* from lov_setstripe */
 727         if ((*val != 0) && (*val != LOV_PATTERN_RAID0)) {
 728                 LCONSOLE_WARN("Unknown stripe pattern: %#x\n", *val);
 729                 *val = 0;
 730         }
 731 }
 732
 733 void lov_fix_desc_qos_maxage(__u32 *val)
 734 {
 735         /* fix qos_maxage */
 736         if (*val == 0)
 737                 *val = QOS_DEFAULT_MAXAGE;
 738 }
 739
 740 void lov_fix_desc(struct lov_desc *desc)
 741 {
 742         lov_fix_desc_stripe_size(&desc->ld_default_stripe_size);
 743         lov_fix_desc_stripe_count(&desc->ld_default_stripe_count);
 744         lov_fix_desc_pattern(&desc->ld_pattern);
 745         lov_fix_desc_qos_maxage(&desc->ld_qos_maxage);
 746 }
 747
 748 int lov_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
 749 {
 750         struct lprocfs_static_vars lvars = { 0 };
 751         struct lov_desc *desc;
 752         struct lov_obd *lov = &obd->u.lov;
 753         int rc;
 754         ENTRY;
 755
 756         if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
 757                 CERROR("LOV setup requires a descriptor\n");
 758                 RETURN(-EINVAL);
 759         }
 760
 761         desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
 762
 763         if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
 764                 CERROR("descriptor size wrong: %d > %d\n",
 765                        (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
 766                 RETURN(-EINVAL);
 767         }
 768
 769         if (desc->ld_magic != LOV_DESC_MAGIC) {
 770                 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
 771                             CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
 772                                    obd->obd_name, desc);
 773                             lustre_swab_lov_desc(desc);
 774                 } else {
 775                         CERROR("%s: Bad lov desc magic: %#x\n",
 776                                obd->obd_name, desc->ld_magic);
 777                         RETURN(-EINVAL);
 778                 }
 779         }
 780
 781         lov_fix_desc(desc);
 782
 783         desc->ld_active_tgt_count = 0;
 784         lov->desc = *desc;
 785         lov->lov_tgt_size = 0;
 786
 787         cfs_sema_init(&lov->lov_lock, 1);
 788         cfs_atomic_set(&lov->lov_refcount, 0);
 789         CFS_INIT_LIST_HEAD(&lov->lov_qos.lq_oss_list);
 790         cfs_init_rwsem(&lov->lov_qos.lq_rw_sem);
 791         lov->lov_sp_me = LUSTRE_SP_CLI;
 792         lov->lov_qos.lq_dirty = 1;
 793         lov->lov_qos.lq_rr.lqr_dirty = 1;
 794         lov->lov_qos.lq_reset = 1;
 795         /* Default priority is toward free space balance */
 796         lov->lov_qos.lq_prio_free = 232;
 797         /* Default threshold for rr (roughly 17%) */
 798         lov->lov_qos.lq_threshold_rr = 43;
 799         /* Init statfs fields */
 800         OBD_ALLOC_PTR(lov->lov_qos.lq_statfs_data);
 801         if (NULL == lov->lov_qos.lq_statfs_data)
 802                 RETURN(-ENOMEM);
 803         cfs_waitq_init(&lov->lov_qos.lq_statfs_waitq);
 804
 805         lov->lov_pools_hash_body = cfs_hash_create("POOLS", HASH_POOLS_CUR_BITS,
 806                                                    HASH_POOLS_MAX_BITS,
 807                                                    &pool_hash_operations, CFS_HASH_REHASH);
 808         CFS_INIT_LIST_HEAD(&lov->lov_pool_list);
 809         lov->lov_pool_count = 0;
 810         rc = lov_ost_pool_init(&lov->lov_packed, 0);
 811         if (rc)
 812                 RETURN(rc);
 813         rc = lov_ost_pool_init(&lov->lov_qos.lq_rr.lqr_pool, 0);
 814         if (rc) {
 815                 lov_ost_pool_free(&lov->lov_packed);
 816                 RETURN(rc);
 817         }
 818
 819         lprocfs_lov_init_vars(&lvars);
 820         lprocfs_obd_setup(obd, lvars.obd_vars);
 821 #ifdef LPROCFS
 822         {
 823                 int rc;
 824
 825                 rc = lprocfs_seq_create(obd->obd_proc_entry, "target_obd",
 826                                         0444, &lov_proc_target_fops, obd);
 827                 if (rc)
 828                         CWARN("Error adding the target_obd file\n");
 829         }
 830 #endif
 831         lov->lov_pool_proc_entry = lprocfs_register("pools",
 832                                                     obd->obd_proc_entry,
 833                                                     NULL, NULL);
 834
 835         RETURN(0);
 836 }
 837
 838 static int lov_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
 839 {
 840         int rc = 0;
 841         struct lov_obd *lov = &obd->u.lov;
 842
 843         ENTRY;
 844
 845         switch (stage) {
 846         case OBD_CLEANUP_EARLY: {
 847                 int i;
 848                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
 849                         if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_active)
 850                                 continue;
 851                         obd_precleanup(class_exp2obd(lov->lov_tgts[i]->ltd_exp),
 852                                        OBD_CLEANUP_EARLY);
 853                 }
 854                 break;
 855         }
 856         case OBD_CLEANUP_EXPORTS:
 857                 rc = obd_llog_finish(obd, 0);
 858                 if (rc != 0)
 859                         CERROR("failed to cleanup llogging subsystems\n");
 860                 break;
 861         }
 862         RETURN(rc);
 863 }
 864
 865 static int lov_cleanup(struct obd_device *obd)
 866 {
 867         struct lov_obd *lov = &obd->u.lov;
 868         cfs_list_t *pos, *tmp;
 869         struct pool_desc *pool;
 870
 871         cfs_list_for_each_safe(pos, tmp, &lov->lov_pool_list) {
 872                 pool = cfs_list_entry(pos, struct pool_desc, pool_list);
 873                 /* free pool structs */
 874                 CDEBUG(D_INFO, "delete pool %p\n", pool);
 875                 lov_pool_del(obd, pool->pool_name);
 876         }
 877         cfs_hash_putref(lov->lov_pools_hash_body);
 878         lov_ost_pool_free(&(lov->lov_qos.lq_rr.lqr_pool));
 879         lov_ost_pool_free(&lov->lov_packed);
 880
 881         if (lov->lov_tgts) {
 882                 int i;
 883                 obd_getref(obd);
 884                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
 885                         if (!lov->lov_tgts[i])
 886                                 continue;
 887
 888                         /* Inactive targets may never have connected */
 889                         if (lov->lov_tgts[i]->ltd_active ||
 890                             cfs_atomic_read(&lov->lov_refcount))
 891                             /* We should never get here - these
 892                                should have been removed in the
 893                              disconnect. */
 894                                 CERROR("lov tgt %d not cleaned!"
 895                                        " deathrow=%d, lovrc=%d\n",
 896                                        i, lov->lov_death_row,
 897                                        cfs_atomic_read(&lov->lov_refcount));
 898                         lov_del_target(obd, i, 0, 0);
 899                 }
 900                 obd_putref(obd);
 901                 OBD_FREE(lov->lov_tgts, sizeof(*lov->lov_tgts) *
 902                          lov->lov_tgt_size);
 903                 lov->lov_tgt_size = 0;
 904         }
 905
 906         /* clear pools parent proc entry only after all pools is killed */
 907         lprocfs_obd_cleanup(obd);
 908
 909         OBD_FREE_PTR(lov->lov_qos.lq_statfs_data);
 910         RETURN(0);
 911 }
 912
 913 int lov_process_config_base(struct obd_device *obd, struct lustre_cfg *lcfg,
 914                             __u32 *indexp, int *genp)
 915 {
 916         struct obd_uuid obd_uuid;
 917         int cmd;
 918         int rc = 0;
 919         ENTRY;
 920
 921         switch(cmd = lcfg->lcfg_command) {
 922         case LCFG_LOV_ADD_OBD:
 923         case LCFG_LOV_ADD_INA:
 924         case LCFG_LOV_DEL_OBD: {
 925                 __u32 index;
 926                 int gen;
 927                 /* lov_modify_tgts add  0:lov_mdsA  1:ost1_UUID  2:0  3:1 */
 928                 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
 929                         GOTO(out, rc = -EINVAL);
 930
 931                 obd_str2uuid(&obd_uuid,  lustre_cfg_buf(lcfg, 1));
 932
 933                 if (sscanf(lustre_cfg_buf(lcfg, 2), "%d", indexp) != 1)
 934                         GOTO(out, rc = -EINVAL);
 935                 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", genp) != 1)
 936                         GOTO(out, rc = -EINVAL);
 937                 index = *indexp;
 938                 gen = *genp;
 939                 if (cmd == LCFG_LOV_ADD_OBD)
 940                         rc = lov_add_target(obd, &obd_uuid, index, gen, 1);
 941                 else if (cmd == LCFG_LOV_ADD_INA)
 942                         rc = lov_add_target(obd, &obd_uuid, index, gen, 0);
 943                 else
 944                         rc = lov_del_target(obd, index, &obd_uuid, gen);
 945                 GOTO(out, rc);
 946         }
 947         case LCFG_PARAM: {
 948                 struct lprocfs_static_vars lvars = { 0 };
 949                 struct lov_desc *desc = &(obd->u.lov.desc);
 950
 951                 if (!desc)
 952                         GOTO(out, rc = -EINVAL);
 953
 954                 lprocfs_lov_init_vars(&lvars);
 955
 956                 rc = class_process_proc_param(PARAM_LOV, lvars.obd_vars,
 957                                               lcfg, obd);
 958                 if (rc > 0)
 959                         rc = 0;
 960                 GOTO(out, rc);
 961         }
 962         case LCFG_POOL_NEW:
 963         case LCFG_POOL_ADD:
 964         case LCFG_POOL_DEL:
 965         case LCFG_POOL_REM:
 966                 GOTO(out, rc);
 967
 968         default: {
 969                 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
 970                 GOTO(out, rc = -EINVAL);
 971
 972         }
 973         }
 974 out:
 975         RETURN(rc);
 976 }
 977
 978 #ifndef log2
 979 #define log2(n) cfs_ffz(~(n))
 980 #endif
 981
 982 static int lov_clear_orphans(struct obd_export *export, struct obdo *src_oa,
 983                              struct lov_stripe_md **ea,
 984                              struct obd_trans_info *oti)
 985 {
 986         struct lov_obd *lov;
 987         struct obdo *tmp_oa;
 988         struct obd_uuid *ost_uuid = NULL;
 989         int rc = 0, i;
 990         ENTRY;
 991
 992         LASSERT(src_oa->o_valid & OBD_MD_FLFLAGS &&
 993                 src_oa->o_flags == OBD_FL_DELORPHAN);
 994
 995         lov = &export->exp_obd->u.lov;
 996
 997         OBDO_ALLOC(tmp_oa);
 998         if (tmp_oa == NULL)
 999                 RETURN(-ENOMEM);
1000
1001         if (oti->oti_ost_uuid) {
1002                 ost_uuid = oti->oti_ost_uuid;
1003                 CDEBUG(D_HA, "clearing orphans only for %s\n",
1004                        ost_uuid->uuid);
1005         }
1006
1007         obd_getref(export->exp_obd);
1008         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
1009                 struct lov_stripe_md obj_md;
