lustre/lov/lov_obd.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2012, Whamcloud, Inc.
  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 #define DEBUG_SUBSYSTEM S_LOV
  45 #ifdef __KERNEL__
  46 #include <libcfs/libcfs.h>
  47 #else
  48 #include <liblustre.h>
  49 #endif
  50
  51 #include <obd_support.h>
  52 #include <lustre_lib.h>
  53 #include <lustre_net.h>
  54 #include <lustre/lustre_idl.h>
  55 #include <lustre_dlm.h>
  56 #include <lustre_mds.h>
  57 #include <lustre_debug.h>
  58 #include <obd_class.h>
  59 #include <obd_lov.h>
  60 #include <obd_ost.h>
  61 #include <lprocfs_status.h>
  62 #include <lustre_param.h>
  63 #include <cl_object.h>
  64 #include <lclient.h> /* for cl_client_lru */
  65 #include <lustre/ll_fiemap.h>
  66 #include <lustre_log.h>
  67
  68 #include "lov_internal.h"
  69
  70 /* Keep a refcount of lov->tgt usage to prevent racing with addition/deletion.
  71    Any function that expects lov_tgts to remain stationary must take a ref. */
  72 static void lov_getref(struct obd_device *obd)
  73 {
  74         struct lov_obd *lov = &obd->u.lov;
  75
  76         /* nobody gets through here until lov_putref is done */
  77         mutex_lock(&lov->lov_lock);
  78         cfs_atomic_inc(&lov->lov_refcount);
  79         mutex_unlock(&lov->lov_lock);
  80         return;
  81 }
  82
  83 static void __lov_del_obd(struct obd_device *obd, struct lov_tgt_desc *tgt);
  84
  85 static void lov_putref(struct obd_device *obd)
  86 {
  87         struct lov_obd *lov = &obd->u.lov;
  88
  89         mutex_lock(&lov->lov_lock);
  90         /* ok to dec to 0 more than once -- ltd_exp's will be null */
  91         if (cfs_atomic_dec_and_test(&lov->lov_refcount) && lov->lov_death_row) {
  92                 CFS_LIST_HEAD(kill);
  93                 int i;
  94                 struct lov_tgt_desc *tgt, *n;
  95                 CDEBUG(D_CONFIG, "destroying %d lov targets\n",
  96                        lov->lov_death_row);
  97                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
  98                         tgt = lov->lov_tgts[i];
  99
 100                         if (!tgt || !tgt->ltd_reap)
 101                                 continue;
 102                         cfs_list_add(&tgt->ltd_kill, &kill);
 103                         /* XXX - right now there is a dependency on ld_tgt_count
 104                          * being the maximum tgt index for computing the
 105                          * mds_max_easize. So we can't shrink it. */
 106                         lov_ost_pool_remove(&lov->lov_packed, i);
 107                         lov->lov_tgts[i] = NULL;
 108                         lov->lov_death_row--;
 109                 }
 110                 mutex_unlock(&lov->lov_lock);
 111
 112                 cfs_list_for_each_entry_safe(tgt, n, &kill, ltd_kill) {
 113                         cfs_list_del(&tgt->ltd_kill);
 114                         /* Disconnect */
 115                         __lov_del_obd(obd, tgt);
 116                 }
 117         } else {
 118                 mutex_unlock(&lov->lov_lock);
 119         }
 120 }
 121
 122 static int lov_set_osc_active(struct obd_device *obd, struct obd_uuid *uuid,
 123                               enum obd_notify_event ev);
 124 static int lov_notify(struct obd_device *obd, struct obd_device *watched,
 125                       enum obd_notify_event ev, void *data);
 126
 127
 128 #define MAX_STRING_SIZE 128
 129 int lov_connect_obd(struct obd_device *obd, __u32 index, int activate,
 130                     struct obd_connect_data *data)
 131 {
 132         struct lov_obd *lov = &obd->u.lov;
 133         struct obd_uuid *tgt_uuid;
 134         struct obd_device *tgt_obd;
 135         static struct obd_uuid lov_osc_uuid = { "LOV_OSC_UUID" };
 136         struct obd_import *imp;
 137 #ifdef __KERNEL__
 138         cfs_proc_dir_entry_t *lov_proc_dir;
 139 #endif
 140         int rc;
 141         ENTRY;
 142
 143         if (!lov->lov_tgts[index])
 144                 RETURN(-EINVAL);
 145
 146         tgt_uuid = &lov->lov_tgts[index]->ltd_uuid;
 147         tgt_obd = lov->lov_tgts[index]->ltd_obd;
 148
 149         if (!tgt_obd->obd_set_up) {
 150                 CERROR("Target %s not set up\n", obd_uuid2str(tgt_uuid));
 151                 RETURN(-EINVAL);
 152         }
 153
 154         /* override the sp_me from lov */
 155         tgt_obd->u.cli.cl_sp_me = lov->lov_sp_me;
 156
 157         if (data && (data->ocd_connect_flags & OBD_CONNECT_INDEX))
 158                 data->ocd_index = index;
 159
 160         /*
 161          * Divine LOV knows that OBDs under it are OSCs.
 162          */
 163         imp = tgt_obd->u.cli.cl_import;
 164
 165         if (activate) {
 166                 tgt_obd->obd_no_recov = 0;
 167                 /* FIXME this is probably supposed to be
 168                    ptlrpc_set_import_active.  Horrible naming. */
 169                 ptlrpc_activate_import(imp);
 170         }
 171
 172         rc = obd_register_observer(tgt_obd, obd);
 173         if (rc) {
 174                 CERROR("Target %s register_observer error %d\n",
 175                        obd_uuid2str(tgt_uuid), rc);
 176                 RETURN(rc);
 177         }
 178
 179
 180         if (imp->imp_invalid) {
 181                 CDEBUG(D_CONFIG, "not connecting OSC %s; administratively "
 182                        "disabled\n", obd_uuid2str(tgt_uuid));
 183                 RETURN(0);
 184         }
 185
 186         rc = obd_connect(NULL, &lov->lov_tgts[index]->ltd_exp, tgt_obd,
 187                          &lov_osc_uuid, data, NULL);
 188         if (rc || !lov->lov_tgts[index]->ltd_exp) {
 189                 CERROR("Target %s connect error %d\n",
 190                        obd_uuid2str(tgt_uuid), rc);
 191                 RETURN(-ENODEV);
 192         }
 193
 194         lov->lov_tgts[index]->ltd_reap = 0;
 195
 196         CDEBUG(D_CONFIG, "Connected tgt idx %d %s (%s) %sactive\n", index,
 197                obd_uuid2str(tgt_uuid), tgt_obd->obd_name, activate ? "":"in");
 198
 199 #ifdef __KERNEL__
 200         lov_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
 201         if (lov_proc_dir) {
 202                 struct obd_device *osc_obd = lov->lov_tgts[index]->ltd_exp->exp_obd;
 203                 cfs_proc_dir_entry_t *osc_symlink;
 204
 205                 LASSERT(osc_obd != NULL);
 206                 LASSERT(osc_obd->obd_magic == OBD_DEVICE_MAGIC);
 207                 LASSERT(osc_obd->obd_type->typ_name != NULL);
 208
 209                 osc_symlink = lprocfs_add_symlink(osc_obd->obd_name,
 210                                                   lov_proc_dir,
 211                                                   "../../../%s/%s",
 212                                                   osc_obd->obd_type->typ_name,
 213                                                   osc_obd->obd_name);
 214                 if (osc_symlink == NULL) {
 215                         CERROR("could not register LOV target "
 216                                 "/proc/fs/lustre/%s/%s/target_obds/%s.",
 217                                 obd->obd_type->typ_name, obd->obd_name,
 218                                 osc_obd->obd_name);
 219                         lprocfs_remove(&lov_proc_dir);
 220                 }
 221         }
 222 #endif
 223
 224         RETURN(0);
 225 }
 226
 227 static int lov_connect(const struct lu_env *env,
 228                        struct obd_export **exp, struct obd_device *obd,
 229                        struct obd_uuid *cluuid, struct obd_connect_data *data,
 230                        void *localdata)
 231 {
 232         struct lov_obd *lov = &obd->u.lov;
 233         struct lov_tgt_desc *tgt;
 234         struct lustre_handle conn;
 235         int i, rc;
 236         ENTRY;
 237
 238         CDEBUG(D_CONFIG, "connect #%d\n", lov->lov_connects);
 239
 240         rc = class_connect(&conn, obd, cluuid);
 241         if (rc)
 242                 RETURN(rc);
 243
 244         *exp = class_conn2export(&conn);
 245
 246         /* Why should there ever be more than 1 connect? */
 247         lov->lov_connects++;
 248         LASSERT(lov->lov_connects == 1);
 249
 250         memset(&lov->lov_ocd, 0, sizeof(lov->lov_ocd));
 251         if (data)
 252                 lov->lov_ocd = *data;
 253
 254         obd_getref(obd);
 255         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
 256                 tgt = lov->lov_tgts[i];
 257                 if (!tgt || obd_uuid_empty(&tgt->ltd_uuid))
 258                         continue;
 259                 /* Flags will be lowest common denominator */
 260                 rc = lov_connect_obd(obd, i, tgt->ltd_activate, &lov->lov_ocd);
 261                 if (rc) {
 262                         CERROR("%s: lov connect tgt %d failed: %d\n",
 263                                obd->obd_name, i, rc);
 264                         continue;
 265                 }
 266                 /* connect to administrative disabled ost */
 267                 if (!lov->lov_tgts[i]->ltd_exp)
 268                         continue;
 269
 270                 rc = lov_notify(obd, lov->lov_tgts[i]->ltd_exp->exp_obd,
 271                                 OBD_NOTIFY_CONNECT, (void *)&i);
 272                 if (rc) {
 273                         CERROR("%s error sending notify %d\n",
 274                                obd->obd_name, rc);
 275                 }
 276         }
 277         obd_putref(obd);
 278
 279         RETURN(0);
 280 }
 281
 282 static int lov_disconnect_obd(struct obd_device *obd, struct lov_tgt_desc *tgt)
 283 {
 284         cfs_proc_dir_entry_t *lov_proc_dir;
 285         struct lov_obd *lov = &obd->u.lov;
 286         struct obd_device *osc_obd;
 287         int rc;
 288         ENTRY;
 289
 290         osc_obd = class_exp2obd(tgt->ltd_exp);
 291         CDEBUG(D_CONFIG, "%s: disconnecting target %s\n",
 292                obd->obd_name, osc_obd->obd_name);
 293
 294         if (tgt->ltd_active) {
 295                 tgt->ltd_active = 0;
 296                 lov->desc.ld_active_tgt_count--;
 297                 tgt->ltd_exp->exp_obd->obd_inactive = 1;
 298         }
 299
 300         lov_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
 301         if (lov_proc_dir) {
 302                 cfs_proc_dir_entry_t *osc_symlink;
 303
 304                 osc_symlink = lprocfs_srch(lov_proc_dir, osc_obd->obd_name);
 305                 if (osc_symlink) {
 306                         lprocfs_remove(&osc_symlink);
 307                 } else {
 308                         CERROR("/proc/fs/lustre/%s/%s/target_obds/%s missing.",
 309                                obd->obd_type->typ_name, obd->obd_name,
 310                                osc_obd->obd_name);
 311                 }
 312         }
 313
 314         if (osc_obd) {
 315                 /* Pass it on to our clients.
 316                  * XXX This should be an argument to disconnect,
 317                  * XXX not a back-door flag on the OBD.  Ah well.
 318                  */
 319                 osc_obd->obd_force = obd->obd_force;
 320                 osc_obd->obd_fail = obd->obd_fail;
 321                 osc_obd->obd_no_recov = obd->obd_no_recov;
 322         }
 323
 324         obd_register_observer(osc_obd, NULL);
 325
 326         rc = obd_disconnect(tgt->ltd_exp);
 327         if (rc) {
 328                 CERROR("Target %s disconnect error %d\n",
 329                        tgt->ltd_uuid.uuid, rc);
 330                 rc = 0;
 331         }
 332
 333         tgt->ltd_exp = NULL;
 334         RETURN(0);
 335 }
 336
 337 static int lov_disconnect(struct obd_export *exp)
 338 {
 339         struct obd_device *obd = class_exp2obd(exp);
 340         struct lov_obd *lov = &obd->u.lov;
 341         int i, rc;
 342         ENTRY;
 343
 344         if (!lov->lov_tgts)
 345                 goto out;
 346
 347         /* Only disconnect the underlying layers on the final disconnect. */
 348         lov->lov_connects--;
 349         if (lov->lov_connects != 0) {
 350                 /* why should there be more than 1 connect? */
 351                 CERROR("disconnect #%d\n", lov->lov_connects);
 352                 goto out;
 353         }
 354
 355         /* Let's hold another reference so lov_del_obd doesn't spin through
 356            putref every time */
 357         obd_getref(obd);
 358
 359         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
 360                 if (lov->lov_tgts[i] && lov->lov_tgts[i]->ltd_exp) {
 361                         /* Disconnection is the last we know about an obd */
 362                         lov_del_target(obd, i, 0, lov->lov_tgts[i]->ltd_gen);
 363                 }
 364         }
 365         obd_putref(obd);
 366
 367 out:
 368         rc = class_disconnect(exp); /* bz 9811 */
 369         RETURN(rc);
 370 }
 371
 372 /* Error codes:
 373  *
 374  *  -EINVAL  : UUID can't be found in the LOV's target list
 375  *  -ENOTCONN: The UUID is found, but the target connection is bad (!)
 376  *  -EBADF   : The UUID is found, but the OBD is the wrong type (!)
