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 static 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                                                    HASH_POOLS_BKT_BITS, 0,
 810                                                    CFS_HASH_MIN_THETA,
 811                                                    CFS_HASH_MAX_THETA,
 812                                                    &pool_hash_operations,
 813                                                    CFS_HASH_DEFAULT);
 814         CFS_INIT_LIST_HEAD(&lov->lov_pool_list);
 815         lov->lov_pool_count = 0;
 816         rc = lov_ost_pool_init(&lov->lov_packed, 0);
 817         if (rc)
 818                 GOTO(out_free_statfs, rc);
 819         rc = lov_ost_pool_init(&lov->lov_qos.lq_rr.lqr_pool, 0);
 820         if (rc)
 821                 GOTO(out_free_lov_packed, rc);
 822
 823         lprocfs_lov_init_vars(&lvars);
 824         lprocfs_obd_setup(obd, lvars.obd_vars);
 825 #ifdef LPROCFS
 826         {
 827                 int rc;
 828
 829                 rc = lprocfs_seq_create(obd->obd_proc_entry, "target_obd",
 830                                         0444, &lov_proc_target_fops, obd);
 831                 if (rc)
 832                         CWARN("Error adding the target_obd file\n");
 833         }
 834 #endif
 835         lov->lov_pool_proc_entry = lprocfs_register("pools",
 836                                                     obd->obd_proc_entry,
 837                                                     NULL, NULL);
 838
 839         RETURN(0);
 840
 841 out_free_lov_packed:
 842         lov_ost_pool_free(&lov->lov_packed);
 843 out_free_statfs:
 844         OBD_FREE_PTR(lov->lov_qos.lq_statfs_data);
 845         return rc;
 846 }
 847
 848 static int lov_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
 849 {
 850         int rc = 0;
 851         struct lov_obd *lov = &obd->u.lov;
 852
 853         ENTRY;
 854
 855         switch (stage) {
 856         case OBD_CLEANUP_EARLY: {
 857                 int i;
 858                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
 859                         if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_active)
 860                                 continue;
 861                         obd_precleanup(class_exp2obd(lov->lov_tgts[i]->ltd_exp),
 862                                        OBD_CLEANUP_EARLY);
 863                 }
 864                 break;
 865         }
 866         case OBD_CLEANUP_EXPORTS:
 867                 rc = obd_llog_finish(obd, 0);
 868                 if (rc != 0)
 869                         CERROR("failed to cleanup llogging subsystems\n");
 870                 break;
 871         }
 872         RETURN(rc);
 873 }
 874
 875 static int lov_cleanup(struct obd_device *obd)
 876 {
 877         struct lov_obd *lov = &obd->u.lov;
 878         cfs_list_t *pos, *tmp;
 879         struct pool_desc *pool;
 880
 881         cfs_list_for_each_safe(pos, tmp, &lov->lov_pool_list) {
 882                 pool = cfs_list_entry(pos, struct pool_desc, pool_list);
 883                 /* free pool structs */
 884                 CDEBUG(D_INFO, "delete pool %p\n", pool);
 885                 lov_pool_del(obd, pool->pool_name);
 886         }
 887         cfs_hash_putref(lov->lov_pools_hash_body);
 888         lov_ost_pool_free(&(lov->lov_qos.lq_rr.lqr_pool));
 889         lov_ost_pool_free(&lov->lov_packed);
 890
 891         if (lov->lov_tgts) {
 892                 int i;
 893                 obd_getref(obd);
 894                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
 895                         if (!lov->lov_tgts[i])
 896                                 continue;
 897
 898                         /* Inactive targets may never have connected */
 899                         if (lov->lov_tgts[i]->ltd_active ||
 900                             cfs_atomic_read(&lov->lov_refcount))
 901                             /* We should never get here - these
 902                                should have been removed in the
 903                              disconnect. */
 904                                 CERROR("lov tgt %d not cleaned!"
 905                                        " deathrow=%d, lovrc=%d\n",
 906                                        i, lov->lov_death_row,
 907                                        cfs_atomic_read(&lov->lov_refcount));
 908                         lov_del_target(obd, i, 0, 0);
 909                 }
 910                 obd_putref(obd);
 911                 OBD_FREE(lov->lov_tgts, sizeof(*lov->lov_tgts) *
 912                          lov->lov_tgt_size);
 913                 lov->lov_tgt_size = 0;
 914         }
 915
 916         /* clear pools parent proc entry only after all pools is killed */
 917         lprocfs_obd_cleanup(obd);
 918
 919         OBD_FREE_PTR(lov->lov_qos.lq_statfs_data);
 920         RETURN(0);
 921 }
 922
 923 int lov_process_config_base(struct obd_device *obd, struct lustre_cfg *lcfg,
 924                             __u32 *indexp, int *genp)
 925 {
 926         struct obd_uuid obd_uuid;
 927         int cmd;
 928         int rc = 0;
 929         ENTRY;
 930
 931         switch(cmd = lcfg->lcfg_command) {
 932         case LCFG_LOV_ADD_OBD:
 933         case LCFG_LOV_ADD_INA:
 934         case LCFG_LOV_DEL_OBD: {
 935                 __u32 index;
 936                 int gen;
 937                 /* lov_modify_tgts add  0:lov_mdsA  1:ost1_UUID  2:0  3:1 */
 938                 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
 939                         GOTO(out, rc = -EINVAL);
 940
 941                 obd_str2uuid(&obd_uuid,  lustre_cfg_buf(lcfg, 1));
 942
 943                 if (sscanf(lustre_cfg_buf(lcfg, 2), "%d", indexp) != 1)
 944                         GOTO(out, rc = -EINVAL);
 945                 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", genp) != 1)
 946                         GOTO(out, rc = -EINVAL);
 947                 index = *indexp;
 948                 gen = *genp;
 949                 if (cmd == LCFG_LOV_ADD_OBD)
 950                         rc = lov_add_target(obd, &obd_uuid, index, gen, 1);
 951                 else if (cmd == LCFG_LOV_ADD_INA)
 952                         rc = lov_add_target(obd, &obd_uuid, index, gen, 0);
 953                 else
 954                         rc = lov_del_target(obd, index, &obd_uuid, gen);
 955                 GOTO(out, rc);
 956         }
 957         case LCFG_PARAM: {
 958                 struct lprocfs_static_vars lvars = { 0 };
 959                 struct lov_desc *desc = &(obd->u.lov.desc);
 960
 961                 if (!desc)
 962                         GOTO(out, rc = -EINVAL);
 963
 964                 lprocfs_lov_init_vars(&lvars);
 965
 966                 rc = class_process_proc_param(PARAM_LOV, lvars.obd_vars,
 967                                               lcfg, obd);
 968                 if (rc > 0)
 969                         rc = 0;
 970                 GOTO(out, rc);
 971         }
 972         case LCFG_POOL_NEW:
 973         case LCFG_POOL_ADD:
 974         case LCFG_POOL_DEL:
 975         case LCFG_POOL_REM:
 976                 GOTO(out, rc);
 977
 978         default: {
 979                 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
 980                 GOTO(out, rc = -EINVAL);
 981
 982         }
 983         }
 984 out:
 985         RETURN(rc);
 986 }
 987
 988 #ifndef log2
 989 #define log2(n) cfs_ffz(~(n))
 990 #endif
 991
 992 static int lov_clear_orphans(struct obd_export *export, struct obdo *src_oa,
 993                              struct lov_stripe_md **ea,
 994                              struct obd_trans_info *oti)
 995 {
 996         struct lov_obd *lov;
 997         struct obdo *tmp_oa;
 998         struct obd_uuid *ost_uuid = NULL;
 999         int rc = 0, i;
1000         ENTRY;
1001
1002         LASSERT(src_oa->o_valid & OBD_MD_FLFLAGS &&
1003                 src_oa->o_flags == OBD_FL_DELORPHAN);
1004
1005         lov = &export->exp_obd->u.lov;
1006
1007         OBDO_ALLOC(tmp_oa);
1008         if (tmp_oa == NULL)
1009                 RETURN(-ENOMEM);
1010
1011         if (oti->oti_ost_uuid) {
1012                 ost_uuid = oti->oti_ost_uuid;
1013                 CDEBUG(D_HA, "clearing orphans only for %s\n",
1014                        ost_uuid->uuid);
1015         }
1016
1017         obd_getref(export->exp_obd);