1010                 struct lov_stripe_md *obj_mdp = &obj_md;
1011                 struct lov_tgt_desc *tgt;
1012                 int err;
1013
1014                 tgt = lov->lov_tgts[i];
1015                 if (!tgt)
1016                         continue;
1017
1018                 /* if called for a specific target, we don't
1019                    care if it is not active. */
1020                 if (!lov->lov_tgts[i]->ltd_active && ost_uuid == NULL) {
1021                         CDEBUG(D_HA, "lov idx %d inactive\n", i);
1022                         continue;
1023                 }
1024
1025                 if (ost_uuid && !obd_uuid_equals(ost_uuid, &tgt->ltd_uuid))
1026                         continue;
1027
1028                 CDEBUG(D_CONFIG,"Clear orphans for %d:%s\n", i,
1029                        obd_uuid2str(ost_uuid));
1030
1031                 memcpy(tmp_oa, src_oa, sizeof(*tmp_oa));
1032
1033                 LASSERT(lov->lov_tgts[i]->ltd_exp);
1034                 /* XXX: LOV STACKING: use real "obj_mdp" sub-data */
1035                 err = obd_create(lov->lov_tgts[i]->ltd_exp,
1036                                  tmp_oa, &obj_mdp, oti);
1037                 if (err) {
1038                         /* This export will be disabled until it is recovered,
1039                            and then orphan recovery will be completed. */
1040                         CERROR("error in orphan recovery on OST idx %d/%d: "
1041                                "rc = %d\n", i, lov->desc.ld_tgt_count, err);
1042                         rc = err;
1043                 }
1044
1045                 if (ost_uuid)
1046                         break;
1047         }
1048         obd_putref(export->exp_obd);
1049
1050         OBDO_FREE(tmp_oa);
1051         RETURN(rc);
1052 }
1053
1054 static int lov_recreate(struct obd_export *exp, struct obdo *src_oa,
1055                         struct lov_stripe_md **ea, struct obd_trans_info *oti)
1056 {
1057         struct lov_stripe_md *obj_mdp, *lsm;
1058         struct lov_obd *lov = &exp->exp_obd->u.lov;
1059         unsigned ost_idx;
1060         int rc, i;
1061         ENTRY;
1062
1063         LASSERT(src_oa->o_valid & OBD_MD_FLFLAGS &&
1064                 src_oa->o_flags & OBD_FL_RECREATE_OBJS);
1065
1066         OBD_ALLOC(obj_mdp, sizeof(*obj_mdp));
1067         if (obj_mdp == NULL)
1068                 RETURN(-ENOMEM);
1069
1070         ost_idx = src_oa->o_nlink;
1071         lsm = *ea;
1072         if (lsm == NULL)
1073                 GOTO(out, rc = -EINVAL);
1074         if (ost_idx >= lov->desc.ld_tgt_count ||
1075             !lov->lov_tgts[ost_idx])
1076                 GOTO(out, rc = -EINVAL);
1077
1078         for (i = 0; i < lsm->lsm_stripe_count; i++) {
1079                 if (lsm->lsm_oinfo[i]->loi_ost_idx == ost_idx) {
1080                         if (lsm->lsm_oinfo[i]->loi_id != src_oa->o_id)
1081                                 GOTO(out, rc = -EINVAL);
1082                         break;
1083                 }
1084         }
1085         if (i == lsm->lsm_stripe_count)
1086                 GOTO(out, rc = -EINVAL);
1087
1088         rc = obd_create(lov->lov_tgts[ost_idx]->ltd_exp, src_oa, &obj_mdp, oti);
1089 out:
1090         OBD_FREE(obj_mdp, sizeof(*obj_mdp));
1091         RETURN(rc);
1092 }
1093
1094 /* the LOV expects oa->o_id to be set to the LOV object id */
1095 static int lov_create(struct obd_export *exp, struct obdo *src_oa,
1096                       struct lov_stripe_md **ea, struct obd_trans_info *oti)
1097 {
1098         struct lov_obd *lov;
1099         struct obd_info oinfo;
1100         struct lov_request_set *set = NULL;
1101         struct lov_request *req;
1102         struct l_wait_info  lwi = { 0 };
1103         int rc = 0;
1104         ENTRY;
1105
1106         LASSERT(ea != NULL);
1107         if (exp == NULL)
1108                 RETURN(-EINVAL);
1109
1110         if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
1111             src_oa->o_flags == OBD_FL_DELORPHAN) {
1112                 rc = lov_clear_orphans(exp, src_oa, ea, oti);
1113                 RETURN(rc);
1114         }
1115
1116         lov = &exp->exp_obd->u.lov;
1117         if (!lov->desc.ld_active_tgt_count)
1118                 RETURN(-EIO);
1119
1120         obd_getref(exp->exp_obd);
1121         /* Recreate a specific object id at the given OST index */
1122         if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
1123             (src_oa->o_flags & OBD_FL_RECREATE_OBJS)) {
1124                  rc = lov_recreate(exp, src_oa, ea, oti);
1125                  GOTO(out, rc);
1126         }
1127
1128         /* issue statfs rpcs if the osfs data is older than qos_maxage - 1s,
1129          * later in alloc_qos(), we will wait for those rpcs to complete if
1130          * the osfs age is older than 2 * qos_maxage */
1131         qos_statfs_update(exp->exp_obd,
1132                           cfs_time_shift_64(-lov->desc.ld_qos_maxage +
1133                                             OBD_STATFS_CACHE_SECONDS),
1134                           0);
1135
1136         rc = lov_prep_create_set(exp, &oinfo, ea, src_oa, oti, &set);
1137         if (rc)
1138                 GOTO(out, rc);
1139
1140         cfs_list_for_each_entry(req, &set->set_list, rq_link) {
1141                 /* XXX: LOV STACKING: use real "obj_mdp" sub-data */
1142                 rc = obd_create_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1143                                       &req->rq_oi, &req->rq_oi.oi_md, oti);
1144         }
1145
1146         /* osc_create have timeout equ obd_timeout/2 so waiting don't be
1147          * longer then this */
1148         l_wait_event(set->set_waitq, lov_finished_set(set), &lwi);
1149
1150         /* we not have ptlrpc set for assign set->interpret and should
1151          * be call interpret function himself. calling from cb_create_update
1152          * not permited because lov_fini_create_set can sleep for long time,
1153          * but we must avoid sleeping in ptlrpcd interpret function. */
1154         rc = lov_fini_create_set(set, ea);
1155 out:
1156         obd_putref(exp->exp_obd);
1157         RETURN(rc);
1158 }
1159
1160 #define ASSERT_LSM_MAGIC(lsmp)                                                  \
1161 do {                                                                            \
1162         LASSERT((lsmp) != NULL);                                                \
1163         LASSERTF(((lsmp)->lsm_magic == LOV_MAGIC_V1 ||                          \
1164                  (lsmp)->lsm_magic == LOV_MAGIC_V3),                            \
1165                  "%p->lsm_magic=%x\n", (lsmp), (lsmp)->lsm_magic);              \
1166 } while (0)
1167
1168 static int lov_destroy(struct obd_export *exp, struct obdo *oa,
1169                        struct lov_stripe_md *lsm, struct obd_trans_info *oti,
1170                        struct obd_export *md_exp, void *capa)
1171 {
1172         struct lov_request_set *set;
1173         struct obd_info oinfo;
1174         struct lov_request *req;
1175         cfs_list_t *pos;
1176         struct lov_obd *lov;
1177         int rc = 0, err = 0;
1178         ENTRY;
1179
1180         ASSERT_LSM_MAGIC(lsm);
1181
1182         if (!exp || !exp->exp_obd)
1183                 RETURN(-ENODEV);
1184
1185         if (oa->o_valid & OBD_MD_FLCOOKIE) {
1186                 LASSERT(oti);
1187                 LASSERT(oti->oti_logcookies);
1188         }
1189
1190         lov = &exp->exp_obd->u.lov;
1191         obd_getref(exp->exp_obd);
1192         rc = lov_prep_destroy_set(exp, &oinfo, oa, lsm, oti, &set);
1193         if (rc)
1194                 GOTO(out, rc);
1195
1196         cfs_list_for_each (pos, &set->set_list) {
1197                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1198
1199                 if (oa->o_valid & OBD_MD_FLCOOKIE)
1200                         oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1201
1202                 err = obd_destroy(lov->lov_tgts[req->rq_idx]->ltd_exp,
1203                                   req->rq_oi.oi_oa, NULL, oti, NULL, capa);
1204                 err = lov_update_common_set(set, req, err);
1205                 if (err) {
1206                         CERROR("error: destroying objid "LPX64" subobj "
1207                                LPX64" on OST idx %d: rc = %d\n",
1208                                oa->o_id, req->rq_oi.oi_oa->o_id,
1209                                req->rq_idx, err);
1210                         if (!rc)
1211                                 rc = err;
1212                 }
1213         }
1214
1215         if (rc == 0) {
1216                 LASSERT(lsm_op_find(lsm->lsm_magic) != NULL);
1217                 rc = lsm_op_find(lsm->lsm_magic)->lsm_destroy(lsm, oa, md_exp);
1218         }
1219         err = lov_fini_destroy_set(set);
1220 out:
1221         obd_putref(exp->exp_obd);
1222         RETURN(rc ? rc : err);
1223 }
1224
1225 static int lov_getattr(struct obd_export *exp, struct obd_info *oinfo)
1226 {
1227         struct lov_request_set *set;
1228         struct lov_request *req;
1229         cfs_list_t *pos;
1230         struct lov_obd *lov;
1231         int err = 0, rc = 0;
1232         ENTRY;
1233
1234         LASSERT(oinfo);
1235         ASSERT_LSM_MAGIC(oinfo->oi_md);
1236
1237         if (!exp || !exp->exp_obd)
1238                 RETURN(-ENODEV);
1239
1240         lov = &exp->exp_obd->u.lov;
1241
1242         rc = lov_prep_getattr_set(exp, oinfo, &set);
1243         if (rc)
1244                 RETURN(rc);
1245
1246         cfs_list_for_each (pos, &set->set_list) {
1247                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1248
1249                 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
1250                        "%u\n", oinfo->oi_oa->o_id, req->rq_stripe,
1251                        req->rq_oi.oi_oa->o_id, req->rq_idx);
1252
1253                 rc = obd_getattr(lov->lov_tgts[req->rq_idx]->ltd_exp,
1254                                  &req->rq_oi);
1255                 err = lov_update_common_set(set, req, rc);
1256                 if (err) {
1257                         CERROR("error: getattr objid "LPX64" subobj "
1258                                LPX64" on OST idx %d: rc = %d\n",
1259                                oinfo->oi_oa->o_id, req->rq_oi.oi_oa->o_id,
1260                                req->rq_idx, err);
1261                         break;
1262                 }
1263         }
1264
1265         rc = lov_fini_getattr_set(set);
1266         if (err)
1267                 rc = err;
1268         RETURN(rc);
1269 }
1270
1271 static int lov_getattr_interpret(struct ptlrpc_request_set *rqset,
1272                                  void *data, int rc)
1273 {
1274         struct lov_request_set *lovset = (struct lov_request_set *)data;
1275         int err;
1276         ENTRY;
1277
1278         /* don't do attribute merge if this aysnc op failed */
1279         if (rc)
1280                 lovset->set_completes = 0;
1281         err = lov_fini_getattr_set(lovset);
1282         RETURN(rc ? rc : err);
1283 }
1284
1285 static int lov_getattr_async(struct obd_export *exp, struct obd_info *oinfo,
