lustre/lov/lov_obd.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * GPL HEADER START
   5  *
   6  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License version 2 only,
  10  * as published by the Free Software Foundation.
  11  *
  12  * This program is distributed in the hope that it will be useful, but
  13  * WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * General Public License version 2 for more details (a copy is included
  16  * in the LICENSE file that accompanied this code).
  17  *
  18  * You should have received a copy of the GNU General Public License
  19  * version 2 along with this program; If not, see
  20  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  21  *
  22  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  23  * CA 95054 USA or visit www.sun.com if you need additional information or
  24  * have any questions.
  25  *
  26  * GPL HEADER END
  27  */
  28 /*
  29  * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
  30  * Use is subject to license terms.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/lov/lov_obd.c
  37  *
  38  * Author: Phil Schwan <phil@clusterfs.com>
  39  * Author: Peter Braam <braam@clusterfs.com>
  40  * Author: Mike Shaver <shaver@clusterfs.com>
  41  * Author: Nathan Rutman <nathan@clusterfs.com>
  42  */
  43
  44 #ifndef EXPORT_SYMTAB
  45 # define EXPORT_SYMTAB
  46 #endif
  47 #define DEBUG_SUBSYSTEM S_LOV
  48 #ifdef __KERNEL__
  49 #include <libcfs/libcfs.h>
  50 #else
  51 #include <liblustre.h>
  52 #endif
  53
  54 #include <obd_support.h>
  55 #include <lustre_lib.h>
  56 #include <lustre_net.h>
  57 #include <lustre/lustre_idl.h>
  58 #include <lustre_dlm.h>
  59 #include <lustre_mds.h>
  60 #include <lustre_debug.h>
  61 #include <obd_class.h>
  62 #include <obd_lov.h>
  63 #include <obd_ost.h>
  64 #include <lprocfs_status.h>
  65 #include <lustre_param.h>
  66 #include <cl_object.h>
  67 #include <lustre/ll_fiemap.h>
  68
  69 #include "lov_internal.h"
  70
  71 /* Keep a refcount of lov->tgt usage to prevent racing with addition/deletion.
  72    Any function that expects lov_tgts to remain stationary must take a ref. */
  73 static void lov_getref(struct obd_device *obd)
  74 {
  75         struct lov_obd *lov = &obd->u.lov;
  76
  77         /* nobody gets through here until lov_putref is done */
  78         cfs_mutex_down(&lov->lov_lock);
  79         cfs_atomic_inc(&lov->lov_refcount);
  80         cfs_mutex_up(&lov->lov_lock);
  81         return;
  82 }
  83
  84 static void __lov_del_obd(struct obd_device *obd, struct lov_tgt_desc *tgt);
  85
  86 static void lov_putref(struct obd_device *obd)
  87 {
  88         struct lov_obd *lov = &obd->u.lov;
  89
  90         cfs_mutex_down(&lov->lov_lock);
  91         /* ok to dec to 0 more than once -- ltd_exp's will be null */
  92         if (cfs_atomic_dec_and_test(&lov->lov_refcount) && lov->lov_death_row) {
  93                 CFS_LIST_HEAD(kill);
  94                 int i;
  95                 struct lov_tgt_desc *tgt, *n;
  96                 CDEBUG(D_CONFIG, "destroying %d lov targets\n",
  97                        lov->lov_death_row);
  98                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
  99                         tgt = lov->lov_tgts[i];
 100
 101                         if (!tgt || !tgt->ltd_reap)
 102                                 continue;
 103                         cfs_list_add(&tgt->ltd_kill, &kill);
 104                         /* XXX - right now there is a dependency on ld_tgt_count
 105                          * being the maximum tgt index for computing the
 106                          * mds_max_easize. So we can't shrink it. */
 107                         lov_ost_pool_remove(&lov->lov_packed, i);
 108                         lov->lov_tgts[i] = NULL;
 109                         lov->lov_death_row--;
 110                 }
 111                 cfs_mutex_up(&lov->lov_lock);
 112
 113                 cfs_list_for_each_entry_safe(tgt, n, &kill, ltd_kill) {
 114                         cfs_list_del(&tgt->ltd_kill);
 115                         /* Disconnect */
 116                         __lov_del_obd(obd, tgt);
 117                 }
 118         } else {
 119                 cfs_mutex_up(&lov->lov_lock);
 120         }
 121 }
 122
 123 static int lov_set_osc_active(struct obd_device *obd, struct obd_uuid *uuid,
 124                               int activate);
 125 static int lov_notify(struct obd_device *obd, struct obd_device *watched,
 126                       enum obd_notify_event ev, void *data);
 127
 128
 129 #define MAX_STRING_SIZE 128
 130 int lov_connect_obd(struct obd_device *obd, __u32 index, int activate,
 131                     struct obd_connect_data *data)
 132 {
 133         struct lov_obd *lov = &obd->u.lov;
 134         struct obd_uuid *tgt_uuid;
 135         struct obd_device *tgt_obd;
 136         static struct obd_uuid lov_osc_uuid = { "LOV_OSC_UUID" };
 137         struct obd_import *imp;
 138 #ifdef __KERNEL__
 139         cfs_proc_dir_entry_t *lov_proc_dir;
 140 #endif
 141         int rc;
 142         ENTRY;
 143
 144         if (!lov->lov_tgts[index])
 145                 RETURN(-EINVAL);
 146
 147         tgt_uuid = &lov->lov_tgts[index]->ltd_uuid;
 148         tgt_obd = lov->lov_tgts[index]->ltd_obd;
 149
 150         if (!tgt_obd->obd_set_up) {
 151                 CERROR("Target %s not set up\n", obd_uuid2str(tgt_uuid));
 152                 RETURN(-EINVAL);
 153         }
 154
 155         /* override the sp_me from lov */
 156         tgt_obd->u.cli.cl_sp_me = lov->lov_sp_me;
 157
 158         if (data && (data->ocd_connect_flags & OBD_CONNECT_INDEX))
 159                 data->ocd_index = index;
 160
 161         /*
 162          * Divine LOV knows that OBDs under it are OSCs.
 163          */
 164         imp = tgt_obd->u.cli.cl_import;
 165
 166         if (activate) {
 167                 tgt_obd->obd_no_recov = 0;
 168                 /* FIXME this is probably supposed to be
 169                    ptlrpc_set_import_active.  Horrible naming. */
 170                 ptlrpc_activate_import(imp);
 171         }
 172
 173         rc = obd_register_observer(tgt_obd, obd);
 174         if (rc) {
 175                 CERROR("Target %s register_observer error %d\n",
 176                        obd_uuid2str(tgt_uuid), rc);
 177                 RETURN(rc);
 178         }
 179
 180
 181         if (imp->imp_invalid) {
 182                 CERROR("not connecting OSC %s; administratively "
 183                        "disabled\n", obd_uuid2str(tgt_uuid));
 184                 RETURN(0);
 185         }
 186
 187         rc = obd_connect(NULL, &lov->lov_tgts[index]->ltd_exp, tgt_obd,
 188                          &lov_osc_uuid, data, NULL);
 189         if (rc || !lov->lov_tgts[index]->ltd_exp) {
 190                 CERROR("Target %s connect error %d\n",
 191                        obd_uuid2str(tgt_uuid), rc);
 192                 RETURN(-ENODEV);
 193         }
 194
 195         lov->lov_tgts[index]->ltd_reap = 0;
 196
 197         CDEBUG(D_CONFIG, "Connected tgt idx %d %s (%s) %sactive\n", index,
 198                obd_uuid2str(tgt_uuid), tgt_obd->obd_name, activate ? "":"in");
 199
 200 #ifdef __KERNEL__
 201         lov_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
 202         if (lov_proc_dir) {
 203                 struct obd_device *osc_obd = lov->lov_tgts[index]->ltd_exp->exp_obd;
 204                 cfs_proc_dir_entry_t *osc_symlink;
 205
 206                 LASSERT(osc_obd != NULL);
 207                 LASSERT(osc_obd->obd_magic == OBD_DEVICE_MAGIC);
 208                 LASSERT(osc_obd->obd_type->typ_name != NULL);
 209
 210                 osc_symlink = lprocfs_add_symlink(osc_obd->obd_name,
 211                                                   lov_proc_dir,
 212                                                   "../../../%s/%s",
 213                                                   osc_obd->obd_type->typ_name,
 214                                                   osc_obd->obd_name);
 215                 if (osc_symlink == NULL) {
 216                         CERROR("could not register LOV target "
 217                                 "/proc/fs/lustre/%s/%s/target_obds/%s.",
 218                                 obd->obd_type->typ_name, obd->obd_name,
 219                                 osc_obd->obd_name);
 220                         lprocfs_remove(&lov_proc_dir);
 221                 }
 222         }
 223 #endif
 224
 225         rc = qos_add_tgt(obd, index);
 226         if (rc)
 227                 CERROR("qos_add_tgt failed %d\n", rc);
 228
 229         RETURN(0);
 230 }
 231
 232 static int lov_connect(const struct lu_env *env,
 233                        struct obd_export **exp, struct obd_device *obd,
 234                        struct obd_uuid *cluuid, struct obd_connect_data *data,
 235                        void *localdata)
 236 {
 237         struct lov_obd *lov = &obd->u.lov;
 238         struct lov_tgt_desc *tgt;
 239         struct lustre_handle conn;
 240         int i, rc;
 241         ENTRY;
 242
 243         CDEBUG(D_CONFIG, "connect #%d\n", lov->lov_connects);
 244
 245         rc = class_connect(&conn, obd, cluuid);
 246         if (rc)
 247                 RETURN(rc);
 248
 249         *exp = class_conn2export(&conn);
 250
 251         /* Why should there ever be more than 1 connect? */
 252         lov->lov_connects++;
 253         LASSERT(lov->lov_connects == 1);
 254
 255         memset(&lov->lov_ocd, 0, sizeof(lov->lov_ocd));
 256         if (data)
 257                 lov->lov_ocd = *data;
 258
 259         obd_getref(obd);
 260         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
 261                 tgt = lov->lov_tgts[i];
 262                 if (!tgt || obd_uuid_empty(&tgt->ltd_uuid))
 263                         continue;
 264                 /* Flags will be lowest common denominator */
 265                 rc = lov_connect_obd(obd, i, tgt->ltd_activate, &lov->lov_ocd);
 266                 if (rc) {
 267                         CERROR("%s: lov connect tgt %d failed: %d\n",
 268                                obd->obd_name, i, rc);
 269                         continue;
 270                 }
 271                 /* connect to administrative disabled ost */
 272                 if (!lov->lov_tgts[i]->ltd_exp)
 273                         continue;
 274
 275                 rc = lov_notify(obd, lov->lov_tgts[i]->ltd_exp->exp_obd,
 276                                 OBD_NOTIFY_CONNECT, (void *)&i);
 277                 if (rc) {
 278                         CERROR("%s error sending notify %d\n",
 279                                obd->obd_name, rc);
 280                 }
 281         }
 282         obd_putref(obd);
 283
 284         RETURN(0);
 285 }
 286
 287 static int lov_disconnect_obd(struct obd_device *obd, struct lov_tgt_desc *tgt)
 288 {
 289         cfs_proc_dir_entry_t *lov_proc_dir;
 290         struct lov_obd *lov = &obd->u.lov;
 291         struct obd_device *osc_obd;
 292         int rc;
 293         ENTRY;
 294
 295         osc_obd = class_exp2obd(tgt->ltd_exp);
 296         CDEBUG(D_CONFIG, "%s: disconnecting target %s\n",
 297                obd->obd_name, osc_obd->obd_name);
 298
 299         if (tgt->ltd_active) {
 300                 tgt->ltd_active = 0;
 301                 lov->desc.ld_active_tgt_count--;
 302                 tgt->ltd_exp->exp_obd->obd_inactive = 1;
 303         }
 304
 305         lov_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
 306         if (lov_proc_dir) {
 307                 cfs_proc_dir_entry_t *osc_symlink;
 308
 309                 osc_symlink = lprocfs_srch(lov_proc_dir, osc_obd->obd_name);
 310                 if (osc_symlink) {
 311                         lprocfs_remove(&osc_symlink);
 312                 } else {
 313                         CERROR("/proc/fs/lustre/%s/%s/target_obds/%s missing.",
 314                                obd->obd_type->typ_name, obd->obd_name,
 315                                osc_obd->obd_name);
 316                 }
 317         }
 318
 319         if (osc_obd) {
 320                 /* Pass it on to our clients.
 321                  * XXX This should be an argument to disconnect,
 322                  * XXX not a back-door flag on the OBD.  Ah well.
 323                  */
 324                 osc_obd->obd_force = obd->obd_force;
 325                 osc_obd->obd_fail = obd->obd_fail;
 326                 osc_obd->obd_no_recov = obd->obd_no_recov;
 327         }
 328
 329         obd_register_observer(osc_obd, NULL);
 330
 331         rc = obd_disconnect(tgt->ltd_exp);
 332         if (rc) {
 333                 CERROR("Target %s disconnect error %d\n",
 334                        tgt->ltd_uuid.uuid, rc);
 335                 rc = 0;
 336         }
 337
 338         qos_del_tgt(obd, tgt);
 339
 340         tgt->ltd_exp = NULL;
 341         RETURN(0);
 342 }
 343
 344 static int lov_disconnect(struct obd_export *exp)
 345 {
 346         struct obd_device *obd = class_exp2obd(exp);
 347         struct lov_obd *lov = &obd->u.lov;
 348         int i, rc;
 349         ENTRY;
 350
 351         if (!lov->lov_tgts)
 352                 goto out;
 353
 354         /* Only disconnect the underlying layers on the final disconnect. */
 355         lov->lov_connects--;
 356         if (lov->lov_connects != 0) {
 357                 /* why should there be more than 1 connect? */
 358                 CERROR("disconnect #%d\n", lov->lov_connects);
 359                 goto out;
 360         }
 361
 362         /* Let's hold another reference so lov_del_obd doesn't spin through
 363            putref every time */
 364         obd_getref(obd);
 365
 366         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
 367                 if (lov->lov_tgts[i] && lov->lov_tgts[i]->ltd_exp) {
 368                         /* Disconnection is the last we know about an obd */
 369                         lov_del_target(obd, i, 0, lov->lov_tgts[i]->ltd_gen);
 370                 }
 371         }
 372         obd_putref(obd);
 373
 374 out:
 375         rc = class_disconnect(exp); /* bz 9811 */
 376         RETURN(rc);
 377 }
 378
 379 /* Error codes:
 380  *
 381  *  -EINVAL  : UUID can't be found in the LOV's target list
 382  *  -ENOTCONN: The UUID is found, but the target connection is bad (!)