1018         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
1019                 struct lov_stripe_md obj_md;
1020                 struct lov_stripe_md *obj_mdp = &obj_md;
1021                 struct lov_tgt_desc *tgt;
1022                 int err;
1023
1024                 tgt = lov->lov_tgts[i];
1025                 if (!tgt)
1026                         continue;
1027
1028                 /* if called for a specific target, we don't
1029                    care if it is not active. */
1030                 if (!lov->lov_tgts[i]->ltd_active && ost_uuid == NULL) {
1031                         CDEBUG(D_HA, "lov idx %d inactive\n", i);
1032                         continue;
1033                 }
1034
1035                 if (ost_uuid && !obd_uuid_equals(ost_uuid, &tgt->ltd_uuid))
1036                         continue;
1037
1038                 CDEBUG(D_CONFIG,"Clear orphans for %d:%s\n", i,
1039                        obd_uuid2str(ost_uuid));
1040
1041                 memcpy(tmp_oa, src_oa, sizeof(*tmp_oa));
1042
1043                 LASSERT(lov->lov_tgts[i]->ltd_exp);
1044                 /* XXX: LOV STACKING: use real "obj_mdp" sub-data */
1045                 err = obd_create(lov->lov_tgts[i]->ltd_exp,
1046                                  tmp_oa, &obj_mdp, oti);
1047                 if (err) {
1048                         /* This export will be disabled until it is recovered,
1049                            and then orphan recovery will be completed. */
1050                         CERROR("error in orphan recovery on OST idx %d/%d: "
1051                                "rc = %d\n", i, lov->desc.ld_tgt_count, err);
1052                         rc = err;
1053                 }
1054
1055                 if (ost_uuid)
1056                         break;
1057         }
1058         obd_putref(export->exp_obd);
1059
1060         OBDO_FREE(tmp_oa);
1061         RETURN(rc);
1062 }
1063
1064 static int lov_recreate(struct obd_export *exp, struct obdo *src_oa,
1065                         struct lov_stripe_md **ea, struct obd_trans_info *oti)
1066 {
1067         struct lov_stripe_md *obj_mdp, *lsm;
1068         struct lov_obd *lov = &exp->exp_obd->u.lov;
1069         unsigned ost_idx;
1070         int rc, i;
1071         ENTRY;
1072
1073         LASSERT(src_oa->o_valid & OBD_MD_FLFLAGS &&
1074                 src_oa->o_flags & OBD_FL_RECREATE_OBJS);
1075
1076         OBD_ALLOC(obj_mdp, sizeof(*obj_mdp));
1077         if (obj_mdp == NULL)
1078                 RETURN(-ENOMEM);
1079
1080         ost_idx = src_oa->o_nlink;
1081         lsm = *ea;
1082         if (lsm == NULL)
1083                 GOTO(out, rc = -EINVAL);
1084         if (ost_idx >= lov->desc.ld_tgt_count ||
1085             !lov->lov_tgts[ost_idx])
1086                 GOTO(out, rc = -EINVAL);
1087
1088         for (i = 0; i < lsm->lsm_stripe_count; i++) {
1089                 if (lsm->lsm_oinfo[i]->loi_ost_idx == ost_idx) {
1090                         if (lsm->lsm_oinfo[i]->loi_id != src_oa->o_id)
1091                                 GOTO(out, rc = -EINVAL);
1092                         break;
1093                 }
1094         }
1095         if (i == lsm->lsm_stripe_count)
1096                 GOTO(out, rc = -EINVAL);
1097
1098         rc = obd_create(lov->lov_tgts[ost_idx]->ltd_exp, src_oa, &obj_mdp, oti);
1099 out:
1100         OBD_FREE(obj_mdp, sizeof(*obj_mdp));
1101         RETURN(rc);
1102 }
1103
1104 /* the LOV expects oa->o_id to be set to the LOV object id */
1105 static int lov_create(struct obd_export *exp, struct obdo *src_oa,
1106                       struct lov_stripe_md **ea, struct obd_trans_info *oti)
1107 {
1108         struct lov_obd *lov;
1109         struct obd_info oinfo;
1110         struct lov_request_set *set = NULL;
1111         struct lov_request *req;
1112         struct l_wait_info  lwi = { 0 };
1113         int rc = 0;
1114         ENTRY;
1115
1116         LASSERT(ea != NULL);
1117         if (exp == NULL)
1118                 RETURN(-EINVAL);
1119
1120         if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
1121             src_oa->o_flags == OBD_FL_DELORPHAN) {
1122                 rc = lov_clear_orphans(exp, src_oa, ea, oti);
1123                 RETURN(rc);
1124         }
1125
1126         lov = &exp->exp_obd->u.lov;
1127         if (!lov->desc.ld_active_tgt_count)
1128                 RETURN(-EIO);
1129
1130         obd_getref(exp->exp_obd);
1131         /* Recreate a specific object id at the given OST index */
1132         if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
1133             (src_oa->o_flags & OBD_FL_RECREATE_OBJS)) {
1134                  rc = lov_recreate(exp, src_oa, ea, oti);
1135                  GOTO(out, rc);
1136         }
1137
1138         /* issue statfs rpcs if the osfs data is older than qos_maxage - 1s,
1139          * later in alloc_qos(), we will wait for those rpcs to complete if
1140          * the osfs age is older than 2 * qos_maxage */
1141         qos_statfs_update(exp->exp_obd,
1142                           cfs_time_shift_64(-lov->desc.ld_qos_maxage +
1143                                             OBD_STATFS_CACHE_SECONDS),
1144                           0);
1145
1146         rc = lov_prep_create_set(exp, &oinfo, ea, src_oa, oti, &set);
1147         if (rc)
1148                 GOTO(out, rc);
1149
1150         cfs_list_for_each_entry(req, &set->set_list, rq_link) {
1151                 /* XXX: LOV STACKING: use real "obj_mdp" sub-data */
1152                 rc = obd_create_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1153                                       &req->rq_oi, &req->rq_oi.oi_md, oti);
1154         }
1155
1156         /* osc_create have timeout equ obd_timeout/2 so waiting don't be
1157          * longer then this */
1158         l_wait_event(set->set_waitq, lov_finished_set(set), &lwi);
1159
1160         /* we not have ptlrpc set for assign set->interpret and should
1161          * be call interpret function himself. calling from cb_create_update
1162          * not permited because lov_fini_create_set can sleep for long time,
1163          * but we must avoid sleeping in ptlrpcd interpret function. */
1164         rc = lov_fini_create_set(set, ea);
1165 out:
1166         obd_putref(exp->exp_obd);
1167         RETURN(rc);
1168 }
1169
1170 #define ASSERT_LSM_MAGIC(lsmp)                                                  \
1171 do {                                                                            \
1172         LASSERT((lsmp) != NULL);                                                \
1173         LASSERTF(((lsmp)->lsm_magic == LOV_MAGIC_V1 ||                          \
1174                  (lsmp)->lsm_magic == LOV_MAGIC_V3),                            \
1175                  "%p->lsm_magic=%x\n", (lsmp), (lsmp)->lsm_magic);              \
1176 } while (0)
1177
1178 static int lov_destroy(struct obd_export *exp, struct obdo *oa,
1179                        struct lov_stripe_md *lsm, struct obd_trans_info *oti,
1180                        struct obd_export *md_exp, void *capa)
1181 {
1182         struct lov_request_set *set;
1183         struct obd_info oinfo;
1184         struct lov_request *req;
1185         cfs_list_t *pos;
1186         struct lov_obd *lov;
1187         int rc = 0, err = 0;
1188         ENTRY;
1189
1190         ASSERT_LSM_MAGIC(lsm);
1191
1192         if (!exp || !exp->exp_obd)
1193                 RETURN(-ENODEV);
1194
1195         if (oa->o_valid & OBD_MD_FLCOOKIE) {
1196                 LASSERT(oti);
1197                 LASSERT(oti->oti_logcookies);
1198         }
1199
1200         lov = &exp->exp_obd->u.lov;
1201         obd_getref(exp->exp_obd);
1202         rc = lov_prep_destroy_set(exp, &oinfo, oa, lsm, oti, &set);
1203         if (rc)
1204                 GOTO(out, rc);
1205
1206         cfs_list_for_each (pos, &set->set_list) {
1207                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1208
1209                 if (oa->o_valid & OBD_MD_FLCOOKIE)
1210                         oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1211
1212                 err = obd_destroy(lov->lov_tgts[req->rq_idx]->ltd_exp,
1213                                   req->rq_oi.oi_oa, NULL, oti, NULL, capa);
1214                 err = lov_update_common_set(set, req, err);
1215                 if (err) {
1216                         CERROR("error: destroying objid "LPX64" subobj "
1217                                LPX64" on OST idx %d: rc = %d\n",