1286                               struct ptlrpc_request_set *rqset)
1287 {
1288         struct lov_request_set *lovset;
1289         struct lov_obd *lov;
1290         cfs_list_t *pos;
1291         struct lov_request *req;
1292         int rc = 0, err;
1293         ENTRY;
1294
1295         LASSERT(oinfo);
1296         ASSERT_LSM_MAGIC(oinfo->oi_md);
1297
1298         if (!exp || !exp->exp_obd)
1299                 RETURN(-ENODEV);
1300
1301         lov = &exp->exp_obd->u.lov;
1302
1303         rc = lov_prep_getattr_set(exp, oinfo, &lovset);
1304         if (rc)
1305                 RETURN(rc);
1306
1307         CDEBUG(D_INFO, "objid "LPX64": %ux%u byte stripes\n",
1308                oinfo->oi_md->lsm_object_id, oinfo->oi_md->lsm_stripe_count,
1309                oinfo->oi_md->lsm_stripe_size);
1310
1311         cfs_list_for_each (pos, &lovset->set_list) {
1312                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1313
1314                 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
1315                        "%u\n", oinfo->oi_oa->o_id, req->rq_stripe,
1316                        req->rq_oi.oi_oa->o_id, req->rq_idx);
1317                 rc = obd_getattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1318                                        &req->rq_oi, rqset);
1319                 if (rc) {
1320                         CERROR("error: getattr objid "LPX64" subobj "
1321                                LPX64" on OST idx %d: rc = %d\n",
1322                                oinfo->oi_oa->o_id, req->rq_oi.oi_oa->o_id,
1323                                req->rq_idx, rc);
1324                         GOTO(out, rc);
1325                 }
1326         }
1327
1328         if (!cfs_list_empty(&rqset->set_requests)) {
1329                 LASSERT(rc == 0);
1330                 LASSERT (rqset->set_interpret == NULL);
1331                 rqset->set_interpret = lov_getattr_interpret;
1332                 rqset->set_arg = (void *)lovset;
1333                 RETURN(rc);
1334         }
1335 out:
1336         if (rc)
1337                 lovset->set_completes = 0;
1338         err = lov_fini_getattr_set(lovset);
1339         RETURN(rc ? rc : err);
1340 }
1341
1342 static int lov_setattr(struct obd_export *exp, struct obd_info *oinfo,
1343                        struct obd_trans_info *oti)
1344 {
1345         struct lov_request_set *set;
1346         struct lov_obd *lov;
1347         cfs_list_t *pos;
1348         struct lov_request *req;
1349         int err = 0, rc = 0;
1350         ENTRY;
1351
1352         LASSERT(oinfo);
1353         ASSERT_LSM_MAGIC(oinfo->oi_md);
1354
1355         if (!exp || !exp->exp_obd)
1356                 RETURN(-ENODEV);
1357
1358         /* for now, we only expect the following updates here */
1359         LASSERT(!(oinfo->oi_oa->o_valid & ~(OBD_MD_FLID | OBD_MD_FLTYPE |
1360                                             OBD_MD_FLMODE | OBD_MD_FLATIME |
1361                                             OBD_MD_FLMTIME | OBD_MD_FLCTIME |
1362                                             OBD_MD_FLFLAGS | OBD_MD_FLSIZE |
1363                                             OBD_MD_FLGROUP | OBD_MD_FLUID |
1364                                             OBD_MD_FLGID | OBD_MD_FLFID |
1365                                             OBD_MD_FLGENER)));
1366         lov = &exp->exp_obd->u.lov;
1367         rc = lov_prep_setattr_set(exp, oinfo, oti, &set);
1368         if (rc)
1369                 RETURN(rc);
1370
1371         cfs_list_for_each (pos, &set->set_list) {
1372                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1373
1374                 rc = obd_setattr(lov->lov_tgts[req->rq_idx]->ltd_exp,
1375                                  &req->rq_oi, NULL);
1376                 err = lov_update_setattr_set(set, req, rc);
1377                 if (err) {
1378                         CERROR("error: setattr objid "LPX64" subobj "
1379                                LPX64" on OST idx %d: rc = %d\n",
1380                                set->set_oi->oi_oa->o_id,
1381                                req->rq_oi.oi_oa->o_id, req->rq_idx, err);
1382                         if (!rc)
1383                                 rc = err;
1384                 }
1385         }
1386         err = lov_fini_setattr_set(set);
1387         if (!rc)
1388                 rc = err;
1389         RETURN(rc);
1390 }
1391
1392 static int lov_setattr_interpret(struct ptlrpc_request_set *rqset,
1393                                  void *data, int rc)
1394 {
1395         struct lov_request_set *lovset = (struct lov_request_set *)data;
1396         int err;
1397         ENTRY;
1398
1399         if (rc)
1400                 lovset->set_completes = 0;
1401         err = lov_fini_setattr_set(lovset);
1402         RETURN(rc ? rc : err);
1403 }
1404
1405 /* If @oti is given, the request goes from MDS and responses from OSTs are not
1406    needed. Otherwise, a client is waiting for responses. */
1407 static int lov_setattr_async(struct obd_export *exp, struct obd_info *oinfo,
1408                              struct obd_trans_info *oti,
1409                              struct ptlrpc_request_set *rqset)
1410 {
1411         struct lov_request_set *set;
1412         struct lov_request *req;
1413         cfs_list_t *pos;
1414         struct lov_obd *lov;
1415         int rc = 0;
1416         ENTRY;
1417
1418         LASSERT(oinfo);
1419         ASSERT_LSM_MAGIC(oinfo->oi_md);
1420         if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE) {
1421                 LASSERT(oti);
1422                 LASSERT(oti->oti_logcookies);
1423         }
1424
1425         if (!exp || !exp->exp_obd)
1426                 RETURN(-ENODEV);
1427
1428         lov = &exp->exp_obd->u.lov;
1429         rc = lov_prep_setattr_set(exp, oinfo, oti, &set);
1430         if (rc)
1431                 RETURN(rc);
1432
1433         CDEBUG(D_INFO, "objid "LPX64": %ux%u byte stripes\n",
1434                oinfo->oi_md->lsm_object_id, oinfo->oi_md->lsm_stripe_count,
1435                oinfo->oi_md->lsm_stripe_size);
1436
1437         cfs_list_for_each (pos, &set->set_list) {
1438                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1439
1440                 if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE)
1441                         oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1442
1443                 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
1444                        "%u\n", oinfo->oi_oa->o_id, req->rq_stripe,
1445                        req->rq_oi.oi_oa->o_id, req->rq_idx);
1446
1447                 rc = obd_setattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1448                                        &req->rq_oi, oti, rqset);
1449                 if (rc) {
1450                         CERROR("error: setattr objid "LPX64" subobj "
1451                                LPX64" on OST idx %d: rc = %d\n",
1452                                set->set_oi->oi_oa->o_id,
1453                                req->rq_oi.oi_oa->o_id,
1454                                req->rq_idx, rc);
1455                         break;
1456                 }
1457         }
1458
1459         /* If we are not waiting for responses on async requests, return. */
1460         if (rc || !rqset || cfs_list_empty(&rqset->set_requests)) {
1461                 int err;
1462                 if (rc)
1463                         set->set_completes = 0;
1464                 err = lov_fini_setattr_set(set);
1465                 RETURN(rc ? rc : err);
1466         }
1467
1468         LASSERT(rqset->set_interpret == NULL);
1469         rqset->set_interpret = lov_setattr_interpret;
1470         rqset->set_arg = (void *)set;
1471
1472         RETURN(0);
1473 }
1474
1475 static int lov_punch_interpret(struct ptlrpc_request_set *rqset,
1476                                void *data, int rc)
1477 {
1478         struct lov_request_set *lovset = (struct lov_request_set *)data;
1479         int err;
1480         ENTRY;
1481
1482         if (rc)
1483                 lovset->set_completes = 0;
1484         err = lov_fini_punch_set(lovset);
1485         RETURN(rc ? rc : err);
1486 }
1487
1488 /* FIXME: maybe we'll just make one node the authoritative attribute node, then
1489  * we can send this 'punch' to just the authoritative node and the nodes
1490  * that the punch will affect. */
1491 static int lov_punch(struct obd_export *exp, struct obd_info *oinfo,
1492                      struct obd_trans_info *oti,
1493                      struct ptlrpc_request_set *rqset)
1494 {
1495         struct lov_request_set *set;
1496         struct lov_obd *lov;
1497         cfs_list_t *pos;
1498         struct lov_request *req;
1499         int rc = 0;
1500         ENTRY;
1501
1502         LASSERT(oinfo);
1503         ASSERT_LSM_MAGIC(oinfo->oi_md);
1504
1505         if (!exp || !exp->exp_obd)
1506                 RETURN(-ENODEV);
1507
1508         lov = &exp->exp_obd->u.lov;
1509         rc = lov_prep_punch_set(exp, oinfo, oti, &set);
1510         if (rc)
1511                 RETURN(rc);
1512
1513         cfs_list_for_each (pos, &set->set_list) {
1514                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1515
1516                 rc = obd_punch(lov->lov_tgts[req->rq_idx]->ltd_exp,
1517                                &req->rq_oi, NULL, rqset);
1518                 if (rc) {
1519                         CERROR("error: punch objid "LPX64" subobj "LPX64
1520                                " on OST idx %d: rc = %d\n",
1521                                set->set_oi->oi_oa->o_id,
1522                                req->rq_oi.oi_oa->o_id, req->rq_idx, rc);
1523                         break;
1524                 }
1525         }
1526
1527         if (rc || cfs_list_empty(&rqset->set_requests)) {
1528                 int err;
1529                 err = lov_fini_punch_set(set);
1530                 RETURN(rc ? rc : err);
1531         }
1532
1533         LASSERT(rqset->set_interpret == NULL);
1534         rqset->set_interpret = lov_punch_interpret;
1535         rqset->set_arg = (void *)set;
1536
1537         RETURN(0);
1538 }
1539
1540 static int lov_sync(struct obd_export *exp, struct obdo *oa,
1541                     struct lov_stripe_md *lsm, obd_off start, obd_off end,
1542                     void *capa)
1543 {
1544         struct lov_request_set *set;
1545         struct obd_info oinfo;
1546         struct lov_obd *lov;
1547         cfs_list_t *pos;
1548         struct lov_request *req;
1549         int err = 0, rc = 0;
1550         ENTRY;
1551
1552         ASSERT_LSM_MAGIC(lsm);
1553
1554         if (!exp->exp_obd)
1555                 RETURN(-ENODEV);
1556
1557         lov = &exp->exp_obd->u.lov;
1558         rc = lov_prep_sync_set(exp, &oinfo, oa, lsm, start, end, &set);
1559         if (rc)
1560                 RETURN(rc);
1561
1562         cfs_list_for_each (pos, &set->set_list) {
1563                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1564
1565                 rc = obd_sync(lov->lov_tgts[req->rq_idx]->ltd_exp,
1566                               req->rq_oi.oi_oa, NULL,
1567                               req->rq_oi.oi_policy.l_extent.start,
1568                               req->rq_oi.oi_policy.l_extent.end, capa);