 377  *  any >= 0 : is log target index
 378  */
 379 static int lov_set_osc_active(struct obd_device *obd, struct obd_uuid *uuid,
 380                               enum obd_notify_event ev)
 381 {
 382         struct lov_obd *lov = &obd->u.lov;
 383         struct lov_tgt_desc *tgt;
 384         int index, activate, active;
 385         ENTRY;
 386
 387         CDEBUG(D_INFO, "Searching in lov %p for uuid %s event(%d)\n",
 388                lov, uuid->uuid, ev);
 389
 390         obd_getref(obd);
 391         for (index = 0; index < lov->desc.ld_tgt_count; index++) {
 392                 tgt = lov->lov_tgts[index];
 393                 if (!tgt || !tgt->ltd_exp)
 394                         continue;
 395
 396                 CDEBUG(D_INFO, "lov idx %d is %s conn "LPX64"\n",
 397                        index, obd_uuid2str(&tgt->ltd_uuid),
 398                        tgt->ltd_exp->exp_handle.h_cookie);
 399                 if (obd_uuid_equals(uuid, &tgt->ltd_uuid))
 400                         break;
 401         }
 402
 403         if (index == lov->desc.ld_tgt_count)
 404                 GOTO(out, index = -EINVAL);
 405
 406         if (ev == OBD_NOTIFY_DEACTIVATE || ev == OBD_NOTIFY_ACTIVATE) {
 407                 activate = (ev == OBD_NOTIFY_ACTIVATE) ? 1 : 0;
 408
 409                 if (lov->lov_tgts[index]->ltd_activate == activate) {
 410                         CDEBUG(D_INFO, "OSC %s already %sactivate!\n",
 411                                uuid->uuid, activate ? "" : "de");
 412                 } else {
 413                         lov->lov_tgts[index]->ltd_activate = activate;
 414                         CDEBUG(D_CONFIG, "%sactivate OSC %s\n",
 415                                activate ? "" : "de", obd_uuid2str(uuid));
 416                 }
 417
 418         } else if (ev == OBD_NOTIFY_INACTIVE || ev == OBD_NOTIFY_ACTIVE) {
 419                 active = (ev == OBD_NOTIFY_ACTIVE) ? 1 : 0;
 420
 421                 if (lov->lov_tgts[index]->ltd_active == active) {
 422                         CDEBUG(D_INFO, "OSC %s already %sactive!\n",
 423                                uuid->uuid, active ? "" : "in");
 424                         GOTO(out, index);
 425                 } else {
 426                         CDEBUG(D_CONFIG, "Marking OSC %s %sactive\n",
 427                                obd_uuid2str(uuid), active ? "" : "in");
 428                 }
 429
 430                 lov->lov_tgts[index]->ltd_active = active;
 431                 if (active) {
 432                         lov->desc.ld_active_tgt_count++;
 433                         lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 0;
 434                 } else {
 435                         lov->desc.ld_active_tgt_count--;
 436                         lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 1;
 437                 }
 438         } else {
 439                 CERROR("Unknown event(%d) for uuid %s", ev, uuid->uuid);
 440         }
 441
 442  out:
 443         obd_putref(obd);
 444         RETURN(index);
 445 }
 446
 447 static int lov_notify(struct obd_device *obd, struct obd_device *watched,
 448                       enum obd_notify_event ev, void *data)
 449 {
 450         int rc = 0;
 451         ENTRY;
 452
 453         if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE ||
 454             ev == OBD_NOTIFY_ACTIVATE || ev == OBD_NOTIFY_DEACTIVATE) {
 455                 struct obd_uuid *uuid;
 456
 457                 LASSERT(watched);
 458
 459                 if (strcmp(watched->obd_type->typ_name, LUSTRE_OSC_NAME)) {
 460                         CERROR("unexpected notification of %s %s!\n",
 461                                watched->obd_type->typ_name,
 462                                watched->obd_name);
 463                         RETURN(-EINVAL);
 464                 }
 465                 uuid = &watched->u.cli.cl_target_uuid;
 466
 467                 /* Set OSC as active before notifying the observer, so the
 468                  * observer can use the OSC normally.
 469                  */
 470                 rc = lov_set_osc_active(obd, uuid, ev);
 471                 if (rc < 0) {
 472                         CERROR("event(%d) of %s failed: %d\n", ev,
 473                                obd_uuid2str(uuid), rc);
 474                         RETURN(rc);
 475                 }
 476                 /* active event should be pass lov target index as data */
 477                 data = &rc;
 478         }
 479
 480         /* Pass the notification up the chain. */
 481         if (watched) {
 482                 rc = obd_notify_observer(obd, watched, ev, data);
 483         } else {
 484                 /* NULL watched means all osc's in the lov (only for syncs) */
 485                 /* sync event should be send lov idx as data */
 486                 struct lov_obd *lov = &obd->u.lov;
 487                 int i, is_sync;
 488
 489                 data = &i;
 490                 is_sync = (ev == OBD_NOTIFY_SYNC) ||
 491                           (ev == OBD_NOTIFY_SYNC_NONBLOCK);
 492
 493                 obd_getref(obd);
 494                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
 495                         if (!lov->lov_tgts[i])
 496                                 continue;
 497
 498                         /* don't send sync event if target not
 499                          * connected/activated */
 500                         if (is_sync &&  !lov->lov_tgts[i]->ltd_active)
 501                                 continue;
 502
 503                         rc = obd_notify_observer(obd, lov->lov_tgts[i]->ltd_obd,
 504                                                  ev, data);
 505                         if (rc) {
 506                                 CERROR("%s: notify %s of %s failed %d\n",
 507                                        obd->obd_name,
 508                                        obd->obd_observer->obd_name,
 509                                        lov->lov_tgts[i]->ltd_obd->obd_name,
 510                                        rc);
 511                         }
 512                 }
 513                 obd_putref(obd);
 514         }
 515
 516         RETURN(rc);
 517 }
 518
 519 static int lov_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
 520                           __u32 index, int gen, int active)
 521 {
 522         struct lov_obd *lov = &obd->u.lov;
 523         struct lov_tgt_desc *tgt;
 524         struct obd_device *tgt_obd;
 525         int rc;
 526         ENTRY;
 527
 528         CDEBUG(D_CONFIG, "uuid:%s idx:%d gen:%d active:%d\n",
 529                uuidp->uuid, index, gen, active);
 530
 531         if (gen <= 0) {
 532                 CERROR("request to add OBD %s with invalid generation: %d\n",
 533                        uuidp->uuid, gen);
 534                 RETURN(-EINVAL);
 535         }
 536
 537         tgt_obd = class_find_client_obd(uuidp, LUSTRE_OSC_NAME,
 538                                         &obd->obd_uuid);
 539         if (tgt_obd == NULL)
 540                 RETURN(-EINVAL);
 541
 542         mutex_lock(&lov->lov_lock);
 543
 544         if ((index < lov->lov_tgt_size) && (lov->lov_tgts[index] != NULL)) {
 545                 tgt = lov->lov_tgts[index];
 546                 CERROR("UUID %s already assigned at LOV target index %d\n",
 547                        obd_uuid2str(&tgt->ltd_uuid), index);
 548                 mutex_unlock(&lov->lov_lock);
 549                 RETURN(-EEXIST);
 550         }
 551
 552         if (index >= lov->lov_tgt_size) {
 553                 /* We need to reallocate the lov target array. */
 554                 struct lov_tgt_desc **newtgts, **old = NULL;
 555                 __u32 newsize, oldsize = 0;
 556
 557                 newsize = max(lov->lov_tgt_size, (__u32)2);
 558                 while (newsize < index + 1)
 559                         newsize = newsize << 1;
 560                 OBD_ALLOC(newtgts, sizeof(*newtgts) * newsize);
 561                 if (newtgts == NULL) {
 562                         mutex_unlock(&lov->lov_lock);
 563                         RETURN(-ENOMEM);
 564                 }
 565
 566                 if (lov->lov_tgt_size) {
 567                         memcpy(newtgts, lov->lov_tgts, sizeof(*newtgts) *
 568                                lov->lov_tgt_size);
 569                         old = lov->lov_tgts;
 570                         oldsize = lov->lov_tgt_size;
 571                 }
 572
 573                 lov->lov_tgts = newtgts;
 574                 lov->lov_tgt_size = newsize;
 575 #ifdef __KERNEL__
 576                 smp_rmb();
 577 #endif
 578                 if (old)
 579                         OBD_FREE(old, sizeof(*old) * oldsize);
 580
 581                 CDEBUG(D_CONFIG, "tgts: %p size: %d\n",
 582                        lov->lov_tgts, lov->lov_tgt_size);
 583         }
 584
 585         OBD_ALLOC_PTR(tgt);
 586         if (!tgt) {
 587                 mutex_unlock(&lov->lov_lock);
 588                 RETURN(-ENOMEM);
 589         }
 590
 591         rc = lov_ost_pool_add(&lov->lov_packed, index, lov->lov_tgt_size);
 592         if (rc) {
 593                 mutex_unlock(&lov->lov_lock);
 594                 OBD_FREE_PTR(tgt);
 595                 RETURN(rc);
 596         }
 597
 598         tgt->ltd_uuid = *uuidp;
 599         tgt->ltd_obd = tgt_obd;
 600         /* XXX - add a sanity check on the generation number. */
 601         tgt->ltd_gen = gen;
 602         tgt->ltd_index = index;
 603         tgt->ltd_activate = active;
 604         lov->lov_tgts[index] = tgt;
 605         if (index >= lov->desc.ld_tgt_count)
 606                 lov->desc.ld_tgt_count = index + 1;
 607
 608         mutex_unlock(&lov->lov_lock);
 609
 610         CDEBUG(D_CONFIG, "idx=%d ltd_gen=%d ld_tgt_count=%d\n",
 611                 index, tgt->ltd_gen, lov->desc.ld_tgt_count);
 612
 613         rc = obd_notify(obd, tgt_obd, OBD_NOTIFY_CREATE, &index);
 614
 615         if (lov->lov_connects == 0) {
 616                 /* lov_connect hasn't been called yet. We'll do the
 617                    lov_connect_obd on this target when that fn first runs,
 618                    because we don't know the connect flags yet. */
 619                 RETURN(0);
 620         }
 621
 622         obd_getref(obd);
 623
 624         rc = lov_connect_obd(obd, index, active, &lov->lov_ocd);
 625         if (rc)
 626                 GOTO(out, rc);
 627
 628         /* connect to administrative disabled ost */
 629         if (!tgt->ltd_exp)
 630                 GOTO(out, rc = 0);
 631
 632         if (lov->lov_cache != NULL) {
 633                 rc = obd_set_info_async(NULL, tgt->ltd_exp,
 634                                 sizeof(KEY_CACHE_SET), KEY_CACHE_SET,
 635                                 sizeof(struct cl_client_cache), lov->lov_cache,
 636                                 NULL);
 637                 if (rc < 0)
 638                         GOTO(out, rc);
 639         }
 640
 641         rc = lov_notify(obd, tgt->ltd_exp->exp_obd,
 642                         active ? OBD_NOTIFY_CONNECT : OBD_NOTIFY_INACTIVE,
 643                         (void *)&index);
 644
 645 out:
 646         if (rc) {
 647                 CERROR("add failed (%d), deleting %s\n", rc,
 648                        obd_uuid2str(&tgt->ltd_uuid));
 649                 lov_del_target(obd, index, 0, 0);
 650         }
 651         obd_putref(obd);
 652         RETURN(rc);
 653 }
 654
 655 /* Schedule a target for deletion */
 656 int lov_del_target(struct obd_device *obd, __u32 index,
 657                    struct obd_uuid *uuidp, int gen)
 658 {
 659         struct lov_obd *lov = &obd->u.lov;
 660         int count = lov->desc.ld_tgt_count;
 661         int rc = 0;
 662         ENTRY;
 663
 664         if (index >= count) {
 665                 CERROR("LOV target index %d >= number of LOV OBDs %d.\n",
 666                        index, count);
 667                 RETURN(-EINVAL);
 668         }
 669
 670         obd_getref(obd);
 671
 672         if (!lov->lov_tgts[index]) {
 673                 CERROR("LOV target at index %d is not setup.\n", index);
 674                 GOTO(out, rc = -EINVAL);
 675         }
 676
 677         if (uuidp && !obd_uuid_equals(uuidp, &lov->lov_tgts[index]->ltd_uuid)) {
 678                 CERROR("LOV target UUID %s at index %d doesn't match %s.\n",
 679                        lov_uuid2str(lov, index), index,
 680                        obd_uuid2str(uuidp));
 681                 GOTO(out, rc = -EINVAL);
 682         }
 683
 684         CDEBUG(D_CONFIG, "uuid: %s idx: %d gen: %d exp: %p active: %d\n",
 685                lov_uuid2str(lov, index), index,
 686                lov->lov_tgts[index]->ltd_gen, lov->lov_tgts[index]->ltd_exp,
 687                lov->lov_tgts[index]->ltd_active);
 688
 689         lov->lov_tgts[index]->ltd_reap = 1;
 690         lov->lov_death_row++;
 691         /* we really delete it from obd_putref */
 692 out:
 693         obd_putref(obd);
 694
 695         RETURN(rc);
 696 }
 697
 698 static void __lov_del_obd(struct obd_device *obd, struct lov_tgt_desc *tgt)
 699 {
 700         struct obd_device *osc_obd;
 701
 702         LASSERT(tgt);
 703         LASSERT(tgt->ltd_reap);
 704
 705         osc_obd = class_exp2obd(tgt->ltd_exp);
 706
 707         CDEBUG(D_CONFIG, "Removing tgt %s : %s\n",
 708                tgt->ltd_uuid.uuid,
 709                osc_obd ? osc_obd->obd_name : "<no obd>");
 710
 711         if (tgt->ltd_exp)
 712                 lov_disconnect_obd(obd, tgt);
 713
 714         OBD_FREE_PTR(tgt);
 715
 716         /* Manual cleanup - no cleanup logs to clean up the osc's.  We must
 717            do it ourselves. And we can't do it from lov_cleanup,
 718            because we just lost our only reference to it. */
 719         if (osc_obd)
 720                 class_manual_cleanup(osc_obd);
 721 }
 722
 723 void lov_fix_desc_stripe_size(__u64 *val)
 724 {
 725         if (*val < PTLRPC_MAX_BRW_SIZE) {
 726                 LCONSOLE_WARN("Increasing default stripe size to min %u\n",
 727                               PTLRPC_MAX_BRW_SIZE);
 728                 *val = PTLRPC_MAX_BRW_SIZE;
 729         } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) {
 730                 *val &= ~(LOV_MIN_STRIPE_SIZE - 1);
 731                 LCONSOLE_WARN("Changing default stripe size to "LPU64" (a "
 732                               "multiple of %u)\n",
 733                               *val, LOV_MIN_STRIPE_SIZE);
 734         }
 735 }
 736
 737 void lov_fix_desc_stripe_count(__u32 *val)
 738 {
 739         if (*val == 0)
 740                 *val = 1;
 741 }
 742
 743 void lov_fix_desc_pattern(__u32 *val)
 744 {
 745         /* from lov_setstripe */
 746         if ((*val != 0) && (*val != LOV_PATTERN_RAID0)) {
 747                 LCONSOLE_WARN("Unknown stripe pattern: %#x\n", *val);