1218                                oa->o_id, req->rq_oi.oi_oa->o_id,
1219                                req->rq_idx, err);
1220                         if (!rc)
1221                                 rc = err;
1222                 }
1223         }
1224
1225         if (rc == 0) {
1226                 LASSERT(lsm_op_find(lsm->lsm_magic) != NULL);
1227                 rc = lsm_op_find(lsm->lsm_magic)->lsm_destroy(lsm, oa, md_exp);
1228         }
1229         err = lov_fini_destroy_set(set);
1230 out:
1231         obd_putref(exp->exp_obd);
1232         RETURN(rc ? rc : err);
1233 }
1234
1235 static int lov_getattr(struct obd_export *exp, struct obd_info *oinfo)
1236 {
1237         struct lov_request_set *set;
1238         struct lov_request *req;
1239         cfs_list_t *pos;
1240         struct lov_obd *lov;
1241         int err = 0, rc = 0;
1242         ENTRY;
1243
1244         LASSERT(oinfo);
1245         ASSERT_LSM_MAGIC(oinfo->oi_md);
1246
1247         if (!exp || !exp->exp_obd)
1248                 RETURN(-ENODEV);
1249
1250         lov = &exp->exp_obd->u.lov;
1251
1252         rc = lov_prep_getattr_set(exp, oinfo, &set);
1253         if (rc)
1254                 RETURN(rc);
1255
1256         cfs_list_for_each (pos, &set->set_list) {
1257                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1258
1259                 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
1260                        "%u\n", oinfo->oi_oa->o_id, req->rq_stripe,
1261                        req->rq_oi.oi_oa->o_id, req->rq_idx);
1262
1263                 rc = obd_getattr(lov->lov_tgts[req->rq_idx]->ltd_exp,
1264                                  &req->rq_oi);
1265                 err = lov_update_common_set(set, req, rc);
1266                 if (err) {
1267                         CERROR("error: getattr objid "LPX64" subobj "
1268                                LPX64" on OST idx %d: rc = %d\n",
1269                                oinfo->oi_oa->o_id, req->rq_oi.oi_oa->o_id,
1270                                req->rq_idx, err);
1271                         break;
1272                 }
1273         }
1274
1275         rc = lov_fini_getattr_set(set);
1276         if (err)
1277                 rc = err;
1278         RETURN(rc);
1279 }
1280
1281 static int lov_getattr_interpret(struct ptlrpc_request_set *rqset,
1282                                  void *data, int rc)
1283 {
1284         struct lov_request_set *lovset = (struct lov_request_set *)data;
1285         int err;
1286         ENTRY;
1287
1288         /* don't do attribute merge if this aysnc op failed */
1289         if (rc)
1290                 lovset->set_completes = 0;
1291         err = lov_fini_getattr_set(lovset);
1292         RETURN(rc ? rc : err);
1293 }
1294
1295 static int lov_getattr_async(struct obd_export *exp, struct obd_info *oinfo,
1296                               struct ptlrpc_request_set *rqset)
1297 {
1298         struct lov_request_set *lovset;
1299         struct lov_obd *lov;
1300         cfs_list_t *pos;
1301         struct lov_request *req;
1302         int rc = 0, err;
1303         ENTRY;
1304
1305         LASSERT(oinfo);
1306         ASSERT_LSM_MAGIC(oinfo->oi_md);
1307
1308         if (!exp || !exp->exp_obd)
1309                 RETURN(-ENODEV);
1310
1311         lov = &exp->exp_obd->u.lov;
1312
1313         rc = lov_prep_getattr_set(exp, oinfo, &lovset);
1314         if (rc)
1315                 RETURN(rc);
1316
1317         CDEBUG(D_INFO, "objid "LPX64": %ux%u byte stripes\n",
1318                oinfo->oi_md->lsm_object_id, oinfo->oi_md->lsm_stripe_count,
1319                oinfo->oi_md->lsm_stripe_size);
1320
1321         cfs_list_for_each (pos, &lovset->set_list) {
1322                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1323
1324                 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
1325                        "%u\n", oinfo->oi_oa->o_id, req->rq_stripe,
1326                        req->rq_oi.oi_oa->o_id, req->rq_idx);
1327                 rc = obd_getattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1328                                        &req->rq_oi, rqset);
1329                 if (rc) {
1330                         CERROR("error: getattr objid "LPX64" subobj "
1331                                LPX64" on OST idx %d: rc = %d\n",
1332                                oinfo->oi_oa->o_id, req->rq_oi.oi_oa->o_id,
1333                                req->rq_idx, rc);
1334                         GOTO(out, rc);
1335                 }
1336         }
1337
1338         if (!cfs_list_empty(&rqset->set_requests)) {
1339                 LASSERT(rc == 0);
1340                 LASSERT (rqset->set_interpret == NULL);
1341                 rqset->set_interpret = lov_getattr_interpret;
1342                 rqset->set_arg = (void *)lovset;
1343                 RETURN(rc);
1344         }
1345 out:
1346         if (rc)
1347                 lovset->set_completes = 0;
1348         err = lov_fini_getattr_set(lovset);
1349         RETURN(rc ? rc : err);
1350 }
1351
1352 static int lov_setattr(struct obd_export *exp, struct obd_info *oinfo,
1353                        struct obd_trans_info *oti)
1354 {
1355         struct lov_request_set *set;
1356         struct lov_obd *lov;
1357         cfs_list_t *pos;
1358         struct lov_request *req;
1359         int err = 0, rc = 0;
1360         ENTRY;
1361
1362         LASSERT(oinfo);
1363         ASSERT_LSM_MAGIC(oinfo->oi_md);
1364
1365         if (!exp || !exp->exp_obd)
1366                 RETURN(-ENODEV);
1367
1368         /* for now, we only expect the following updates here */
1369         LASSERT(!(oinfo->oi_oa->o_valid & ~(OBD_MD_FLID | OBD_MD_FLTYPE |
1370                                             OBD_MD_FLMODE | OBD_MD_FLATIME |
1371                                             OBD_MD_FLMTIME | OBD_MD_FLCTIME |
1372                                             OBD_MD_FLFLAGS | OBD_MD_FLSIZE |
1373                                             OBD_MD_FLGROUP | OBD_MD_FLUID |
1374                                             OBD_MD_FLGID | OBD_MD_FLFID |
1375                                             OBD_MD_FLGENER)));
1376         lov = &exp->exp_obd->u.lov;
1377         rc = lov_prep_setattr_set(exp, oinfo, oti, &set);
1378         if (rc)
1379                 RETURN(rc);
1380
1381         cfs_list_for_each (pos, &set->set_list) {
1382                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1383
1384                 rc = obd_setattr(lov->lov_tgts[req->rq_idx]->ltd_exp,
1385                                  &req->rq_oi, NULL);
1386                 err = lov_update_setattr_set(set, req, rc);
1387                 if (err) {
1388                         CERROR("error: setattr objid "LPX64" subobj "
1389                                LPX64" on OST idx %d: rc = %d\n",
1390                                set->set_oi->oi_oa->o_id,
1391                                req->rq_oi.oi_oa->o_id, req->rq_idx, err);
1392                         if (!rc)
1393                                 rc = err;
1394                 }
1395         }
1396         err = lov_fini_setattr_set(set);
1397         if (!rc)
1398                 rc = err;
1399         RETURN(rc);
1400 }
1401
1402 static int lov_setattr_interpret(struct ptlrpc_request_set *rqset,
1403                                  void *data, int rc)
1404 {
1405         struct lov_request_set *lovset = (struct lov_request_set *)data;
1406         int err;
1407         ENTRY;
1408
1409         if (rc)
1410                 lovset->set_completes = 0;
1411         err = lov_fini_setattr_set(lovset);
1412         RETURN(rc ? rc : err);
1413 }
1414
1415 /* If @oti is given, the request goes from MDS and responses from OSTs are not
1416    needed. Otherwise, a client is waiting for responses. */
1417 static int lov_setattr_async(struct obd_export *exp, struct obd_info *oinfo,
1418                              struct obd_trans_info *oti,
1419                              struct ptlrpc_request_set *rqset)
1420 {
1421         struct lov_request_set *set;
1422         struct lov_request *req;
1423         cfs_list_t *pos;
1424         struct lov_obd *lov;
1425         int rc = 0;
1426         ENTRY;
1427
1428         LASSERT(oinfo);
1429         ASSERT_LSM_MAGIC(oinfo->oi_md);
1430         if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE) {
1431                 LASSERT(oti);