1569                 err = lov_update_common_set(set, req, rc);
1570                 if (err) {
1571                         CERROR("error: fsync objid "LPX64" subobj "LPX64
1572                                " on OST idx %d: rc = %d\n",
1573                                set->set_oi->oi_oa->o_id,
1574                                req->rq_oi.oi_oa->o_id, req->rq_idx, rc);
1575                         if (!rc)
1576                                 rc = err;
1577                 }
1578         }
1579         err = lov_fini_sync_set(set);
1580         if (!rc)
1581                 rc = err;
1582         RETURN(rc);
1583 }
1584
1585 static int lov_brw_check(struct lov_obd *lov, struct obd_info *lov_oinfo,
1586                          obd_count oa_bufs, struct brw_page *pga)
1587 {
1588         struct obd_info oinfo = { { { 0 } } };
1589         int i, rc = 0;
1590
1591         oinfo.oi_oa = lov_oinfo->oi_oa;
1592
1593         /* The caller just wants to know if there's a chance that this
1594          * I/O can succeed */
1595         for (i = 0; i < oa_bufs; i++) {
1596                 int stripe = lov_stripe_number(lov_oinfo->oi_md, pga[i].off);
1597                 int ost = lov_oinfo->oi_md->lsm_oinfo[stripe]->loi_ost_idx;
1598                 obd_off start, end;
1599
1600                 if (!lov_stripe_intersects(lov_oinfo->oi_md, i, pga[i].off,
1601                                            pga[i].off + pga[i].count - 1,
1602                                            &start, &end))
1603                         continue;
1604
1605                 if (!lov->lov_tgts[ost] || !lov->lov_tgts[ost]->ltd_active) {
1606                         CDEBUG(D_HA, "lov idx %d inactive\n", ost);
1607                         return -EIO;
1608                 }
1609
1610                 rc = obd_brw(OBD_BRW_CHECK, lov->lov_tgts[ost]->ltd_exp, &oinfo,
1611                              1, &pga[i], NULL);
1612                 if (rc)
1613                         break;
1614         }
1615         return rc;
1616 }
1617
1618 static int lov_brw(int cmd, struct obd_export *exp, struct obd_info *oinfo,
1619                    obd_count oa_bufs, struct brw_page *pga,
1620                    struct obd_trans_info *oti)
1621 {
1622         struct lov_request_set *set;
1623         struct lov_request *req;
1624         cfs_list_t *pos;
1625         struct lov_obd *lov = &exp->exp_obd->u.lov;
1626         int err, rc = 0;
1627         ENTRY;
1628
1629         ASSERT_LSM_MAGIC(oinfo->oi_md);
1630
1631         if (cmd == OBD_BRW_CHECK) {
1632                 rc = lov_brw_check(lov, oinfo, oa_bufs, pga);
1633                 RETURN(rc);
1634         }
1635
1636         rc = lov_prep_brw_set(exp, oinfo, oa_bufs, pga, oti, &set);
1637         if (rc)
1638                 RETURN(rc);
1639
1640         cfs_list_for_each (pos, &set->set_list) {
1641                 struct obd_export *sub_exp;
1642                 struct brw_page *sub_pga;
1643                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1644
1645                 sub_exp = lov->lov_tgts[req->rq_idx]->ltd_exp;
1646                 sub_pga = set->set_pga + req->rq_pgaidx;
1647                 rc = obd_brw(cmd, sub_exp, &req->rq_oi, req->rq_oabufs,
1648                              sub_pga, oti);
1649                 if (rc)
1650                         break;
1651                 lov_update_common_set(set, req, rc);
1652         }
1653
1654         err = lov_fini_brw_set(set);
1655         if (!rc)
1656                 rc = err;
1657         RETURN(rc);
1658 }
1659
1660 static int lov_enqueue_interpret(struct ptlrpc_request_set *rqset,
1661                                  void *data, int rc)
1662 {
1663         struct lov_request_set *lovset = (struct lov_request_set *)data;
1664         ENTRY;
1665         rc = lov_fini_enqueue_set(lovset, lovset->set_ei->ei_mode, rc, rqset);
1666         RETURN(rc);
1667 }
1668
1669 static int lov_enqueue(struct obd_export *exp, struct obd_info *oinfo,
1670                        struct ldlm_enqueue_info *einfo,
1671                        struct ptlrpc_request_set *rqset)
1672 {
1673         ldlm_mode_t mode = einfo->ei_mode;
1674         struct lov_request_set *set;
1675         struct lov_request *req;
1676         cfs_list_t *pos;
1677         struct lov_obd *lov;
1678         ldlm_error_t rc;
1679         ENTRY;
1680
1681         LASSERT(oinfo);
1682         ASSERT_LSM_MAGIC(oinfo->oi_md);
1683         LASSERT(mode == (mode & -mode));
1684
1685         /* we should never be asked to replay a lock this way. */
1686         LASSERT((oinfo->oi_flags & LDLM_FL_REPLAY) == 0);
1687
1688         if (!exp || !exp->exp_obd)
1689                 RETURN(-ENODEV);
1690
1691         lov = &exp->exp_obd->u.lov;
1692         rc = lov_prep_enqueue_set(exp, oinfo, einfo, &set);
1693         if (rc)
1694                 RETURN(rc);
1695
1696         cfs_list_for_each (pos, &set->set_list) {
1697                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1698
1699                 rc = obd_enqueue(lov->lov_tgts[req->rq_idx]->ltd_exp,
1700                                  &req->rq_oi, einfo, rqset);
1701                 if (rc != ELDLM_OK)
1702                         GOTO(out, rc);
1703         }
1704
1705         if (rqset && !cfs_list_empty(&rqset->set_requests)) {
1706                 LASSERT(rc == 0);
1707                 LASSERT(rqset->set_interpret == NULL);
1708                 rqset->set_interpret = lov_enqueue_interpret;
1709                 rqset->set_arg = (void *)set;
1710                 RETURN(rc);
1711         }
1712 out:
1713         rc = lov_fini_enqueue_set(set, mode, rc, rqset);
1714         RETURN(rc);
1715 }
1716
1717 static int lov_change_cbdata(struct obd_export *exp,
1718                              struct lov_stripe_md *lsm, ldlm_iterator_t it,
1719                              void *data)
1720 {
1721         struct lov_obd *lov;
1722         int rc = 0, i;
1723         ENTRY;
1724
1725         ASSERT_LSM_MAGIC(lsm);
1726
1727         if (!exp || !exp->exp_obd)
1728                 RETURN(-ENODEV);
1729
1730         lov = &exp->exp_obd->u.lov;
1731         for (i = 0; i < lsm->lsm_stripe_count; i++) {
1732                 struct lov_stripe_md submd;
1733                 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
1734
1735                 if (!lov->lov_tgts[loi->loi_ost_idx]) {
1736                         CDEBUG(D_HA, "lov idx %d NULL \n", loi->loi_ost_idx);
1737                         continue;
1738                 }
1739
1740                 LASSERT_SEQ_IS_MDT(loi->loi_seq);
1741                 submd.lsm_object_id = loi->loi_id;
1742                 submd.lsm_object_seq = loi->loi_seq;
1743                 submd.lsm_stripe_count = 0;
1744                 rc = obd_change_cbdata(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1745                                        &submd, it, data);
1746         }
1747         RETURN(rc);
1748 }
1749
1750 /* find any ldlm lock of the inode in lov
1751  * return 0    not find
1752  *        1    find one
1753  *      < 0    error */
1754 static int lov_find_cbdata(struct obd_export *exp,
1755                            struct lov_stripe_md *lsm, ldlm_iterator_t it,
1756                            void *data)
1757 {
1758         struct lov_obd *lov;
1759         int rc = 0, i;
1760         ENTRY;
1761
1762         ASSERT_LSM_MAGIC(lsm);
1763
1764         if (!exp || !exp->exp_obd)
1765                 RETURN(-ENODEV);
1766
1767         lov = &exp->exp_obd->u.lov;
1768         for (i = 0; i < lsm->lsm_stripe_count; i++) {
1769                 struct lov_stripe_md submd;
1770                 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
1771
1772                 if (!lov->lov_tgts[loi->loi_ost_idx]) {
1773                         CDEBUG(D_HA, "lov idx %d NULL \n", loi->loi_ost_idx);
1774                         continue;
1775                 }
1776
1777                 LASSERT_SEQ_IS_MDT(loi->loi_seq);
1778                 submd.lsm_object_id = loi->loi_id;
1779                 submd.lsm_object_seq = loi->loi_seq;
1780                 submd.lsm_stripe_count = 0;
1781                 rc = obd_find_cbdata(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1782                                      &submd, it, data);
1783                 if (rc != 0)
1784                         RETURN(rc);
1785         }
1786         RETURN(rc);
1787 }
1788
1789 static int lov_cancel(struct obd_export *exp, struct lov_stripe_md *lsm,
1790                       __u32 mode, struct lustre_handle *lockh)
1791 {
1792         struct lov_request_set *set;
1793         struct obd_info oinfo;
1794         struct lov_request *req;
1795         cfs_list_t *pos;
1796         struct lov_obd *lov;
1797         struct lustre_handle *lov_lockhp;
1798         int err = 0, rc = 0;
1799         ENTRY;
1800
1801         ASSERT_LSM_MAGIC(lsm);
1802
1803         if (!exp || !exp->exp_obd)
1804                 RETURN(-ENODEV);
1805
1806         LASSERT_SEQ_IS_MDT(lsm->lsm_object_seq);
1807         LASSERT(lockh);
1808         lov = &exp->exp_obd->u.lov;
1809         rc = lov_prep_cancel_set(exp, &oinfo, lsm, mode, lockh, &set);
1810         if (rc)
1811                 RETURN(rc);
1812
1813         cfs_list_for_each (pos, &set->set_list) {
1814                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1815                 lov_lockhp = set->set_lockh->llh_handles + req->rq_stripe;
1816
1817                 rc = obd_cancel(lov->lov_tgts[req->rq_idx]->ltd_exp,
1818                                 req->rq_oi.oi_md, mode, lov_lockhp);
1819                 rc = lov_update_common_set(set, req, rc);
1820                 if (rc) {
1821                         CERROR("error: cancel objid "LPX64" subobj "
1822                                LPX64" on OST idx %d: rc = %d\n",
1823                                lsm->lsm_object_id,
1824                                req->rq_oi.oi_md->lsm_object_id,
1825                                req->rq_idx, rc);
1826                         err = rc;
1827                 }
1828
1829         }
1830         lov_fini_cancel_set(set);
1831         RETURN(err);
1832 }
1833
1834 static int lov_cancel_unused(struct obd_export *exp,
1835                              struct lov_stripe_md *lsm,
1836                              ldlm_cancel_flags_t flags, void *opaque)
1837 {
1838         struct lov_obd *lov;
1839         int rc = 0, i;
1840         ENTRY;
1841
1842         if (!exp || !exp->exp_obd)
1843                 RETURN(-ENODEV);
1844
1845         lov = &exp->exp_obd->u.lov;
1846         if (lsm == NULL) {
1847                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
1848                         int err;
1849                         if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
1850                                 continue;
1851
1852                         err = obd_cancel_unused(lov->lov_tgts[i]->ltd_exp, NULL,
1853                                                 flags, opaque);
1854                         if (!rc)
1855                                 rc = err;
1856                 }
1857                 RETURN(rc);
1858         }
1859
1860         ASSERT_LSM_MAGIC(lsm);