 748                 *val = 0;
 749         }
 750 }
 751
 752 void lov_fix_desc_qos_maxage(__u32 *val)
 753 {
 754         /* fix qos_maxage */
 755         if (*val == 0)
 756                 *val = QOS_DEFAULT_MAXAGE;
 757 }
 758
 759 void lov_fix_desc(struct lov_desc *desc)
 760 {
 761         lov_fix_desc_stripe_size(&desc->ld_default_stripe_size);
 762         lov_fix_desc_stripe_count(&desc->ld_default_stripe_count);
 763         lov_fix_desc_pattern(&desc->ld_pattern);
 764         lov_fix_desc_qos_maxage(&desc->ld_qos_maxage);
 765 }
 766
 767 int lov_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
 768 {
 769         struct lprocfs_static_vars lvars = { 0 };
 770         struct lov_desc *desc;
 771         struct lov_obd *lov = &obd->u.lov;
 772         int rc;
 773         ENTRY;
 774
 775         if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
 776                 CERROR("LOV setup requires a descriptor\n");
 777                 RETURN(-EINVAL);
 778         }
 779
 780         desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
 781
 782         if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
 783                 CERROR("descriptor size wrong: %d > %d\n",
 784                        (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
 785                 RETURN(-EINVAL);
 786         }
 787
 788         if (desc->ld_magic != LOV_DESC_MAGIC) {
 789                 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
 790                             CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
 791                                    obd->obd_name, desc);
 792                             lustre_swab_lov_desc(desc);
 793                 } else {
 794                         CERROR("%s: Bad lov desc magic: %#x\n",
 795                                obd->obd_name, desc->ld_magic);
 796                         RETURN(-EINVAL);
 797                 }
 798         }
 799
 800         lov_fix_desc(desc);
 801
 802         desc->ld_active_tgt_count = 0;
 803         lov->desc = *desc;
 804         lov->lov_tgt_size = 0;
 805
 806         mutex_init(&lov->lov_lock);
 807         cfs_atomic_set(&lov->lov_refcount, 0);
 808         lov->lov_sp_me = LUSTRE_SP_CLI;
 809
 810         lov->lov_pools_hash_body = cfs_hash_create("POOLS", HASH_POOLS_CUR_BITS,
 811                                                    HASH_POOLS_MAX_BITS,
 812                                                    HASH_POOLS_BKT_BITS, 0,
 813                                                    CFS_HASH_MIN_THETA,
 814                                                    CFS_HASH_MAX_THETA,
 815                                                    &pool_hash_operations,
 816                                                    CFS_HASH_DEFAULT);
 817         CFS_INIT_LIST_HEAD(&lov->lov_pool_list);
 818         lov->lov_pool_count = 0;
 819         rc = lov_ost_pool_init(&lov->lov_packed, 0);
 820         if (rc)
 821                 GOTO(out, rc);
 822
 823         lprocfs_lov_init_vars(&lvars);
 824         lprocfs_obd_setup(obd, lvars.obd_vars);
 825 #ifdef LPROCFS
 826         {
 827                 int rc;
 828
 829                 rc = lprocfs_seq_create(obd->obd_proc_entry, "target_obd",
 830                                         0444, &lov_proc_target_fops, obd);
 831                 if (rc)
 832                         CWARN("Error adding the target_obd file\n");
 833         }
 834 #endif
 835         lov->lov_pool_proc_entry = lprocfs_register("pools",
 836                                                     obd->obd_proc_entry,
 837                                                     NULL, NULL);
 838
 839         RETURN(0);
 840
 841 out:
 842         return rc;
 843 }
 844
 845 static int lov_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
 846 {
 847         int rc = 0;
 848         struct lov_obd *lov = &obd->u.lov;
 849
 850         ENTRY;
 851
 852         switch (stage) {
 853         case OBD_CLEANUP_EARLY: {
 854                 int i;
 855                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
 856                         if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_active)
 857                                 continue;
 858                         obd_precleanup(class_exp2obd(lov->lov_tgts[i]->ltd_exp),
 859                                        OBD_CLEANUP_EARLY);
 860                 }
 861                 break;
 862         }
 863         case OBD_CLEANUP_EXPORTS:
 864                 rc = obd_llog_finish(obd, 0);
 865                 if (rc != 0)
 866                         CERROR("failed to cleanup llogging subsystems\n");
 867                 break;
 868         }
 869         RETURN(rc);
 870 }
 871
 872 static int lov_cleanup(struct obd_device *obd)
 873 {
 874         struct lov_obd *lov = &obd->u.lov;
 875         cfs_list_t *pos, *tmp;
 876         struct pool_desc *pool;
 877         ENTRY;
 878
 879         cfs_list_for_each_safe(pos, tmp, &lov->lov_pool_list) {
 880                 pool = cfs_list_entry(pos, struct pool_desc, pool_list);
 881                 /* free pool structs */
 882                 CDEBUG(D_INFO, "delete pool %p\n", pool);
 883                 /* In the function below, .hs_keycmp resolves to
 884                  * pool_hashkey_keycmp() */
 885                 /* coverity[overrun-buffer-val] */
 886                 lov_pool_del(obd, pool->pool_name);
 887         }
 888         cfs_hash_putref(lov->lov_pools_hash_body);
 889         lov_ost_pool_free(&lov->lov_packed);
 890
 891         lprocfs_obd_cleanup(obd);
 892         if (lov->lov_tgts) {
 893                 int i;
 894                 obd_getref(obd);
 895                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
 896                         if (!lov->lov_tgts[i])
 897                                 continue;
 898
 899                         /* Inactive targets may never have connected */
 900                         if (lov->lov_tgts[i]->ltd_active ||
 901                             cfs_atomic_read(&lov->lov_refcount))
 902                             /* We should never get here - these
 903                                should have been removed in the
 904                              disconnect. */
 905                                 CERROR("lov tgt %d not cleaned!"
 906                                        " deathrow=%d, lovrc=%d\n",
 907                                        i, lov->lov_death_row,
 908                                        cfs_atomic_read(&lov->lov_refcount));
 909                         lov_del_target(obd, i, 0, 0);
 910                 }
 911                 obd_putref(obd);
 912                 OBD_FREE(lov->lov_tgts, sizeof(*lov->lov_tgts) *
 913                          lov->lov_tgt_size);
 914                 lov->lov_tgt_size = 0;
 915         }
 916         RETURN(0);
 917 }
 918
 919 int lov_process_config_base(struct obd_device *obd, struct lustre_cfg *lcfg,
 920                             __u32 *indexp, int *genp)
 921 {
 922         struct obd_uuid obd_uuid;
 923         int cmd;
 924         int rc = 0;
 925         ENTRY;
 926
 927         switch(cmd = lcfg->lcfg_command) {
 928         case LCFG_LOV_ADD_OBD:
 929         case LCFG_LOV_ADD_INA:
 930         case LCFG_LOV_DEL_OBD: {
 931                 __u32 index;
 932                 int gen;
 933                 /* lov_modify_tgts add  0:lov_mdsA  1:ost1_UUID  2:0  3:1 */
 934                 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
 935                         GOTO(out, rc = -EINVAL);
 936
 937                 obd_str2uuid(&obd_uuid,  lustre_cfg_buf(lcfg, 1));
 938
 939                 if (sscanf(lustre_cfg_buf(lcfg, 2), "%d", indexp) != 1)
 940                         GOTO(out, rc = -EINVAL);
 941                 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", genp) != 1)
 942                         GOTO(out, rc = -EINVAL);
 943                 index = *indexp;
 944                 gen = *genp;
 945                 if (cmd == LCFG_LOV_ADD_OBD)
 946                         rc = lov_add_target(obd, &obd_uuid, index, gen, 1);
 947                 else if (cmd == LCFG_LOV_ADD_INA)
 948                         rc = lov_add_target(obd, &obd_uuid, index, gen, 0);
 949                 else
 950                         rc = lov_del_target(obd, index, &obd_uuid, gen);
 951                 GOTO(out, rc);
 952         }
 953         case LCFG_PARAM: {
 954                 struct lprocfs_static_vars lvars = { 0 };
 955                 struct lov_desc *desc = &(obd->u.lov.desc);
 956
 957                 if (!desc)
 958                         GOTO(out, rc = -EINVAL);
 959
 960                 lprocfs_lov_init_vars(&lvars);
 961
 962                 rc = class_process_proc_param(PARAM_LOV, lvars.obd_vars,
 963                                               lcfg, obd);
 964                 if (rc > 0)
 965                         rc = 0;
 966                 GOTO(out, rc);
 967         }
 968         case LCFG_POOL_NEW:
 969         case LCFG_POOL_ADD:
 970         case LCFG_POOL_DEL:
 971         case LCFG_POOL_REM:
 972                 GOTO(out, rc);
 973
 974         default: {
 975                 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
 976                 GOTO(out, rc = -EINVAL);
 977
 978         }
 979         }
 980 out:
 981         RETURN(rc);
 982 }
 983
 984 static int lov_recreate(struct obd_export *exp, struct obdo *src_oa,
 985                         struct lov_stripe_md **ea, struct obd_trans_info *oti)
 986 {
 987         struct lov_stripe_md *obj_mdp, *lsm;
 988         struct lov_obd *lov = &exp->exp_obd->u.lov;
 989         unsigned ost_idx;
 990         int rc, i;
 991         ENTRY;
 992
 993         LASSERT(src_oa->o_valid & OBD_MD_FLFLAGS &&
 994                 src_oa->o_flags & OBD_FL_RECREATE_OBJS);
 995
 996         OBD_ALLOC(obj_mdp, sizeof(*obj_mdp));
 997         if (obj_mdp == NULL)
 998                 RETURN(-ENOMEM);
 999
1000         ost_idx = src_oa->o_nlink;
1001         lsm = *ea;
1002         if (lsm == NULL)
1003                 GOTO(out, rc = -EINVAL);
1004         if (ost_idx >= lov->desc.ld_tgt_count ||
1005             !lov->lov_tgts[ost_idx])
1006                 GOTO(out, rc = -EINVAL);
1007
1008         for (i = 0; i < lsm->lsm_stripe_count; i++) {
1009                 if (lsm->lsm_oinfo[i]->loi_ost_idx == ost_idx) {
1010                         if (lsm->lsm_oinfo[i]->loi_id != src_oa->o_id)
1011                                 GOTO(out, rc = -EINVAL);
1012                         break;
1013                 }
1014         }
1015         if (i == lsm->lsm_stripe_count)
1016                 GOTO(out, rc = -EINVAL);
1017
1018         rc = obd_create(NULL, lov->lov_tgts[ost_idx]->ltd_exp,
1019                         src_oa, &obj_mdp, oti);
1020 out:
1021         OBD_FREE(obj_mdp, sizeof(*obj_mdp));
1022         RETURN(rc);
1023 }
1024
1025 /* the LOV expects oa->o_id to be set to the LOV object id */
1026 static int lov_create(const struct lu_env *env, struct obd_export *exp,
1027                       struct obdo *src_oa, struct lov_stripe_md **ea,
1028                       struct obd_trans_info *oti)
1029 {
1030         struct lov_obd *lov;
1031         int rc = 0;
1032         ENTRY;
1033
1034         LASSERT(ea != NULL);
1035         if (exp == NULL)
1036                 RETURN(-EINVAL);
1037
1038         if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
1039             src_oa->o_flags == OBD_FL_DELORPHAN) {
1040                 /* should be used with LOV anymore */
1041                 LBUG();
1042         }
1043
1044         lov = &exp->exp_obd->u.lov;
1045         if (!lov->desc.ld_active_tgt_count)
1046                 RETURN(-EIO);
1047
1048         obd_getref(exp->exp_obd);
1049         /* Recreate a specific object id at the given OST index */
1050         if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
1051             (src_oa->o_flags & OBD_FL_RECREATE_OBJS)) {
1052                  rc = lov_recreate(exp, src_oa, ea, oti);
1053         }
1054
1055         obd_putref(exp->exp_obd);
1056         RETURN(rc);
1057 }
1058
1059 #define ASSERT_LSM_MAGIC(lsmp)                                                  \
1060 do {                                                                            \
1061         LASSERT((lsmp) != NULL);                                                \
1062         LASSERTF(((lsmp)->lsm_magic == LOV_MAGIC_V1 ||                          \
1063                  (lsmp)->lsm_magic == LOV_MAGIC_V3),                            \
1064                  "%p->lsm_magic=%x\n", (lsmp), (lsmp)->lsm_magic);              \
1065 } while (0)
1066
1067 static int lov_destroy(const struct lu_env *env, struct obd_export *exp,
1068                        struct obdo *oa, struct lov_stripe_md *lsm,
1069                        struct obd_trans_info *oti, struct obd_export *md_exp,
1070                        void *capa)
1071 {
1072         struct lov_request_set *set;
1073         struct obd_info oinfo;
1074         struct lov_request *req;
1075         cfs_list_t *pos;
1076         struct lov_obd *lov;
1077         int rc = 0, err = 0;
1078         ENTRY;
1079
1080         ASSERT_LSM_MAGIC(lsm);
1081
1082         if (!exp || !exp->exp_obd)
1083                 RETURN(-ENODEV);
1084
1085         if (oa->o_valid & OBD_MD_FLCOOKIE) {
1086                 LASSERT(oti);
1087                 LASSERT(oti->oti_logcookies);
1088         }
1089
1090         lov = &exp->exp_obd->u.lov;
1091         obd_getref(exp->exp_obd);
1092         rc = lov_prep_destroy_set(exp, &oinfo, oa, lsm, oti, &set);
1093         if (rc)
1094                 GOTO(out, rc);
1095
1096         cfs_list_for_each (pos, &set->set_list) {
1097                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1098
1099                 if (oa->o_valid & OBD_MD_FLCOOKIE)
1100                         oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1101
1102                 err = obd_destroy(env, lov->lov_tgts[req->rq_idx]->ltd_exp,
1103                                   req->rq_oi.oi_oa, NULL, oti, NULL, capa);
1104                 err = lov_update_common_set(set, req, err);
1105                 if (err) {
1106                         CERROR("error: destroying objid "LPX64" subobj "
1107                                LPX64" on OST idx %d: rc = %d\n",
1108                                oa->o_id, req->rq_oi.oi_oa->o_id,
1109                                req->rq_idx, err);
1110                         if (!rc)
1111                                 rc = err;
1112                 }
1113         }
1114
1115         if (rc == 0) {
1116                 LASSERT(lsm_op_find(lsm->lsm_magic) != NULL);
1117                 rc = lsm_op_find(lsm->lsm_magic)->lsm_destroy(lsm, oa, md_exp);
1118         }
1119         err = lov_fini_destroy_set(set);