 383  *  -EBADF   : The UUID is found, but the OBD is the wrong type (!)
 384  *  any >= 0 : is log target index
 385  */
 386 static int lov_set_osc_active(struct obd_device *obd, struct obd_uuid *uuid,
 387                               int activate)
 388 {
 389         struct lov_obd *lov = &obd->u.lov;
 390         struct lov_tgt_desc *tgt;
 391         int index;
 392         ENTRY;
 393
 394         CDEBUG(D_INFO, "Searching in lov %p for uuid %s (activate=%d)\n",
 395                lov, uuid->uuid, activate);
 396
 397         obd_getref(obd);
 398         for (index = 0; index < lov->desc.ld_tgt_count; index++) {
 399                 tgt = lov->lov_tgts[index];
 400                 if (!tgt || !tgt->ltd_exp)
 401                         continue;
 402
 403                 CDEBUG(D_INFO, "lov idx %d is %s conn "LPX64"\n",
 404                        index, obd_uuid2str(&tgt->ltd_uuid),
 405                        tgt->ltd_exp->exp_handle.h_cookie);
 406                 if (obd_uuid_equals(uuid, &tgt->ltd_uuid))
 407                         break;
 408         }
 409
 410         if (index == lov->desc.ld_tgt_count)
 411                 GOTO(out, index = -EINVAL);
 412
 413         if (lov->lov_tgts[index]->ltd_active == activate) {
 414                 CDEBUG(D_INFO, "OSC %s already %sactive!\n", uuid->uuid,
 415                        activate ? "" : "in");
 416                 GOTO(out, index);
 417         }
 418
 419         CDEBUG(D_CONFIG, "Marking OSC %s %sactive\n", obd_uuid2str(uuid),
 420                activate ? "" : "in");
 421
 422         lov->lov_tgts[index]->ltd_active = activate;
 423
 424         if (activate) {
 425                 lov->desc.ld_active_tgt_count++;
 426                 lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 0;
 427         } else {
 428                 lov->desc.ld_active_tgt_count--;
 429                 lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 1;
 430         }
 431         /* remove any old qos penalty */
 432         lov->lov_tgts[index]->ltd_qos.ltq_penalty = 0;
 433
 434  out:
 435         obd_putref(obd);
 436         RETURN(index);
 437 }
 438
 439 static int lov_notify(struct obd_device *obd, struct obd_device *watched,
 440                       enum obd_notify_event ev, void *data)
 441 {
 442         int rc = 0;
 443         ENTRY;
 444
 445         if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE) {
 446                 struct obd_uuid *uuid;
 447
 448                 LASSERT(watched);
 449
 450                 if (strcmp(watched->obd_type->typ_name, LUSTRE_OSC_NAME)) {
 451                         CERROR("unexpected notification of %s %s!\n",
 452                                watched->obd_type->typ_name,
 453                                watched->obd_name);
 454                         RETURN(-EINVAL);
 455                 }
 456                 uuid = &watched->u.cli.cl_target_uuid;
 457
 458                 /* Set OSC as active before notifying the observer, so the
 459                  * observer can use the OSC normally.
 460                  */
 461                 rc = lov_set_osc_active(obd, uuid, ev == OBD_NOTIFY_ACTIVE);
 462                 if (rc < 0) {
 463                         CERROR("%sactivation of %s failed: %d\n",
 464                                (ev == OBD_NOTIFY_ACTIVE) ? "" : "de",
 465                                obd_uuid2str(uuid), rc);
 466                         RETURN(rc);
 467                 }
 468                 /* active event should be pass lov target index as data */
 469                 data = &rc;
 470         }
 471
 472         /* Pass the notification up the chain. */
 473         if (watched) {
 474                 rc = obd_notify_observer(obd, watched, ev, data);
 475         } else {
 476                 /* NULL watched means all osc's in the lov (only for syncs) */
 477                 /* sync event should be send lov idx as data */
 478                 struct lov_obd *lov = &obd->u.lov;
 479                 int i, is_sync;
 480
 481                 data = &i;
 482                 is_sync = (ev == OBD_NOTIFY_SYNC) ||
 483                           (ev == OBD_NOTIFY_SYNC_NONBLOCK);
 484
 485                 obd_getref(obd);
 486                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
 487                         if (!lov->lov_tgts[i])
 488                                 continue;
 489
 490                         /* don't send sync event if target not
 491                          * connected/activated */
 492                         if (is_sync &&  !lov->lov_tgts[i]->ltd_active)
 493                                 continue;
 494
 495                         rc = obd_notify_observer(obd, lov->lov_tgts[i]->ltd_obd,
 496                                                  ev, data);
 497                         if (rc) {
 498                                 CERROR("%s: notify %s of %s failed %d\n",
 499                                        obd->obd_name,
 500                                        obd->obd_observer->obd_name,
 501                                        lov->lov_tgts[i]->ltd_obd->obd_name,
 502                                        rc);
 503                         }
 504                 }
 505                 obd_putref(obd);
 506         }
 507
 508         RETURN(rc);
 509 }
 510
 511 int lov_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
 512                    __u32 index, int gen, int active)
 513 {
 514         struct lov_obd *lov = &obd->u.lov;
 515         struct lov_tgt_desc *tgt;
 516         struct obd_device *tgt_obd;
 517         int rc;
 518         ENTRY;
 519
 520         CDEBUG(D_CONFIG, "uuid:%s idx:%d gen:%d active:%d\n",
 521                uuidp->uuid, index, gen, active);
 522
 523         if (gen <= 0) {
 524                 CERROR("request to add OBD %s with invalid generation: %d\n",
 525                        uuidp->uuid, gen);
 526                 RETURN(-EINVAL);
 527         }
 528
 529         tgt_obd = class_find_client_obd(uuidp, LUSTRE_OSC_NAME,
 530                                         &obd->obd_uuid);
 531         if (tgt_obd == NULL)
 532                 RETURN(-EINVAL);
 533
 534         cfs_mutex_down(&lov->lov_lock);
 535
 536         if ((index < lov->lov_tgt_size) && (lov->lov_tgts[index] != NULL)) {
 537                 tgt = lov->lov_tgts[index];
 538                 CERROR("UUID %s already assigned at LOV target index %d\n",
 539                        obd_uuid2str(&tgt->ltd_uuid), index);
 540                 cfs_mutex_up(&lov->lov_lock);
 541                 RETURN(-EEXIST);
 542         }
 543
 544         if (index >= lov->lov_tgt_size) {
 545                 /* We need to reallocate the lov target array. */
 546                 struct lov_tgt_desc **newtgts, **old = NULL;
 547                 __u32 newsize, oldsize = 0;
 548
 549                 newsize = max(lov->lov_tgt_size, (__u32)2);
 550                 while (newsize < index + 1)
 551                         newsize = newsize << 1;
 552                 OBD_ALLOC(newtgts, sizeof(*newtgts) * newsize);
 553                 if (newtgts == NULL) {
 554                         cfs_mutex_up(&lov->lov_lock);
 555                         RETURN(-ENOMEM);
 556                 }
 557
 558                 if (lov->lov_tgt_size) {
 559                         memcpy(newtgts, lov->lov_tgts, sizeof(*newtgts) *
 560                                lov->lov_tgt_size);
 561                         old = lov->lov_tgts;
 562                         oldsize = lov->lov_tgt_size;
 563                 }
 564
 565                 lov->lov_tgts = newtgts;
 566                 lov->lov_tgt_size = newsize;
 567 #ifdef __KERNEL__
 568                 smp_rmb();
 569 #endif
 570                 if (old)
 571                         OBD_FREE(old, sizeof(*old) * oldsize);
 572
 573                 CDEBUG(D_CONFIG, "tgts: %p size: %d\n",
 574                        lov->lov_tgts, lov->lov_tgt_size);
 575         }
 576
 577         OBD_ALLOC_PTR(tgt);
 578         if (!tgt) {
 579                 cfs_mutex_up(&lov->lov_lock);
 580                 RETURN(-ENOMEM);
 581         }
 582
 583         rc = lov_ost_pool_add(&lov->lov_packed, index, lov->lov_tgt_size);
 584         if (rc) {
 585                 cfs_mutex_up(&lov->lov_lock);
 586                 OBD_FREE_PTR(tgt);
 587                 RETURN(rc);
 588         }
 589
 590         tgt->ltd_uuid = *uuidp;
 591         tgt->ltd_obd = tgt_obd;
 592         /* XXX - add a sanity check on the generation number. */
 593         tgt->ltd_gen = gen;
 594         tgt->ltd_index = index;
 595         tgt->ltd_activate = active;
 596         lov->lov_tgts[index] = tgt;
 597         if (index >= lov->desc.ld_tgt_count)
 598                 lov->desc.ld_tgt_count = index + 1;
 599
 600         cfs_mutex_up(&lov->lov_lock);
 601
 602         CDEBUG(D_CONFIG, "idx=%d ltd_gen=%d ld_tgt_count=%d\n",
 603                 index, tgt->ltd_gen, lov->desc.ld_tgt_count);
 604
 605         rc = obd_notify(obd, tgt_obd, OBD_NOTIFY_CREATE, &index);
 606
 607         if (lov->lov_connects == 0) {
 608                 /* lov_connect hasn't been called yet. We'll do the
 609                    lov_connect_obd on this target when that fn first runs,
 610                    because we don't know the connect flags yet. */
 611                 RETURN(0);
 612         }
 613
 614         obd_getref(obd);
 615
 616         rc = lov_connect_obd(obd, index, active, &lov->lov_ocd);
 617         if (rc)
 618                 GOTO(out, rc);
 619
 620         /* connect to administrative disabled ost */
 621         if (!tgt->ltd_exp)
 622                 GOTO(out, rc = 0);
 623
 624         rc = lov_notify(obd, tgt->ltd_exp->exp_obd,
 625                         active ? OBD_NOTIFY_CONNECT : OBD_NOTIFY_INACTIVE,
 626                         (void *)&index);
 627
 628 out:
 629         if (rc) {
 630                 CERROR("add failed (%d), deleting %s\n", rc,
 631                        obd_uuid2str(&tgt->ltd_uuid));
 632                 lov_del_target(obd, index, 0, 0);
 633         }
 634         obd_putref(obd);
 635         RETURN(rc);
 636 }
 637
 638 /* Schedule a target for deletion */
 639 int lov_del_target(struct obd_device *obd, __u32 index,
 640                    struct obd_uuid *uuidp, int gen)
 641 {
 642         struct lov_obd *lov = &obd->u.lov;
 643         int count = lov->desc.ld_tgt_count;
 644         int rc = 0;
 645         ENTRY;
 646
 647         if (index >= count) {
 648                 CERROR("LOV target index %d >= number of LOV OBDs %d.\n",
 649                        index, count);
 650                 RETURN(-EINVAL);
 651         }
 652
 653         obd_getref(obd);
 654
 655         if (!lov->lov_tgts[index]) {
 656                 CERROR("LOV target at index %d is not setup.\n", index);
 657                 GOTO(out, rc = -EINVAL);
 658         }
 659
 660         if (uuidp && !obd_uuid_equals(uuidp, &lov->lov_tgts[index]->ltd_uuid)) {
 661                 CERROR("LOV target UUID %s at index %d doesn't match %s.\n",
 662                        lov_uuid2str(lov, index), index,
 663                        obd_uuid2str(uuidp));
 664                 GOTO(out, rc = -EINVAL);
 665         }
 666
 667         CDEBUG(D_CONFIG, "uuid: %s idx: %d gen: %d exp: %p active: %d\n",
 668                lov_uuid2str(lov, index), index,
 669                lov->lov_tgts[index]->ltd_gen, lov->lov_tgts[index]->ltd_exp,
 670                lov->lov_tgts[index]->ltd_active);
 671
 672         lov->lov_tgts[index]->ltd_reap = 1;
 673         lov->lov_death_row++;
 674         /* we really delete it from obd_putref */
 675 out:
 676         obd_putref(obd);
 677
 678         RETURN(rc);
 679 }
 680
 681 static void __lov_del_obd(struct obd_device *obd, struct lov_tgt_desc *tgt)
 682 {
 683         struct obd_device *osc_obd;
 684
 685         LASSERT(tgt);
 686         LASSERT(tgt->ltd_reap);
 687
 688         osc_obd = class_exp2obd(tgt->ltd_exp);
 689
 690         CDEBUG(D_CONFIG, "Removing tgt %s : %s\n",
 691                tgt->ltd_uuid.uuid,
 692                osc_obd ? osc_obd->obd_name : "<no obd>");
 693
 694         if (tgt->ltd_exp)
 695                 lov_disconnect_obd(obd, tgt);
 696
 697         OBD_FREE_PTR(tgt);
 698
 699         /* Manual cleanup - no cleanup logs to clean up the osc's.  We must
 700            do it ourselves. And we can't do it from lov_cleanup,
 701            because we just lost our only reference to it. */
 702         if (osc_obd)
 703                 class_manual_cleanup(osc_obd);
 704 }
 705
 706 void lov_fix_desc_stripe_size(__u64 *val)
 707 {
 708         if (*val < PTLRPC_MAX_BRW_SIZE) {
 709                 LCONSOLE_WARN("Increasing default stripe size to min %u\n",
 710                               PTLRPC_MAX_BRW_SIZE);
 711                 *val = PTLRPC_MAX_BRW_SIZE;
 712         } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) {
 713                 *val &= ~(LOV_MIN_STRIPE_SIZE - 1);
 714                 LCONSOLE_WARN("Changing default stripe size to "LPU64" (a "
 715                               "multiple of %u)\n",
 716                               *val, LOV_MIN_STRIPE_SIZE);
 717         }
 718 }
 719
 720 void lov_fix_desc_stripe_count(__u32 *val)
 721 {
 722         if (*val == 0)
 723                 *val = 1;
 724 }
 725
 726 void lov_fix_desc_pattern(__u32 *val)
 727 {
 728         /* from lov_setstripe */
 729         if ((*val != 0) && (*val != LOV_PATTERN_RAID0)) {
 730                 LCONSOLE_WARN("Unknown stripe pattern: %#x\n", *val);
 731                 *val = 0;
 732         }
 733 }
 734
 735 void lov_fix_desc_qos_maxage(__u32 *val)
 736 {
 737         /* fix qos_maxage */
 738         if (*val == 0)
 739                 *val = QOS_DEFAULT_MAXAGE;
 740 }
 741
 742 void lov_fix_desc(struct lov_desc *desc)
 743 {
 744         lov_fix_desc_stripe_size(&desc->ld_default_stripe_size);
 745         lov_fix_desc_stripe_count(&desc->ld_default_stripe_count);
 746         lov_fix_desc_pattern(&desc->ld_pattern);
 747         lov_fix_desc_qos_maxage(&desc->ld_qos_maxage);
 748 }
 749
 750 int lov_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
 751 {
 752         struct lprocfs_static_vars lvars = { 0 };
 753         struct lov_desc *desc;
 754         struct lov_obd *lov = &obd->u.lov;
 755         int rc;
 756         ENTRY;
 757
 758         if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
 759                 CERROR("LOV setup requires a descriptor\n");
 760                 RETURN(-EINVAL);
 761         }
 762
 763         desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
 764
 765         if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