1861
1862         LASSERT_SEQ_IS_MDT(lsm->lsm_object_seq);
1863         for (i = 0; i < lsm->lsm_stripe_count; i++) {
1864                 struct lov_stripe_md submd;
1865                 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
1866                 int err;
1867
1868                 if (!lov->lov_tgts[loi->loi_ost_idx]) {
1869                         CDEBUG(D_HA, "lov idx %d NULL\n", loi->loi_ost_idx);
1870                         continue;
1871                 }
1872
1873                 if (!lov->lov_tgts[loi->loi_ost_idx]->ltd_active)
1874                         CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
1875
1876                 submd.lsm_object_id = loi->loi_id;
1877                 submd.lsm_object_seq = loi->loi_seq;
1878                 submd.lsm_stripe_count = 0;
1879                 err = obd_cancel_unused(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1880                                         &submd, flags, opaque);
1881                 if (err && lov->lov_tgts[loi->loi_ost_idx]->ltd_active) {
1882                         CERROR("error: cancel unused objid "LPX64" subobj "LPX64
1883                                " on OST idx %d: rc = %d\n", lsm->lsm_object_id,
1884                                loi->loi_id, loi->loi_ost_idx, err);
1885                         if (!rc)
1886                                 rc = err;
1887                 }
1888         }
1889         RETURN(rc);
1890 }
1891
1892 int lov_statfs_interpret(struct ptlrpc_request_set *rqset, void *data, int rc)
1893 {
1894         struct lov_request_set *lovset = (struct lov_request_set *)data;
1895         int err;
1896         ENTRY;
1897
1898         if (rc)
1899                 lovset->set_completes = 0;
1900
1901         err = lov_fini_statfs_set(lovset);
1902         RETURN(rc ? rc : err);
1903 }
1904
1905 static int lov_statfs_async(struct obd_device *obd, struct obd_info *oinfo,
1906                             __u64 max_age, struct ptlrpc_request_set *rqset)
1907 {
1908         struct lov_request_set *set;
1909         struct lov_request *req;
1910         cfs_list_t *pos;
1911         struct lov_obd *lov;
1912         int rc = 0;
1913         ENTRY;
1914
1915         LASSERT(oinfo != NULL);
1916         LASSERT(oinfo->oi_osfs != NULL);
1917
1918         lov = &obd->u.lov;
1919         rc = lov_prep_statfs_set(obd, oinfo, &set);
1920         if (rc)
1921                 RETURN(rc);
1922
1923         cfs_list_for_each (pos, &set->set_list) {
1924                 struct obd_device *osc_obd;
1925
1926                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1927
1928                 osc_obd = class_exp2obd(lov->lov_tgts[req->rq_idx]->ltd_exp);
1929                 rc = obd_statfs_async(osc_obd, &req->rq_oi, max_age, rqset);
1930                 if (rc)
1931                         break;
1932         }
1933
1934         if (rc || cfs_list_empty(&rqset->set_requests)) {
1935                 int err;
1936                 if (rc)
1937                         set->set_completes = 0;
1938                 err = lov_fini_statfs_set(set);
1939                 RETURN(rc ? rc : err);
1940         }
1941
1942         LASSERT(rqset->set_interpret == NULL);
1943         rqset->set_interpret = lov_statfs_interpret;
1944         rqset->set_arg = (void *)set;
1945         RETURN(0);
1946 }
1947
1948 static int lov_statfs(struct obd_device *obd, struct obd_statfs *osfs,
1949                       __u64 max_age, __u32 flags)
1950 {
1951         struct ptlrpc_request_set *set = NULL;
1952         struct obd_info oinfo = { { { 0 } } };
1953         int rc = 0;
1954         ENTRY;
1955
1956
1957         /* for obdclass we forbid using obd_statfs_rqset, but prefer using async
1958          * statfs requests */
1959         set = ptlrpc_prep_set();
1960         if (set == NULL)
1961                 RETURN(-ENOMEM);
1962
1963         oinfo.oi_osfs = osfs;
1964         oinfo.oi_flags = flags;
1965         rc = lov_statfs_async(obd, &oinfo, max_age, set);
1966         if (rc == 0)
1967                 rc = ptlrpc_set_wait(set);
1968         ptlrpc_set_destroy(set);
1969
1970         RETURN(rc);
1971 }
1972
1973 static int lov_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1974                          void *karg, void *uarg)
1975 {
1976         struct obd_device *obddev = class_exp2obd(exp);
1977         struct lov_obd *lov = &obddev->u.lov;
1978         int i = 0, rc = 0, count = lov->desc.ld_tgt_count;
1979         struct obd_uuid *uuidp;
1980         ENTRY;
1981
1982         switch (cmd) {
1983         case IOC_OBD_STATFS: {
1984                 struct obd_ioctl_data *data = karg;
1985                 struct obd_device *osc_obd;
1986                 struct obd_statfs stat_buf = {0};
1987                 __u32 index;
1988
1989                 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
1990                 if ((index >= count))
1991                         RETURN(-ENODEV);
1992
1993                 if (!lov->lov_tgts[index])
1994                         /* Try again with the next index */
1995                         RETURN(-EAGAIN);
1996                 if (!lov->lov_tgts[index]->ltd_active)
1997                         RETURN(-ENODATA);
1998
1999                 osc_obd = class_exp2obd(lov->lov_tgts[index]->ltd_exp);
2000                 if (!osc_obd)
2001                         RETURN(-EINVAL);
2002
2003                 /* copy UUID */
2004                 if (cfs_copy_to_user(data->ioc_pbuf2, obd2cli_tgt(osc_obd),
2005                                      min((int) data->ioc_plen2,
2006                                          (int) sizeof(struct obd_uuid))))
2007                         RETURN(-EFAULT);
2008
2009                 /* got statfs data */
2010                 rc = obd_statfs(osc_obd, &stat_buf,
2011                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
2012                                 0);
2013                 if (rc)
2014                         RETURN(rc);
2015                 if (cfs_copy_to_user(data->ioc_pbuf1, &stat_buf,
2016                                      min((int) data->ioc_plen1,
2017                                          (int) sizeof(stat_buf))))
2018                         RETURN(-EFAULT);
2019                 break;
2020         }
2021         case OBD_IOC_LOV_GET_CONFIG: {
2022                 struct obd_ioctl_data *data;
2023                 struct lov_desc *desc;
2024                 char *buf = NULL;
2025                 __u32 *genp;
2026
2027                 len = 0;
2028                 if (obd_ioctl_getdata(&buf, &len, (void *)uarg))
2029                         RETURN(-EINVAL);
2030
2031                 data = (struct obd_ioctl_data *)buf;
2032
2033                 if (sizeof(*desc) > data->ioc_inllen1) {
2034                         obd_ioctl_freedata(buf, len);
2035                         RETURN(-EINVAL);
2036                 }
2037
2038                 if (sizeof(uuidp->uuid) * count > data->ioc_inllen2) {
2039                         obd_ioctl_freedata(buf, len);
2040                         RETURN(-EINVAL);
2041                 }
2042
2043                 if (sizeof(__u32) * count > data->ioc_inllen3) {
2044                         obd_ioctl_freedata(buf, len);
2045                         RETURN(-EINVAL);
2046                 }
2047
2048                 desc = (struct lov_desc *)data->ioc_inlbuf1;
2049                 memcpy(desc, &(lov->desc), sizeof(*desc));
2050
2051                 uuidp = (struct obd_uuid *)data->ioc_inlbuf2;
2052                 genp = (__u32 *)data->ioc_inlbuf3;
2053                 /* the uuid will be empty for deleted OSTs */
2054                 for (i = 0; i < count; i++, uuidp++, genp++) {
2055                         if (!lov->lov_tgts[i])
2056                                 continue;
2057                         *uuidp = lov->lov_tgts[i]->ltd_uuid;
2058                         *genp = lov->lov_tgts[i]->ltd_gen;
2059                 }
2060
2061                 if (cfs_copy_to_user((void *)uarg, buf, len))
2062                         rc = -EFAULT;
2063                 obd_ioctl_freedata(buf, len);
2064                 break;
2065         }
2066         case LL_IOC_LOV_SETSTRIPE:
2067                 rc = lov_setstripe(exp, len, karg, uarg);
2068                 break;
2069         case LL_IOC_LOV_GETSTRIPE:
2070                 rc = lov_getstripe(exp, karg, uarg);
2071                 break;
2072         case LL_IOC_LOV_SETEA:
2073                 rc = lov_setea(exp, karg, uarg);
2074                 break;
2075         case OBD_IOC_QUOTACTL: {
2076                 struct if_quotactl *qctl = karg;
2077                 struct lov_tgt_desc *tgt = NULL;
2078                 struct obd_quotactl *oqctl;
2079
2080                 if (qctl->qc_valid == QC_OSTIDX) {
2081                         if (qctl->qc_idx < 0 || count <= qctl->qc_idx)
2082                                 RETURN(-EINVAL);
2083
2084                         tgt = lov->lov_tgts[qctl->qc_idx];
2085                         if (!tgt || !tgt->ltd_exp)
2086                                 RETURN(-EINVAL);
2087                 } else if (qctl->qc_valid == QC_UUID) {
2088                         for (i = 0; i < count; i++) {
2089                                 tgt = lov->lov_tgts[i];
2090                                 if (!tgt ||
2091                                     !obd_uuid_equals(&tgt->ltd_uuid,
2092                                                      &qctl->obd_uuid))
2093                                         continue;
2094
2095                                 if (tgt->ltd_exp == NULL)
2096                                         RETURN(-EINVAL);
2097
2098                                 break;
2099                         }
2100                 } else {
2101                         RETURN(-EINVAL);
2102                 }
2103
2104                 if (i >= count)
2105                         RETURN(-EAGAIN);
2106
2107                 LASSERT(tgt && tgt->ltd_exp);
2108                 OBD_ALLOC_PTR(oqctl);
2109                 if (!oqctl)
2110                         RETURN(-ENOMEM);
2111
2112                 QCTL_COPY(oqctl, qctl);
2113                 rc = obd_quotactl(tgt->ltd_exp, oqctl);
2114                 if (rc == 0) {
2115                         QCTL_COPY(qctl, oqctl);
2116                         qctl->qc_valid = QC_OSTIDX;
2117                         qctl->obd_uuid = tgt->ltd_uuid;
2118                 }
2119                 OBD_FREE_PTR(oqctl);
2120                 break;
2121         }
2122         default: {
2123                 int set = 0;
2124
2125                 if (count == 0)
2126                         RETURN(-ENOTTY);
2127
2128                 for (i = 0; i < count; i++) {
2129                         int err;
2130
2131                         /* OST was disconnected */
2132                         if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
2133                                 continue;
2134
2135                         err = obd_iocontrol(cmd, lov->lov_tgts[i]->ltd_exp,
2136                                             len, karg, uarg);
2137                         if (err == -ENODATA && cmd == OBD_IOC_POLL_QUOTACHECK) {