1120 out:
1121         obd_putref(exp->exp_obd);
1122         RETURN(rc ? rc : err);
1123 }
1124
1125 static int lov_getattr(const struct lu_env *env, struct obd_export *exp,
1126                        struct obd_info *oinfo)
1127 {
1128         struct lov_request_set *set;
1129         struct lov_request *req;
1130         cfs_list_t *pos;
1131         struct lov_obd *lov;
1132         int err = 0, rc = 0;
1133         ENTRY;
1134
1135         LASSERT(oinfo);
1136         ASSERT_LSM_MAGIC(oinfo->oi_md);
1137
1138         if (!exp || !exp->exp_obd)
1139                 RETURN(-ENODEV);
1140
1141         lov = &exp->exp_obd->u.lov;
1142
1143         rc = lov_prep_getattr_set(exp, oinfo, &set);
1144         if (rc)
1145                 RETURN(rc);
1146
1147         cfs_list_for_each (pos, &set->set_list) {
1148                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1149
1150                 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
1151                        "%u\n", oinfo->oi_oa->o_id, req->rq_stripe,
1152                        req->rq_oi.oi_oa->o_id, req->rq_idx);
1153
1154                 rc = obd_getattr(env, lov->lov_tgts[req->rq_idx]->ltd_exp,
1155                                  &req->rq_oi);
1156                 err = lov_update_common_set(set, req, rc);
1157                 if (err) {
1158                         CERROR("error: getattr objid "LPX64" subobj "
1159                                LPX64" on OST idx %d: rc = %d\n",
1160                                oinfo->oi_oa->o_id, req->rq_oi.oi_oa->o_id,
1161                                req->rq_idx, err);
1162                         break;
1163                 }
1164         }
1165
1166         rc = lov_fini_getattr_set(set);
1167         if (err)
1168                 rc = err;
1169         RETURN(rc);
1170 }
1171
1172 static int lov_getattr_interpret(struct ptlrpc_request_set *rqset,
1173                                  void *data, int rc)
1174 {
1175         struct lov_request_set *lovset = (struct lov_request_set *)data;
1176         int err;
1177         ENTRY;
1178
1179         /* don't do attribute merge if this aysnc op failed */
1180         if (rc)
1181                 cfs_atomic_set(&lovset->set_completes, 0);
1182         err = lov_fini_getattr_set(lovset);
1183         RETURN(rc ? rc : err);
1184 }
1185
1186 static int lov_getattr_async(struct obd_export *exp, struct obd_info *oinfo,
1187                               struct ptlrpc_request_set *rqset)
1188 {
1189         struct lov_request_set *lovset;
1190         struct lov_obd *lov;
1191         cfs_list_t *pos;
1192         struct lov_request *req;
1193         int rc = 0, err;
1194         ENTRY;
1195
1196         LASSERT(oinfo);
1197         ASSERT_LSM_MAGIC(oinfo->oi_md);
1198
1199         if (!exp || !exp->exp_obd)
1200                 RETURN(-ENODEV);
1201
1202         lov = &exp->exp_obd->u.lov;
1203
1204         rc = lov_prep_getattr_set(exp, oinfo, &lovset);
1205         if (rc)
1206                 RETURN(rc);
1207
1208         CDEBUG(D_INFO, "objid "LPX64": %ux%u byte stripes\n",
1209                oinfo->oi_md->lsm_object_id, oinfo->oi_md->lsm_stripe_count,
1210                oinfo->oi_md->lsm_stripe_size);
1211
1212         cfs_list_for_each (pos, &lovset->set_list) {
1213                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1214
1215                 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
1216                        "%u\n", oinfo->oi_oa->o_id, req->rq_stripe,
1217                        req->rq_oi.oi_oa->o_id, req->rq_idx);
1218                 rc = obd_getattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1219                                        &req->rq_oi, rqset);
1220                 if (rc) {
1221                         CERROR("error: getattr objid "LPX64" subobj "
1222                                LPX64" on OST idx %d: rc = %d\n",
1223                                oinfo->oi_oa->o_id, req->rq_oi.oi_oa->o_id,
1224                                req->rq_idx, rc);
1225                         GOTO(out, rc);
1226                 }
1227         }
1228
1229         if (!cfs_list_empty(&rqset->set_requests)) {
1230                 LASSERT(rc == 0);
1231                 LASSERT (rqset->set_interpret == NULL);
1232                 rqset->set_interpret = lov_getattr_interpret;
1233                 rqset->set_arg = (void *)lovset;
1234                 RETURN(rc);
1235         }
1236 out:
1237         if (rc)
1238                 cfs_atomic_set(&lovset->set_completes, 0);
1239         err = lov_fini_getattr_set(lovset);
1240         RETURN(rc ? rc : err);
1241 }
1242
1243 static int lov_setattr(const struct lu_env *env, struct obd_export *exp,
1244                        struct obd_info *oinfo, struct obd_trans_info *oti)
1245 {
1246         struct lov_request_set *set;
1247         struct lov_obd *lov;
1248         cfs_list_t *pos;
1249         struct lov_request *req;
1250         int err = 0, rc = 0;
1251         ENTRY;
1252
1253         LASSERT(oinfo);
1254         ASSERT_LSM_MAGIC(oinfo->oi_md);
1255
1256         if (!exp || !exp->exp_obd)
1257                 RETURN(-ENODEV);
1258
1259         /* for now, we only expect the following updates here */
1260         LASSERT(!(oinfo->oi_oa->o_valid & ~(OBD_MD_FLID | OBD_MD_FLTYPE |
1261                                             OBD_MD_FLMODE | OBD_MD_FLATIME |
1262                                             OBD_MD_FLMTIME | OBD_MD_FLCTIME |
1263                                             OBD_MD_FLFLAGS | OBD_MD_FLSIZE |
1264                                             OBD_MD_FLGROUP | OBD_MD_FLUID |
1265                                             OBD_MD_FLGID | OBD_MD_FLFID |
1266                                             OBD_MD_FLGENER)));
1267         lov = &exp->exp_obd->u.lov;
1268         rc = lov_prep_setattr_set(exp, oinfo, oti, &set);
1269         if (rc)
1270                 RETURN(rc);
1271
1272         cfs_list_for_each (pos, &set->set_list) {
1273                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1274
1275                 rc = obd_setattr(env, lov->lov_tgts[req->rq_idx]->ltd_exp,
1276                                  &req->rq_oi, NULL);
1277                 err = lov_update_setattr_set(set, req, rc);
1278                 if (err) {
1279                         CERROR("error: setattr objid "LPX64" subobj "
1280                                LPX64" on OST idx %d: rc = %d\n",
1281                                set->set_oi->oi_oa->o_id,
1282                                req->rq_oi.oi_oa->o_id, req->rq_idx, err);
1283                         if (!rc)
1284                                 rc = err;
1285                 }
1286         }
1287         err = lov_fini_setattr_set(set);
1288         if (!rc)
1289                 rc = err;
1290         RETURN(rc);
1291 }
1292
1293 static int lov_setattr_interpret(struct ptlrpc_request_set *rqset,
1294                                  void *data, int rc)
1295 {
1296         struct lov_request_set *lovset = (struct lov_request_set *)data;
1297         int err;
1298         ENTRY;
1299
1300         if (rc)
1301                 cfs_atomic_set(&lovset->set_completes, 0);
1302         err = lov_fini_setattr_set(lovset);
1303         RETURN(rc ? rc : err);
1304 }
1305
1306 /* If @oti is given, the request goes from MDS and responses from OSTs are not
1307    needed. Otherwise, a client is waiting for responses. */
1308 static int lov_setattr_async(struct obd_export *exp, struct obd_info *oinfo,
1309                              struct obd_trans_info *oti,
1310                              struct ptlrpc_request_set *rqset)
1311 {
1312         struct lov_request_set *set;
1313         struct lov_request *req;
1314         cfs_list_t *pos;
1315         struct lov_obd *lov;
1316         int rc = 0;
1317         ENTRY;
1318
1319         LASSERT(oinfo);
1320         ASSERT_LSM_MAGIC(oinfo->oi_md);
1321         if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE) {
1322                 LASSERT(oti);
1323                 LASSERT(oti->oti_logcookies);
1324         }
1325
1326         if (!exp || !exp->exp_obd)
1327                 RETURN(-ENODEV);
1328
1329         lov = &exp->exp_obd->u.lov;
1330         rc = lov_prep_setattr_set(exp, oinfo, oti, &set);
1331         if (rc)
1332                 RETURN(rc);
1333
1334         CDEBUG(D_INFO, "objid "LPX64": %ux%u byte stripes\n",
1335                oinfo->oi_md->lsm_object_id, oinfo->oi_md->lsm_stripe_count,
1336                oinfo->oi_md->lsm_stripe_size);
1337
1338         cfs_list_for_each (pos, &set->set_list) {
1339                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1340
1341                 if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE)
1342                         oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1343
1344                 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
1345                        "%u\n", oinfo->oi_oa->o_id, req->rq_stripe,
1346                        req->rq_oi.oi_oa->o_id, req->rq_idx);
1347
1348                 rc = obd_setattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1349                                        &req->rq_oi, oti, rqset);
1350                 if (rc) {
1351                         CERROR("error: setattr objid "LPX64" subobj "
1352                                LPX64" on OST idx %d: rc = %d\n",
1353                                set->set_oi->oi_oa->o_id,
1354                                req->rq_oi.oi_oa->o_id,
1355                                req->rq_idx, rc);
1356                         break;
1357                 }
1358         }
1359
1360         /* If we are not waiting for responses on async requests, return. */
1361         if (rc || !rqset || cfs_list_empty(&rqset->set_requests)) {
1362                 int err;
1363                 if (rc)
1364                         cfs_atomic_set(&set->set_completes, 0);
1365                 err = lov_fini_setattr_set(set);
1366                 RETURN(rc ? rc : err);
1367         }
1368
1369         LASSERT(rqset->set_interpret == NULL);
1370         rqset->set_interpret = lov_setattr_interpret;
1371         rqset->set_arg = (void *)set;
1372
1373         RETURN(0);
1374 }
1375
1376 static int lov_punch_interpret(struct ptlrpc_request_set *rqset,
1377                                void *data, int rc)
1378 {
1379         struct lov_request_set *lovset = (struct lov_request_set *)data;
1380         int err;
1381         ENTRY;
1382
1383         if (rc)
1384                 cfs_atomic_set(&lovset->set_completes, 0);
1385         err = lov_fini_punch_set(lovset);
1386         RETURN(rc ? rc : err);
1387 }
1388
1389 /* FIXME: maybe we'll just make one node the authoritative attribute node, then
1390  * we can send this 'punch' to just the authoritative node and the nodes
1391  * that the punch will affect. */
1392 static int lov_punch(const struct lu_env *env, struct obd_export *exp,
1393                      struct obd_info *oinfo, struct obd_trans_info *oti,
1394                      struct ptlrpc_request_set *rqset)
1395 {
1396         struct lov_request_set *set;
1397         struct lov_obd *lov;
1398         cfs_list_t *pos;
1399         struct lov_request *req;
1400         int rc = 0;
1401         ENTRY;
1402
1403         LASSERT(oinfo);
1404         ASSERT_LSM_MAGIC(oinfo->oi_md);
1405
1406         if (!exp || !exp->exp_obd)
1407                 RETURN(-ENODEV);
1408
1409         lov = &exp->exp_obd->u.lov;
1410         rc = lov_prep_punch_set(exp, oinfo, oti, &set);
1411         if (rc)
1412                 RETURN(rc);
1413
1414         cfs_list_for_each (pos, &set->set_list) {
1415                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1416
1417                 rc = obd_punch(env, lov->lov_tgts[req->rq_idx]->ltd_exp,
1418                                &req->rq_oi, NULL, rqset);
1419                 if (rc) {
1420                         CERROR("error: punch objid "LPX64" subobj "LPX64
1421                                " on OST idx %d: rc = %d\n",
1422                                set->set_oi->oi_oa->o_id,
1423                                req->rq_oi.oi_oa->o_id, req->rq_idx, rc);
1424                         break;
1425                 }
1426         }
1427
1428         if (rc || cfs_list_empty(&rqset->set_requests)) {
1429                 int err;
1430                 err = lov_fini_punch_set(set);
1431                 RETURN(rc ? rc : err);
1432         }
1433
1434         LASSERT(rqset->set_interpret == NULL);
1435         rqset->set_interpret = lov_punch_interpret;
1436         rqset->set_arg = (void *)set;
1437
1438         RETURN(0);
1439 }
1440
1441 static int lov_sync_interpret(struct ptlrpc_request_set *rqset,
1442                               void *data, int rc)
1443 {
1444         struct lov_request_set *lovset = data;
1445         int err;
1446         ENTRY;
1447
1448         if (rc)
1449                 cfs_atomic_set(&lovset->set_completes, 0);
1450         err = lov_fini_sync_set(lovset);
1451         RETURN(rc ?: err);
1452 }
1453
1454 static int lov_sync(const struct lu_env *env, struct obd_export *exp,
1455                     struct obd_info *oinfo, obd_off start, obd_off end,
1456                     struct ptlrpc_request_set *rqset)
1457 {
1458         struct lov_request_set *set = NULL;
1459         struct lov_obd *lov;
1460         cfs_list_t *pos;
1461         struct lov_request *req;
1462         int rc = 0;
1463         ENTRY;
1464
1465         ASSERT_LSM_MAGIC(oinfo->oi_md);
1466         LASSERT(rqset != NULL);
1467
1468         if (!exp->exp_obd)
1469                 RETURN(-ENODEV);
1470
1471         lov = &exp->exp_obd->u.lov;
1472         rc = lov_prep_sync_set(exp, oinfo, start, end, &set);
1473         if (rc)
1474                 RETURN(rc);
1475
1476         CDEBUG(D_INFO, "fsync objid "LPX64" ["LPX64", "LPX64"]\n",
1477                set->set_oi->oi_oa->o_id, start, end);
1478
1479         cfs_list_for_each (pos, &set->set_list) {
1480                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1481
1482                 rc = obd_sync(env, lov->lov_tgts[req->rq_idx]->ltd_exp,
1483                               &req->rq_oi, req->rq_oi.oi_policy.l_extent.start,
1484                               req->rq_oi.oi_policy.l_extent.end, rqset);
1485                 if (rc) {
1486                         CERROR("error: fsync objid "LPX64" subobj "LPX64
1487                                " on OST idx %d: rc = %d\n",
1488                                set->set_oi->oi_oa->o_id,
1489                                req->rq_oi.oi_oa->o_id, req->rq_idx, rc);
1490                         break;
1491                 }
1492         }
1493
1494         /* If we are not waiting for responses on async requests, return. */
1495         if (rc || cfs_list_empty(&rqset->set_requests)) {