 766                 CERROR("descriptor size wrong: %d > %d\n",
 767                        (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
 768                 RETURN(-EINVAL);
 769         }
 770
 771         if (desc->ld_magic != LOV_DESC_MAGIC) {
 772                 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
 773                             CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
 774                                    obd->obd_name, desc);
 775                             lustre_swab_lov_desc(desc);
 776                 } else {
 777                         CERROR("%s: Bad lov desc magic: %#x\n",
 778                                obd->obd_name, desc->ld_magic);
 779                         RETURN(-EINVAL);
 780                 }
 781         }
 782
 783         lov_fix_desc(desc);
 784
 785         desc->ld_active_tgt_count = 0;
 786         lov->desc = *desc;
 787         lov->lov_tgt_size = 0;
 788
 789         cfs_sema_init(&lov->lov_lock, 1);
 790         cfs_atomic_set(&lov->lov_refcount, 0);
 791         CFS_INIT_LIST_HEAD(&lov->lov_qos.lq_oss_list);
 792         cfs_init_rwsem(&lov->lov_qos.lq_rw_sem);
 793         lov->lov_sp_me = LUSTRE_SP_CLI;
 794         lov->lov_qos.lq_dirty = 1;
 795         lov->lov_qos.lq_rr.lqr_dirty = 1;
 796         lov->lov_qos.lq_reset = 1;
 797         /* Default priority is toward free space balance */
 798         lov->lov_qos.lq_prio_free = 232;
 799         /* Default threshold for rr (roughly 17%) */
 800         lov->lov_qos.lq_threshold_rr = 43;
 801         /* Init statfs fields */
 802         OBD_ALLOC_PTR(lov->lov_qos.lq_statfs_data);
 803         if (NULL == lov->lov_qos.lq_statfs_data)
 804                 RETURN(-ENOMEM);
 805         cfs_waitq_init(&lov->lov_qos.lq_statfs_waitq);
 806
 807         lov->lov_pools_hash_body = cfs_hash_create("POOLS", HASH_POOLS_CUR_BITS,
 808                                                    HASH_POOLS_MAX_BITS,
 809                                                    &pool_hash_operations, CFS_HASH_REHASH);
 810         CFS_INIT_LIST_HEAD(&lov->lov_pool_list);
 811         lov->lov_pool_count = 0;
 812         rc = lov_ost_pool_init(&lov->lov_packed, 0);
 813         if (rc)
 814                 RETURN(rc);
 815         rc = lov_ost_pool_init(&lov->lov_qos.lq_rr.lqr_pool, 0);
 816         if (rc) {
 817                 lov_ost_pool_free(&lov->lov_packed);
 818                 RETURN(rc);
 819         }
 820
 821         lprocfs_lov_init_vars(&lvars);
 822         lprocfs_obd_setup(obd, lvars.obd_vars);
 823 #ifdef LPROCFS
 824         {
 825                 int rc;
 826
 827                 rc = lprocfs_seq_create(obd->obd_proc_entry, "target_obd",
 828                                         0444, &lov_proc_target_fops, obd);
 829                 if (rc)
 830                         CWARN("Error adding the target_obd file\n");
 831         }
 832 #endif
 833         lov->lov_pool_proc_entry = lprocfs_register("pools",
 834                                                     obd->obd_proc_entry,
 835                                                     NULL, NULL);
 836
 837         RETURN(0);
 838 }
 839
 840 static int lov_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
 841 {
 842         int rc = 0;
 843         struct lov_obd *lov = &obd->u.lov;
 844
 845         ENTRY;
 846
 847         switch (stage) {
 848         case OBD_CLEANUP_EARLY: {
 849                 int i;
 850                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
 851                         if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_active)
 852                                 continue;
 853                         obd_precleanup(class_exp2obd(lov->lov_tgts[i]->ltd_exp),
 854                                        OBD_CLEANUP_EARLY);
 855                 }
 856                 break;
 857         }
 858         case OBD_CLEANUP_EXPORTS:
 859                 rc = obd_llog_finish(obd, 0);
 860                 if (rc != 0)
 861                         CERROR("failed to cleanup llogging subsystems\n");
 862                 break;
 863         }
 864         RETURN(rc);
 865 }
 866
 867 static int lov_cleanup(struct obd_device *obd)
 868 {
 869         struct lov_obd *lov = &obd->u.lov;
 870         cfs_list_t *pos, *tmp;
 871         struct pool_desc *pool;
 872
 873         cfs_list_for_each_safe(pos, tmp, &lov->lov_pool_list) {
 874                 pool = cfs_list_entry(pos, struct pool_desc, pool_list);
 875                 /* free pool structs */
 876                 CDEBUG(D_INFO, "delete pool %p\n", pool);
 877                 lov_pool_del(obd, pool->pool_name);
 878         }
 879         cfs_hash_putref(lov->lov_pools_hash_body);
 880         lov_ost_pool_free(&(lov->lov_qos.lq_rr.lqr_pool));
 881         lov_ost_pool_free(&lov->lov_packed);
 882
 883         if (lov->lov_tgts) {
 884                 int i;
 885                 obd_getref(obd);
 886                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
 887                         if (!lov->lov_tgts[i])
 888                                 continue;
 889
 890                         /* Inactive targets may never have connected */
 891                         if (lov->lov_tgts[i]->ltd_active ||
 892                             cfs_atomic_read(&lov->lov_refcount))
 893                             /* We should never get here - these
 894                                should have been removed in the
 895                              disconnect. */
 896                                 CERROR("lov tgt %d not cleaned!"
 897                                        " deathrow=%d, lovrc=%d\n",
 898                                        i, lov->lov_death_row,
 899                                        cfs_atomic_read(&lov->lov_refcount));
 900                         lov_del_target(obd, i, 0, 0);
 901                 }
 902                 obd_putref(obd);
 903                 OBD_FREE(lov->lov_tgts, sizeof(*lov->lov_tgts) *
 904                          lov->lov_tgt_size);
 905                 lov->lov_tgt_size = 0;
 906         }
 907
 908         /* clear pools parent proc entry only after all pools is killed */
 909         lprocfs_obd_cleanup(obd);
 910
 911         OBD_FREE_PTR(lov->lov_qos.lq_statfs_data);
 912         RETURN(0);
 913 }
 914
 915 int lov_process_config_base(struct obd_device *obd, struct lustre_cfg *lcfg,
 916                             __u32 *indexp, int *genp)
 917 {
 918         struct obd_uuid obd_uuid;
 919         int cmd;
 920         int rc = 0;
 921         ENTRY;
 922
 923         switch(cmd = lcfg->lcfg_command) {
 924         case LCFG_LOV_ADD_OBD:
 925         case LCFG_LOV_ADD_INA:
 926         case LCFG_LOV_DEL_OBD: {
 927                 __u32 index;
 928                 int gen;
 929                 /* lov_modify_tgts add  0:lov_mdsA  1:ost1_UUID  2:0  3:1 */
 930                 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
 931                         GOTO(out, rc = -EINVAL);
 932
 933                 obd_str2uuid(&obd_uuid,  lustre_cfg_buf(lcfg, 1));
 934
 935                 if (sscanf(lustre_cfg_buf(lcfg, 2), "%d", indexp) != 1)
 936                         GOTO(out, rc = -EINVAL);
 937                 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", genp) != 1)
 938                         GOTO(out, rc = -EINVAL);
 939                 index = *indexp;
 940                 gen = *genp;
 941                 if (cmd == LCFG_LOV_ADD_OBD)
 942                         rc = lov_add_target(obd, &obd_uuid, index, gen, 1);
 943                 else if (cmd == LCFG_LOV_ADD_INA)
 944                         rc = lov_add_target(obd, &obd_uuid, index, gen, 0);
 945                 else
 946                         rc = lov_del_target(obd, index, &obd_uuid, gen);
 947                 GOTO(out, rc);
 948         }
 949         case LCFG_PARAM: {
 950                 struct lprocfs_static_vars lvars = { 0 };
 951                 struct lov_desc *desc = &(obd->u.lov.desc);
 952
 953                 if (!desc)
 954                         GOTO(out, rc = -EINVAL);
 955
 956                 lprocfs_lov_init_vars(&lvars);
 957
 958                 rc = class_process_proc_param(PARAM_LOV, lvars.obd_vars,
 959                                               lcfg, obd);
 960                 if (rc > 0)
 961                         rc = 0;
 962                 GOTO(out, rc);
 963         }
 964         case LCFG_POOL_NEW:
 965         case LCFG_POOL_ADD:
 966         case LCFG_POOL_DEL:
 967         case LCFG_POOL_REM:
 968                 GOTO(out, rc);
 969
 970         default: {
 971                 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
 972                 GOTO(out, rc = -EINVAL);
 973
 974         }
 975         }
 976 out:
 977         RETURN(rc);
 978 }
 979
 980 #ifndef log2
 981 #define log2(n) cfs_ffz(~(n))
 982 #endif
 983
 984 static int lov_clear_orphans(struct obd_export *export, struct obdo *src_oa,
 985                              struct lov_stripe_md **ea,
 986                              struct obd_trans_info *oti)
 987 {
 988         struct lov_obd *lov;
 989         struct obdo *tmp_oa;
 990         struct obd_uuid *ost_uuid = NULL;
 991         int rc = 0, i;
 992         ENTRY;
 993
 994         LASSERT(src_oa->o_valid & OBD_MD_FLFLAGS &&
 995                 src_oa->o_flags == OBD_FL_DELORPHAN);
 996
 997         lov = &export->exp_obd->u.lov;
 998
 999         OBDO_ALLOC(tmp_oa);
1000         if (tmp_oa == NULL)
1001                 RETURN(-ENOMEM);
1002
1003         if (oti->oti_ost_uuid) {
1004                 ost_uuid = oti->oti_ost_uuid;
1005                 CDEBUG(D_HA, "clearing orphans only for %s\n",
1006                        ost_uuid->uuid);
1007         }
1008
1009         obd_getref(export->exp_obd);
1010         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
1011                 struct lov_stripe_md obj_md;
1012                 struct lov_stripe_md *obj_mdp = &obj_md;
1013                 struct lov_tgt_desc *tgt;
1014                 int err;
1015
1016                 tgt = lov->lov_tgts[i];
1017                 if (!tgt)
1018                         continue;
1019
1020                 /* if called for a specific target, we don't
1021                    care if it is not active. */
1022                 if (!lov->lov_tgts[i]->ltd_active && ost_uuid == NULL) {
1023                         CDEBUG(D_HA, "lov idx %d inactive\n", i);
1024                         continue;
1025                 }
1026
1027                 if (ost_uuid && !obd_uuid_equals(ost_uuid, &tgt->ltd_uuid))
1028                         continue;
1029
1030                 CDEBUG(D_CONFIG,"Clear orphans for %d:%s\n", i,
1031                        obd_uuid2str(ost_uuid));
1032
1033                 memcpy(tmp_oa, src_oa, sizeof(*tmp_oa));
1034
1035                 LASSERT(lov->lov_tgts[i]->ltd_exp);
1036                 /* XXX: LOV STACKING: use real "obj_mdp" sub-data */
1037                 err = obd_create(lov->lov_tgts[i]->ltd_exp,
1038                                  tmp_oa, &obj_mdp, oti);
1039                 if (err) {
1040                         /* This export will be disabled until it is recovered,
1041                            and then orphan recovery will be completed. */
1042                         CERROR("error in orphan recovery on OST idx %d/%d: "
1043                                "rc = %d\n", i, lov->desc.ld_tgt_count, err);
1044                         rc = err;
1045                 }
1046
1047                 if (ost_uuid)
1048                         break;
1049         }
1050         obd_putref(export->exp_obd);
1051
1052         OBDO_FREE(tmp_oa);
1053         RETURN(rc);
1054 }
1055
1056 static int lov_recreate(struct obd_export *exp, struct obdo *src_oa,
1057                         struct lov_stripe_md **ea, struct obd_trans_info *oti)
1058 {
1059         struct lov_stripe_md *obj_mdp, *lsm;
1060         struct lov_obd *lov = &exp->exp_obd->u.lov;
1061         unsigned ost_idx;
1062         int rc, i;
1063         ENTRY;
1064
1065         LASSERT(src_oa->o_valid & OBD_MD_FLFLAGS &&
1066                 src_oa->o_flags & OBD_FL_RECREATE_OBJS);
1067
1068         OBD_ALLOC(obj_mdp, sizeof(*obj_mdp));
1069         if (obj_mdp == NULL)
1070                 RETURN(-ENOMEM);
1071
1072         ost_idx = src_oa->o_nlink;
1073         lsm = *ea;
1074         if (lsm == NULL)
1075                 GOTO(out, rc = -EINVAL);
1076         if (ost_idx >= lov->desc.ld_tgt_count ||
1077             !lov->lov_tgts[ost_idx])
1078                 GOTO(out, rc = -EINVAL);
1079
1080         for (i = 0; i < lsm->lsm_stripe_count; i++) {
1081                 if (lsm->lsm_oinfo[i]->loi_ost_idx == ost_idx) {
1082                         if (lsm->lsm_oinfo[i]->loi_id != src_oa->o_id)
1083                                 GOTO(out, rc = -EINVAL);
1084                         break;
1085                 }
1086         }
1087         if (i == lsm->lsm_stripe_count)
1088                 GOTO(out, rc = -EINVAL);
1089
1090         rc = obd_create(lov->lov_tgts[ost_idx]->ltd_exp, src_oa, &obj_mdp, oti);
1091 out:
1092         OBD_FREE(obj_mdp, sizeof(*obj_mdp));
1093         RETURN(rc);
1094 }
1095
1096 /* the LOV expects oa->o_id to be set to the LOV object id */
1097 static int lov_create(struct obd_export *exp, struct obdo *src_oa,
1098                       struct lov_stripe_md **ea, struct obd_trans_info *oti)
1099 {
1100         struct lov_obd *lov;
1101         struct obd_info oinfo;
1102         struct lov_request_set *set = NULL;
1103         struct lov_request *req;
1104         struct l_wait_info  lwi = { 0 };
1105         int rc = 0;
1106         ENTRY;
1107
1108         LASSERT(ea != NULL);
1109         if (exp == NULL)
1110                 RETURN(-EINVAL);
1111
1112         if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
1113             src_oa->o_flags == OBD_FL_DELORPHAN) {
1114                 rc = lov_clear_orphans(exp, src_oa, ea, oti);
1115                 RETURN(rc);
1116         }
1117
1118         lov = &exp->exp_obd->u.lov;
1119         if (!lov->desc.ld_active_tgt_count)
1120                 RETURN(-EIO);
1121
1122         obd_getref(exp->exp_obd);
1123         /* Recreate a specific object id at the given OST index */
1124         if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
1125             (src_oa->o_flags & OBD_FL_RECREATE_OBJS)) {
1126                  rc = lov_recreate(exp, src_oa, ea, oti);
1127                  GOTO(out, rc);
1128         }
1129
1130         /* issue statfs rpcs if the osfs data is older than qos_maxage - 1s,
1131          * later in alloc_qos(), we will wait for those rpcs to complete if