2138                                 RETURN(err);
2139                         } else if (err) {
2140                                 if (lov->lov_tgts[i]->ltd_active) {
2141                                         CDEBUG(err == -ENOTTY ?
2142                                                D_IOCTL : D_WARNING,
2143                                                "iocontrol OSC %s on OST "
2144                                                "idx %d cmd %x: err = %d\n",
2145                                                lov_uuid2str(lov, i),
2146                                                i, cmd, err);
2147                                         if (!rc)
2148                                                 rc = err;
2149                                 }
2150                         } else {
2151                                 set = 1;
2152                         }
2153                 }
2154                 if (!set && !rc)
2155                         rc = -EIO;
2156         }
2157         }
2158
2159         RETURN(rc);
2160 }
2161
2162 #define FIEMAP_BUFFER_SIZE 4096
2163
2164 /**
2165  * Non-zero fe_logical indicates that this is a continuation FIEMAP
2166  * call. The local end offset and the device are sent in the first
2167  * fm_extent. This function calculates the stripe number from the index.
2168  * This function returns a stripe_no on which mapping is to be restarted.
2169  *
2170  * This function returns fm_end_offset which is the in-OST offset at which
2171  * mapping should be restarted. If fm_end_offset=0 is returned then caller
2172  * will re-calculate proper offset in next stripe.
2173  * Note that the first extent is passed to lov_get_info via the value field.
2174  *
2175  * \param fiemap fiemap request header
2176  * \param lsm striping information for the file
2177  * \param fm_start logical start of mapping
2178  * \param fm_end logical end of mapping
2179  * \param start_stripe starting stripe will be returned in this
2180  */
2181 obd_size fiemap_calc_fm_end_offset(struct ll_user_fiemap *fiemap,
2182                                    struct lov_stripe_md *lsm, obd_size fm_start,
2183                                    obd_size fm_end, int *start_stripe)
2184 {
2185         obd_size local_end = fiemap->fm_extents[0].fe_logical;
2186         obd_off lun_start, lun_end;
2187         obd_size fm_end_offset;
2188         int stripe_no = -1, i;
2189
2190         if (fiemap->fm_extent_count == 0 ||
2191             fiemap->fm_extents[0].fe_logical == 0)
2192                 return 0;
2193
2194         /* Find out stripe_no from ost_index saved in the fe_device */
2195         for (i = 0; i < lsm->lsm_stripe_count; i++) {
2196                 if (lsm->lsm_oinfo[i]->loi_ost_idx ==
2197                                         fiemap->fm_extents[0].fe_device) {
2198                         stripe_no = i;
2199                         break;
2200                 }
2201         }
2202
2203         /* If we have finished mapping on previous device, shift logical
2204          * offset to start of next device */
2205         if ((lov_stripe_intersects(lsm, stripe_no, fm_start, fm_end,
2206                                    &lun_start, &lun_end)) != 0 &&
2207                                    local_end < lun_end) {
2208                 fm_end_offset = local_end;
2209                 *start_stripe = stripe_no;
2210         } else {
2211                 /* This is a special value to indicate that caller should
2212                  * calculate offset in next stripe. */
2213                 fm_end_offset = 0;
2214                 *start_stripe = (stripe_no + 1) % lsm->lsm_stripe_count;
2215         }
2216
2217         return fm_end_offset;
2218 }
2219
2220 /**
2221  * We calculate on which OST the mapping will end. If the length of mapping
2222  * is greater than (stripe_size * stripe_count) then the last_stripe will
2223  * will be one just before start_stripe. Else we check if the mapping
2224  * intersects each OST and find last_stripe.
2225  * This function returns the last_stripe and also sets the stripe_count
2226  * over which the mapping is spread
2227  *
2228  * \param lsm striping information for the file
2229  * \param fm_start logical start of mapping
2230  * \param fm_end logical end of mapping
2231  * \param start_stripe starting stripe of the mapping
2232  * \param stripe_count the number of stripes across which to map is returned
2233  *
2234  * \retval last_stripe return the last stripe of the mapping
2235  */
2236 int fiemap_calc_last_stripe(struct lov_stripe_md *lsm, obd_size fm_start,
2237                             obd_size fm_end, int start_stripe,
2238                             int *stripe_count)
2239 {
2240         int last_stripe;
2241         obd_off obd_start, obd_end;
2242         int i, j;
2243
2244         if (fm_end - fm_start > lsm->lsm_stripe_size * lsm->lsm_stripe_count) {
2245                 last_stripe = (start_stripe < 1 ? lsm->lsm_stripe_count - 1 :
2246                                                               start_stripe - 1);
2247                 *stripe_count = lsm->lsm_stripe_count;
2248         } else {
2249                 for (j = 0, i = start_stripe; j < lsm->lsm_stripe_count;
2250                      i = (i + 1) % lsm->lsm_stripe_count, j++) {
2251                         if ((lov_stripe_intersects(lsm, i, fm_start, fm_end,
2252                                                    &obd_start, &obd_end)) == 0)
2253                                 break;
2254                 }
2255                 *stripe_count = j;
2256                 last_stripe = (start_stripe + j - 1) %lsm->lsm_stripe_count;
2257         }
2258
2259         return last_stripe;
2260 }
2261
2262 /**
2263  * Set fe_device and copy extents from local buffer into main return buffer.
2264  *
2265  * \param fiemap fiemap request header
2266  * \param lcl_fm_ext array of local fiemap extents to be copied
2267  * \param ost_index OST index to be written into the fm_device field for each
2268                     extent
2269  * \param ext_count number of extents to be copied
2270  * \param current_extent where to start copying in main extent array
2271  */
2272 void fiemap_prepare_and_copy_exts(struct ll_user_fiemap *fiemap,
2273                                   struct ll_fiemap_extent *lcl_fm_ext,
2274                                   int ost_index, unsigned int ext_count,
2275                                   int current_extent)
2276 {
2277         char *to;
2278         int ext;
2279
2280         for (ext = 0; ext < ext_count; ext++) {
2281                 lcl_fm_ext[ext].fe_device = ost_index;
2282                 lcl_fm_ext[ext].fe_flags |= FIEMAP_EXTENT_NET;
2283         }
2284
2285         /* Copy fm_extent's from fm_local to return buffer */
2286         to = (char *)fiemap + fiemap_count_to_size(current_extent);
2287         memcpy(to, lcl_fm_ext, ext_count * sizeof(struct ll_fiemap_extent));
2288 }
2289
2290 /**
2291  * Break down the FIEMAP request and send appropriate calls to individual OSTs.
2292  * This also handles the restarting of FIEMAP calls in case mapping overflows
2293  * the available number of extents in single call.
2294  */
2295 static int lov_fiemap(struct lov_obd *lov, __u32 keylen, void *key,
2296                       __u32 *vallen, void *val, struct lov_stripe_md *lsm)
2297 {
2298         struct ll_fiemap_info_key *fm_key = key;
2299         struct ll_user_fiemap *fiemap = val;
2300         struct ll_user_fiemap *fm_local = NULL;
2301         struct ll_fiemap_extent *lcl_fm_ext;
2302         int count_local;
2303         unsigned int get_num_extents = 0;
2304         int ost_index = 0, actual_start_stripe, start_stripe;
2305         obd_size fm_start, fm_end, fm_length, fm_end_offset = 0;
2306         obd_size curr_loc;
2307         int current_extent = 0, rc = 0, i;
2308         int ost_eof = 0; /* EOF for object */
2309         int ost_done = 0; /* done with required mapping for this OST? */
2310         int last_stripe;
2311         int cur_stripe = 0, cur_stripe_wrap = 0, stripe_count;
2312         unsigned int buffer_size = FIEMAP_BUFFER_SIZE;
2313
2314         if (lsm == NULL)
2315                 GOTO(out, rc = 0);
2316
2317         if (fiemap_count_to_size(fm_key->fiemap.fm_extent_count) < buffer_size)
2318                 buffer_size = fiemap_count_to_size(fm_key->fiemap.fm_extent_count);
2319
2320         OBD_ALLOC(fm_local, buffer_size);
2321         if (fm_local == NULL)
2322                 GOTO(out, rc = -ENOMEM);
2323         lcl_fm_ext = &fm_local->fm_extents[0];
2324
2325         count_local = fiemap_size_to_count(buffer_size);
2326
2327         memcpy(fiemap, &fm_key->fiemap, sizeof(*fiemap));
2328         fm_start = fiemap->fm_start;
2329         fm_length = fiemap->fm_length;
2330         /* Calculate start stripe, last stripe and length of mapping */
2331         actual_start_stripe = start_stripe = lov_stripe_number(lsm, fm_start);
2332         fm_end = (fm_length == ~0ULL ? fm_key->oa.o_size :
2333                                                 fm_start + fm_length - 1);
2334         /* If fm_length != ~0ULL but fm_start+fm_length-1 exceeds file size */
2335         if (fm_end > fm_key->oa.o_size)
2336                 fm_end = fm_key->oa.o_size;
2337
2338         last_stripe = fiemap_calc_last_stripe(lsm, fm_start, fm_end,
2339                                             actual_start_stripe, &stripe_count);
2340
2341         fm_end_offset = fiemap_calc_fm_end_offset(fiemap, lsm, fm_start, fm_end,
2342                                                   &start_stripe);
2343
2344         if (fiemap->fm_extent_count == 0) {
2345                 get_num_extents = 1;
2346                 count_local = 0;
2347         }
2348
2349         /* Check each stripe */
2350         for (cur_stripe = start_stripe, i = 0; i < stripe_count;
2351              i++, cur_stripe = (cur_stripe + 1) % lsm->lsm_stripe_count) {
2352                 obd_size req_fm_len; /* Stores length of required mapping */
2353                 obd_size len_mapped_single_call;
2354                 obd_off lun_start, lun_end, obd_object_end;
2355                 unsigned int ext_count;
2356
2357                 cur_stripe_wrap = cur_stripe;
2358
2359                 /* Find out range of mapping on this stripe */
2360                 if ((lov_stripe_intersects(lsm, cur_stripe, fm_start, fm_end,
2361                                            &lun_start, &obd_object_end)) == 0)
2362                         continue;
2363
2364                 /* If this is a continuation FIEMAP call and we are on
2365                  * starting stripe then lun_start needs to be set to
2366                  * fm_end_offset */
2367                 if (fm_end_offset != 0 && cur_stripe == start_stripe)
2368                         lun_start = fm_end_offset;
2369
2370                 if (fm_length != ~0ULL) {
2371                         /* Handle fm_start + fm_length overflow */
2372                         if (fm_start + fm_length < fm_start)
2373                                 fm_length = ~0ULL - fm_start;