1496                 int err = lov_fini_sync_set(set);
1497
1498                 RETURN(rc ?: err);
1499         }
1500
1501         LASSERT(rqset->set_interpret == NULL);
1502         rqset->set_interpret = lov_sync_interpret;
1503         rqset->set_arg = (void *)set;
1504
1505         RETURN(0);
1506 }
1507
1508 static int lov_brw_check(struct lov_obd *lov, struct obd_info *lov_oinfo,
1509                          obd_count oa_bufs, struct brw_page *pga)
1510 {
1511         struct obd_info oinfo = { { { 0 } } };
1512         int i, rc = 0;
1513
1514         oinfo.oi_oa = lov_oinfo->oi_oa;
1515
1516         /* The caller just wants to know if there's a chance that this
1517          * I/O can succeed */
1518         for (i = 0; i < oa_bufs; i++) {
1519                 int stripe = lov_stripe_number(lov_oinfo->oi_md, pga[i].off);
1520                 int ost = lov_oinfo->oi_md->lsm_oinfo[stripe]->loi_ost_idx;
1521                 obd_off start, end;
1522
1523                 if (!lov_stripe_intersects(lov_oinfo->oi_md, i, pga[i].off,
1524                                            pga[i].off + pga[i].count - 1,
1525                                            &start, &end))
1526                         continue;
1527
1528                 if (!lov->lov_tgts[ost] || !lov->lov_tgts[ost]->ltd_active) {
1529                         CDEBUG(D_HA, "lov idx %d inactive\n", ost);
1530                         return -EIO;
1531                 }
1532
1533                 rc = obd_brw(OBD_BRW_CHECK, lov->lov_tgts[ost]->ltd_exp, &oinfo,
1534                              1, &pga[i], NULL);
1535                 if (rc)
1536                         break;
1537         }
1538         return rc;
1539 }
1540
1541 static int lov_brw(int cmd, struct obd_export *exp, struct obd_info *oinfo,
1542                    obd_count oa_bufs, struct brw_page *pga,
1543                    struct obd_trans_info *oti)
1544 {
1545         struct lov_request_set *set;
1546         struct lov_request *req;
1547         cfs_list_t *pos;
1548         struct lov_obd *lov = &exp->exp_obd->u.lov;
1549         int err, rc = 0;
1550         ENTRY;
1551
1552         ASSERT_LSM_MAGIC(oinfo->oi_md);
1553
1554         if (cmd == OBD_BRW_CHECK) {
1555                 rc = lov_brw_check(lov, oinfo, oa_bufs, pga);
1556                 RETURN(rc);
1557         }
1558
1559         rc = lov_prep_brw_set(exp, oinfo, oa_bufs, pga, oti, &set);
1560         if (rc)
1561                 RETURN(rc);
1562
1563         cfs_list_for_each (pos, &set->set_list) {
1564                 struct obd_export *sub_exp;
1565                 struct brw_page *sub_pga;
1566                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1567
1568                 sub_exp = lov->lov_tgts[req->rq_idx]->ltd_exp;
1569                 sub_pga = set->set_pga + req->rq_pgaidx;
1570                 rc = obd_brw(cmd, sub_exp, &req->rq_oi, req->rq_oabufs,
1571                              sub_pga, oti);
1572                 if (rc)
1573                         break;
1574                 lov_update_common_set(set, req, rc);
1575         }
1576
1577         err = lov_fini_brw_set(set);
1578         if (!rc)
1579                 rc = err;
1580         RETURN(rc);
1581 }
1582
1583 static int lov_enqueue_interpret(struct ptlrpc_request_set *rqset,
1584                                  void *data, int rc)
1585 {
1586         struct lov_request_set *lovset = (struct lov_request_set *)data;
1587         ENTRY;
1588         rc = lov_fini_enqueue_set(lovset, lovset->set_ei->ei_mode, rc, rqset);
1589         RETURN(rc);
1590 }
1591
1592 static int lov_enqueue(struct obd_export *exp, struct obd_info *oinfo,
1593                        struct ldlm_enqueue_info *einfo,
1594                        struct ptlrpc_request_set *rqset)
1595 {
1596         ldlm_mode_t mode = einfo->ei_mode;
1597         struct lov_request_set *set;
1598         struct lov_request *req;
1599         cfs_list_t *pos;
1600         struct lov_obd *lov;
1601         ldlm_error_t rc;
1602         ENTRY;
1603
1604         LASSERT(oinfo);
1605         ASSERT_LSM_MAGIC(oinfo->oi_md);
1606         LASSERT(mode == (mode & -mode));
1607
1608         /* we should never be asked to replay a lock this way. */
1609         LASSERT((oinfo->oi_flags & LDLM_FL_REPLAY) == 0);
1610
1611         if (!exp || !exp->exp_obd)
1612                 RETURN(-ENODEV);
1613
1614         lov = &exp->exp_obd->u.lov;
1615         rc = lov_prep_enqueue_set(exp, oinfo, einfo, &set);
1616         if (rc)
1617                 RETURN(rc);
1618
1619         cfs_list_for_each (pos, &set->set_list) {
1620                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1621
1622                 rc = obd_enqueue(lov->lov_tgts[req->rq_idx]->ltd_exp,
1623                                  &req->rq_oi, einfo, rqset);
1624                 if (rc != ELDLM_OK)
1625                         GOTO(out, rc);
1626         }
1627
1628         if (rqset && !cfs_list_empty(&rqset->set_requests)) {
1629                 LASSERT(rc == 0);
1630                 LASSERT(rqset->set_interpret == NULL);
1631                 rqset->set_interpret = lov_enqueue_interpret;
1632                 rqset->set_arg = (void *)set;
1633                 RETURN(rc);
1634         }
1635 out:
1636         rc = lov_fini_enqueue_set(set, mode, rc, rqset);
1637         RETURN(rc);
1638 }
1639
1640 static int lov_change_cbdata(struct obd_export *exp,
1641                              struct lov_stripe_md *lsm, ldlm_iterator_t it,
1642                              void *data)
1643 {
1644         struct lov_obd *lov;
1645         int rc = 0, i;
1646         ENTRY;
1647
1648         ASSERT_LSM_MAGIC(lsm);
1649
1650         if (!exp || !exp->exp_obd)
1651                 RETURN(-ENODEV);
1652
1653         lov = &exp->exp_obd->u.lov;
1654         for (i = 0; i < lsm->lsm_stripe_count; i++) {
1655                 struct lov_stripe_md submd;
1656                 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
1657
1658                 if (!lov->lov_tgts[loi->loi_ost_idx]) {
1659                         CDEBUG(D_HA, "lov idx %d NULL \n", loi->loi_ost_idx);
1660                         continue;
1661                 }
1662
1663                 LASSERT_SEQ_IS_MDT(loi->loi_seq);
1664                 submd.lsm_object_id = loi->loi_id;
1665                 submd.lsm_object_seq = loi->loi_seq;
1666                 submd.lsm_stripe_count = 0;
1667                 rc = obd_change_cbdata(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1668                                        &submd, it, data);
1669         }
1670         RETURN(rc);
1671 }
1672
1673 /* find any ldlm lock of the inode in lov
1674  * return 0    not find
1675  *        1    find one
1676  *      < 0    error */
1677 static int lov_find_cbdata(struct obd_export *exp,
1678                            struct lov_stripe_md *lsm, ldlm_iterator_t it,
1679                            void *data)
1680 {
1681         struct lov_obd *lov;
1682         int rc = 0, i;
1683         ENTRY;
1684
1685         ASSERT_LSM_MAGIC(lsm);
1686
1687         if (!exp || !exp->exp_obd)
1688                 RETURN(-ENODEV);
1689
1690         lov = &exp->exp_obd->u.lov;
1691         for (i = 0; i < lsm->lsm_stripe_count; i++) {
1692                 struct lov_stripe_md submd;
1693                 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
1694
1695                 if (!lov->lov_tgts[loi->loi_ost_idx]) {
1696                         CDEBUG(D_HA, "lov idx %d NULL \n", loi->loi_ost_idx);
1697                         continue;
1698                 }
1699
1700                 LASSERT_SEQ_IS_MDT(loi->loi_seq);
1701                 submd.lsm_object_id = loi->loi_id;
1702                 submd.lsm_object_seq = loi->loi_seq;
1703                 submd.lsm_stripe_count = 0;
1704                 rc = obd_find_cbdata(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1705                                      &submd, it, data);
1706                 if (rc != 0)
1707                         RETURN(rc);
1708         }
1709         RETURN(rc);
1710 }
1711
1712 static int lov_cancel(struct obd_export *exp, struct lov_stripe_md *lsm,
1713                       __u32 mode, struct lustre_handle *lockh)
1714 {
1715         struct lov_request_set *set;
1716         struct obd_info oinfo;
1717         struct lov_request *req;
1718         cfs_list_t *pos;
1719         struct lov_obd *lov;
1720         struct lustre_handle *lov_lockhp;
1721         int err = 0, rc = 0;
1722         ENTRY;
1723
1724         ASSERT_LSM_MAGIC(lsm);
1725
1726         if (!exp || !exp->exp_obd)
1727                 RETURN(-ENODEV);
1728
1729         LASSERT_SEQ_IS_MDT(lsm->lsm_object_seq);
1730         LASSERT(lockh);
1731         lov = &exp->exp_obd->u.lov;
1732         rc = lov_prep_cancel_set(exp, &oinfo, lsm, mode, lockh, &set);
1733         if (rc)
1734                 RETURN(rc);
1735
1736         cfs_list_for_each (pos, &set->set_list) {
1737                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1738                 lov_lockhp = set->set_lockh->llh_handles + req->rq_stripe;
1739
1740                 rc = obd_cancel(lov->lov_tgts[req->rq_idx]->ltd_exp,
1741                                 req->rq_oi.oi_md, mode, lov_lockhp);
1742                 rc = lov_update_common_set(set, req, rc);
1743                 if (rc) {
1744                         CERROR("error: cancel objid "LPX64" subobj "
1745                                LPX64" on OST idx %d: rc = %d\n",
1746                                lsm->lsm_object_id,
1747                                req->rq_oi.oi_md->lsm_object_id,
1748                                req->rq_idx, rc);
1749                         err = rc;
1750                 }
1751
1752         }
1753         lov_fini_cancel_set(set);
1754         RETURN(err);
1755 }
1756
1757 static int lov_cancel_unused(struct obd_export *exp,
1758                              struct lov_stripe_md *lsm,
1759                              ldlm_cancel_flags_t flags, void *opaque)
1760 {
1761         struct lov_obd *lov;
1762         int rc = 0, i;
1763         ENTRY;
1764
1765         if (!exp || !exp->exp_obd)
1766                 RETURN(-ENODEV);
1767
1768         lov = &exp->exp_obd->u.lov;
1769         if (lsm == NULL) {
1770                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
1771                         int err;
1772                         if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
1773                                 continue;
1774
1775                         err = obd_cancel_unused(lov->lov_tgts[i]->ltd_exp, NULL,
1776                                                 flags, opaque);
1777                         if (!rc)
1778                                 rc = err;
1779                 }
1780                 RETURN(rc);
1781         }
1782
1783         ASSERT_LSM_MAGIC(lsm);
1784
1785         LASSERT_SEQ_IS_MDT(lsm->lsm_object_seq);
1786         for (i = 0; i < lsm->lsm_stripe_count; i++) {
1787                 struct lov_stripe_md submd;
1788                 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
1789                 int err;
1790
1791                 if (!lov->lov_tgts[loi->loi_ost_idx]) {
1792                         CDEBUG(D_HA, "lov idx %d NULL\n", loi->loi_ost_idx);
1793                         continue;
1794                 }
1795
1796                 if (!lov->lov_tgts[loi->loi_ost_idx]->ltd_active)
1797                         CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
1798
1799                 submd.lsm_object_id = loi->loi_id;
1800                 submd.lsm_object_seq = loi->loi_seq;
1801                 submd.lsm_stripe_count = 0;
1802                 err = obd_cancel_unused(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1803                                         &submd, flags, opaque);
1804                 if (err && lov->lov_tgts[loi->loi_ost_idx]->ltd_active) {
1805                         CERROR("error: cancel unused objid "LPX64" subobj "LPX64
1806                                " on OST idx %d: rc = %d\n", lsm->lsm_object_id,
1807                                loi->loi_id, loi->loi_ost_idx, err);
1808                         if (!rc)
1809                                 rc = err;
1810                 }
1811         }
1812         RETURN(rc);
1813 }
1814
1815 int lov_statfs_interpret(struct ptlrpc_request_set *rqset, void *data, int rc)
1816 {
1817         struct lov_request_set *lovset = (struct lov_request_set *)data;
1818         int err;
1819         ENTRY;
1820
1821         if (rc)
1822                 cfs_atomic_set(&lovset->set_completes, 0);
1823
1824         err = lov_fini_statfs_set(lovset);
1825         RETURN(rc ? rc : err);
1826 }
1827
1828 static int lov_statfs_async(struct obd_export *exp, struct obd_info *oinfo,
1829                             __u64 max_age, struct ptlrpc_request_set *rqset)
1830 {
1831         struct obd_device      *obd = class_exp2obd(exp);
1832         struct lov_request_set *set;
1833         struct lov_request *req;
1834         cfs_list_t *pos;
1835         struct lov_obd *lov;
1836         int rc = 0;
1837         ENTRY;
1838
1839         LASSERT(oinfo != NULL);
1840         LASSERT(oinfo->oi_osfs != NULL);
1841
1842         lov = &obd->u.lov;
1843         rc = lov_prep_statfs_set(obd, oinfo, &set);
1844         if (rc)
1845                 RETURN(rc);
1846
1847         cfs_list_for_each (pos, &set->set_list) {
1848                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1849                 rc = obd_statfs_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1850                                       &req->rq_oi, max_age, rqset);
1851                 if (rc)
1852                         break;
1853         }
1854
1855         if (rc || cfs_list_empty(&rqset->set_requests)) {
1856                 int err;
1857                 if (rc)
1858                         cfs_atomic_set(&set->set_completes, 0);
1859                 err = lov_fini_statfs_set(set);
1860                 RETURN(rc ? rc : err);
1861         }
1862
1863         LASSERT(rqset->set_interpret == NULL);
1864         rqset->set_interpret = lov_statfs_interpret;
1865         rqset->set_arg = (void *)set;
1866         RETURN(0);
1867 }
1868
1869 static int lov_statfs(const struct lu_env *env, struct obd_export *exp,
1870                       struct obd_statfs *osfs, __u64 max_age, __u32 flags)
1871 {
1872         struct ptlrpc_request_set *set = NULL;
1873         struct obd_info oinfo = { { { 0 } } };
1874         int rc = 0;
1875         ENTRY;
1876
1877
1878         /* for obdclass we forbid using obd_statfs_rqset, but prefer using async
1879          * statfs requests */
1880         set = ptlrpc_prep_set();
1881         if (set == NULL)
1882                 RETURN(-ENOMEM);
1883
1884         oinfo.oi_osfs = osfs;