1432                 LASSERT(oti->oti_logcookies);
1433         }
1434
1435         if (!exp || !exp->exp_obd)
1436                 RETURN(-ENODEV);
1437
1438         lov = &exp->exp_obd->u.lov;
1439         rc = lov_prep_setattr_set(exp, oinfo, oti, &set);
1440         if (rc)
1441                 RETURN(rc);
1442
1443         CDEBUG(D_INFO, "objid "LPX64": %ux%u byte stripes\n",
1444                oinfo->oi_md->lsm_object_id, oinfo->oi_md->lsm_stripe_count,
1445                oinfo->oi_md->lsm_stripe_size);
1446
1447         cfs_list_for_each (pos, &set->set_list) {
1448                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1449
1450                 if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE)
1451                         oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1452
1453                 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
1454                        "%u\n", oinfo->oi_oa->o_id, req->rq_stripe,
1455                        req->rq_oi.oi_oa->o_id, req->rq_idx);
1456
1457                 rc = obd_setattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1458                                        &req->rq_oi, oti, rqset);
1459                 if (rc) {
1460                         CERROR("error: setattr objid "LPX64" subobj "
1461                                LPX64" on OST idx %d: rc = %d\n",
1462                                set->set_oi->oi_oa->o_id,
1463                                req->rq_oi.oi_oa->o_id,
1464                                req->rq_idx, rc);
1465                         break;
1466                 }
1467         }
1468
1469         /* If we are not waiting for responses on async requests, return. */
1470         if (rc || !rqset || cfs_list_empty(&rqset->set_requests)) {
1471                 int err;
1472                 if (rc)
1473                         set->set_completes = 0;
1474                 err = lov_fini_setattr_set(set);
1475                 RETURN(rc ? rc : err);
1476         }
1477
1478         LASSERT(rqset->set_interpret == NULL);
1479         rqset->set_interpret = lov_setattr_interpret;
1480         rqset->set_arg = (void *)set;
1481
1482         RETURN(0);
1483 }
1484
1485 static int lov_punch_interpret(struct ptlrpc_request_set *rqset,
1486                                void *data, int rc)
1487 {
1488         struct lov_request_set *lovset = (struct lov_request_set *)data;
1489         int err;
1490         ENTRY;
1491
1492         if (rc)
1493                 lovset->set_completes = 0;
1494         err = lov_fini_punch_set(lovset);
1495         RETURN(rc ? rc : err);
1496 }
1497
1498 /* FIXME: maybe we'll just make one node the authoritative attribute node, then
1499  * we can send this 'punch' to just the authoritative node and the nodes
1500  * that the punch will affect. */
1501 static int lov_punch(struct obd_export *exp, struct obd_info *oinfo,
1502                      struct obd_trans_info *oti,
1503                      struct ptlrpc_request_set *rqset)
1504 {
1505         struct lov_request_set *set;
1506         struct lov_obd *lov;
1507         cfs_list_t *pos;
1508         struct lov_request *req;
1509         int rc = 0;
1510         ENTRY;
1511
1512         LASSERT(oinfo);
1513         ASSERT_LSM_MAGIC(oinfo->oi_md);
1514
1515         if (!exp || !exp->exp_obd)
1516                 RETURN(-ENODEV);
1517
1518         lov = &exp->exp_obd->u.lov;
1519         rc = lov_prep_punch_set(exp, oinfo, oti, &set);
1520         if (rc)
1521                 RETURN(rc);
1522
1523         cfs_list_for_each (pos, &set->set_list) {
1524                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1525
1526                 rc = obd_punch(lov->lov_tgts[req->rq_idx]->ltd_exp,
1527                                &req->rq_oi, NULL, rqset);
1528                 if (rc) {
1529                         CERROR("error: punch objid "LPX64" subobj "LPX64
1530                                " on OST idx %d: rc = %d\n",
1531                                set->set_oi->oi_oa->o_id,
1532                                req->rq_oi.oi_oa->o_id, req->rq_idx, rc);
1533                         break;
1534                 }
1535         }
1536
1537         if (rc || cfs_list_empty(&rqset->set_requests)) {
1538                 int err;
1539                 err = lov_fini_punch_set(set);
1540                 RETURN(rc ? rc : err);
1541         }
1542
1543         LASSERT(rqset->set_interpret == NULL);
1544         rqset->set_interpret = lov_punch_interpret;
1545         rqset->set_arg = (void *)set;
1546
1547         RETURN(0);
1548 }
1549
1550 static int lov_sync(struct obd_export *exp, struct obdo *oa,
1551                     struct lov_stripe_md *lsm, obd_off start, obd_off end,
1552                     void *capa)
1553 {
1554         struct lov_request_set *set;
1555         struct obd_info oinfo;
1556         struct lov_obd *lov;
1557         cfs_list_t *pos;
1558         struct lov_request *req;
1559         int err = 0, rc = 0;
1560         ENTRY;
1561
1562         ASSERT_LSM_MAGIC(lsm);
1563
1564         if (!exp->exp_obd)
1565                 RETURN(-ENODEV);
1566
1567         lov = &exp->exp_obd->u.lov;
1568         rc = lov_prep_sync_set(exp, &oinfo, oa, lsm, start, end, &set);
1569         if (rc)
1570                 RETURN(rc);
1571
1572         cfs_list_for_each (pos, &set->set_list) {
1573                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1574
1575                 rc = obd_sync(lov->lov_tgts[req->rq_idx]->ltd_exp,
1576                               req->rq_oi.oi_oa, NULL,
1577                               req->rq_oi.oi_policy.l_extent.start,
1578                               req->rq_oi.oi_policy.l_extent.end, capa);
1579                 err = lov_update_common_set(set, req, rc);
1580                 if (err) {
1581                         CERROR("error: fsync objid "LPX64" subobj "LPX64
1582                                " on OST idx %d: rc = %d\n",
1583                                set->set_oi->oi_oa->o_id,
1584                                req->rq_oi.oi_oa->o_id, req->rq_idx, rc);
1585                         if (!rc)
1586                                 rc = err;
1587                 }
1588         }
1589         err = lov_fini_sync_set(set);
1590         if (!rc)
1591                 rc = err;
1592         RETURN(rc);
1593 }
1594
1595 static int lov_sync_fs(struct obd_export *exp, struct obd_info *dummy,
1596                        int wait)
1597 {
1598         struct lov_obd *lov;
1599         struct obd_info oinfo = { { { 0 } } };
1600         struct lov_request *req;
1601         struct lov_request_set *set;
1602         struct l_wait_info  lwi = { 0 };
1603         cfs_list_t *pos;
1604         int rc = 0;
1605         ENTRY;
1606
1607         lov = &exp->exp_obd->u.lov;
1608         rc  = lov_prep_sync_fs_set(exp, &oinfo, &set);
1609         if (rc)
1610                 RETURN(rc);
1611
1612         cfs_list_for_each(pos, &set->set_list) {
1613                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1614                 rc = obd_sync_fs(lov->lov_tgts[req->rq_idx]->ltd_exp,
1615                                  &req->rq_oi, wait);
1616                 if (rc)
1617                         break;
1618         }
1619         /* if wait then check if all sync_fs IO's are done */
1620         if (wait)
1621                 l_wait_event(set->set_waitq, lov_finished_set(set), &lwi);
1622
1623         rc = lov_fini_sync_fs_set(set);
1624
1625         RETURN(rc);
1626 }
1627
1628 static int lov_brw_check(struct lov_obd *lov, struct obd_info *lov_oinfo,
1629                          obd_count oa_bufs, struct brw_page *pga)
1630 {
1631         struct obd_info oinfo = { { { 0 } } };
1632         int i, rc = 0;
1633
1634         oinfo.oi_oa = lov_oinfo->oi_oa;
1635
1636         /* The caller just wants to know if there's a chance that this
1637          * I/O can succeed */
1638         for (i = 0; i < oa_bufs; i++) {
1639                 int stripe = lov_stripe_number(lov_oinfo->oi_md, pga[i].off);
1640                 int ost = lov_oinfo->oi_md->lsm_oinfo[stripe]->loi_ost_idx;
1641                 obd_off start, end;
1642
1643                 if (!lov_stripe_intersects(lov_oinfo->oi_md, i, pga[i].off,
1644                                            pga[i].off + pga[i].count - 1,
1645                                            &start, &end))