1132          * the osfs age is older than 2 * qos_maxage */
1133         qos_statfs_update(exp->exp_obd,
1134                           cfs_time_shift_64(-lov->desc.ld_qos_maxage +
1135                                             OBD_STATFS_CACHE_SECONDS),
1136                           0);
1137
1138         rc = lov_prep_create_set(exp, &oinfo, ea, src_oa, oti, &set);
1139         if (rc)
1140                 GOTO(out, rc);
1141
1142         cfs_list_for_each_entry(req, &set->set_list, rq_link) {
1143                 /* XXX: LOV STACKING: use real "obj_mdp" sub-data */
1144                 rc = obd_create_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1145                                       &req->rq_oi, &req->rq_oi.oi_md, oti);
1146         }
1147
1148         /* osc_create have timeout equ obd_timeout/2 so waiting don't be
1149          * longer then this */
1150         l_wait_event(set->set_waitq, lov_finished_set(set), &lwi);
1151
1152         /* we not have ptlrpc set for assign set->interpret and should
1153          * be call interpret function himself. calling from cb_create_update
1154          * not permited because lov_fini_create_set can sleep for long time,
1155          * but we must avoid sleeping in ptlrpcd interpret function. */
1156         rc = lov_fini_create_set(set, ea);
1157 out:
1158         obd_putref(exp->exp_obd);
1159         RETURN(rc);
1160 }
1161
1162 #define ASSERT_LSM_MAGIC(lsmp)                                                  \
1163 do {                                                                            \
1164         LASSERT((lsmp) != NULL);                                                \
1165         LASSERTF(((lsmp)->lsm_magic == LOV_MAGIC_V1 ||                          \
1166                  (lsmp)->lsm_magic == LOV_MAGIC_V3),                            \
1167                  "%p->lsm_magic=%x\n", (lsmp), (lsmp)->lsm_magic);              \
1168 } while (0)
1169
1170 static int lov_destroy(struct obd_export *exp, struct obdo *oa,
1171                        struct lov_stripe_md *lsm, struct obd_trans_info *oti,
1172                        struct obd_export *md_exp, void *capa)
1173 {
1174         struct lov_request_set *set;
1175         struct obd_info oinfo;
1176         struct lov_request *req;
1177         cfs_list_t *pos;
1178         struct lov_obd *lov;
1179         int rc = 0, err = 0;
1180         ENTRY;
1181
1182         ASSERT_LSM_MAGIC(lsm);
1183
1184         if (!exp || !exp->exp_obd)
1185                 RETURN(-ENODEV);
1186
1187         if (oa->o_valid & OBD_MD_FLCOOKIE) {
1188                 LASSERT(oti);
1189                 LASSERT(oti->oti_logcookies);
1190         }
1191
1192         lov = &exp->exp_obd->u.lov;
1193         obd_getref(exp->exp_obd);
1194         rc = lov_prep_destroy_set(exp, &oinfo, oa, lsm, oti, &set);
1195         if (rc)
1196                 GOTO(out, rc);
1197
1198         cfs_list_for_each (pos, &set->set_list) {
1199                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1200
1201                 if (oa->o_valid & OBD_MD_FLCOOKIE)
1202                         oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1203
1204                 err = obd_destroy(lov->lov_tgts[req->rq_idx]->ltd_exp,
1205                                   req->rq_oi.oi_oa, NULL, oti, NULL, capa);
1206                 err = lov_update_common_set(set, req, err);
1207                 if (err) {
1208                         CERROR("error: destroying objid "LPX64" subobj "
1209                                LPX64" on OST idx %d: rc = %d\n",
1210                                oa->o_id, req->rq_oi.oi_oa->o_id,
1211                                req->rq_idx, err);
1212                         if (!rc)
1213                                 rc = err;
1214                 }
1215         }
1216
1217         if (rc == 0) {
1218                 LASSERT(lsm_op_find(lsm->lsm_magic) != NULL);
1219                 rc = lsm_op_find(lsm->lsm_magic)->lsm_destroy(lsm, oa, md_exp);
1220         }
1221         err = lov_fini_destroy_set(set);
1222 out:
1223         obd_putref(exp->exp_obd);
1224         RETURN(rc ? rc : err);
1225 }
1226
1227 static int lov_getattr(struct obd_export *exp, struct obd_info *oinfo)
1228 {
1229         struct lov_request_set *set;
1230         struct lov_request *req;
1231         cfs_list_t *pos;
1232         struct lov_obd *lov;
1233         int err = 0, rc = 0;
1234         ENTRY;
1235
1236         LASSERT(oinfo);
1237         ASSERT_LSM_MAGIC(oinfo->oi_md);
1238
1239         if (!exp || !exp->exp_obd)
1240                 RETURN(-ENODEV);
1241
1242         lov = &exp->exp_obd->u.lov;
1243
1244         rc = lov_prep_getattr_set(exp, oinfo, &set);
1245         if (rc)
1246                 RETURN(rc);
1247
1248         cfs_list_for_each (pos, &set->set_list) {
1249                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1250
1251                 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
1252                        "%u\n", oinfo->oi_oa->o_id, req->rq_stripe,
1253                        req->rq_oi.oi_oa->o_id, req->rq_idx);
1254
1255                 rc = obd_getattr(lov->lov_tgts[req->rq_idx]->ltd_exp,
1256                                  &req->rq_oi);
1257                 err = lov_update_common_set(set, req, rc);
1258                 if (err) {
1259                         CERROR("error: getattr objid "LPX64" subobj "
1260                                LPX64" on OST idx %d: rc = %d\n",
1261                                oinfo->oi_oa->o_id, req->rq_oi.oi_oa->o_id,
1262                                req->rq_idx, err);
1263                         break;
1264                 }
1265         }
1266
1267         rc = lov_fini_getattr_set(set);
1268         if (err)
1269                 rc = err;
1270         RETURN(rc);
1271 }
1272
1273 static int lov_getattr_interpret(struct ptlrpc_request_set *rqset,
1274                                  void *data, int rc)
1275 {
1276         struct lov_request_set *lovset = (struct lov_request_set *)data;
1277         int err;
1278         ENTRY;
1279
1280         /* don't do attribute merge if this aysnc op failed */
1281         if (rc)
1282                 lovset->set_completes = 0;
1283         err = lov_fini_getattr_set(lovset);
1284         RETURN(rc ? rc : err);
1285 }
1286
1287 static int lov_getattr_async(struct obd_export *exp, struct obd_info *oinfo,
1288                               struct ptlrpc_request_set *rqset)
1289 {
1290         struct lov_request_set *lovset;
1291         struct lov_obd *lov;
1292         cfs_list_t *pos;
1293         struct lov_request *req;
1294         int rc = 0, err;
1295         ENTRY;
1296
1297         LASSERT(oinfo);
1298         ASSERT_LSM_MAGIC(oinfo->oi_md);
1299
1300         if (!exp || !exp->exp_obd)
1301                 RETURN(-ENODEV);
1302
1303         lov = &exp->exp_obd->u.lov;
1304
1305         rc = lov_prep_getattr_set(exp, oinfo, &lovset);
1306         if (rc)
1307                 RETURN(rc);
1308
1309         CDEBUG(D_INFO, "objid "LPX64": %ux%u byte stripes\n",
1310                oinfo->oi_md->lsm_object_id, oinfo->oi_md->lsm_stripe_count,
1311                oinfo->oi_md->lsm_stripe_size);
1312
1313         cfs_list_for_each (pos, &lovset->set_list) {
1314                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1315
1316                 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
1317                        "%u\n", oinfo->oi_oa->o_id, req->rq_stripe,
1318                        req->rq_oi.oi_oa->o_id, req->rq_idx);
1319                 rc = obd_getattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1320                                        &req->rq_oi, rqset);
1321                 if (rc) {
1322                         CERROR("error: getattr objid "LPX64" subobj "
1323                                LPX64" on OST idx %d: rc = %d\n",
1324                                oinfo->oi_oa->o_id, req->rq_oi.oi_oa->o_id,
1325                                req->rq_idx, rc);
1326                         GOTO(out, rc);
1327                 }
1328         }
1329
1330         if (!cfs_list_empty(&rqset->set_requests)) {
1331                 LASSERT(rc == 0);
1332                 LASSERT (rqset->set_interpret == NULL);
1333                 rqset->set_interpret = lov_getattr_interpret;
1334                 rqset->set_arg = (void *)lovset;
1335                 RETURN(rc);
1336         }
1337 out:
1338         if (rc)
1339                 lovset->set_completes = 0;
1340         err = lov_fini_getattr_set(lovset);
1341         RETURN(rc ? rc : err);
1342 }
1343
1344 static int lov_setattr(struct obd_export *exp, struct obd_info *oinfo,
1345                        struct obd_trans_info *oti)
1346 {
1347         struct lov_request_set *set;
1348         struct lov_obd *lov;
1349         cfs_list_t *pos;
1350         struct lov_request *req;
1351         int err = 0, rc = 0;
1352         ENTRY;
1353
1354         LASSERT(oinfo);
1355         ASSERT_LSM_MAGIC(oinfo->oi_md);
1356
1357         if (!exp || !exp->exp_obd)
1358                 RETURN(-ENODEV);
1359
1360         /* for now, we only expect the following updates here */
1361         LASSERT(!(oinfo->oi_oa->o_valid & ~(OBD_MD_FLID | OBD_MD_FLTYPE |
1362                                             OBD_MD_FLMODE | OBD_MD_FLATIME |
1363                                             OBD_MD_FLMTIME | OBD_MD_FLCTIME |
1364                                             OBD_MD_FLFLAGS | OBD_MD_FLSIZE |
1365                                             OBD_MD_FLGROUP | OBD_MD_FLUID |
1366                                             OBD_MD_FLGID | OBD_MD_FLFID |
1367                                             OBD_MD_FLGENER)));
1368         lov = &exp->exp_obd->u.lov;
1369         rc = lov_prep_setattr_set(exp, oinfo, oti, &set);
1370         if (rc)
1371                 RETURN(rc);
1372
1373         cfs_list_for_each (pos, &set->set_list) {
1374                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1375
1376                 rc = obd_setattr(lov->lov_tgts[req->rq_idx]->ltd_exp,
1377                                  &req->rq_oi, NULL);
1378                 err = lov_update_setattr_set(set, req, rc);
1379                 if (err) {
1380                         CERROR("error: setattr objid "LPX64" subobj "
1381                                LPX64" on OST idx %d: rc = %d\n",
1382                                set->set_oi->oi_oa->o_id,
1383                                req->rq_oi.oi_oa->o_id, req->rq_idx, err);
1384                         if (!rc)
1385                                 rc = err;
1386                 }
1387         }
1388         err = lov_fini_setattr_set(set);
1389         if (!rc)
1390                 rc = err;
1391         RETURN(rc);
1392 }
1393
1394 static int lov_setattr_interpret(struct ptlrpc_request_set *rqset,
1395                                  void *data, int rc)
1396 {
1397         struct lov_request_set *lovset = (struct lov_request_set *)data;
1398         int err;
1399         ENTRY;
1400
1401         if (rc)
1402                 lovset->set_completes = 0;
1403         err = lov_fini_setattr_set(lovset);
1404         RETURN(rc ? rc : err);
1405 }
1406
1407 /* If @oti is given, the request goes from MDS and responses from OSTs are not
1408    needed. Otherwise, a client is waiting for responses. */
1409 static int lov_setattr_async(struct obd_export *exp, struct obd_info *oinfo,
1410                              struct obd_trans_info *oti,
1411                              struct ptlrpc_request_set *rqset)
1412 {
1413         struct lov_request_set *set;
1414         struct lov_request *req;
1415         cfs_list_t *pos;
1416         struct lov_obd *lov;
1417         int rc = 0;
1418         ENTRY;
1419
1420         LASSERT(oinfo);
1421         ASSERT_LSM_MAGIC(oinfo->oi_md);
1422         if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE) {
1423                 LASSERT(oti);
1424                 LASSERT(oti->oti_logcookies);
1425         }
1426
1427         if (!exp || !exp->exp_obd)
1428                 RETURN(-ENODEV);
1429
1430         lov = &exp->exp_obd->u.lov;
1431         rc = lov_prep_setattr_set(exp, oinfo, oti, &set);
1432         if (rc)
1433                 RETURN(rc);
1434
1435         CDEBUG(D_INFO, "objid "LPX64": %ux%u byte stripes\n",
1436                oinfo->oi_md->lsm_object_id, oinfo->oi_md->lsm_stripe_count,
1437                oinfo->oi_md->lsm_stripe_size);
1438
1439         cfs_list_for_each (pos, &set->set_list) {
1440                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1441
1442                 if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE)
1443                         oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1444
1445                 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
1446                        "%u\n", oinfo->oi_oa->o_id, req->rq_stripe,
1447                        req->rq_oi.oi_oa->o_id, req->rq_idx);
1448
1449                 rc = obd_setattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1450                                        &req->rq_oi, oti, rqset);
1451                 if (rc) {
1452                         CERROR("error: setattr objid "LPX64" subobj "
1453                                LPX64" on OST idx %d: rc = %d\n",
1454                                set->set_oi->oi_oa->o_id,
1455                                req->rq_oi.oi_oa->o_id,
1456                                req->rq_idx, rc);
1457                         break;
1458                 }
1459         }
1460
1461         /* If we are not waiting for responses on async requests, return. */
1462         if (rc || !rqset || cfs_list_empty(&rqset->set_requests)) {
1463                 int err;
1464                 if (rc)
1465                         set->set_completes = 0;
1466                 err = lov_fini_setattr_set(set);
1467                 RETURN(rc ? rc : err);
1468         }
1469
1470         LASSERT(rqset->set_interpret == NULL);
1471         rqset->set_interpret = lov_setattr_interpret;
1472         rqset->set_arg = (void *)set;
1473
1474         RETURN(0);
1475 }
1476
1477 static int lov_punch_interpret(struct ptlrpc_request_set *rqset,
1478                                void *data, int rc)
1479 {
1480         struct lov_request_set *lovset = (struct lov_request_set *)data;
1481         int err;
1482         ENTRY;
1483
1484         if (rc)
1485                 lovset->set_completes = 0;
1486         err = lov_fini_punch_set(lovset);
1487         RETURN(rc ? rc : err);
1488 }
1489
1490 /* FIXME: maybe we'll just make one node the authoritative attribute node, then