1885         oinfo.oi_flags = flags;
1886         rc = lov_statfs_async(exp, &oinfo, max_age, set);
1887         if (rc == 0)
1888                 rc = ptlrpc_set_wait(set);
1889         ptlrpc_set_destroy(set);
1890
1891         RETURN(rc);
1892 }
1893
1894 static int lov_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1895                          void *karg, void *uarg)
1896 {
1897         struct obd_device *obddev = class_exp2obd(exp);
1898         struct lov_obd *lov = &obddev->u.lov;
1899         int i = 0, rc = 0, count = lov->desc.ld_tgt_count;
1900         struct obd_uuid *uuidp;
1901         ENTRY;
1902
1903         switch (cmd) {
1904         case IOC_OBD_STATFS: {
1905                 struct obd_ioctl_data *data = karg;
1906                 struct obd_device *osc_obd;
1907                 struct obd_statfs stat_buf = {0};
1908                 __u32 index;
1909                 __u32 flags;
1910
1911                 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
1912                 if ((index >= count))
1913                         RETURN(-ENODEV);
1914
1915                 if (!lov->lov_tgts[index])
1916                         /* Try again with the next index */
1917                         RETURN(-EAGAIN);
1918                 if (!lov->lov_tgts[index]->ltd_active)
1919                         RETURN(-ENODATA);
1920
1921                 osc_obd = class_exp2obd(lov->lov_tgts[index]->ltd_exp);
1922                 if (!osc_obd)
1923                         RETURN(-EINVAL);
1924
1925                 /* copy UUID */
1926                 if (cfs_copy_to_user(data->ioc_pbuf2, obd2cli_tgt(osc_obd),
1927                                      min((int) data->ioc_plen2,
1928                                          (int) sizeof(struct obd_uuid))))
1929                         RETURN(-EFAULT);
1930
1931                 flags = uarg ? *(__u32*)uarg : 0;
1932                 /* got statfs data */
1933                 rc = obd_statfs(NULL, lov->lov_tgts[index]->ltd_exp, &stat_buf,
1934                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1935                                 flags);
1936                 if (rc)
1937                         RETURN(rc);
1938                 if (cfs_copy_to_user(data->ioc_pbuf1, &stat_buf,
1939                                      min((int) data->ioc_plen1,
1940                                          (int) sizeof(stat_buf))))
1941                         RETURN(-EFAULT);
1942                 break;
1943         }
1944         case OBD_IOC_LOV_GET_CONFIG: {
1945                 struct obd_ioctl_data *data;
1946                 struct lov_desc *desc;
1947                 char *buf = NULL;
1948                 __u32 *genp;
1949
1950                 len = 0;
1951                 if (obd_ioctl_getdata(&buf, &len, (void *)uarg))
1952                         RETURN(-EINVAL);
1953
1954                 data = (struct obd_ioctl_data *)buf;
1955
1956                 if (sizeof(*desc) > data->ioc_inllen1) {
1957                         obd_ioctl_freedata(buf, len);
1958                         RETURN(-EINVAL);
1959                 }
1960
1961                 if (sizeof(uuidp->uuid) * count > data->ioc_inllen2) {
1962                         obd_ioctl_freedata(buf, len);
1963                         RETURN(-EINVAL);
1964                 }
1965
1966                 if (sizeof(__u32) * count > data->ioc_inllen3) {
1967                         obd_ioctl_freedata(buf, len);
1968                         RETURN(-EINVAL);
1969                 }
1970
1971                 desc = (struct lov_desc *)data->ioc_inlbuf1;
1972                 memcpy(desc, &(lov->desc), sizeof(*desc));
1973
1974                 uuidp = (struct obd_uuid *)data->ioc_inlbuf2;
1975                 genp = (__u32 *)data->ioc_inlbuf3;
1976                 /* the uuid will be empty for deleted OSTs */
1977                 for (i = 0; i < count; i++, uuidp++, genp++) {
1978                         if (!lov->lov_tgts[i])
1979                                 continue;
1980                         *uuidp = lov->lov_tgts[i]->ltd_uuid;
1981                         *genp = lov->lov_tgts[i]->ltd_gen;
1982                 }
1983
1984                 if (cfs_copy_to_user((void *)uarg, buf, len))
1985                         rc = -EFAULT;
1986                 obd_ioctl_freedata(buf, len);
1987                 break;
1988         }
1989         case LL_IOC_LOV_SETSTRIPE:
1990                 rc = lov_setstripe(exp, len, karg, uarg);
1991                 break;
1992         case LL_IOC_LOV_GETSTRIPE:
1993                 rc = lov_getstripe(exp, karg, uarg);
1994                 break;
1995         case LL_IOC_LOV_SETEA:
1996                 rc = lov_setea(exp, karg, uarg);
1997                 break;
1998         case OBD_IOC_QUOTACTL: {
1999                 struct if_quotactl *qctl = karg;
2000                 struct lov_tgt_desc *tgt = NULL;
2001                 struct obd_quotactl *oqctl;
2002
2003                 if (qctl->qc_valid == QC_OSTIDX) {
2004                         if (qctl->qc_idx < 0 || count <= qctl->qc_idx)
2005                                 RETURN(-EINVAL);
2006
2007                         tgt = lov->lov_tgts[qctl->qc_idx];
2008                         if (!tgt || !tgt->ltd_exp)
2009                                 RETURN(-EINVAL);
2010                 } else if (qctl->qc_valid == QC_UUID) {
2011                         for (i = 0; i < count; i++) {
2012                                 tgt = lov->lov_tgts[i];
2013                                 if (!tgt ||
2014                                     !obd_uuid_equals(&tgt->ltd_uuid,
2015                                                      &qctl->obd_uuid))
2016                                         continue;
2017
2018                                 if (tgt->ltd_exp == NULL)
2019                                         RETURN(-EINVAL);
2020
2021                                 break;
2022                         }
2023                 } else {
2024                         RETURN(-EINVAL);
2025                 }
2026
2027                 if (i >= count)
2028                         RETURN(-EAGAIN);
2029
2030                 LASSERT(tgt && tgt->ltd_exp);
2031                 OBD_ALLOC_PTR(oqctl);
2032                 if (!oqctl)
2033                         RETURN(-ENOMEM);
2034
2035                 QCTL_COPY(oqctl, qctl);
2036                 rc = obd_quotactl(tgt->ltd_exp, oqctl);
2037                 if (rc == 0) {
2038                         QCTL_COPY(qctl, oqctl);
2039                         qctl->qc_valid = QC_OSTIDX;
2040                         qctl->obd_uuid = tgt->ltd_uuid;
2041                 }
2042                 OBD_FREE_PTR(oqctl);
2043                 break;
2044         }
2045         default: {
2046                 int set = 0;
2047
2048                 if (count == 0)
2049                         RETURN(-ENOTTY);
2050
2051                 for (i = 0; i < count; i++) {
2052                         int err;
2053                         struct obd_device *osc_obd;
2054
2055                         /* OST was disconnected */
2056                         if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
2057                                 continue;
2058
2059                         /* ll_umount_begin() sets force flag but for lov, not
2060                          * osc. Let's pass it through */
2061                         osc_obd = class_exp2obd(lov->lov_tgts[i]->ltd_exp);
2062                         osc_obd->obd_force = obddev->obd_force;
2063                         err = obd_iocontrol(cmd, lov->lov_tgts[i]->ltd_exp,
2064                                             len, karg, uarg);
2065                         if (err == -ENODATA && cmd == OBD_IOC_POLL_QUOTACHECK) {
2066                                 RETURN(err);
2067                         } else if (err) {
2068                                 if (lov->lov_tgts[i]->ltd_active) {
2069                                         CDEBUG(err == -ENOTTY ?
2070                                                D_IOCTL : D_WARNING,
2071                                                "iocontrol OSC %s on OST "
2072                                                "idx %d cmd %x: err = %d\n",
2073                                                lov_uuid2str(lov, i),
2074                                                i, cmd, err);
2075                                         if (!rc)
2076                                                 rc = err;
2077                                 }
2078                         } else {
2079                                 set = 1;
2080                         }
2081                 }
2082                 if (!set && !rc)
2083                         rc = -EIO;
2084         }
2085         }
2086
2087         RETURN(rc);
2088 }
2089
2090 #define FIEMAP_BUFFER_SIZE 4096
2091
2092 /**
2093  * Non-zero fe_logical indicates that this is a continuation FIEMAP
2094  * call. The local end offset and the device are sent in the first
2095  * fm_extent. This function calculates the stripe number from the index.
2096  * This function returns a stripe_no on which mapping is to be restarted.
2097  *
2098  * This function returns fm_end_offset which is the in-OST offset at which
2099  * mapping should be restarted. If fm_end_offset=0 is returned then caller
2100  * will re-calculate proper offset in next stripe.
2101  * Note that the first extent is passed to lov_get_info via the value field.
2102  *
2103  * \param fiemap fiemap request header
2104  * \param lsm striping information for the file
2105  * \param fm_start logical start of mapping
2106  * \param fm_end logical end of mapping
2107  * \param start_stripe starting stripe will be returned in this
2108  */
2109 obd_size fiemap_calc_fm_end_offset(struct ll_user_fiemap *fiemap,
2110                                    struct lov_stripe_md *lsm, obd_size fm_start,
2111                                    obd_size fm_end, int *start_stripe)
2112 {
2113         obd_size local_end = fiemap->fm_extents[0].fe_logical;
2114         obd_off lun_start, lun_end;
2115         obd_size fm_end_offset;
2116         int stripe_no = -1, i;
2117
2118         if (fiemap->fm_extent_count == 0 ||
2119             fiemap->fm_extents[0].fe_logical == 0)
2120                 return 0;
2121
2122         /* Find out stripe_no from ost_index saved in the fe_device */
2123         for (i = 0; i < lsm->lsm_stripe_count; i++) {
2124                 if (lsm->lsm_oinfo[i]->loi_ost_idx ==
2125                                         fiemap->fm_extents[0].fe_device) {
2126                         stripe_no = i;
2127                         break;
2128                 }
2129         }
2130         if (stripe_no == -1)
2131                 return -EINVAL;
2132
2133         /* If we have finished mapping on previous device, shift logical
2134          * offset to start of next device */
2135         if ((lov_stripe_intersects(lsm, stripe_no, fm_start, fm_end,
2136                                    &lun_start, &lun_end)) != 0 &&
2137                                    local_end < lun_end) {
2138                 fm_end_offset = local_end;
2139                 *start_stripe = stripe_no;
2140         } else {
2141                 /* This is a special value to indicate that caller should
2142                  * calculate offset in next stripe. */
2143                 fm_end_offset = 0;
2144                 *start_stripe = (stripe_no + 1) % lsm->lsm_stripe_count;
2145         }
2146
2147         return fm_end_offset;
2148 }
2149
2150 /**
2151  * We calculate on which OST the mapping will end. If the length of mapping
2152  * is greater than (stripe_size * stripe_count) then the last_stripe will
2153  * will be one just before start_stripe. Else we check if the mapping
2154  * intersects each OST and find last_stripe.
2155  * This function returns the last_stripe and also sets the stripe_count
2156  * over which the mapping is spread
2157  *
2158  * \param lsm striping information for the file
2159  * \param fm_start logical start of mapping
2160  * \param fm_end logical end of mapping
2161  * \param start_stripe starting stripe of the mapping
2162  * \param stripe_count the number of stripes across which to map is returned
2163  *
2164  * \retval last_stripe return the last stripe of the mapping
2165  */
2166 int fiemap_calc_last_stripe(struct lov_stripe_md *lsm, obd_size fm_start,
2167                             obd_size fm_end, int start_stripe,
2168                             int *stripe_count)
2169 {
2170         int last_stripe;
2171         obd_off obd_start, obd_end;
2172         int i, j;
2173
2174         if (fm_end - fm_start > lsm->lsm_stripe_size * lsm->lsm_stripe_count) {
2175                 last_stripe = (start_stripe < 1 ? lsm->lsm_stripe_count - 1 :
2176                                                               start_stripe - 1);
2177                 *stripe_count = lsm->lsm_stripe_count;
2178         } else {
2179                 for (j = 0, i = start_stripe; j < lsm->lsm_stripe_count;
2180                      i = (i + 1) % lsm->lsm_stripe_count, j++) {
2181                         if ((lov_stripe_intersects(lsm, i, fm_start, fm_end,
2182                                                    &obd_start, &obd_end)) == 0)
2183                                 break;
2184                 }
2185                 *stripe_count = j;
2186                 last_stripe = (start_stripe + j - 1) %lsm->lsm_stripe_count;
2187         }
2188
2189         return last_stripe;
2190 }
2191
2192 /**
2193  * Set fe_device and copy extents from local buffer into main return buffer.
2194  *
2195  * \param fiemap fiemap request header
2196  * \param lcl_fm_ext array of local fiemap extents to be copied
2197  * \param ost_index OST index to be written into the fm_device field for each
2198                     extent
2199  * \param ext_count number of extents to be copied
2200  * \param current_extent where to start copying in main extent array
2201  */
2202 void fiemap_prepare_and_copy_exts(struct ll_user_fiemap *fiemap,
2203                                   struct ll_fiemap_extent *lcl_fm_ext,
2204                                   int ost_index, unsigned int ext_count,
2205                                   int current_extent)
2206 {
2207         char *to;
2208         int ext;
2209
2210         for (ext = 0; ext < ext_count; ext++) {
2211                 lcl_fm_ext[ext].fe_device = ost_index;
2212                 lcl_fm_ext[ext].fe_flags |= FIEMAP_EXTENT_NET;
2213         }
2214
2215         /* Copy fm_extent's from fm_local to return buffer */
2216         to = (char *)fiemap + fiemap_count_to_size(current_extent);
2217         memcpy(to, lcl_fm_ext, ext_count * sizeof(struct ll_fiemap_extent));
2218 }
2219
2220 /**
2221  * Break down the FIEMAP request and send appropriate calls to individual OSTs.