1646                         continue;
1647
1648                 if (!lov->lov_tgts[ost] || !lov->lov_tgts[ost]->ltd_active) {
1649                         CDEBUG(D_HA, "lov idx %d inactive\n", ost);
1650                         return -EIO;
1651                 }
1652
1653                 rc = obd_brw(OBD_BRW_CHECK, lov->lov_tgts[ost]->ltd_exp, &oinfo,
1654                              1, &pga[i], NULL);
1655                 if (rc)
1656                         break;
1657         }
1658         return rc;
1659 }
1660
1661 static int lov_brw(int cmd, struct obd_export *exp, struct obd_info *oinfo,
1662                    obd_count oa_bufs, struct brw_page *pga,
1663                    struct obd_trans_info *oti)
1664 {
1665         struct lov_request_set *set;
1666         struct lov_request *req;
1667         cfs_list_t *pos;
1668         struct lov_obd *lov = &exp->exp_obd->u.lov;
1669         int err, rc = 0;
1670         ENTRY;
1671
1672         ASSERT_LSM_MAGIC(oinfo->oi_md);
1673
1674         if (cmd == OBD_BRW_CHECK) {
1675                 rc = lov_brw_check(lov, oinfo, oa_bufs, pga);
1676                 RETURN(rc);
1677         }
1678
1679         rc = lov_prep_brw_set(exp, oinfo, oa_bufs, pga, oti, &set);
1680         if (rc)
1681                 RETURN(rc);
1682
1683         cfs_list_for_each (pos, &set->set_list) {
1684                 struct obd_export *sub_exp;
1685                 struct brw_page *sub_pga;
1686                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1687
1688                 sub_exp = lov->lov_tgts[req->rq_idx]->ltd_exp;
1689                 sub_pga = set->set_pga + req->rq_pgaidx;
1690                 rc = obd_brw(cmd, sub_exp, &req->rq_oi, req->rq_oabufs,
1691                              sub_pga, oti);
1692                 if (rc)
1693                         break;
1694                 lov_update_common_set(set, req, rc);
1695         }
1696
1697         err = lov_fini_brw_set(set);
1698         if (!rc)
1699                 rc = err;
1700         RETURN(rc);
1701 }
1702
1703 static int lov_enqueue_interpret(struct ptlrpc_request_set *rqset,
1704                                  void *data, int rc)
1705 {
1706         struct lov_request_set *lovset = (struct lov_request_set *)data;
1707         ENTRY;
1708         rc = lov_fini_enqueue_set(lovset, lovset->set_ei->ei_mode, rc, rqset);
1709         RETURN(rc);
1710 }
1711
1712 static int lov_enqueue(struct obd_export *exp, struct obd_info *oinfo,
1713                        struct ldlm_enqueue_info *einfo,
1714                        struct ptlrpc_request_set *rqset)
1715 {
1716         ldlm_mode_t mode = einfo->ei_mode;
1717         struct lov_request_set *set;
1718         struct lov_request *req;
1719         cfs_list_t *pos;
1720         struct lov_obd *lov;
1721         ldlm_error_t rc;
1722         ENTRY;
1723
1724         LASSERT(oinfo);
1725         ASSERT_LSM_MAGIC(oinfo->oi_md);
1726         LASSERT(mode == (mode & -mode));
1727
1728         /* we should never be asked to replay a lock this way. */
1729         LASSERT((oinfo->oi_flags & LDLM_FL_REPLAY) == 0);
1730
1731         if (!exp || !exp->exp_obd)
1732                 RETURN(-ENODEV);
1733
1734         lov = &exp->exp_obd->u.lov;
1735         rc = lov_prep_enqueue_set(exp, oinfo, einfo, &set);
1736         if (rc)
1737                 RETURN(rc);
1738
1739         cfs_list_for_each (pos, &set->set_list) {
1740                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1741
1742                 rc = obd_enqueue(lov->lov_tgts[req->rq_idx]->ltd_exp,
1743                                  &req->rq_oi, einfo, rqset);
1744                 if (rc != ELDLM_OK)
1745                         GOTO(out, rc);
1746         }
1747
1748         if (rqset && !cfs_list_empty(&rqset->set_requests)) {
1749                 LASSERT(rc == 0);
1750                 LASSERT(rqset->set_interpret == NULL);
1751                 rqset->set_interpret = lov_enqueue_interpret;
1752                 rqset->set_arg = (void *)set;
1753                 RETURN(rc);
1754         }
1755 out:
1756         rc = lov_fini_enqueue_set(set, mode, rc, rqset);
1757         RETURN(rc);
1758 }
1759
1760 static int lov_change_cbdata(struct obd_export *exp,
1761                              struct lov_stripe_md *lsm, ldlm_iterator_t it,
1762                              void *data)
1763 {
1764         struct lov_obd *lov;
1765         int rc = 0, i;
1766         ENTRY;
1767
1768         ASSERT_LSM_MAGIC(lsm);
1769
1770         if (!exp || !exp->exp_obd)
1771                 RETURN(-ENODEV);
1772
1773         lov = &exp->exp_obd->u.lov;
1774         for (i = 0; i < lsm->lsm_stripe_count; i++) {
1775                 struct lov_stripe_md submd;
1776                 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
1777
1778                 if (!lov->lov_tgts[loi->loi_ost_idx]) {
1779                         CDEBUG(D_HA, "lov idx %d NULL \n", loi->loi_ost_idx);
1780                         continue;
1781                 }
1782
1783                 LASSERT_SEQ_IS_MDT(loi->loi_seq);
1784                 submd.lsm_object_id = loi->loi_id;
1785                 submd.lsm_object_seq = loi->loi_seq;
1786                 submd.lsm_stripe_count = 0;
1787                 rc = obd_change_cbdata(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1788                                        &submd, it, data);
1789         }
1790         RETURN(rc);
1791 }
1792
1793 /* find any ldlm lock of the inode in lov
1794  * return 0    not find
1795  *        1    find one
1796  *      < 0    error */
1797 static int lov_find_cbdata(struct obd_export *exp,
1798                            struct lov_stripe_md *lsm, ldlm_iterator_t it,
1799                            void *data)
1800 {
1801         struct lov_obd *lov;
1802         int rc = 0, i;
1803         ENTRY;
1804
1805         ASSERT_LSM_MAGIC(lsm);
1806
1807         if (!exp || !exp->exp_obd)
1808                 RETURN(-ENODEV);
1809
1810         lov = &exp->exp_obd->u.lov;
1811         for (i = 0; i < lsm->lsm_stripe_count; i++) {
1812                 struct lov_stripe_md submd;
1813                 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
1814
1815                 if (!lov->lov_tgts[loi->loi_ost_idx]) {
1816                         CDEBUG(D_HA, "lov idx %d NULL \n", loi->loi_ost_idx);
1817                         continue;
1818                 }
1819
1820                 LASSERT_SEQ_IS_MDT(loi->loi_seq);
1821                 submd.lsm_object_id = loi->loi_id;
1822                 submd.lsm_object_seq = loi->loi_seq;
1823                 submd.lsm_stripe_count = 0;
1824                 rc = obd_find_cbdata(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1825                                      &submd, it, data);
1826                 if (rc != 0)
1827                         RETURN(rc);
1828         }
1829         RETURN(rc);
1830 }
1831
1832 static int lov_cancel(struct obd_export *exp, struct lov_stripe_md *lsm,
1833                       __u32 mode, struct lustre_handle *lockh)
1834 {
1835         struct lov_request_set *set;
1836         struct obd_info oinfo;
1837         struct lov_request *req;
1838         cfs_list_t *pos;
1839         struct lov_obd *lov;
1840         struct lustre_handle *lov_lockhp;
1841         int err = 0, rc = 0;
1842         ENTRY;
1843
1844         ASSERT_LSM_MAGIC(lsm);
1845
1846         if (!exp || !exp->exp_obd)
1847                 RETURN(-ENODEV);
1848
1849         LASSERT_SEQ_IS_MDT(lsm->lsm_object_seq);
1850         LASSERT(lockh);
1851         lov = &exp->exp_obd->u.lov;
1852         rc = lov_prep_cancel_set(exp, &oinfo, lsm, mode, lockh, &set);
1853         if (rc)
1854                 RETURN(rc);
1855
1856         cfs_list_for_each (pos, &set->set_list) {
1857                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1858                 lov_lockhp = set->set_lockh->llh_handles + req->rq_stripe;
1859
1860                 rc = obd_cancel(lov->lov_tgts[req->rq_idx]->ltd_exp,
1861                                 req->rq_oi.oi_md, mode, lov_lockhp);
1862                 rc = lov_update_common_set(set, req, rc);
1863                 if (rc) {
1864                         CERROR("error: cancel objid "LPX64" subobj "
1865                                LPX64" on OST idx %d: rc = %d\n",
1866                                lsm->lsm_object_id,
1867                                req->rq_oi.oi_md->lsm_object_id,