1491  * we can send this 'punch' to just the authoritative node and the nodes
1492  * that the punch will affect. */
1493 static int lov_punch(struct obd_export *exp, struct obd_info *oinfo,
1494                      struct obd_trans_info *oti,
1495                      struct ptlrpc_request_set *rqset)
1496 {
1497         struct lov_request_set *set;
1498         struct lov_obd *lov;
1499         cfs_list_t *pos;
1500         struct lov_request *req;
1501         int rc = 0;
1502         ENTRY;
1503
1504         LASSERT(oinfo);
1505         ASSERT_LSM_MAGIC(oinfo->oi_md);
1506
1507         if (!exp || !exp->exp_obd)
1508                 RETURN(-ENODEV);
1509
1510         lov = &exp->exp_obd->u.lov;
1511         rc = lov_prep_punch_set(exp, oinfo, oti, &set);
1512         if (rc)
1513                 RETURN(rc);
1514
1515         cfs_list_for_each (pos, &set->set_list) {
1516                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1517
1518                 rc = obd_punch(lov->lov_tgts[req->rq_idx]->ltd_exp,
1519                                &req->rq_oi, NULL, rqset);
1520                 if (rc) {
1521                         CERROR("error: punch objid "LPX64" subobj "LPX64
1522                                " on OST idx %d: rc = %d\n",
1523                                set->set_oi->oi_oa->o_id,
1524                                req->rq_oi.oi_oa->o_id, req->rq_idx, rc);
1525                         break;
1526                 }
1527         }
1528
1529         if (rc || cfs_list_empty(&rqset->set_requests)) {
1530                 int err;
1531                 err = lov_fini_punch_set(set);
1532                 RETURN(rc ? rc : err);
1533         }
1534
1535         LASSERT(rqset->set_interpret == NULL);
1536         rqset->set_interpret = lov_punch_interpret;
1537         rqset->set_arg = (void *)set;
1538
1539         RETURN(0);
1540 }
1541
1542 static int lov_sync(struct obd_export *exp, struct obdo *oa,
1543                     struct lov_stripe_md *lsm, obd_off start, obd_off end,
1544                     void *capa)
1545 {
1546         struct lov_request_set *set;
1547         struct obd_info oinfo;
1548         struct lov_obd *lov;
1549         cfs_list_t *pos;
1550         struct lov_request *req;
1551         int err = 0, rc = 0;
1552         ENTRY;
1553
1554         ASSERT_LSM_MAGIC(lsm);
1555
1556         if (!exp->exp_obd)
1557                 RETURN(-ENODEV);
1558
1559         lov = &exp->exp_obd->u.lov;
1560         rc = lov_prep_sync_set(exp, &oinfo, oa, lsm, start, end, &set);
1561         if (rc)
1562                 RETURN(rc);
1563
1564         cfs_list_for_each (pos, &set->set_list) {
1565                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1566
1567                 rc = obd_sync(lov->lov_tgts[req->rq_idx]->ltd_exp,
1568                               req->rq_oi.oi_oa, NULL,
1569                               req->rq_oi.oi_policy.l_extent.start,
1570                               req->rq_oi.oi_policy.l_extent.end, capa);
1571                 err = lov_update_common_set(set, req, rc);
1572                 if (err) {
1573                         CERROR("error: fsync objid "LPX64" subobj "LPX64
1574                                " on OST idx %d: rc = %d\n",
1575                                set->set_oi->oi_oa->o_id,
1576                                req->rq_oi.oi_oa->o_id, req->rq_idx, rc);
1577                         if (!rc)
1578                                 rc = err;
1579                 }
1580         }
1581         err = lov_fini_sync_set(set);
1582         if (!rc)
1583                 rc = err;
1584         RETURN(rc);
1585 }
1586
1587 static int lov_brw_check(struct lov_obd *lov, struct obd_info *lov_oinfo,
1588                          obd_count oa_bufs, struct brw_page *pga)
1589 {
1590         struct obd_info oinfo = { { { 0 } } };
1591         int i, rc = 0;
1592
1593         oinfo.oi_oa = lov_oinfo->oi_oa;
1594
1595         /* The caller just wants to know if there's a chance that this
1596          * I/O can succeed */
1597         for (i = 0; i < oa_bufs; i++) {
1598                 int stripe = lov_stripe_number(lov_oinfo->oi_md, pga[i].off);
1599                 int ost = lov_oinfo->oi_md->lsm_oinfo[stripe]->loi_ost_idx;
1600                 obd_off start, end;
1601
1602                 if (!lov_stripe_intersects(lov_oinfo->oi_md, i, pga[i].off,
1603                                            pga[i].off + pga[i].count - 1,
1604                                            &start, &end))
1605                         continue;
1606
1607                 if (!lov->lov_tgts[ost] || !lov->lov_tgts[ost]->ltd_active) {
1608                         CDEBUG(D_HA, "lov idx %d inactive\n", ost);
1609                         return -EIO;
1610                 }
1611
1612                 rc = obd_brw(OBD_BRW_CHECK, lov->lov_tgts[ost]->ltd_exp, &oinfo,
1613                              1, &pga[i], NULL);
1614                 if (rc)
1615                         break;
1616         }
1617         return rc;
1618 }
1619
1620 static int lov_brw(int cmd, struct obd_export *exp, struct obd_info *oinfo,
1621                    obd_count oa_bufs, struct brw_page *pga,
1622                    struct obd_trans_info *oti)
1623 {
1624         struct lov_request_set *set;
1625         struct lov_request *req;
1626         cfs_list_t *pos;
1627         struct lov_obd *lov = &exp->exp_obd->u.lov;
1628         int err, rc = 0;
1629         ENTRY;
1630
1631         ASSERT_LSM_MAGIC(oinfo->oi_md);
1632
1633         if (cmd == OBD_BRW_CHECK) {
1634                 rc = lov_brw_check(lov, oinfo, oa_bufs, pga);
1635                 RETURN(rc);
1636         }
1637
1638         rc = lov_prep_brw_set(exp, oinfo, oa_bufs, pga, oti, &set);
1639         if (rc)
1640                 RETURN(rc);
1641
1642         cfs_list_for_each (pos, &set->set_list) {
1643                 struct obd_export *sub_exp;
1644                 struct brw_page *sub_pga;
1645                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1646
1647                 sub_exp = lov->lov_tgts[req->rq_idx]->ltd_exp;
1648                 sub_pga = set->set_pga + req->rq_pgaidx;
1649                 rc = obd_brw(cmd, sub_exp, &req->rq_oi, req->rq_oabufs,
1650                              sub_pga, oti);
1651                 if (rc)
1652                         break;
1653                 lov_update_common_set(set, req, rc);
1654         }
1655
1656         err = lov_fini_brw_set(set);
1657         if (!rc)
1658                 rc = err;
1659         RETURN(rc);
1660 }
1661
1662 static int lov_enqueue_interpret(struct ptlrpc_request_set *rqset,
1663                                  void *data, int rc)
1664 {
1665         struct lov_request_set *lovset = (struct lov_request_set *)data;
1666         ENTRY;
1667         rc = lov_fini_enqueue_set(lovset, lovset->set_ei->ei_mode, rc, rqset);
1668         RETURN(rc);
1669 }
1670
1671 static int lov_enqueue(struct obd_export *exp, struct obd_info *oinfo,
1672                        struct ldlm_enqueue_info *einfo,
1673                        struct ptlrpc_request_set *rqset)
1674 {
1675         ldlm_mode_t mode = einfo->ei_mode;
1676         struct lov_request_set *set;
1677         struct lov_request *req;
1678         cfs_list_t *pos;
1679         struct lov_obd *lov;
1680         ldlm_error_t rc;
1681         ENTRY;
1682
1683         LASSERT(oinfo);
1684         ASSERT_LSM_MAGIC(oinfo->oi_md);
1685         LASSERT(mode == (mode & -mode));
1686
1687         /* we should never be asked to replay a lock this way. */
1688         LASSERT((oinfo->oi_flags & LDLM_FL_REPLAY) == 0);
1689
1690         if (!exp || !exp->exp_obd)
1691                 RETURN(-ENODEV);
1692
1693         lov = &exp->exp_obd->u.lov;
1694         rc = lov_prep_enqueue_set(exp, oinfo, einfo, &set);
1695         if (rc)
1696                 RETURN(rc);
1697
1698         cfs_list_for_each (pos, &set->set_list) {
1699                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1700
1701                 rc = obd_enqueue(lov->lov_tgts[req->rq_idx]->ltd_exp,
1702                                  &req->rq_oi, einfo, rqset);
1703                 if (rc != ELDLM_OK)
1704                         GOTO(out, rc);
1705         }
1706
1707         if (rqset && !cfs_list_empty(&rqset->set_requests)) {
1708                 LASSERT(rc == 0);
1709                 LASSERT(rqset->set_interpret == NULL);
1710                 rqset->set_interpret = lov_enqueue_interpret;
1711                 rqset->set_arg = (void *)set;
1712                 RETURN(rc);
1713         }
1714 out:
1715         rc = lov_fini_enqueue_set(set, mode, rc, rqset);
1716         RETURN(rc);
1717 }
1718
1719 static int lov_change_cbdata(struct obd_export *exp,
1720                              struct lov_stripe_md *lsm, ldlm_iterator_t it,
1721                              void *data)
1722 {
1723         struct lov_obd *lov;
1724         int rc = 0, i;
1725         ENTRY;
1726
1727         ASSERT_LSM_MAGIC(lsm);
1728
1729         if (!exp || !exp->exp_obd)
1730                 RETURN(-ENODEV);
1731
1732         lov = &exp->exp_obd->u.lov;
1733         for (i = 0; i < lsm->lsm_stripe_count; i++) {
1734                 struct lov_stripe_md submd;
1735                 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
1736
1737                 if (!lov->lov_tgts[loi->loi_ost_idx]) {
1738                         CDEBUG(D_HA, "lov idx %d NULL \n", loi->loi_ost_idx);
1739                         continue;
1740                 }
1741
1742                 LASSERT_SEQ_IS_MDT(loi->loi_seq);
1743                 submd.lsm_object_id = loi->loi_id;
1744                 submd.lsm_object_seq = loi->loi_seq;
1745                 submd.lsm_stripe_count = 0;
1746                 rc = obd_change_cbdata(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1747                                        &submd, it, data);
1748         }
1749         RETURN(rc);
1750 }
1751
1752 /* find any ldlm lock of the inode in lov
1753  * return 0    not find
1754  *        1    find one
1755  *      < 0    error */
1756 static int lov_find_cbdata(struct obd_export *exp,
1757                            struct lov_stripe_md *lsm, ldlm_iterator_t it,
1758                            void *data)
1759 {
1760         struct lov_obd *lov;
1761         int rc = 0, i;
1762         ENTRY;
1763
1764         ASSERT_LSM_MAGIC(lsm);
1765
1766         if (!exp || !exp->exp_obd)
1767                 RETURN(-ENODEV);
1768
1769         lov = &exp->exp_obd->u.lov;
1770         for (i = 0; i < lsm->lsm_stripe_count; i++) {
1771                 struct lov_stripe_md submd;
1772                 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
1773
1774                 if (!lov->lov_tgts[loi->loi_ost_idx]) {
1775                         CDEBUG(D_HA, "lov idx %d NULL \n", loi->loi_ost_idx);
1776                         continue;
1777                 }
1778
1779                 LASSERT_SEQ_IS_MDT(loi->loi_seq);
1780                 submd.lsm_object_id = loi->loi_id;
1781                 submd.lsm_object_seq = loi->loi_seq;
1782                 submd.lsm_stripe_count = 0;
1783                 rc = obd_find_cbdata(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1784                                      &submd, it, data);
1785                 if (rc != 0)
1786                         RETURN(rc);
1787         }
1788         RETURN(rc);
1789 }
1790
1791 static int lov_cancel(struct obd_export *exp, struct lov_stripe_md *lsm,
1792                       __u32 mode, struct lustre_handle *lockh)
1793 {
1794         struct lov_request_set *set;
1795         struct obd_info oinfo;
1796         struct lov_request *req;
1797         cfs_list_t *pos;
1798         struct lov_obd *lov;
1799         struct lustre_handle *lov_lockhp;
1800         int err = 0, rc = 0;
1801         ENTRY;
1802
1803         ASSERT_LSM_MAGIC(lsm);
1804
1805         if (!exp || !exp->exp_obd)
1806                 RETURN(-ENODEV);
1807
1808         LASSERT_SEQ_IS_MDT(lsm->lsm_object_seq);
1809         LASSERT(lockh);
1810         lov = &exp->exp_obd->u.lov;
1811         rc = lov_prep_cancel_set(exp, &oinfo, lsm, mode, lockh, &set);
1812         if (rc)
1813                 RETURN(rc);
1814
1815         cfs_list_for_each (pos, &set->set_list) {
1816                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1817                 lov_lockhp = set->set_lockh->llh_handles + req->rq_stripe;
1818
1819                 rc = obd_cancel(lov->lov_tgts[req->rq_idx]->ltd_exp,
1820                                 req->rq_oi.oi_md, mode, lov_lockhp);
1821                 rc = lov_update_common_set(set, req, rc);
1822                 if (rc) {
1823                         CERROR("error: cancel objid "LPX64" subobj "
1824                                LPX64" on OST idx %d: rc = %d\n",
1825                                lsm->lsm_object_id,
1826                                req->rq_oi.oi_md->lsm_object_id,
1827                                req->rq_idx, rc);
1828                         err = rc;
1829                 }
1830
1831         }
1832         lov_fini_cancel_set(set);
1833         RETURN(err);
1834 }
1835
1836 static int lov_cancel_unused(struct obd_export *exp,
1837                              struct lov_stripe_md *lsm,
1838                              ldlm_cancel_flags_t flags, void *opaque)
1839 {
1840         struct lov_obd *lov;
1841         int rc = 0, i;
1842         ENTRY;
1843
1844         if (!exp || !exp->exp_obd)
1845                 RETURN(-ENODEV);
1846
1847         lov = &exp->exp_obd->u.lov;
1848         if (lsm == NULL) {
1849                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
1850                         int err;
1851                         if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
1852                                 continue;
1853
1854                         err = obd_cancel_unused(lov->lov_tgts[i]->ltd_exp, NULL,
1855                                                 flags, opaque);
1856                         if (!rc)
1857                                 rc = err;
1858                 }
1859                 RETURN(rc);
1860         }
1861
1862         ASSERT_LSM_MAGIC(lsm);
1863
1864         LASSERT_SEQ_IS_MDT(lsm->lsm_object_seq);
1865         for (i = 0; i < lsm->lsm_stripe_count; i++) {
1866                 struct lov_stripe_md submd;
1867                 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
1868                 int err;
1869
1870                 if (!lov->lov_tgts[loi->loi_ost_idx]) {
1871                         CDEBUG(D_HA, "lov idx %d NULL\n", loi->loi_ost_idx);
1872                         continue;
1873                 }
1874