2222  * This also handles the restarting of FIEMAP calls in case mapping overflows
2223  * the available number of extents in single call.
2224  */
2225 static int lov_fiemap(struct lov_obd *lov, __u32 keylen, void *key,
2226                       __u32 *vallen, void *val, struct lov_stripe_md *lsm)
2227 {
2228         struct ll_fiemap_info_key *fm_key = key;
2229         struct ll_user_fiemap *fiemap = val;
2230         struct ll_user_fiemap *fm_local = NULL;
2231         struct ll_fiemap_extent *lcl_fm_ext;
2232         int count_local;
2233         unsigned int get_num_extents = 0;
2234         int ost_index = 0, actual_start_stripe, start_stripe;
2235         obd_size fm_start, fm_end, fm_length, fm_end_offset;
2236         obd_size curr_loc;
2237         int current_extent = 0, rc = 0, i;
2238         int ost_eof = 0; /* EOF for object */
2239         int ost_done = 0; /* done with required mapping for this OST? */
2240         int last_stripe;
2241         int cur_stripe = 0, cur_stripe_wrap = 0, stripe_count;
2242         unsigned int buffer_size = FIEMAP_BUFFER_SIZE;
2243
2244         if (lsm == NULL)
2245                 GOTO(out, rc = 0);
2246
2247         if (fiemap_count_to_size(fm_key->fiemap.fm_extent_count) < buffer_size)
2248                 buffer_size = fiemap_count_to_size(fm_key->fiemap.fm_extent_count);
2249
2250         OBD_ALLOC_LARGE(fm_local, buffer_size);
2251         if (fm_local == NULL)
2252                 GOTO(out, rc = -ENOMEM);
2253         lcl_fm_ext = &fm_local->fm_extents[0];
2254
2255         count_local = fiemap_size_to_count(buffer_size);
2256
2257         memcpy(fiemap, &fm_key->fiemap, sizeof(*fiemap));
2258         fm_start = fiemap->fm_start;
2259         fm_length = fiemap->fm_length;
2260         /* Calculate start stripe, last stripe and length of mapping */
2261         actual_start_stripe = start_stripe = lov_stripe_number(lsm, fm_start);
2262         fm_end = (fm_length == ~0ULL ? fm_key->oa.o_size :
2263                                                 fm_start + fm_length - 1);
2264         /* If fm_length != ~0ULL but fm_start+fm_length-1 exceeds file size */
2265         if (fm_end > fm_key->oa.o_size)
2266                 fm_end = fm_key->oa.o_size;
2267
2268         last_stripe = fiemap_calc_last_stripe(lsm, fm_start, fm_end,
2269                                             actual_start_stripe, &stripe_count);
2270
2271         fm_end_offset = fiemap_calc_fm_end_offset(fiemap, lsm, fm_start,
2272                                                   fm_end, &start_stripe);
2273         if (fm_end_offset == -EINVAL)
2274                 return -EINVAL;
2275
2276         if (fiemap->fm_extent_count == 0) {
2277                 get_num_extents = 1;
2278                 count_local = 0;
2279         }
2280
2281         /* Check each stripe */
2282         for (cur_stripe = start_stripe, i = 0; i < stripe_count;
2283              i++, cur_stripe = (cur_stripe + 1) % lsm->lsm_stripe_count) {
2284                 obd_size req_fm_len; /* Stores length of required mapping */
2285                 obd_size len_mapped_single_call;
2286                 obd_off lun_start, lun_end, obd_object_end;
2287                 unsigned int ext_count;
2288
2289                 cur_stripe_wrap = cur_stripe;
2290
2291                 /* Find out range of mapping on this stripe */
2292                 if ((lov_stripe_intersects(lsm, cur_stripe, fm_start, fm_end,
2293                                            &lun_start, &obd_object_end)) == 0)
2294                         continue;
2295
2296                 /* If this is a continuation FIEMAP call and we are on
2297                  * starting stripe then lun_start needs to be set to
2298                  * fm_end_offset */
2299                 if (fm_end_offset != 0 && cur_stripe == start_stripe)
2300                         lun_start = fm_end_offset;
2301
2302                 if (fm_length != ~0ULL) {
2303                         /* Handle fm_start + fm_length overflow */
2304                         if (fm_start + fm_length < fm_start)
2305                                 fm_length = ~0ULL - fm_start;
2306                         lun_end = lov_size_to_stripe(lsm, fm_start + fm_length,
2307                                                      cur_stripe);
2308                 } else {
2309                         lun_end = ~0ULL;
2310                 }
2311
2312                 if (lun_start == lun_end)
2313                         continue;
2314
2315                 req_fm_len = obd_object_end - lun_start;
2316                 fm_local->fm_length = 0;
2317                 len_mapped_single_call = 0;
2318
2319                 /* If the output buffer is very large and the objects have many
2320                  * extents we may need to loop on a single OST repeatedly */
2321                 ost_eof = 0;
2322                 ost_done = 0;
2323                 do {
2324                         if (get_num_extents == 0) {
2325                                 /* Don't get too many extents. */
2326                                 if (current_extent + count_local >
2327                                     fiemap->fm_extent_count)
2328                                         count_local = fiemap->fm_extent_count -
2329                                                                  current_extent;
2330                         }
2331
2332                         lun_start += len_mapped_single_call;
2333                         fm_local->fm_length = req_fm_len - len_mapped_single_call;
2334                         req_fm_len = fm_local->fm_length;
2335                         fm_local->fm_extent_count = count_local;
2336                         fm_local->fm_mapped_extents = 0;
2337                         fm_local->fm_flags = fiemap->fm_flags;
2338
2339                         fm_key->oa.o_id = lsm->lsm_oinfo[cur_stripe]->loi_id;
2340                         fm_key->oa.o_seq = lsm->lsm_oinfo[cur_stripe]->loi_seq;
2341                         ost_index = lsm->lsm_oinfo[cur_stripe]->loi_ost_idx;
2342
2343                         if (ost_index < 0 || ost_index >=lov->desc.ld_tgt_count)
2344                                 GOTO(out, rc = -EINVAL);
2345
2346                         /* If OST is inactive, return extent with UNKNOWN flag */
2347                         if (!lov->lov_tgts[ost_index]->ltd_active) {
2348                                 fm_local->fm_flags |= FIEMAP_EXTENT_LAST;
2349                                 fm_local->fm_mapped_extents = 1;
2350
2351                                 lcl_fm_ext[0].fe_logical = lun_start;
2352                                 lcl_fm_ext[0].fe_length = obd_object_end -
2353                                                                       lun_start;
2354                                 lcl_fm_ext[0].fe_flags |= FIEMAP_EXTENT_UNKNOWN;
2355
2356                                 goto inactive_tgt;
2357                         }
2358
2359                         fm_local->fm_start = lun_start;
2360                         fm_local->fm_flags &= ~FIEMAP_FLAG_DEVICE_ORDER;
2361                         memcpy(&fm_key->fiemap, fm_local, sizeof(*fm_local));
2362                         *vallen=fiemap_count_to_size(fm_local->fm_extent_count);
2363                         rc = obd_get_info(NULL,
2364                                           lov->lov_tgts[ost_index]->ltd_exp,
2365                                           keylen, key, vallen, fm_local, lsm);
2366                         if (rc != 0)
2367                                 GOTO(out, rc);
2368
2369 inactive_tgt:
2370                         ext_count = fm_local->fm_mapped_extents;
2371                         if (ext_count == 0) {
2372                                 ost_done = 1;
2373                                 /* If last stripe has hole at the end,
2374                                  * then we need to return */
2375                                 if (cur_stripe_wrap == last_stripe) {
2376                                         fiemap->fm_mapped_extents = 0;
2377                                         goto finish;
2378                                 }
2379                                 break;
2380                         }
2381
2382                         /* If we just need num of extents then go to next device */
2383                         if (get_num_extents) {
2384                                 current_extent += ext_count;
2385                                 break;
2386                         }
2387
2388                         len_mapped_single_call = lcl_fm_ext[ext_count-1].fe_logical -
2389                                   lun_start + lcl_fm_ext[ext_count - 1].fe_length;
2390
2391                         /* Have we finished mapping on this device? */
2392                         if (req_fm_len <= len_mapped_single_call)
2393                                 ost_done = 1;
2394
2395                         /* Clear the EXTENT_LAST flag which can be present on
2396                          * last extent */
2397                         if (lcl_fm_ext[ext_count-1].fe_flags & FIEMAP_EXTENT_LAST)
2398                                 lcl_fm_ext[ext_count - 1].fe_flags &=
2399                                                             ~FIEMAP_EXTENT_LAST;
2400
2401                         curr_loc = lov_stripe_size(lsm,
2402                                            lcl_fm_ext[ext_count - 1].fe_logical+
2403                                            lcl_fm_ext[ext_count - 1].fe_length,
2404                                            cur_stripe);
2405                         if (curr_loc >= fm_key->oa.o_size)
2406                                 ost_eof = 1;
2407
2408                         fiemap_prepare_and_copy_exts(fiemap, lcl_fm_ext,
2409                                                      ost_index, ext_count,
2410                                                      current_extent);
2411
2412                         current_extent += ext_count;
2413
2414                         /* Ran out of available extents? */
2415                         if (current_extent >= fiemap->fm_extent_count)
2416                                 goto finish;
2417                 } while (ost_done == 0 && ost_eof == 0);
2418
2419                 if (cur_stripe_wrap == last_stripe)
2420                         goto finish;
2421         }
2422
2423 finish:
2424         /* Indicate that we are returning device offsets unless file just has
2425          * single stripe */
2426         if (lsm->lsm_stripe_count > 1)
2427                 fiemap->fm_flags |= FIEMAP_FLAG_DEVICE_ORDER;
2428
2429         if (get_num_extents)
2430                 goto skip_last_device_calc;
2431
2432         /* Check if we have reached the last stripe and whether mapping for that
2433          * stripe is done. */
2434         if (cur_stripe_wrap == last_stripe) {
2435                 if (ost_done || ost_eof)
2436                         fiemap->fm_extents[current_extent - 1].fe_flags |=
2437                                                              FIEMAP_EXTENT_LAST;
2438         }
2439
2440 skip_last_device_calc:
2441         fiemap->fm_mapped_extents = current_extent;
2442
2443 out:
2444         OBD_FREE_LARGE(fm_local, buffer_size);
2445         return rc;
2446 }
2447
2448 static int lov_get_info(const struct lu_env *env, struct obd_export *exp,
2449                         __u32 keylen, void *key, __u32 *vallen, void *val,
2450                         struct lov_stripe_md *lsm)
2451 {
2452         struct obd_device *obddev = class_exp2obd(exp);
2453         struct lov_obd *lov = &obddev->u.lov;
2454         int i, rc;
2455         ENTRY;
2456
2457         if (!vallen || !val)
2458                 RETURN(-EFAULT);
2459
2460         obd_getref(obddev);
2461
2462         if (KEY_IS(KEY_LOCK_TO_STRIPE)) {
2463                 struct {
2464                         char name[16];
2465                         struct ldlm_lock *lock;
2466                 } *data = key;
2467                 struct ldlm_res_id *res_id = &data->lock->l_resource->lr_name;
2468                 struct lov_oinfo *loi;
2469                 __u32 *stripe = val;
2470
2471                 if (*vallen < sizeof(*stripe))
2472                         GOTO(out, rc = -EFAULT);
2473                 *vallen = sizeof(*stripe);
2474
2475                 /* XXX This is another one of those bits that will need to
2476                  * change if we ever actually support nested LOVs.  It uses
2477                  * the lock's export to find out which stripe it is. */
2478                 /* XXX - it's assumed all the locks for deleted OSTs have
2479                  * been cancelled. Also, the export for deleted OSTs will
2480                  * be NULL and won't match the lock's export. */
2481                 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2482                         loi = lsm->lsm_oinfo[i];
2483                         if (!lov->lov_tgts[loi->loi_ost_idx])
2484                                 continue;
2485                         if (lov->lov_tgts[loi->loi_ost_idx]->ltd_exp ==
2486                             data->lock->l_conn_export &&
2487                             osc_res_name_eq(loi->loi_id, loi->loi_seq, res_id)) {
2488                                 *stripe = i;
2489                                 GOTO(out, rc = 0);
2490                         }
2491                 }
2492                 LDLM_ERROR(data->lock, "lock on inode without such object");
2493                 dump_lsm(D_ERROR, lsm);
2494                 GOTO(out, rc = -ENXIO);
2495         } else if (KEY_IS(KEY_LAST_ID)) {
2496                 struct obd_id_info *info = val;
2497                 __u32 size = sizeof(obd_id);
2498                 struct lov_tgt_desc *tgt;
2499
2500                 LASSERT(*vallen == sizeof(struct obd_id_info));
2501                 tgt = lov->lov_tgts[info->idx];
2502
2503                 if (!tgt || !tgt->ltd_active)
2504                         GOTO(out, rc = -ESRCH);
2505
2506                 rc = obd_get_info(env, tgt->ltd_exp, keylen, key,
2507                                   &size, info->data, NULL);
2508                 GOTO(out, rc = 0);
2509         } else if (KEY_IS(KEY_LOVDESC)) {
2510                 struct lov_desc *desc_ret = val;
2511                 *desc_ret = lov->desc;
2512
2513                 GOTO(out, rc = 0);
2514         } else if (KEY_IS(KEY_FIEMAP)) {
2515                 rc = lov_fiemap(lov, keylen, key, vallen, val, lsm);
2516                 GOTO(out, rc);
2517         } else if (KEY_IS(KEY_CONNECT_FLAG)) {
2518                 struct lov_tgt_desc *tgt;
2519                 __u64 ost_idx = *((__u64*)val);
2520
2521                 LASSERT(*vallen == sizeof(__u64));
2522                 LASSERT(ost_idx < lov->desc.ld_tgt_count);
2523                 tgt = lov->lov_tgts[ost_idx];
2524