1868                                req->rq_idx, rc);
1869                         err = rc;
1870                 }
1871
1872         }
1873         lov_fini_cancel_set(set);
1874         RETURN(err);
1875 }
1876
1877 static int lov_cancel_unused(struct obd_export *exp,
1878                              struct lov_stripe_md *lsm,
1879                              ldlm_cancel_flags_t flags, void *opaque)
1880 {
1881         struct lov_obd *lov;
1882         int rc = 0, i;
1883         ENTRY;
1884
1885         if (!exp || !exp->exp_obd)
1886                 RETURN(-ENODEV);
1887
1888         lov = &exp->exp_obd->u.lov;
1889         if (lsm == NULL) {
1890                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
1891                         int err;
1892                         if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
1893                                 continue;
1894
1895                         err = obd_cancel_unused(lov->lov_tgts[i]->ltd_exp, NULL,
1896                                                 flags, opaque);
1897                         if (!rc)
1898                                 rc = err;
1899                 }
1900                 RETURN(rc);
1901         }
1902
1903         ASSERT_LSM_MAGIC(lsm);
1904
1905         LASSERT_SEQ_IS_MDT(lsm->lsm_object_seq);
1906         for (i = 0; i < lsm->lsm_stripe_count; i++) {
1907                 struct lov_stripe_md submd;
1908                 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
1909                 int err;
1910
1911                 if (!lov->lov_tgts[loi->loi_ost_idx]) {
1912                         CDEBUG(D_HA, "lov idx %d NULL\n", loi->loi_ost_idx);
1913                         continue;
1914                 }
1915
1916                 if (!lov->lov_tgts[loi->loi_ost_idx]->ltd_active)
1917                         CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
1918
1919                 submd.lsm_object_id = loi->loi_id;
1920                 submd.lsm_object_seq = loi->loi_seq;
1921                 submd.lsm_stripe_count = 0;
1922                 err = obd_cancel_unused(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1923                                         &submd, flags, opaque);
1924                 if (err && lov->lov_tgts[loi->loi_ost_idx]->ltd_active) {
1925                         CERROR("error: cancel unused objid "LPX64" subobj "LPX64
1926                                " on OST idx %d: rc = %d\n", lsm->lsm_object_id,
1927                                loi->loi_id, loi->loi_ost_idx, err);
1928                         if (!rc)
1929                                 rc = err;
1930                 }
1931         }
1932         RETURN(rc);
1933 }
1934
1935 int lov_statfs_interpret(struct ptlrpc_request_set *rqset, void *data, int rc)
1936 {
1937         struct lov_request_set *lovset = (struct lov_request_set *)data;
1938         int err;
1939         ENTRY;
1940
1941         if (rc)
1942                 lovset->set_completes = 0;
1943
1944         err = lov_fini_statfs_set(lovset);
1945         RETURN(rc ? rc : err);
1946 }
1947
1948 static int lov_statfs_async(struct obd_device *obd, struct obd_info *oinfo,
1949                             __u64 max_age, struct ptlrpc_request_set *rqset)
1950 {
1951         struct lov_request_set *set;
1952         struct lov_request *req;
1953         cfs_list_t *pos;
1954         struct lov_obd *lov;
1955         int rc = 0;
1956         ENTRY;
1957
1958         LASSERT(oinfo != NULL);
1959         LASSERT(oinfo->oi_osfs != NULL);
1960
1961         lov = &obd->u.lov;
1962         rc = lov_prep_statfs_set(obd, oinfo, &set);
1963         if (rc)
1964                 RETURN(rc);
1965
1966         cfs_list_for_each (pos, &set->set_list) {
1967                 struct obd_device *osc_obd;
1968
1969                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1970
1971                 osc_obd = class_exp2obd(lov->lov_tgts[req->rq_idx]->ltd_exp);
1972                 rc = obd_statfs_async(osc_obd, &req->rq_oi, max_age, rqset);
1973                 if (rc)
1974                         break;
1975         }
1976
1977         if (rc || cfs_list_empty(&rqset->set_requests)) {
1978                 int err;
1979                 if (rc)
1980                         set->set_completes = 0;
1981                 err = lov_fini_statfs_set(set);
1982                 RETURN(rc ? rc : err);
1983         }
1984
1985         LASSERT(rqset->set_interpret == NULL);
1986         rqset->set_interpret = lov_statfs_interpret;
1987         rqset->set_arg = (void *)set;
1988         RETURN(0);
1989 }
1990
1991 static int lov_statfs(struct obd_device *obd, struct obd_statfs *osfs,
1992                       __u64 max_age, __u32 flags)
1993 {
1994         struct ptlrpc_request_set *set = NULL;
1995         struct obd_info oinfo = { { { 0 } } };
1996         int rc = 0;
1997         ENTRY;
1998
1999
2000         /* for obdclass we forbid using obd_statfs_rqset, but prefer using async
2001          * statfs requests */
2002         set = ptlrpc_prep_set();
2003         if (set == NULL)
2004                 RETURN(-ENOMEM);
2005
2006         oinfo.oi_osfs = osfs;
2007         oinfo.oi_flags = flags;
2008         rc = lov_statfs_async(obd, &oinfo, max_age, set);
2009         if (rc == 0)
2010                 rc = ptlrpc_set_wait(set);
2011         ptlrpc_set_destroy(set);
2012
2013         RETURN(rc);
2014 }
2015
2016 static int lov_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
2017                          void *karg, void *uarg)
2018 {
2019         struct obd_device *obddev = class_exp2obd(exp);
2020         struct lov_obd *lov = &obddev->u.lov;
2021         int i = 0, rc = 0, count = lov->desc.ld_tgt_count;
2022         struct obd_uuid *uuidp;
2023         ENTRY;
2024
2025         switch (cmd) {
2026         case IOC_OBD_STATFS: {
2027                 struct obd_ioctl_data *data = karg;
2028                 struct obd_device *osc_obd;
2029                 struct obd_statfs stat_buf = {0};
2030                 __u32 index;
2031
2032                 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
2033                 if ((index >= count))
2034                         RETURN(-ENODEV);
2035
2036                 if (!lov->lov_tgts[index])
2037                         /* Try again with the next index */
2038                         RETURN(-EAGAIN);
2039                 if (!lov->lov_tgts[index]->ltd_active)
2040                         RETURN(-ENODATA);
2041
2042                 osc_obd = class_exp2obd(lov->lov_tgts[index]->ltd_exp);
2043                 if (!osc_obd)
2044                         RETURN(-EINVAL);
2045
2046                 /* copy UUID */
2047                 if (cfs_copy_to_user(data->ioc_pbuf2, obd2cli_tgt(osc_obd),
2048                                      min((int) data->ioc_plen2,
2049                                          (int) sizeof(struct obd_uuid))))
2050                         RETURN(-EFAULT);
2051
2052                 /* got statfs data */
2053                 rc = obd_statfs(osc_obd, &stat_buf,
2054                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
2055                                 0);
2056                 if (rc)
2057                         RETURN(rc);
2058                 if (cfs_copy_to_user(data->ioc_pbuf1, &stat_buf,
2059                                      min((int) data->ioc_plen1,
2060                                          (int) sizeof(stat_buf))))
2061                         RETURN(-EFAULT);
2062                 break;
2063         }
2064         case OBD_IOC_LOV_GET_CONFIG: {
2065                 struct obd_ioctl_data *data;
2066                 struct lov_desc *desc;
2067                 char *buf = NULL;
2068                 __u32 *genp;
2069
2070                 len = 0;
2071                 if (obd_ioctl_getdata(&buf, &len, (void *)uarg))
2072                         RETURN(-EINVAL);
2073
2074                 data = (struct obd_ioctl_data *)buf;
2075
2076                 if (sizeof(*desc) > data->ioc_inllen1) {
2077                         obd_ioctl_freedata(buf, len);
2078                         RETURN(-EINVAL);
2079                 }
2080
2081                 if (sizeof(uuidp->uuid) * count > data->ioc_inllen2) {
2082                         obd_ioctl_freedata(buf, len);
2083                         RETURN(-EINVAL);
2084                 }
2085