1875                 if (!lov->lov_tgts[loi->loi_ost_idx]->ltd_active)
1876                         CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
1877
1878                 submd.lsm_object_id = loi->loi_id;
1879                 submd.lsm_object_seq = loi->loi_seq;
1880                 submd.lsm_stripe_count = 0;
1881                 err = obd_cancel_unused(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1882                                         &submd, flags, opaque);
1883                 if (err && lov->lov_tgts[loi->loi_ost_idx]->ltd_active) {
1884                         CERROR("error: cancel unused objid "LPX64" subobj "LPX64
1885                                " on OST idx %d: rc = %d\n", lsm->lsm_object_id,
1886                                loi->loi_id, loi->loi_ost_idx, err);
1887                         if (!rc)
1888                                 rc = err;
1889                 }
1890         }
1891         RETURN(rc);
1892 }
1893
1894 int lov_statfs_interpret(struct ptlrpc_request_set *rqset, void *data, int rc)
1895 {
1896         struct lov_request_set *lovset = (struct lov_request_set *)data;
1897         int err;
1898         ENTRY;
1899
1900         if (rc)
1901                 lovset->set_completes = 0;
1902
1903         err = lov_fini_statfs_set(lovset);
1904         RETURN(rc ? rc : err);
1905 }
1906
1907 static int lov_statfs_async(struct obd_device *obd, struct obd_info *oinfo,
1908                             __u64 max_age, struct ptlrpc_request_set *rqset)
1909 {
1910         struct lov_request_set *set;
1911         struct lov_request *req;
1912         cfs_list_t *pos;
1913         struct lov_obd *lov;
1914         int rc = 0;
1915         ENTRY;
1916
1917         LASSERT(oinfo != NULL);
1918         LASSERT(oinfo->oi_osfs != NULL);
1919
1920         lov = &obd->u.lov;
1921         rc = lov_prep_statfs_set(obd, oinfo, &set);
1922         if (rc)
1923                 RETURN(rc);
1924
1925         cfs_list_for_each (pos, &set->set_list) {
1926                 struct obd_device *osc_obd;
1927
1928                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1929
1930                 osc_obd = class_exp2obd(lov->lov_tgts[req->rq_idx]->ltd_exp);
1931                 rc = obd_statfs_async(osc_obd, &req->rq_oi, max_age, rqset);
1932                 if (rc)
1933                         break;
1934         }
1935
1936         if (rc || cfs_list_empty(&rqset->set_requests)) {
1937                 int err;
1938                 if (rc)
1939                         set->set_completes = 0;
1940                 err = lov_fini_statfs_set(set);
1941                 RETURN(rc ? rc : err);
1942         }
1943
1944         LASSERT(rqset->set_interpret == NULL);
1945         rqset->set_interpret = lov_statfs_interpret;
1946         rqset->set_arg = (void *)set;
1947         RETURN(0);
1948 }
1949
1950 static int lov_statfs(struct obd_device *obd, struct obd_statfs *osfs,
1951                       __u64 max_age, __u32 flags)
1952 {
1953         struct ptlrpc_request_set *set = NULL;
1954         struct obd_info oinfo = { { { 0 } } };
1955         int rc = 0;
1956         ENTRY;
1957
1958
1959         /* for obdclass we forbid using obd_statfs_rqset, but prefer using async
1960          * statfs requests */
1961         set = ptlrpc_prep_set();
1962         if (set == NULL)
1963                 RETURN(-ENOMEM);
1964
1965         oinfo.oi_osfs = osfs;
1966         oinfo.oi_flags = flags;
1967         rc = lov_statfs_async(obd, &oinfo, max_age, set);
1968         if (rc == 0)
1969                 rc = ptlrpc_set_wait(set);
1970         ptlrpc_set_destroy(set);
1971
1972         RETURN(rc);
1973 }
1974
1975 static int lov_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1976                          void *karg, void *uarg)
1977 {
1978         struct obd_device *obddev = class_exp2obd(exp);
1979         struct lov_obd *lov = &obddev->u.lov;
1980         int i = 0, rc = 0, count = lov->desc.ld_tgt_count;
1981         struct obd_uuid *uuidp;
1982         ENTRY;
1983
1984         switch (cmd) {
1985         case IOC_OBD_STATFS: {
1986                 struct obd_ioctl_data *data = karg;
1987                 struct obd_device *osc_obd;
1988                 struct obd_statfs stat_buf = {0};
1989                 __u32 index;
1990
1991                 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
1992                 if ((index >= count))
1993                         RETURN(-ENODEV);
1994
1995                 if (!lov->lov_tgts[index])
1996                         /* Try again with the next index */
1997                         RETURN(-EAGAIN);
1998                 if (!lov->lov_tgts[index]->ltd_active)
1999                         RETURN(-ENODATA);
2000
2001                 osc_obd = class_exp2obd(lov->lov_tgts[index]->ltd_exp);
2002                 if (!osc_obd)
2003                         RETURN(-EINVAL);
2004
2005                 /* copy UUID */
2006                 if (cfs_copy_to_user(data->ioc_pbuf2, obd2cli_tgt(osc_obd),
2007                                      min((int) data->ioc_plen2,
2008                                          (int) sizeof(struct obd_uuid))))
2009                         RETURN(-EFAULT);
2010
2011                 /* got statfs data */
2012                 rc = obd_statfs(osc_obd, &stat_buf,
2013                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
2014                                 0);
2015                 if (rc)
2016                         RETURN(rc);
2017                 if (cfs_copy_to_user(data->ioc_pbuf1, &stat_buf,
2018                                      min((int) data->ioc_plen1,
2019                                          (int) sizeof(stat_buf))))
2020                         RETURN(-EFAULT);
2021                 break;
2022         }
2023         case OBD_IOC_LOV_GET_CONFIG: {
2024                 struct obd_ioctl_data *data;
2025                 struct lov_desc *desc;
2026                 char *buf = NULL;
2027                 __u32 *genp;
2028
2029                 len = 0;
2030                 if (obd_ioctl_getdata(&buf, &len, (void *)uarg))
2031                         RETURN(-EINVAL);
2032
2033                 data = (struct obd_ioctl_data *)buf;
2034
2035                 if (sizeof(*desc) > data->ioc_inllen1) {
2036                         obd_ioctl_freedata(buf, len);
2037                         RETURN(-EINVAL);
2038                 }
2039
2040                 if (sizeof(uuidp->uuid) * count > data->ioc_inllen2) {
2041                         obd_ioctl_freedata(buf, len);
2042                         RETURN(-EINVAL);
2043                 }
2044
2045                 if (sizeof(__u32) * count > data->ioc_inllen3) {
2046                         obd_ioctl_freedata(buf, len);
2047                         RETURN(-EINVAL);
2048                 }
2049
2050                 desc = (struct lov_desc *)data->ioc_inlbuf1;
2051                 memcpy(desc, &(lov->desc), sizeof(*desc));
2052
2053                 uuidp = (struct obd_uuid *)data->ioc_inlbuf2;
2054                 genp = (__u32 *)data->ioc_inlbuf3;
2055                 /* the uuid will be empty for deleted OSTs */
2056                 for (i = 0; i < count; i++, uuidp++, genp++) {
2057                         if (!lov->lov_tgts[i])
2058                                 continue;
2059                         *uuidp = lov->lov_tgts[i]->ltd_uuid;
2060                         *genp = lov->lov_tgts[i]->ltd_gen;
2061                 }
2062
2063                 if (cfs_copy_to_user((void *)uarg, buf, len))
2064                         rc = -EFAULT;
2065                 obd_ioctl_freedata(buf, len);
2066                 break;
2067         }
2068         case LL_IOC_LOV_SETSTRIPE:
2069                 rc = lov_setstripe(exp, len, karg, uarg);
2070                 break;
2071         case LL_IOC_LOV_GETSTRIPE:
2072                 rc = lov_getstripe(exp, karg, uarg);
2073                 break;
2074         case LL_IOC_LOV_SETEA:
2075                 rc = lov_setea(exp, karg, uarg);
2076                 break;
2077         case OBD_IOC_QUOTACTL: {
2078                 struct if_quotactl *qctl = karg;
2079                 struct lov_tgt_desc *tgt = NULL;
2080                 struct obd_quotactl *oqctl;
2081
2082                 if (qctl->qc_valid == QC_OSTIDX) {
2083                         if (qctl->qc_idx < 0 || count <= qctl->qc_idx)
2084                                 RETURN(-EINVAL);
2085
2086                         tgt = lov->lov_tgts[qctl->qc_idx];
2087                         if (!tgt || !tgt->ltd_exp)
2088                                 RETURN(-EINVAL);
2089                 } else if (qctl->qc_valid == QC_UUID) {
2090                         for (i = 0; i < count; i++) {
2091                                 tgt = lov->lov_tgts[i];
2092                                 if (!tgt ||
2093                                     !obd_uuid_equals(&tgt->ltd_uuid,
2094                                                      &qctl->obd_uuid))
2095                                         continue;
2096
2097                                 if (tgt->ltd_exp == NULL)
2098                                         RETURN(-EINVAL);
2099
2100                                 break;
2101                         }
2102                 } else {
2103                         RETURN(-EINVAL);
2104                 }
2105
2106                 if (i >= count)
2107                         RETURN(-EAGAIN);
2108
2109                 LASSERT(tgt && tgt->ltd_exp);
2110                 OBD_ALLOC_PTR(oqctl);
2111                 if (!oqctl)
2112                         RETURN(-ENOMEM);
2113
2114                 QCTL_COPY(oqctl, qctl);
2115                 rc = obd_quotactl(tgt->ltd_exp, oqctl);
2116                 if (rc == 0) {
2117                         QCTL_COPY(qctl, oqctl);
2118                         qctl->qc_valid = QC_OSTIDX;
2119                         qctl->obd_uuid = tgt->ltd_uuid;
2120                 }
2121                 OBD_FREE_PTR(oqctl);
2122                 break;
2123         }
2124         default: {
2125                 int set = 0;
2126
2127                 if (count == 0)
2128                         RETURN(-ENOTTY);
2129
2130                 for (i = 0; i < count; i++) {
2131                         int err;
2132
2133                         /* OST was disconnected */
2134                         if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
2135                                 continue;
2136
2137                         err = obd_iocontrol(cmd, lov->lov_tgts[i]->ltd_exp,
2138                                             len, karg, uarg);
2139                         if (err == -ENODATA && cmd == OBD_IOC_POLL_QUOTACHECK) {
2140                                 RETURN(err);
2141                         } else if (err) {
2142                                 if (lov->lov_tgts[i]->ltd_active) {
2143                                         CDEBUG(err == -ENOTTY ?
2144                                                D_IOCTL : D_WARNING,
2145                                                "iocontrol OSC %s on OST "
2146                                                "idx %d cmd %x: err = %d\n",
2147                                                lov_uuid2str(lov, i),
2148                                                i, cmd, err);
2149                                         if (!rc)
2150                                                 rc = err;
2151                                 }
2152                         } else {
2153                                 set = 1;
2154                         }
2155                 }
2156                 if (!set && !rc)
2157                         rc = -EIO;
2158         }
2159         }
2160
2161         RETURN(rc);
2162 }
2163
2164 #define FIEMAP_BUFFER_SIZE 4096
2165
2166 /**
2167  * Non-zero fe_logical indicates that this is a continuation FIEMAP
2168  * call. The local end offset and the device are sent in the first
2169  * fm_extent. This function calculates the stripe number from the index.
2170  * This function returns a stripe_no on which mapping is to be restarted.
2171  *
2172  * This function returns fm_end_offset which is the in-OST offset at which
2173  * mapping should be restarted. If fm_end_offset=0 is returned then caller
2174  * will re-calculate proper offset in next stripe.
2175  * Note that the first extent is passed to lov_get_info via the value field.
2176  *
2177  * \param fiemap fiemap request header
2178  * \param lsm striping information for the file
2179  * \param fm_start logical start of mapping
2180  * \param fm_end logical end of mapping
2181  * \param start_stripe starting stripe will be returned in this
2182  */
2183 obd_size fiemap_calc_fm_end_offset(struct ll_user_fiemap *fiemap,
2184                                    struct lov_stripe_md *lsm, obd_size fm_start,
2185                                    obd_size fm_end, int *start_stripe)
2186 {
2187         obd_size local_end = fiemap->fm_extents[0].fe_logical;
2188         obd_off lun_start, lun_end;
2189         obd_size fm_end_offset;
2190         int stripe_no = -1, i;
2191
2192         if (fiemap->fm_extent_count == 0 ||
2193             fiemap->fm_extents[0].fe_logical == 0)
2194                 return 0;
2195
2196         /* Find out stripe_no from ost_index saved in the fe_device */
2197         for (i = 0; i < lsm->lsm_stripe_count; i++) {
2198                 if (lsm->lsm_oinfo[i]->loi_ost_idx ==
2199                                         fiemap->fm_extents[0].fe_device) {
2200                         stripe_no = i;
2201                         break;
2202                 }
2203         }
2204
2205         /* If we have finished mapping on previous device, shift logical
2206          * offset to start of next device */
2207         if ((lov_stripe_intersects(lsm, stripe_no, fm_start, fm_end,
2208                                    &lun_start, &lun_end)) != 0 &&
2209                                    local_end < lun_end) {
2210                 fm_end_offset = local_end;
2211                 *start_stripe = stripe_no;
2212         } else {
2213                 /* This is a special value to indicate that caller should
2214                  * calculate offset in next stripe. */
2215                 fm_end_offset = 0;
2216                 *start_stripe = (stripe_no + 1) % lsm->lsm_stripe_count;
2217         }
2218
2219         return fm_end_offset;
2220 }
2221
2222 /**
2223  * We calculate on which OST the mapping will end. If the length of mapping
2224  * is greater than (stripe_size * stripe_count) then the last_stripe will
2225  * will be one just before start_stripe. Else we check if the mapping
2226  * intersects each OST and find last_stripe.