2525                 if (!tgt || !tgt->ltd_exp)
2526                         GOTO(out, rc = -ESRCH);
2527
2528                 *((__u64*)val) = tgt->ltd_exp->exp_connect_flags;
2529                 GOTO(out, rc = 0);
2530         } else if (KEY_IS(KEY_TGT_COUNT)) {
2531                 *((int *)val) = lov->desc.ld_tgt_count;
2532                 GOTO(out, rc = 0);
2533         }
2534
2535         rc = -EINVAL;
2536
2537 out:
2538         obd_putref(obddev);
2539         RETURN(rc);
2540 }
2541
2542 static int lov_set_info_async(const struct lu_env *env, struct obd_export *exp,
2543                               obd_count keylen, void *key, obd_count vallen,
2544                               void *val, struct ptlrpc_request_set *set)
2545 {
2546         struct obd_device *obddev = class_exp2obd(exp);
2547         struct lov_obd *lov = &obddev->u.lov;
2548         obd_count count;
2549         int i, rc = 0, err;
2550         struct lov_tgt_desc *tgt;
2551         unsigned incr, check_uuid,
2552                  do_inactive, no_set;
2553         unsigned next_id = 0,  mds_con = 0, capa = 0;
2554         ENTRY;
2555
2556         incr = check_uuid = do_inactive = no_set = 0;
2557         if (set == NULL) {
2558                 no_set = 1;
2559                 set = ptlrpc_prep_set();
2560                 if (!set)
2561                         RETURN(-ENOMEM);
2562         }
2563
2564         obd_getref(obddev);
2565         count = lov->desc.ld_tgt_count;
2566
2567         if (KEY_IS(KEY_NEXT_ID)) {
2568                 count = vallen / sizeof(struct obd_id_info);
2569                 vallen = sizeof(obd_id);
2570                 incr = sizeof(struct obd_id_info);
2571                 do_inactive = 1;
2572                 next_id = 1;
2573         } else if (KEY_IS(KEY_CHECKSUM)) {
2574                 do_inactive = 1;
2575         } else if (KEY_IS(KEY_EVICT_BY_NID)) {
2576                 /* use defaults:  do_inactive = incr = 0; */
2577         } else if (KEY_IS(KEY_MDS_CONN)) {
2578                 mds_con = 1;
2579         } else if (KEY_IS(KEY_CAPA_KEY)) {
2580                 capa = 1;
2581         } else if (KEY_IS(KEY_CACHE_SET)) {
2582                 LASSERT(lov->lov_cache == NULL);
2583                 lov->lov_cache = val;
2584                 do_inactive = 1;
2585         }
2586
2587         for (i = 0; i < count; i++, val = (char *)val + incr) {
2588                 if (next_id) {
2589                         tgt = lov->lov_tgts[((struct obd_id_info*)val)->idx];
2590                 } else {
2591                         tgt = lov->lov_tgts[i];
2592                 }
2593                 /* OST was disconnected */
2594                 if (!tgt || !tgt->ltd_exp)
2595                         continue;
2596
2597                 /* OST is inactive and we don't want inactive OSCs */
2598                 if (!tgt->ltd_active && !do_inactive)
2599                         continue;
2600
2601                 if (mds_con) {
2602                         struct mds_group_info *mgi;
2603
2604                         LASSERT(vallen == sizeof(*mgi));
2605                         mgi = (struct mds_group_info *)val;
2606
2607                         /* Only want a specific OSC */
2608                         if (mgi->uuid && !obd_uuid_equals(mgi->uuid,
2609                                                 &tgt->ltd_uuid))
2610                                 continue;
2611
2612                         err = obd_set_info_async(env, tgt->ltd_exp,
2613                                          keylen, key, sizeof(int),
2614                                          &mgi->group, set);
2615                 } else if (next_id) {
2616                         err = obd_set_info_async(env, tgt->ltd_exp,
2617                                          keylen, key, vallen,
2618                                          ((struct obd_id_info*)val)->data, set);
2619                 } else if (capa) {
2620                         struct mds_capa_info *info = (struct mds_capa_info*)val;
2621
2622                         LASSERT(vallen == sizeof(*info));
2623
2624                          /* Only want a specific OSC */
2625                         if (info->uuid &&
2626                             !obd_uuid_equals(info->uuid, &tgt->ltd_uuid))
2627                                 continue;
2628
2629                         err = obd_set_info_async(env, tgt->ltd_exp, keylen,
2630                                                  key, sizeof(*info->capa),
2631                                                  info->capa, set);
2632                 } else {
2633                         /* Only want a specific OSC */
2634                         if (check_uuid &&
2635                             !obd_uuid_equals(val, &tgt->ltd_uuid))
2636                                 continue;
2637
2638                         err = obd_set_info_async(env, tgt->ltd_exp,
2639                                          keylen, key, vallen, val, set);
2640                 }
2641
2642                 if (!rc)
2643                         rc = err;
2644         }
2645
2646         obd_putref(obddev);
2647         if (no_set) {
2648                 err = ptlrpc_set_wait(set);
2649                 if (!rc)
2650                         rc = err;
2651                 ptlrpc_set_destroy(set);
2652         }
2653         RETURN(rc);
2654 }
2655
2656 static int lov_extent_calc(struct obd_export *exp, struct lov_stripe_md *lsm,
2657                            int cmd, __u64 *offset)
2658 {
2659         __u32 ssize = lsm->lsm_stripe_size;
2660         __u64 start;
2661
2662         start = *offset;
2663         lov_do_div64(start, ssize);
2664         start = start * ssize;
2665
2666         CDEBUG(D_DLMTRACE, "offset "LPU64", stripe %u, start "LPU64
2667                            ", end "LPU64"\n", *offset, ssize, start,
2668                            start + ssize - 1);
2669         if (cmd == OBD_CALC_STRIPE_END) {
2670                 *offset = start + ssize - 1;
2671         } else if (cmd == OBD_CALC_STRIPE_START) {
2672                 *offset = start;
2673         } else {
2674                 LBUG();
2675         }
2676
2677         RETURN(0);
2678 }
2679
2680 void lov_stripe_lock(struct lov_stripe_md *md)
2681 {
2682         LASSERT(md->lsm_lock_owner != cfs_curproc_pid());
2683         spin_lock(&md->lsm_lock);
2684         LASSERT(md->lsm_lock_owner == 0);
2685         md->lsm_lock_owner = cfs_curproc_pid();
2686 }
2687 EXPORT_SYMBOL(lov_stripe_lock);
2688
2689 void lov_stripe_unlock(struct lov_stripe_md *md)
2690 {
2691         LASSERT(md->lsm_lock_owner == cfs_curproc_pid());
2692         md->lsm_lock_owner = 0;
2693         spin_unlock(&md->lsm_lock);
2694 }
2695 EXPORT_SYMBOL(lov_stripe_unlock);
2696
2697 static int lov_quotactl(struct obd_device *obd, struct obd_export *exp,
2698                         struct obd_quotactl *oqctl)
2699 {
2700         struct lov_obd      *lov = &obd->u.lov;
2701         struct lov_tgt_desc *tgt;
2702         __u64                curspace = 0;
2703         __u64                bhardlimit = 0;
2704         int                  i, rc = 0;
2705         ENTRY;
2706
2707         if (oqctl->qc_cmd != LUSTRE_Q_QUOTAON &&
2708             oqctl->qc_cmd != LUSTRE_Q_QUOTAOFF &&
2709             oqctl->qc_cmd != Q_GETOQUOTA &&
2710             oqctl->qc_cmd != Q_INITQUOTA &&
2711             oqctl->qc_cmd != LUSTRE_Q_SETQUOTA &&
2712             oqctl->qc_cmd != Q_FINVALIDATE) {
2713                 CERROR("bad quota opc %x for lov obd", oqctl->qc_cmd);
2714                 RETURN(-EFAULT);
2715         }
2716
2717         /* for lov tgt */
2718         obd_getref(obd);
2719         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2720                 int err;
2721
2722                 tgt = lov->lov_tgts[i];
2723
2724                 if (!tgt)
2725                         continue;
2726
2727                 if (!tgt->ltd_active || tgt->ltd_reap) {
2728                         if (oqctl->qc_cmd == Q_GETOQUOTA &&
2729                             lov->lov_tgts[i]->ltd_activate) {
2730                                 rc = -EREMOTEIO;
2731                                 CERROR("ost %d is inactive\n", i);
2732                         } else {
2733                                 CDEBUG(D_HA, "ost %d is inactive\n", i);
2734                         }
2735                         continue;
2736                 }
2737
2738                 err = obd_quotactl(tgt->ltd_exp, oqctl);
2739                 if (err) {
2740                         if (tgt->ltd_active && !rc)
2741                                 rc = err;
2742                         continue;
2743                 }
2744
2745                 if (oqctl->qc_cmd == Q_GETOQUOTA) {
2746                         curspace += oqctl->qc_dqblk.dqb_curspace;
2747                         bhardlimit += oqctl->qc_dqblk.dqb_bhardlimit;
2748                 }
2749         }
2750         obd_putref(obd);
2751
2752         if (oqctl->qc_cmd == Q_GETOQUOTA) {
2753                 oqctl->qc_dqblk.dqb_curspace = curspace;
2754                 oqctl->qc_dqblk.dqb_bhardlimit = bhardlimit;
2755         }
2756         RETURN(rc);
2757 }
2758
2759 static int lov_quotacheck(struct obd_device *obd, struct obd_export *exp,
2760                           struct obd_quotactl *oqctl)
2761 {
2762         struct lov_obd *lov = &obd->u.lov;
2763         int             i, rc = 0;
2764         ENTRY;
2765
2766         obd_getref(obd);
2767
2768         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2769                 if (!lov->lov_tgts[i])
2770                         continue;
2771
2772                 /* Skip quota check on the administratively disabled OSTs. */
2773                 if (!lov->lov_tgts[i]->ltd_activate) {
2774                         CWARN("lov idx %d was administratively disabled, "
2775                               "skip quotacheck on it.\n", i);
2776                         continue;
2777                 }
2778
2779                 if (!lov->lov_tgts[i]->ltd_active) {
2780                         CERROR("lov idx %d inactive\n", i);
2781                         rc = -EIO;
2782                         goto out;
2783                 }
2784         }
2785
2786         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2787                 int err;
2788
2789                 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_activate)
2790                         continue;
2791
2792                 err = obd_quotacheck(lov->lov_tgts[i]->ltd_exp, oqctl);
2793                 if (err && !rc)
2794                         rc = err;
2795         }
2796
2797 out:
2798         obd_putref(obd);
2799
2800         RETURN(rc);
2801 }
2802
2803 struct obd_ops lov_obd_ops = {
2804         .o_owner               = THIS_MODULE,
2805         .o_setup               = lov_setup,
2806         .o_precleanup          = lov_precleanup,
2807         .o_cleanup             = lov_cleanup,
2808         //.o_process_config      = lov_process_config,
2809         .o_connect             = lov_connect,
2810         .o_disconnect          = lov_disconnect,
2811         .o_statfs              = lov_statfs,
2812         .o_statfs_async        = lov_statfs_async,
2813         .o_packmd              = lov_packmd,
2814         .o_unpackmd            = lov_unpackmd,
2815         .o_create              = lov_create,
2816         .o_destroy             = lov_destroy,
2817         .o_getattr             = lov_getattr,
2818         .o_getattr_async       = lov_getattr_async,
2819         .o_setattr             = lov_setattr,
2820         .o_setattr_async       = lov_setattr_async,
2821         .o_brw                 = lov_brw,
2822         .o_merge_lvb           = lov_merge_lvb,
2823         .o_adjust_kms          = lov_adjust_kms,
2824         .o_punch               = lov_punch,
2825         .o_sync                = lov_sync,
2826         .o_enqueue             = lov_enqueue,
2827         .o_change_cbdata       = lov_change_cbdata,
2828         .o_find_cbdata         = lov_find_cbdata,
2829         .o_cancel              = lov_cancel,
2830         .o_cancel_unused       = lov_cancel_unused,
2831         .o_iocontrol           = lov_iocontrol,
2832         .o_get_info            = lov_get_info,
2833         .o_set_info_async      = lov_set_info_async,
2834         .o_extent_calc         = lov_extent_calc,
2835         .o_llog_init           = lov_llog_init,
2836         .o_llog_finish         = lov_llog_finish,
2837         .o_notify              = lov_notify,
2838         .o_pool_new            = lov_pool_new,
2839         .o_pool_rem            = lov_pool_remove,
2840         .o_pool_add            = lov_pool_add,
2841         .o_pool_del            = lov_pool_del,
2842         .o_getref              = lov_getref,
2843         .o_putref              = lov_putref,
2844         .o_quotactl            = lov_quotactl,
2845         .o_quotacheck          = lov_quotacheck,
2846 };
2847
2848 cfs_mem_cache_t *lov_oinfo_slab;
2849
2850 extern struct lu_kmem_descr lov_caches[];
2851
2852 int __init lov_init(void)
2853 {
2854         struct lprocfs_static_vars lvars = { 0 };
2855         int rc, rc2;
2856         ENTRY;
2857
2858         /* print an address of _any_ initialized kernel symbol from this
2859          * module, to allow debugging with gdb that doesn't support data
2860          * symbols from modules.*/
2861         CDEBUG(D_INFO, "Lustre LOV module (%p).\n", &lov_caches);
2862
2863         rc = lu_kmem_init(lov_caches);
2864         if (rc)
2865                 return rc;
2866
2867         lov_oinfo_slab = cfs_mem_cache_create("lov_oinfo",
2868                                               sizeof(struct lov_oinfo),
2869                                               0, CFS_SLAB_HWCACHE_ALIGN);
2870         if (lov_oinfo_slab == NULL) {
2871                 lu_kmem_fini(lov_caches);
2872                 return -ENOMEM;
2873         }
2874         lprocfs_lov_init_vars(&lvars);
2875
2876         rc = class_register_type(&lov_obd_ops, NULL, lvars.module_vars,
2877                                  LUSTRE_LOV_NAME, &lov_device_type);
2878
2879         if (rc) {
2880                 rc2 = cfs_mem_cache_destroy(lov_oinfo_slab);
2881                 LASSERT(rc2 == 0);
2882                 lu_kmem_fini(lov_caches);
2883         }
2884
2885         RETURN(rc);
2886 }
2887
2888 #ifdef __KERNEL__
2889 static void /*__exit*/ lov_exit(void)
2890 {
2891         int rc;
2892
2893         class_unregister_type(LUSTRE_LOV_NAME);
2894         rc = cfs_mem_cache_destroy(lov_oinfo_slab);
2895         LASSERT(rc == 0);
2896
2897         lu_kmem_fini(lov_caches);
2898 }
2899
2900 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2901 MODULE_DESCRIPTION("Lustre Logical Object Volume OBD driver");
2902 MODULE_LICENSE("GPL");
2903
2904 cfs_module(lov, LUSTRE_VERSION_STRING, lov_init, lov_exit);
2905 #endif