2086                 if (sizeof(__u32) * count > data->ioc_inllen3) {
2087                         obd_ioctl_freedata(buf, len);
2088                         RETURN(-EINVAL);
2089                 }
2090
2091                 desc = (struct lov_desc *)data->ioc_inlbuf1;
2092                 memcpy(desc, &(lov->desc), sizeof(*desc));
2093
2094                 uuidp = (struct obd_uuid *)data->ioc_inlbuf2;
2095                 genp = (__u32 *)data->ioc_inlbuf3;
2096                 /* the uuid will be empty for deleted OSTs */
2097                 for (i = 0; i < count; i++, uuidp++, genp++) {
2098                         if (!lov->lov_tgts[i])
2099                                 continue;
2100                         *uuidp = lov->lov_tgts[i]->ltd_uuid;
2101                         *genp = lov->lov_tgts[i]->ltd_gen;
2102                 }
2103
2104                 if (cfs_copy_to_user((void *)uarg, buf, len))
2105                         rc = -EFAULT;
2106                 obd_ioctl_freedata(buf, len);
2107                 break;
2108         }
2109         case LL_IOC_LOV_SETSTRIPE:
2110                 rc = lov_setstripe(exp, len, karg, uarg);
2111                 break;
2112         case LL_IOC_LOV_GETSTRIPE:
2113                 rc = lov_getstripe(exp, karg, uarg);
2114                 break;
2115         case LL_IOC_LOV_SETEA:
2116                 rc = lov_setea(exp, karg, uarg);
2117                 break;
2118         case OBD_IOC_QUOTACTL: {
2119                 struct if_quotactl *qctl = karg;
2120                 struct lov_tgt_desc *tgt = NULL;
2121                 struct obd_quotactl *oqctl;
2122
2123                 if (qctl->qc_valid == QC_OSTIDX) {
2124                         if (qctl->qc_idx < 0 || count <= qctl->qc_idx)
2125                                 RETURN(-EINVAL);
2126
2127                         tgt = lov->lov_tgts[qctl->qc_idx];
2128                         if (!tgt || !tgt->ltd_exp)
2129                                 RETURN(-EINVAL);
2130                 } else if (qctl->qc_valid == QC_UUID) {
2131                         for (i = 0; i < count; i++) {
2132                                 tgt = lov->lov_tgts[i];
2133                                 if (!tgt ||
2134                                     !obd_uuid_equals(&tgt->ltd_uuid,
2135                                                      &qctl->obd_uuid))
2136                                         continue;
2137
2138                                 if (tgt->ltd_exp == NULL)
2139                                         RETURN(-EINVAL);
2140
2141                                 break;
2142                         }
2143                 } else {
2144                         RETURN(-EINVAL);
2145                 }
2146
2147                 if (i >= count)
2148                         RETURN(-EAGAIN);
2149
2150                 LASSERT(tgt && tgt->ltd_exp);
2151                 OBD_ALLOC_PTR(oqctl);
2152                 if (!oqctl)
2153                         RETURN(-ENOMEM);
2154
2155                 QCTL_COPY(oqctl, qctl);
2156                 rc = obd_quotactl(tgt->ltd_exp, oqctl);
2157                 if (rc == 0) {
2158                         QCTL_COPY(qctl, oqctl);
2159                         qctl->qc_valid = QC_OSTIDX;
2160                         qctl->obd_uuid = tgt->ltd_uuid;
2161                 }
2162                 OBD_FREE_PTR(oqctl);
2163                 break;
2164         }
2165         default: {
2166                 int set = 0;
2167
2168                 if (count == 0)
2169                         RETURN(-ENOTTY);
2170
2171                 for (i = 0; i < count; i++) {
2172                         int err;
2173                         struct obd_device *osc_obd;
2174
2175                         /* OST was disconnected */
2176                         if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
2177                                 continue;
2178
2179                         /* ll_umount_begin() sets force flag but for lov, not
2180                          * osc. Let's pass it through */
2181                         osc_obd = class_exp2obd(lov->lov_tgts[i]->ltd_exp);
2182                         osc_obd->obd_force = obddev->obd_force;
2183                         err = obd_iocontrol(cmd, lov->lov_tgts[i]->ltd_exp,
2184                                             len, karg, uarg);
2185                         if (err == -ENODATA && cmd == OBD_IOC_POLL_QUOTACHECK) {
2186                                 RETURN(err);
2187                         } else if (err) {
2188                                 if (lov->lov_tgts[i]->ltd_active) {
2189                                         CDEBUG(err == -ENOTTY ?
2190                                                D_IOCTL : D_WARNING,
2191                                                "iocontrol OSC %s on OST "
2192                                                "idx %d cmd %x: err = %d\n",
2193                                                lov_uuid2str(lov, i),
2194                                                i, cmd, err);
2195                                         if (!rc)
2196                                                 rc = err;
2197                                 }
2198                         } else {
2199                                 set = 1;
2200                         }
2201                 }
2202                 if (!set && !rc)
2203                         rc = -EIO;
2204         }
2205         }
2206
2207         RETURN(rc);
2208 }
2209
2210 #define FIEMAP_BUFFER_SIZE 4096
2211
2212 /**
2213  * Non-zero fe_logical indicates that this is a continuation FIEMAP
2214  * call. The local end offset and the device are sent in the first
2215  * fm_extent. This function calculates the stripe number from the index.
2216  * This function returns a stripe_no on which mapping is to be restarted.
2217  *
2218  * This function returns fm_end_offset which is the in-OST offset at which
2219  * mapping should be restarted. If fm_end_offset=0 is returned then caller
2220  * will re-calculate proper offset in next stripe.
2221  * Note that the first extent is passed to lov_get_info via the value field.
2222  *
2223  * \param fiemap fiemap request header
2224  * \param lsm striping information for the file
2225  * \param fm_start logical start of mapping
2226  * \param fm_end logical end of mapping
2227  * \param start_stripe starting stripe will be returned in this
2228  */
2229 obd_size fiemap_calc_fm_end_offset(struct ll_user_fiemap *fiemap,
2230                                    struct lov_stripe_md *lsm, obd_size fm_start,
2231                                    obd_size fm_end, int *start_stripe)
2232 {
2233         obd_size local_end = fiemap->fm_extents[0].fe_logical;
2234         obd_off lun_start, lun_end;
2235         obd_size fm_end_offset;
2236         int stripe_no = -1, i;
2237
2238         if (fiemap->fm_extent_count == 0 ||
2239             fiemap->fm_extents[0].fe_logical == 0)
2240                 return 0;
2241
2242         /* Find out stripe_no from ost_index saved in the fe_device */
2243         for (i = 0; i < lsm->lsm_stripe_count; i++) {
2244                 if (lsm->lsm_oinfo[i]->loi_ost_idx ==
2245                                         fiemap->fm_extents[0].fe_device) {
2246                         stripe_no = i;
2247                         break;
2248                 }
2249         }
2250
2251         /* If we have finished mapping on previous device, shift logical
2252          * offset to start of next device */
2253         if ((lov_stripe_intersects(lsm, stripe_no, fm_start, fm_end,
2254                                    &lun_start, &lun_end)) != 0 &&
2255                                    local_end < lun_end) {
2256                 fm_end_offset = local_end;
2257                 *start_stripe = stripe_no;
2258         } else {
2259                 /* This is a special value to indicate that caller should
2260                  * calculate offset in next stripe. */
2261                 fm_end_offset = 0;
2262                 *start_stripe = (stripe_no + 1) % lsm->lsm_stripe_count;
2263         }
2264
2265         return fm_end_offset;
2266 }
2267
2268 /**
2269  * We calculate on which OST the mapping will end. If the length of mapping
2270  * is greater than (stripe_size * stripe_count) then the last_stripe will
2271  * will be one just before start_stripe. Else we check if the mapping
2272  * intersects each OST and find last_stripe.