2227  * This function returns the last_stripe and also sets the stripe_count
2228  * over which the mapping is spread
2229  *
2230  * \param lsm striping information for the file
2231  * \param fm_start logical start of mapping
2232  * \param fm_end logical end of mapping
2233  * \param start_stripe starting stripe of the mapping
2234  * \param stripe_count the number of stripes across which to map is returned
2235  *
2236  * \retval last_stripe return the last stripe of the mapping
2237  */
2238 int fiemap_calc_last_stripe(struct lov_stripe_md *lsm, obd_size fm_start,
2239                             obd_size fm_end, int start_stripe,
2240                             int *stripe_count)
2241 {
2242         int last_stripe;
2243         obd_off obd_start, obd_end;
2244         int i, j;
2245
2246         if (fm_end - fm_start > lsm->lsm_stripe_size * lsm->lsm_stripe_count) {
2247                 last_stripe = (start_stripe < 1 ? lsm->lsm_stripe_count - 1 :
2248                                                               start_stripe - 1);
2249                 *stripe_count = lsm->lsm_stripe_count;
2250         } else {
2251                 for (j = 0, i = start_stripe; j < lsm->lsm_stripe_count;
2252                      i = (i + 1) % lsm->lsm_stripe_count, j++) {
2253                         if ((lov_stripe_intersects(lsm, i, fm_start, fm_end,
2254                                                    &obd_start, &obd_end)) == 0)
2255                                 break;
2256                 }
2257                 *stripe_count = j;
2258                 last_stripe = (start_stripe + j - 1) %lsm->lsm_stripe_count;
2259         }
2260
2261         return last_stripe;
2262 }
2263
2264 /**
2265  * Set fe_device and copy extents from local buffer into main return buffer.
2266  *
2267  * \param fiemap fiemap request header
2268  * \param lcl_fm_ext array of local fiemap extents to be copied
2269  * \param ost_index OST index to be written into the fm_device field for each
2270                     extent
2271  * \param ext_count number of extents to be copied
2272  * \param current_extent where to start copying in main extent array
2273  */
2274 void fiemap_prepare_and_copy_exts(struct ll_user_fiemap *fiemap,
2275                                   struct ll_fiemap_extent *lcl_fm_ext,
2276                                   int ost_index, unsigned int ext_count,
2277                                   int current_extent)
2278 {
2279         char *to;
2280         int ext;
2281
2282         for (ext = 0; ext < ext_count; ext++) {
2283                 lcl_fm_ext[ext].fe_device = ost_index;
2284                 lcl_fm_ext[ext].fe_flags |= FIEMAP_EXTENT_NET;
2285         }
2286
2287         /* Copy fm_extent's from fm_local to return buffer */
2288         to = (char *)fiemap + fiemap_count_to_size(current_extent);
2289         memcpy(to, lcl_fm_ext, ext_count * sizeof(struct ll_fiemap_extent));
2290 }
2291
2292 /**
2293  * Break down the FIEMAP request and send appropriate calls to individual OSTs.
2294  * This also handles the restarting of FIEMAP calls in case mapping overflows
2295  * the available number of extents in single call.
2296  */
2297 static int lov_fiemap(struct lov_obd *lov, __u32 keylen, void *key,
2298                       __u32 *vallen, void *val, struct lov_stripe_md *lsm)
2299 {
2300         struct ll_fiemap_info_key *fm_key = key;
2301         struct ll_user_fiemap *fiemap = val;
2302         struct ll_user_fiemap *fm_local = NULL;
2303         struct ll_fiemap_extent *lcl_fm_ext;
2304         int count_local;
2305         unsigned int get_num_extents = 0;
2306         int ost_index = 0, actual_start_stripe, start_stripe;
2307         obd_size fm_start, fm_end, fm_length, fm_end_offset = 0;
2308         obd_size curr_loc;
2309         int current_extent = 0, rc = 0, i;
2310         int ost_eof = 0; /* EOF for object */
2311         int ost_done = 0; /* done with required mapping for this OST? */
2312         int last_stripe;
2313         int cur_stripe = 0, cur_stripe_wrap = 0, stripe_count;
2314         unsigned int buffer_size = FIEMAP_BUFFER_SIZE;
2315
2316         if (lsm == NULL)
2317                 GOTO(out, rc = 0);
2318
2319         if (fiemap_count_to_size(fm_key->fiemap.fm_extent_count) < buffer_size)
2320                 buffer_size = fiemap_count_to_size(fm_key->fiemap.fm_extent_count);
2321
2322         OBD_ALLOC(fm_local, buffer_size);
2323         if (fm_local == NULL)
2324                 GOTO(out, rc = -ENOMEM);
2325         lcl_fm_ext = &fm_local->fm_extents[0];
2326
2327         count_local = fiemap_size_to_count(buffer_size);
2328
2329         memcpy(fiemap, &fm_key->fiemap, sizeof(*fiemap));
2330         fm_start = fiemap->fm_start;
2331         fm_length = fiemap->fm_length;
2332         /* Calculate start stripe, last stripe and length of mapping */
2333         actual_start_stripe = start_stripe = lov_stripe_number(lsm, fm_start);
2334         fm_end = (fm_length == ~0ULL ? fm_key->oa.o_size :
2335                                                 fm_start + fm_length - 1);
2336         /* If fm_length != ~0ULL but fm_start+fm_length-1 exceeds file size */
2337         if (fm_end > fm_key->oa.o_size)
2338                 fm_end = fm_key->oa.o_size;
2339
2340         last_stripe = fiemap_calc_last_stripe(lsm, fm_start, fm_end,
2341                                             actual_start_stripe, &stripe_count);
2342
2343         fm_end_offset = fiemap_calc_fm_end_offset(fiemap, lsm, fm_start, fm_end,
2344                                                   &start_stripe);
2345
2346         if (fiemap->fm_extent_count == 0) {
2347                 get_num_extents = 1;
2348                 count_local = 0;
2349         }
2350
2351         /* Check each stripe */
2352         for (cur_stripe = start_stripe, i = 0; i < stripe_count;
2353              i++, cur_stripe = (cur_stripe + 1) % lsm->lsm_stripe_count) {
2354                 obd_size req_fm_len; /* Stores length of required mapping */
2355                 obd_size len_mapped_single_call;
2356                 obd_off lun_start, lun_end, obd_object_end;
2357                 unsigned int ext_count;
2358
2359                 cur_stripe_wrap = cur_stripe;
2360
2361                 /* Find out range of mapping on this stripe */
2362                 if ((lov_stripe_intersects(lsm, cur_stripe, fm_start, fm_end,
2363                                            &lun_start, &obd_object_end)) == 0)
2364                         continue;
2365
2366                 /* If this is a continuation FIEMAP call and we are on
2367                  * starting stripe then lun_start needs to be set to
2368                  * fm_end_offset */
2369                 if (fm_end_offset != 0 && cur_stripe == start_stripe)
2370                         lun_start = fm_end_offset;
2371
2372                 if (fm_length != ~0ULL) {
2373                         /* Handle fm_start + fm_length overflow */
2374                         if (fm_start + fm_length < fm_start)
2375                                 fm_length = ~0ULL - fm_start;
2376                         lun_end = lov_size_to_stripe(lsm, fm_start + fm_length,
2377                                                      cur_stripe);
2378                 } else {
2379                         lun_end = ~0ULL;
2380                 }
2381
2382                 if (lun_start == lun_end)
2383                         continue;
2384
2385                 req_fm_len = obd_object_end - lun_start;
2386                 fm_local->fm_length = 0;
2387                 len_mapped_single_call = 0;
2388
2389                 /* If the output buffer is very large and the objects have many
2390                  * extents we may need to loop on a single OST repeatedly */
2391                 ost_eof = 0;
2392                 ost_done = 0;
2393                 do {
2394                         if (get_num_extents == 0) {
2395                                 /* Don't get too many extents. */
2396                                 if (current_extent + count_local >
2397                                     fiemap->fm_extent_count)
2398                                         count_local = fiemap->fm_extent_count -
2399                                                                  current_extent;
2400                         }
2401
2402                         lun_start += len_mapped_single_call;
2403                         fm_local->fm_length = req_fm_len - len_mapped_single_call;
2404                         req_fm_len = fm_local->fm_length;
2405                         fm_local->fm_extent_count = count_local;
2406                         fm_local->fm_mapped_extents = 0;
2407                         fm_local->fm_flags = fiemap->fm_flags;
2408
2409                         fm_key->oa.o_id = lsm->lsm_oinfo[cur_stripe]->loi_id;
2410                         fm_key->oa.o_seq = lsm->lsm_oinfo[cur_stripe]->loi_seq;
2411                         ost_index = lsm->lsm_oinfo[cur_stripe]->loi_ost_idx;
2412
2413                         if (ost_index < 0 || ost_index >=lov->desc.ld_tgt_count)
2414                                 GOTO(out, rc = -EINVAL);
2415
2416                         /* If OST is inactive, return extent with UNKNOWN flag */
2417                         if (!lov->lov_tgts[ost_index]->ltd_active) {
2418                                 fm_local->fm_flags |= FIEMAP_EXTENT_LAST;
2419                                 fm_local->fm_mapped_extents = 1;
2420
2421                                 lcl_fm_ext[0].fe_logical = lun_start;
2422                                 lcl_fm_ext[0].fe_length = obd_object_end -
2423                                                                       lun_start;
2424                                 lcl_fm_ext[0].fe_flags |= FIEMAP_EXTENT_UNKNOWN;
2425
2426                                 goto inactive_tgt;
2427                         }
2428
2429                         fm_local->fm_start = lun_start;
2430                         fm_local->fm_flags &= ~FIEMAP_FLAG_DEVICE_ORDER;
2431                         memcpy(&fm_key->fiemap, fm_local, sizeof(*fm_local));
2432                         *vallen=fiemap_count_to_size(fm_local->fm_extent_count);
2433                         rc = obd_get_info(lov->lov_tgts[ost_index]->ltd_exp,
2434                                           keylen, key, vallen, fm_local, lsm);
2435                         if (rc != 0)
2436                                 GOTO(out, rc);
2437
2438 inactive_tgt:
2439                         ext_count = fm_local->fm_mapped_extents;
2440                         if (ext_count == 0) {
2441                                 ost_done = 1;
2442                                 /* If last stripe has hole at the end,
2443                                  * then we need to return */
2444                                 if (cur_stripe_wrap == last_stripe) {
2445                                         fiemap->fm_mapped_extents = 0;
2446                                         goto finish;
2447                                 }
2448                                 break;
2449                         }
2450
2451                         /* If we just need num of extents then go to next device */
2452                         if (get_num_extents) {
2453                                 current_extent += ext_count;
2454                                 break;
2455                         }
2456
2457                         len_mapped_single_call = lcl_fm_ext[ext_count-1].fe_logical -
2458                                   lun_start + lcl_fm_ext[ext_count - 1].fe_length;
2459
2460                         /* Have we finished mapping on this device? */
2461                         if (req_fm_len <= len_mapped_single_call)
2462                                 ost_done = 1;
2463
2464                         /* Clear the EXTENT_LAST flag which can be present on
2465                          * last extent */
2466                         if (lcl_fm_ext[ext_count-1].fe_flags & FIEMAP_EXTENT_LAST)
2467                                 lcl_fm_ext[ext_count - 1].fe_flags &=
2468                                                             ~FIEMAP_EXTENT_LAST;
2469
2470                         curr_loc = lov_stripe_size(lsm,
2471                                            lcl_fm_ext[ext_count - 1].fe_logical+
2472                                            lcl_fm_ext[ext_count - 1].fe_length,
2473                                            cur_stripe);
2474                         if (curr_loc >= fm_key->oa.o_size)
2475                                 ost_eof = 1;
2476
2477                         fiemap_prepare_and_copy_exts(fiemap, lcl_fm_ext,
2478                                                      ost_index, ext_count,
2479                                                      current_extent);
2480
2481                         current_extent += ext_count;
2482
2483                         /* Ran out of available extents? */
2484                         if (current_extent >= fiemap->fm_extent_count)
2485                                 goto finish;
2486                 } while (ost_done == 0 && ost_eof == 0);
2487
2488                 if (cur_stripe_wrap == last_stripe)
2489                         goto finish;
2490         }
2491
2492 finish:
2493         /* Indicate that we are returning device offsets unless file just has
2494          * single stripe */
2495         if (lsm->lsm_stripe_count > 1)
2496                 fiemap->fm_flags |= FIEMAP_FLAG_DEVICE_ORDER;
2497
2498         if (get_num_extents)
2499                 goto skip_last_device_calc;
2500
2501         /* Check if we have reached the last stripe and whether mapping for that
2502          * stripe is done. */
2503         if (cur_stripe_wrap == last_stripe) {
2504                 if (ost_done || ost_eof)
2505                         fiemap->fm_extents[current_extent - 1].fe_flags |=
2506                                                              FIEMAP_EXTENT_LAST;
2507         }
2508
2509 skip_last_device_calc:
2510         fiemap->fm_mapped_extents = current_extent;
2511
2512 out:
2513         OBD_FREE(fm_local, buffer_size);
2514         return rc;
2515 }
2516
2517 static int lov_get_info(struct obd_export *exp, __u32 keylen,
2518                         void *key, __u32 *vallen, void *val,
2519                         struct lov_stripe_md *lsm)
2520 {
2521         struct obd_device *obddev = class_exp2obd(exp);
2522         struct lov_obd *lov = &obddev->u.lov;
2523         int i, rc;
2524         ENTRY;
2525
2526         if (!vallen || !val)
2527                 RETURN(-EFAULT);
2528