2374                         lun_end = lov_size_to_stripe(lsm, fm_start + fm_length,
2375                                                      cur_stripe);
2376                 } else {
2377                         lun_end = ~0ULL;
2378                 }
2379
2380                 if (lun_start == lun_end)
2381                         continue;
2382
2383                 req_fm_len = obd_object_end - lun_start;
2384                 fm_local->fm_length = 0;
2385                 len_mapped_single_call = 0;
2386
2387                 /* If the output buffer is very large and the objects have many
2388                  * extents we may need to loop on a single OST repeatedly */
2389                 ost_eof = 0;
2390                 ost_done = 0;
2391                 do {
2392                         if (get_num_extents == 0) {
2393                                 /* Don't get too many extents. */
2394                                 if (current_extent + count_local >
2395                                     fiemap->fm_extent_count)
2396                                         count_local = fiemap->fm_extent_count -
2397                                                                  current_extent;
2398                         }
2399
2400                         lun_start += len_mapped_single_call;
2401                         fm_local->fm_length = req_fm_len - len_mapped_single_call;
2402                         req_fm_len = fm_local->fm_length;
2403                         fm_local->fm_extent_count = count_local;
2404                         fm_local->fm_mapped_extents = 0;
2405                         fm_local->fm_flags = fiemap->fm_flags;
2406
2407                         fm_key->oa.o_id = lsm->lsm_oinfo[cur_stripe]->loi_id;
2408                         fm_key->oa.o_seq = lsm->lsm_oinfo[cur_stripe]->loi_seq;
2409                         ost_index = lsm->lsm_oinfo[cur_stripe]->loi_ost_idx;
2410
2411                         if (ost_index < 0 || ost_index >=lov->desc.ld_tgt_count)
2412                                 GOTO(out, rc = -EINVAL);
2413
2414                         /* If OST is inactive, return extent with UNKNOWN flag */
2415                         if (!lov->lov_tgts[ost_index]->ltd_active) {
2416                                 fm_local->fm_flags |= FIEMAP_EXTENT_LAST;
2417                                 fm_local->fm_mapped_extents = 1;
2418
2419                                 lcl_fm_ext[0].fe_logical = lun_start;
2420                                 lcl_fm_ext[0].fe_length = obd_object_end -
2421                                                                       lun_start;
2422                                 lcl_fm_ext[0].fe_flags |= FIEMAP_EXTENT_UNKNOWN;
2423
2424                                 goto inactive_tgt;
2425                         }
2426
2427                         fm_local->fm_start = lun_start;
2428                         fm_local->fm_flags &= ~FIEMAP_FLAG_DEVICE_ORDER;
2429                         memcpy(&fm_key->fiemap, fm_local, sizeof(*fm_local));
2430                         *vallen=fiemap_count_to_size(fm_local->fm_extent_count);
2431                         rc = obd_get_info(lov->lov_tgts[ost_index]->ltd_exp,
2432                                           keylen, key, vallen, fm_local, lsm);
2433                         if (rc != 0)
2434                                 GOTO(out, rc);
2435
2436 inactive_tgt:
2437                         ext_count = fm_local->fm_mapped_extents;
2438                         if (ext_count == 0) {
2439                                 ost_done = 1;
2440                                 /* If last stripe has hole at the end,
2441                                  * then we need to return */
2442                                 if (cur_stripe_wrap == last_stripe) {
2443                                         fiemap->fm_mapped_extents = 0;
2444                                         goto finish;
2445                                 }
2446                                 break;
2447                         }
2448
2449                         /* If we just need num of extents then go to next device */
2450                         if (get_num_extents) {
2451                                 current_extent += ext_count;
2452                                 break;
2453                         }
2454
2455                         len_mapped_single_call = lcl_fm_ext[ext_count-1].fe_logical -
2456                                   lun_start + lcl_fm_ext[ext_count - 1].fe_length;
2457
2458                         /* Have we finished mapping on this device? */
2459                         if (req_fm_len <= len_mapped_single_call)
2460                                 ost_done = 1;
2461
2462                         /* Clear the EXTENT_LAST flag which can be present on
2463                          * last extent */
2464                         if (lcl_fm_ext[ext_count-1].fe_flags & FIEMAP_EXTENT_LAST)
2465                                 lcl_fm_ext[ext_count - 1].fe_flags &=
2466                                                             ~FIEMAP_EXTENT_LAST;
2467
2468                         curr_loc = lov_stripe_size(lsm,
2469                                            lcl_fm_ext[ext_count - 1].fe_logical+
2470                                            lcl_fm_ext[ext_count - 1].fe_length,
2471                                            cur_stripe);
2472                         if (curr_loc >= fm_key->oa.o_size)
2473                                 ost_eof = 1;
2474
2475                         fiemap_prepare_and_copy_exts(fiemap, lcl_fm_ext,
2476                                                      ost_index, ext_count,
2477                                                      current_extent);
2478
2479                         current_extent += ext_count;
2480
2481                         /* Ran out of available extents? */
2482                         if (current_extent >= fiemap->fm_extent_count)
2483                                 goto finish;
2484                 } while (ost_done == 0 && ost_eof == 0);
2485
2486                 if (cur_stripe_wrap == last_stripe)
2487                         goto finish;
2488         }
2489
2490 finish:
2491         /* Indicate that we are returning device offsets unless file just has
2492          * single stripe */
2493         if (lsm->lsm_stripe_count > 1)
2494                 fiemap->fm_flags |= FIEMAP_FLAG_DEVICE_ORDER;
2495
2496         if (get_num_extents)
2497                 goto skip_last_device_calc;
2498
2499         /* Check if we have reached the last stripe and whether mapping for that
2500          * stripe is done. */
2501         if (cur_stripe_wrap == last_stripe) {
2502                 if (ost_done || ost_eof)
2503                         fiemap->fm_extents[current_extent - 1].fe_flags |=
2504                                                              FIEMAP_EXTENT_LAST;
2505         }
2506
2507 skip_last_device_calc:
2508         fiemap->fm_mapped_extents = current_extent;
2509
2510 out:
2511         OBD_FREE(fm_local, buffer_size);
2512         return rc;
2513 }
2514
2515 static int lov_get_info(struct obd_export *exp, __u32 keylen,
2516                         void *key, __u32 *vallen, void *val,
2517                         struct lov_stripe_md *lsm)
2518 {
2519         struct obd_device *obddev = class_exp2obd(exp);
2520         struct lov_obd *lov = &obddev->u.lov;
2521         int i, rc;
2522         ENTRY;
2523
2524         if (!vallen || !val)
2525                 RETURN(-EFAULT);
2526
2527         obd_getref(obddev);
2528
2529         if (KEY_IS(KEY_LOCK_TO_STRIPE)) {
2530                 struct {
2531                         char name[16];
2532                         struct ldlm_lock *lock;
2533                 } *data = key;
2534                 struct ldlm_res_id *res_id = &data->lock->l_resource->lr_name;
2535                 struct lov_oinfo *loi;
2536                 __u32 *stripe = val;
2537
2538                 if (*vallen < sizeof(*stripe))
2539                         GOTO(out, rc = -EFAULT);
2540                 *vallen = sizeof(*stripe);
2541
2542                 /* XXX This is another one of those bits that will need to
2543                  * change if we ever actually support nested LOVs.  It uses
2544                  * the lock's export to find out which stripe it is. */
2545                 /* XXX - it's assumed all the locks for deleted OSTs have
2546                  * been cancelled. Also, the export for deleted OSTs will
2547                  * be NULL and won't match the lock's export. */
2548                 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2549                         loi = lsm->lsm_oinfo[i];
2550                         if (!lov->lov_tgts[loi->loi_ost_idx])
2551                                 continue;
2552                         if (lov->lov_tgts[loi->loi_ost_idx]->ltd_exp ==
2553                             data->lock->l_conn_export &&
2554                             osc_res_name_eq(loi->loi_id, loi->loi_seq, res_id)) {
2555                                 *stripe = i;
2556                                 GOTO(out, rc = 0);
2557                         }
2558                 }
2559                 LDLM_ERROR(data->lock, "lock on inode without such object");
2560                 dump_lsm(D_ERROR, lsm);
2561                 GOTO(out, rc = -ENXIO);
2562         } else if (KEY_IS(KEY_LAST_ID)) {
2563                 struct obd_id_info *info = val;
2564                 __u32 size = sizeof(obd_id);
2565                 struct lov_tgt_desc *tgt;
2566
2567                 LASSERT(*vallen == sizeof(struct obd_id_info));
2568                 tgt = lov->lov_tgts[info->idx];
2569
2570                 if (!tgt || !tgt->ltd_active)
2571                         GOTO(out, rc = -ESRCH);
2572
2573                 rc = obd_get_info(tgt->ltd_exp, keylen, key, &size, info->data, NULL);
2574                 GOTO(out, rc = 0);
2575         } else if (KEY_IS(KEY_LOVDESC)) {
2576                 struct lov_desc *desc_ret = val;
2577                 *desc_ret = lov->desc;
2578
2579                 GOTO(out, rc = 0);
2580         } else if (KEY_IS(KEY_FIEMAP)) {
2581                 rc = lov_fiemap(lov, keylen, key, vallen, val, lsm);
2582                 GOTO(out, rc);
2583         } else if (KEY_IS(KEY_CONNECT_FLAG)) {
2584                 struct lov_tgt_desc *tgt;
2585                 __u64 ost_idx = *((__u64*)val);
2586
2587                 LASSERT(*vallen == sizeof(__u64));
2588                 LASSERT(ost_idx < lov->desc.ld_tgt_count);
2589                 tgt = lov->lov_tgts[ost_idx];
2590
2591                 if (!tgt || !tgt->ltd_exp)
2592                         GOTO(out, rc = -ESRCH);
2593
2594                 *((__u64*)val) = tgt->ltd_exp->exp_connect_flags;
2595                 GOTO(out, rc = 0);
2596         }
2597
2598         rc = -EINVAL;
2599
2600 out:
2601         obd_putref(obddev);
2602         RETURN(rc);
2603 }
2604
2605 static int lov_set_info_async(struct obd_export *exp, obd_count keylen,