2529         obd_getref(obddev);
2530
2531         if (KEY_IS(KEY_LOCK_TO_STRIPE)) {
2532                 struct {
2533                         char name[16];
2534                         struct ldlm_lock *lock;
2535                 } *data = key;
2536                 struct ldlm_res_id *res_id = &data->lock->l_resource->lr_name;
2537                 struct lov_oinfo *loi;
2538                 __u32 *stripe = val;
2539
2540                 if (*vallen < sizeof(*stripe))
2541                         GOTO(out, rc = -EFAULT);
2542                 *vallen = sizeof(*stripe);
2543
2544                 /* XXX This is another one of those bits that will need to
2545                  * change if we ever actually support nested LOVs.  It uses
2546                  * the lock's export to find out which stripe it is. */
2547                 /* XXX - it's assumed all the locks for deleted OSTs have
2548                  * been cancelled. Also, the export for deleted OSTs will
2549                  * be NULL and won't match the lock's export. */
2550                 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2551                         loi = lsm->lsm_oinfo[i];
2552                         if (!lov->lov_tgts[loi->loi_ost_idx])
2553                                 continue;
2554                         if (lov->lov_tgts[loi->loi_ost_idx]->ltd_exp ==
2555                             data->lock->l_conn_export &&
2556                             osc_res_name_eq(loi->loi_id, loi->loi_seq, res_id)) {
2557                                 *stripe = i;
2558                                 GOTO(out, rc = 0);
2559                         }
2560                 }
2561                 LDLM_ERROR(data->lock, "lock on inode without such object");
2562                 dump_lsm(D_ERROR, lsm);
2563                 GOTO(out, rc = -ENXIO);
2564         } else if (KEY_IS(KEY_LAST_ID)) {
2565                 struct obd_id_info *info = val;
2566                 __u32 size = sizeof(obd_id);
2567                 struct lov_tgt_desc *tgt;
2568
2569                 LASSERT(*vallen == sizeof(struct obd_id_info));
2570                 tgt = lov->lov_tgts[info->idx];
2571
2572                 if (!tgt || !tgt->ltd_active)
2573                         GOTO(out, rc = -ESRCH);
2574
2575                 rc = obd_get_info(tgt->ltd_exp, keylen, key, &size, info->data, NULL);
2576                 GOTO(out, rc = 0);
2577         } else if (KEY_IS(KEY_LOVDESC)) {
2578                 struct lov_desc *desc_ret = val;
2579                 *desc_ret = lov->desc;
2580
2581                 GOTO(out, rc = 0);
2582         } else if (KEY_IS(KEY_FIEMAP)) {
2583                 rc = lov_fiemap(lov, keylen, key, vallen, val, lsm);
2584                 GOTO(out, rc);
2585         } else if (KEY_IS(KEY_CONNECT_FLAG)) {
2586                 struct lov_tgt_desc *tgt;
2587                 __u64 ost_idx = *((__u64*)val);
2588
2589                 LASSERT(*vallen == sizeof(__u64));
2590                 LASSERT(ost_idx < lov->desc.ld_tgt_count);
2591                 tgt = lov->lov_tgts[ost_idx];
2592
2593                 if (!tgt || !tgt->ltd_exp)
2594                         GOTO(out, rc = -ESRCH);
2595
2596                 *((__u64*)val) = tgt->ltd_exp->exp_connect_flags;
2597                 GOTO(out, rc = 0);
2598         }
2599
2600         rc = -EINVAL;
2601
2602 out:
2603         obd_putref(obddev);
2604         RETURN(rc);
2605 }
2606
2607 static int lov_set_info_async(struct obd_export *exp, obd_count keylen,
2608                               void *key, obd_count vallen, void *val,
2609                               struct ptlrpc_request_set *set)
2610 {
2611         struct obd_device *obddev = class_exp2obd(exp);
2612         struct lov_obd *lov = &obddev->u.lov;
2613         obd_count count;
2614         int i, rc = 0, err;
2615         struct lov_tgt_desc *tgt;
2616         unsigned incr, check_uuid,
2617                  do_inactive, no_set;
2618         unsigned next_id = 0,  mds_con = 0, capa = 0;
2619         ENTRY;
2620
2621         incr = check_uuid = do_inactive = no_set = 0;
2622         if (set == NULL) {
2623                 no_set = 1;
2624                 set = ptlrpc_prep_set();
2625                 if (!set)
2626                         RETURN(-ENOMEM);
2627         }
2628
2629         obd_getref(obddev);
2630         count = lov->desc.ld_tgt_count;
2631
2632         if (KEY_IS(KEY_NEXT_ID)) {
2633                 count = vallen / sizeof(struct obd_id_info);
2634                 vallen = sizeof(obd_id);
2635                 incr = sizeof(struct obd_id_info);
2636                 do_inactive = 1;
2637                 next_id = 1;
2638         } else if (KEY_IS(KEY_CHECKSUM)) {
2639                 do_inactive = 1;
2640         } else if (KEY_IS(KEY_EVICT_BY_NID)) {
2641                 /* use defaults:  do_inactive = incr = 0; */
2642         } else if (KEY_IS(KEY_MDS_CONN)) {
2643                 mds_con = 1;
2644         } else if (KEY_IS(KEY_CAPA_KEY)) {
2645                 capa = 1;
2646         }
2647
2648         for (i = 0; i < count; i++, val = (char *)val + incr) {
2649                 if (next_id) {
2650                         tgt = lov->lov_tgts[((struct obd_id_info*)val)->idx];
2651                 } else {
2652                         tgt = lov->lov_tgts[i];
2653                 }
2654                 /* OST was disconnected */
2655                 if (!tgt || !tgt->ltd_exp)
2656                         continue;
2657
2658                 /* OST is inactive and we don't want inactive OSCs */
2659                 if (!tgt->ltd_active && !do_inactive)
2660                         continue;
2661
2662                 if (mds_con) {
2663                         struct mds_group_info *mgi;
2664
2665                         LASSERT(vallen == sizeof(*mgi));
2666                         mgi = (struct mds_group_info *)val;
2667
2668                         /* Only want a specific OSC */
2669                         if (mgi->uuid && !obd_uuid_equals(mgi->uuid,
2670                                                 &tgt->ltd_uuid))
2671                                 continue;
2672
2673                         err = obd_set_info_async(tgt->ltd_exp,
2674                                          keylen, key, sizeof(int),
2675                                          &mgi->group, set);
2676                 } else if (next_id) {
2677                         err = obd_set_info_async(tgt->ltd_exp,
2678                                          keylen, key, vallen,
2679                                          ((struct obd_id_info*)val)->data, set);
2680                 } else if (capa) {
2681                         struct mds_capa_info *info = (struct mds_capa_info*)val;
2682
2683                         LASSERT(vallen == sizeof(*info));
2684
2685                          /* Only want a specific OSC */
2686                         if (info->uuid &&
2687                             !obd_uuid_equals(info->uuid, &tgt->ltd_uuid))
2688                                 continue;
2689
2690                         err = obd_set_info_async(tgt->ltd_exp, keylen, key,
2691                                                  sizeof(*info->capa),
2692                                                  info->capa, set);
2693
2694                 } else {
2695                         /* Only want a specific OSC */
2696                         if (check_uuid &&
2697                             !obd_uuid_equals(val, &tgt->ltd_uuid))
2698                                 continue;
2699
2700                         err = obd_set_info_async(tgt->ltd_exp,
2701                                          keylen, key, vallen, val, set);
2702                 }
2703
2704                 if (!rc)
2705                         rc = err;
2706         }
2707
2708         obd_putref(obddev);
2709         if (no_set) {
2710                 err = ptlrpc_set_wait(set);
2711                 if (!rc)
2712                         rc = err;
2713                 ptlrpc_set_destroy(set);
2714         }
2715         RETURN(rc);
2716 }
2717
2718 int lov_test_and_clear_async_rc(struct lov_stripe_md *lsm)
2719 {
2720         int i, rc = 0;
2721         ENTRY;
2722
2723         for (i = 0; i < lsm->lsm_stripe_count; i++) {
2724                 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
2725                 if (loi->loi_ar.ar_rc && !rc)
2726                         rc = loi->loi_ar.ar_rc;
2727                 loi->loi_ar.ar_rc = 0;
2728         }
2729         RETURN(rc);
2730 }
2731 EXPORT_SYMBOL(lov_test_and_clear_async_rc);
2732
2733
2734 static int lov_extent_calc(struct obd_export *exp, struct lov_stripe_md *lsm,
2735                            int cmd, __u64 *offset)
2736 {
2737         __u32 ssize = lsm->lsm_stripe_size;
2738         __u64 start;
2739
2740         start = *offset;
2741         do_div(start, ssize);
2742         start = start * ssize;
2743
2744         CDEBUG(D_DLMTRACE, "offset "LPU64", stripe %u, start "LPU64
2745                            ", end "LPU64"\n", *offset, ssize, start,
2746                            start + ssize - 1);
2747         if (cmd == OBD_CALC_STRIPE_END) {
2748                 *offset = start + ssize - 1;
2749         } else if (cmd == OBD_CALC_STRIPE_START) {
2750                 *offset = start;
2751         } else {
2752                 LBUG();
2753         }
2754
2755         RETURN(0);
2756 }
2757
2758 void lov_stripe_lock(struct lov_stripe_md *md)
2759 {
2760         LASSERT(md->lsm_lock_owner != cfs_curproc_pid());
2761         cfs_spin_lock(&md->lsm_lock);
2762         LASSERT(md->lsm_lock_owner == 0);
2763         md->lsm_lock_owner = cfs_curproc_pid();
2764 }
2765 EXPORT_SYMBOL(lov_stripe_lock);
2766
2767 void lov_stripe_unlock(struct lov_stripe_md *md)
2768 {
2769         LASSERT(md->lsm_lock_owner == cfs_curproc_pid());
2770         md->lsm_lock_owner = 0;
2771         cfs_spin_unlock(&md->lsm_lock);
2772 }
2773 EXPORT_SYMBOL(lov_stripe_unlock);
2774
2775
2776 struct obd_ops lov_obd_ops = {
2777         .o_owner               = THIS_MODULE,
2778         .o_setup               = lov_setup,
2779         .o_precleanup          = lov_precleanup,
2780         .o_cleanup             = lov_cleanup,
2781         //.o_process_config      = lov_process_config,
2782         .o_connect             = lov_connect,
2783         .o_disconnect          = lov_disconnect,
2784         .o_statfs              = lov_statfs,
2785         .o_statfs_async        = lov_statfs_async,
2786         .o_packmd              = lov_packmd,
2787         .o_unpackmd            = lov_unpackmd,
2788         .o_create              = lov_create,
2789         .o_destroy             = lov_destroy,
2790         .o_getattr             = lov_getattr,
2791         .o_getattr_async       = lov_getattr_async,
2792         .o_setattr             = lov_setattr,
2793         .o_setattr_async       = lov_setattr_async,
2794         .o_brw                 = lov_brw,
2795         .o_merge_lvb           = lov_merge_lvb,
2796         .o_adjust_kms          = lov_adjust_kms,
2797         .o_punch               = lov_punch,
2798         .o_sync                = lov_sync,
2799         .o_enqueue             = lov_enqueue,
2800         .o_change_cbdata       = lov_change_cbdata,
2801         .o_find_cbdata         = lov_find_cbdata,
2802         .o_cancel              = lov_cancel,
2803         .o_cancel_unused       = lov_cancel_unused,
2804         .o_iocontrol           = lov_iocontrol,
2805         .o_get_info            = lov_get_info,
2806         .o_set_info_async      = lov_set_info_async,
2807         .o_extent_calc         = lov_extent_calc,
2808         .o_llog_init           = lov_llog_init,
2809         .o_llog_finish         = lov_llog_finish,
2810         .o_notify              = lov_notify,
2811         .o_pool_new            = lov_pool_new,
2812         .o_pool_rem            = lov_pool_remove,
2813         .o_pool_add            = lov_pool_add,
2814         .o_pool_del            = lov_pool_del,
2815         .o_getref              = lov_getref,
2816         .o_putref              = lov_putref,
2817 };
2818
2819 static quota_interface_t *quota_interface;
2820 extern quota_interface_t lov_quota_interface;
2821
2822 cfs_mem_cache_t *lov_oinfo_slab;
2823
2824 extern struct lu_kmem_descr lov_caches[];
2825
2826 int __init lov_init(void)
2827 {
2828         struct lprocfs_static_vars lvars = { 0 };
2829         int rc, rc2;
2830         ENTRY;
2831
2832         /* print an address of _any_ initialized kernel symbol from this
2833          * module, to allow debugging with gdb that doesn't support data
2834          * symbols from modules.*/
2835         CDEBUG(D_CONSOLE, "Lustre LOV module (%p).\n", &lov_caches);
2836
2837         rc = lu_kmem_init(lov_caches);
2838         if (rc)
2839                 return rc;
2840
2841         lov_oinfo_slab = cfs_mem_cache_create("lov_oinfo",
2842                                               sizeof(struct lov_oinfo),
2843                                               0, CFS_SLAB_HWCACHE_ALIGN);
2844         if (lov_oinfo_slab == NULL) {
2845                 lu_kmem_fini(lov_caches);
2846                 return -ENOMEM;
2847         }
2848         lprocfs_lov_init_vars(&lvars);
2849
2850         cfs_request_module("lquota");
2851         quota_interface = PORTAL_SYMBOL_GET(lov_quota_interface);
2852         init_obd_quota_ops(quota_interface, &lov_obd_ops);
2853
2854         rc = class_register_type(&lov_obd_ops, NULL, lvars.module_vars,
2855                                  LUSTRE_LOV_NAME, &lov_device_type);
2856
2857         if (rc) {
2858                 if (quota_interface)
2859                         PORTAL_SYMBOL_PUT(lov_quota_interface);
2860                 rc2 = cfs_mem_cache_destroy(lov_oinfo_slab);
2861                 LASSERT(rc2 == 0);
2862                 lu_kmem_fini(lov_caches);
2863         }
2864
2865         RETURN(rc);
2866 }
2867
2868 #ifdef __KERNEL__
2869 static void /*__exit*/ lov_exit(void)
2870 {
2871         int rc;
2872
2873         lu_device_type_fini(&lov_device_type);
2874         lu_kmem_fini(lov_caches);
2875
2876         if (quota_interface)
2877                 PORTAL_SYMBOL_PUT(lov_quota_interface);
2878
2879         class_unregister_type(LUSTRE_LOV_NAME);
2880         rc = cfs_mem_cache_destroy(lov_oinfo_slab);
2881         LASSERT(rc == 0);
2882 }
2883
2884 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2885 MODULE_DESCRIPTION("Lustre Logical Object Volume OBD driver");
2886 MODULE_LICENSE("GPL");
2887
2888 cfs_module(lov, LUSTRE_VERSION_STRING, lov_init, lov_exit);
2889 #endif