2606                               void *key, obd_count vallen, void *val,
2607                               struct ptlrpc_request_set *set)
2608 {
2609         struct obd_device *obddev = class_exp2obd(exp);
2610         struct lov_obd *lov = &obddev->u.lov;
2611         obd_count count;
2612         int i, rc = 0, err;
2613         struct lov_tgt_desc *tgt;
2614         unsigned incr, check_uuid,
2615                  do_inactive, no_set;
2616         unsigned next_id = 0,  mds_con = 0, capa = 0;
2617         ENTRY;
2618
2619         incr = check_uuid = do_inactive = no_set = 0;
2620         if (set == NULL) {
2621                 no_set = 1;
2622                 set = ptlrpc_prep_set();
2623                 if (!set)
2624                         RETURN(-ENOMEM);
2625         }
2626
2627         obd_getref(obddev);
2628         count = lov->desc.ld_tgt_count;
2629
2630         if (KEY_IS(KEY_NEXT_ID)) {
2631                 count = vallen / sizeof(struct obd_id_info);
2632                 vallen = sizeof(obd_id);
2633                 incr = sizeof(struct obd_id_info);
2634                 do_inactive = 1;
2635                 next_id = 1;
2636         } else if (KEY_IS(KEY_CHECKSUM)) {
2637                 do_inactive = 1;
2638         } else if (KEY_IS(KEY_EVICT_BY_NID)) {
2639                 /* use defaults:  do_inactive = incr = 0; */
2640         } else if (KEY_IS(KEY_MDS_CONN)) {
2641                 mds_con = 1;
2642         } else if (KEY_IS(KEY_CAPA_KEY)) {
2643                 capa = 1;
2644         }
2645
2646         for (i = 0; i < count; i++, val = (char *)val + incr) {
2647                 if (next_id) {
2648                         tgt = lov->lov_tgts[((struct obd_id_info*)val)->idx];
2649                 } else {
2650                         tgt = lov->lov_tgts[i];
2651                 }
2652                 /* OST was disconnected */
2653                 if (!tgt || !tgt->ltd_exp)
2654                         continue;
2655
2656                 /* OST is inactive and we don't want inactive OSCs */
2657                 if (!tgt->ltd_active && !do_inactive)
2658                         continue;
2659
2660                 if (mds_con) {
2661                         struct mds_group_info *mgi;
2662
2663                         LASSERT(vallen == sizeof(*mgi));
2664                         mgi = (struct mds_group_info *)val;
2665
2666                         /* Only want a specific OSC */
2667                         if (mgi->uuid && !obd_uuid_equals(mgi->uuid,
2668                                                 &tgt->ltd_uuid))
2669                                 continue;
2670
2671                         err = obd_set_info_async(tgt->ltd_exp,
2672                                          keylen, key, sizeof(int),
2673                                          &mgi->group, set);
2674                 } else if (next_id) {
2675                         err = obd_set_info_async(tgt->ltd_exp,
2676                                          keylen, key, vallen,
2677                                          ((struct obd_id_info*)val)->data, set);
2678                 } else if (capa) {
2679                         struct mds_capa_info *info = (struct mds_capa_info*)val;
2680
2681                         LASSERT(vallen == sizeof(*info));
2682
2683                          /* Only want a specific OSC */
2684                         if (info->uuid &&
2685                             !obd_uuid_equals(info->uuid, &tgt->ltd_uuid))
2686                                 continue;
2687
2688                         err = obd_set_info_async(tgt->ltd_exp, keylen, key,
2689                                                  sizeof(*info->capa),
2690                                                  info->capa, set);
2691
2692                 } else {
2693                         /* Only want a specific OSC */
2694                         if (check_uuid &&
2695                             !obd_uuid_equals(val, &tgt->ltd_uuid))
2696                                 continue;
2697
2698                         err = obd_set_info_async(tgt->ltd_exp,
2699                                          keylen, key, vallen, val, set);
2700                 }
2701
2702                 if (!rc)
2703                         rc = err;
2704         }
2705
2706         obd_putref(obddev);
2707         if (no_set) {
2708                 err = ptlrpc_set_wait(set);
2709                 if (!rc)
2710                         rc = err;
2711                 ptlrpc_set_destroy(set);
2712         }
2713         RETURN(rc);
2714 }
2715
2716 int lov_test_and_clear_async_rc(struct lov_stripe_md *lsm)
2717 {
2718         int i, rc = 0;
2719         ENTRY;
2720
2721         for (i = 0; i < lsm->lsm_stripe_count; i++) {
2722                 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
2723                 if (loi->loi_ar.ar_rc && !rc)
2724                         rc = loi->loi_ar.ar_rc;
2725                 loi->loi_ar.ar_rc = 0;
2726         }
2727         RETURN(rc);
2728 }
2729 EXPORT_SYMBOL(lov_test_and_clear_async_rc);
2730
2731
2732 static int lov_extent_calc(struct obd_export *exp, struct lov_stripe_md *lsm,
2733                            int cmd, __u64 *offset)
2734 {
2735         __u32 ssize = lsm->lsm_stripe_size;
2736         __u64 start;
2737
2738         start = *offset;
2739         do_div(start, ssize);
2740         start = start * ssize;
2741
2742         CDEBUG(D_DLMTRACE, "offset "LPU64", stripe %u, start "LPU64
2743                            ", end "LPU64"\n", *offset, ssize, start,
2744                            start + ssize - 1);
2745         if (cmd == OBD_CALC_STRIPE_END) {
2746                 *offset = start + ssize - 1;
2747         } else if (cmd == OBD_CALC_STRIPE_START) {
2748                 *offset = start;
2749         } else {
2750                 LBUG();
2751         }
2752
2753         RETURN(0);
2754 }
2755
2756 void lov_stripe_lock(struct lov_stripe_md *md)
2757 {
2758         LASSERT(md->lsm_lock_owner != cfs_curproc_pid());
2759         cfs_spin_lock(&md->lsm_lock);
2760         LASSERT(md->lsm_lock_owner == 0);
2761         md->lsm_lock_owner = cfs_curproc_pid();
2762 }
2763 EXPORT_SYMBOL(lov_stripe_lock);
2764
2765 void lov_stripe_unlock(struct lov_stripe_md *md)
2766 {
2767         LASSERT(md->lsm_lock_owner == cfs_curproc_pid());
2768         md->lsm_lock_owner = 0;
2769         cfs_spin_unlock(&md->lsm_lock);
2770 }
2771 EXPORT_SYMBOL(lov_stripe_unlock);
2772
2773
2774 struct obd_ops lov_obd_ops = {
2775         .o_owner               = THIS_MODULE,
2776         .o_setup               = lov_setup,
2777         .o_precleanup          = lov_precleanup,
2778         .o_cleanup             = lov_cleanup,
2779         //.o_process_config      = lov_process_config,
2780         .o_connect             = lov_connect,
2781         .o_disconnect          = lov_disconnect,
2782         .o_statfs              = lov_statfs,
2783         .o_statfs_async        = lov_statfs_async,
2784         .o_packmd              = lov_packmd,
2785         .o_unpackmd            = lov_unpackmd,
2786         .o_create              = lov_create,
2787         .o_destroy             = lov_destroy,
2788         .o_getattr             = lov_getattr,
2789         .o_getattr_async       = lov_getattr_async,
2790         .o_setattr             = lov_setattr,
2791         .o_setattr_async       = lov_setattr_async,
2792         .o_brw                 = lov_brw,
2793         .o_merge_lvb           = lov_merge_lvb,
2794         .o_adjust_kms          = lov_adjust_kms,
2795         .o_punch               = lov_punch,
2796         .o_sync                = lov_sync,
2797         .o_enqueue             = lov_enqueue,
2798         .o_change_cbdata       = lov_change_cbdata,
2799         .o_find_cbdata         = lov_find_cbdata,
2800         .o_cancel              = lov_cancel,
2801         .o_cancel_unused       = lov_cancel_unused,
2802         .o_iocontrol           = lov_iocontrol,
2803         .o_get_info            = lov_get_info,
2804         .o_set_info_async      = lov_set_info_async,
2805         .o_extent_calc         = lov_extent_calc,
2806         .o_llog_init           = lov_llog_init,
2807         .o_llog_finish         = lov_llog_finish,
2808         .o_notify              = lov_notify,
2809         .o_pool_new            = lov_pool_new,
2810         .o_pool_rem            = lov_pool_remove,
2811         .o_pool_add            = lov_pool_add,
2812         .o_pool_del            = lov_pool_del,
2813         .o_getref              = lov_getref,
2814         .o_putref              = lov_putref,
2815 };
2816
2817 static quota_interface_t *quota_interface;
2818 extern quota_interface_t lov_quota_interface;
2819
2820 cfs_mem_cache_t *lov_oinfo_slab;
2821
2822 extern struct lu_kmem_descr lov_caches[];
2823
2824 int __init lov_init(void)
2825 {
2826         struct lprocfs_static_vars lvars = { 0 };
2827         int rc, rc2;
2828         ENTRY;
2829
2830         /* print an address of _any_ initialized kernel symbol from this
2831          * module, to allow debugging with gdb that doesn't support data
2832          * symbols from modules.*/
2833         CDEBUG(D_CONSOLE, "Lustre LOV module (%p).\n", &lov_caches);
2834
2835         rc = lu_kmem_init(lov_caches);
2836         if (rc)
2837                 return rc;
2838
2839         lov_oinfo_slab = cfs_mem_cache_create("lov_oinfo",
2840                                               sizeof(struct lov_oinfo),
2841                                               0, CFS_SLAB_HWCACHE_ALIGN);
2842         if (lov_oinfo_slab == NULL) {
2843                 lu_kmem_fini(lov_caches);
2844                 return -ENOMEM;
2845         }
2846         lprocfs_lov_init_vars(&lvars);
2847
2848         cfs_request_module("lquota");
2849         quota_interface = PORTAL_SYMBOL_GET(lov_quota_interface);
2850         init_obd_quota_ops(quota_interface, &lov_obd_ops);
2851
2852         rc = class_register_type(&lov_obd_ops, NULL, lvars.module_vars,
2853                                  LUSTRE_LOV_NAME, &lov_device_type);
2854
2855         if (rc) {
2856                 if (quota_interface)
2857                         PORTAL_SYMBOL_PUT(lov_quota_interface);
2858                 rc2 = cfs_mem_cache_destroy(lov_oinfo_slab);
2859                 LASSERT(rc2 == 0);
2860                 lu_kmem_fini(lov_caches);
2861         }
2862
2863         RETURN(rc);
2864 }
2865
2866 #ifdef __KERNEL__
2867 static void /*__exit*/ lov_exit(void)
2868 {
2869         int rc;
2870
2871         lu_device_type_fini(&lov_device_type);
2872         lu_kmem_fini(lov_caches);
2873
2874         if (quota_interface)
2875                 PORTAL_SYMBOL_PUT(lov_quota_interface);
2876
2877         class_unregister_type(LUSTRE_LOV_NAME);
2878         rc = cfs_mem_cache_destroy(lov_oinfo_slab);
2879         LASSERT(rc == 0);
2880 }
2881
2882 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2883 MODULE_DESCRIPTION("Lustre Logical Object Volume OBD driver");
2884 MODULE_LICENSE("GPL");
2885
2886 cfs_module(lov, LUSTRE_VERSION_STRING, lov_init, lov_exit);
2887 #endif