2273  * This function returns the last_stripe and also sets the stripe_count
2274  * over which the mapping is spread
2275  *
2276  * \param lsm striping information for the file
2277  * \param fm_start logical start of mapping
2278  * \param fm_end logical end of mapping
2279  * \param start_stripe starting stripe of the mapping
2280  * \param stripe_count the number of stripes across which to map is returned
2281  *
2282  * \retval last_stripe return the last stripe of the mapping
2283  */
2284 int fiemap_calc_last_stripe(struct lov_stripe_md *lsm, obd_size fm_start,
2285                             obd_size fm_end, int start_stripe,
2286                             int *stripe_count)
2287 {
2288         int last_stripe;
2289         obd_off obd_start, obd_end;
2290         int i, j;
2291
2292         if (fm_end - fm_start > lsm->lsm_stripe_size * lsm->lsm_stripe_count) {
2293                 last_stripe = (start_stripe < 1 ? lsm->lsm_stripe_count - 1 :
2294                                                               start_stripe - 1);
2295                 *stripe_count = lsm->lsm_stripe_count;
2296         } else {
2297                 for (j = 0, i = start_stripe; j < lsm->lsm_stripe_count;
2298                      i = (i + 1) % lsm->lsm_stripe_count, j++) {
2299                         if ((lov_stripe_intersects(lsm, i, fm_start, fm_end,
2300                                                    &obd_start, &obd_end)) == 0)
2301                                 break;
2302                 }
2303                 *stripe_count = j;
2304                 last_stripe = (start_stripe + j - 1) %lsm->lsm_stripe_count;
2305         }
2306
2307         return last_stripe;
2308 }
2309
2310 /**
2311  * Set fe_device and copy extents from local buffer into main return buffer.
2312  *
2313  * \param fiemap fiemap request header
2314  * \param lcl_fm_ext array of local fiemap extents to be copied
2315  * \param ost_index OST index to be written into the fm_device field for each
2316                     extent
2317  * \param ext_count number of extents to be copied
2318  * \param current_extent where to start copying in main extent array
2319  */
2320 void fiemap_prepare_and_copy_exts(struct ll_user_fiemap *fiemap,
2321                                   struct ll_fiemap_extent *lcl_fm_ext,
2322                                   int ost_index, unsigned int ext_count,
2323                                   int current_extent)
2324 {
2325         char *to;
2326         int ext;
2327
2328         for (ext = 0; ext < ext_count; ext++) {
2329                 lcl_fm_ext[ext].fe_device = ost_index;
2330                 lcl_fm_ext[ext].fe_flags |= FIEMAP_EXTENT_NET;
2331         }
2332
2333         /* Copy fm_extent's from fm_local to return buffer */
2334         to = (char *)fiemap + fiemap_count_to_size(current_extent);
2335         memcpy(to, lcl_fm_ext, ext_count * sizeof(struct ll_fiemap_extent));
2336 }
2337
2338 /**
2339  * Break down the FIEMAP request and send appropriate calls to individual OSTs.
2340  * This also handles the restarting of FIEMAP calls in case mapping overflows
2341  * the available number of extents in single call.
2342  */
2343 static int lov_fiemap(struct lov_obd *lov, __u32 keylen, void *key,
2344                       __u32 *vallen, void *val, struct lov_stripe_md *lsm)
2345 {
2346         struct ll_fiemap_info_key *fm_key = key;
2347         struct ll_user_fiemap *fiemap = val;
2348         struct ll_user_fiemap *fm_local = NULL;
2349         struct ll_fiemap_extent *lcl_fm_ext;
2350         int count_local;
2351         unsigned int get_num_extents = 0;
2352         int ost_index = 0, actual_start_stripe, start_stripe;
2353         obd_size fm_start, fm_end, fm_length, fm_end_offset = 0;
2354         obd_size curr_loc;
2355         int current_extent = 0, rc = 0, i;
2356         int ost_eof = 0; /* EOF for object */
2357         int ost_done = 0; /* done with required mapping for this OST? */
2358         int last_stripe;
2359         int cur_stripe = 0, cur_stripe_wrap = 0, stripe_count;
2360         unsigned int buffer_size = FIEMAP_BUFFER_SIZE;
2361
2362         if (lsm == NULL)
2363                 GOTO(out, rc = 0);
2364
2365         if (fiemap_count_to_size(fm_key->fiemap.fm_extent_count) < buffer_size)
2366                 buffer_size = fiemap_count_to_size(fm_key->fiemap.fm_extent_count);
2367
2368         OBD_ALLOC(fm_local, buffer_size);
2369         if (fm_local == NULL)
2370                 GOTO(out, rc = -ENOMEM);
2371         lcl_fm_ext = &fm_local->fm_extents[0];
2372
2373         count_local = fiemap_size_to_count(buffer_size);
2374
2375         memcpy(fiemap, &fm_key->fiemap, sizeof(*fiemap));
2376         fm_start = fiemap->fm_start;
2377         fm_length = fiemap->fm_length;
2378         /* Calculate start stripe, last stripe and length of mapping */
2379         actual_start_stripe = start_stripe = lov_stripe_number(lsm, fm_start);
2380         fm_end = (fm_length == ~0ULL ? fm_key->oa.o_size :
2381                                                 fm_start + fm_length - 1);
2382         /* If fm_length != ~0ULL but fm_start+fm_length-1 exceeds file size */
2383         if (fm_end > fm_key->oa.o_size)
2384                 fm_end = fm_key->oa.o_size;
2385
2386         last_stripe = fiemap_calc_last_stripe(lsm, fm_start, fm_end,
2387                                             actual_start_stripe, &stripe_count);
2388
2389         fm_end_offset = fiemap_calc_fm_end_offset(fiemap, lsm, fm_start, fm_end,
2390                                                   &start_stripe);
2391
2392         if (fiemap->fm_extent_count == 0) {
2393                 get_num_extents = 1;
2394                 count_local = 0;
2395         }
2396
2397         /* Check each stripe */
2398         for (cur_stripe = start_stripe, i = 0; i < stripe_count;
2399              i++, cur_stripe = (cur_stripe + 1) % lsm->lsm_stripe_count) {
2400                 obd_size req_fm_len; /* Stores length of required mapping */
2401                 obd_size len_mapped_single_call;
2402                 obd_off lun_start, lun_end, obd_object_end;
2403                 unsigned int ext_count;
2404
2405                 cur_stripe_wrap = cur_stripe;
2406
2407                 /* Find out range of mapping on this stripe */
2408                 if ((lov_stripe_intersects(lsm, cur_stripe, fm_start, fm_end,
2409                                            &lun_start, &obd_object_end)) == 0)
2410                         continue;
2411
2412                 /* If this is a continuation FIEMAP call and we are on
2413                  * starting stripe then lun_start needs to be set to
2414                  * fm_end_offset */
2415                 if (fm_end_offset != 0 && cur_stripe == start_stripe)
2416                         lun_start = fm_end_offset;
2417
2418                 if (fm_length != ~0ULL) {
2419                         /* Handle fm_start + fm_length overflow */
2420                         if (fm_start + fm_length < fm_start)
2421                                 fm_length = ~0ULL - fm_start;
2422                         lun_end = lov_size_to_stripe(lsm, fm_start + fm_length,
2423                                                      cur_stripe);
2424                 } else {
2425                         lun_end = ~0ULL;
2426                 }
2427
2428                 if (lun_start == lun_end)
2429                         continue;
2430
2431                 req_fm_len = obd_object_end - lun_start;
2432                 fm_local->fm_length = 0;
2433                 len_mapped_single_call = 0;
2434
2435                 /* If the output buffer is very large and the objects have many
2436                  * extents we may need to loop on a single OST repeatedly */
2437                 ost_eof = 0;
2438                 ost_done = 0;
2439                 do {
2440                         if (get_num_extents == 0) {
2441                                 /* Don't get too many extents. */
2442                                 if (current_extent + count_local >
2443                                     fiemap->fm_extent_count)
2444                                         count_local = fiemap->fm_extent_count -
2445                                                                  current_extent;
2446                         }
2447
2448                         lun_start += len_mapped_single_call;
2449                         fm_local->fm_length = req_fm_len - len_mapped_single_call;
2450                         req_fm_len = fm_local->fm_length;
2451                         fm_local->fm_extent_count = count_local;
2452                         fm_local->fm_mapped_extents = 0;
2453                         fm_local->fm_flags = fiemap->fm_flags;
2454
2455                         fm_key->oa.o_id = lsm->lsm_oinfo[cur_stripe]->loi_id;
2456                         fm_key->oa.o_seq = lsm->lsm_oinfo[cur_stripe]->loi_seq;
2457                         ost_index = lsm->lsm_oinfo[cur_stripe]->loi_ost_idx;
2458
2459                         if (ost_index < 0 || ost_index >=lov->desc.ld_tgt_count)
2460                                 GOTO(out, rc = -EINVAL);
2461
2462                         /* If OST is inactive, return extent with UNKNOWN flag */
2463                         if (!lov->lov_tgts[ost_index]->ltd_active) {
2464                                 fm_local->fm_flags |= FIEMAP_EXTENT_LAST;
2465                                 fm_local->fm_mapped_extents = 1;
2466
2467                                 lcl_fm_ext[0].fe_logical = lun_start;
2468                                 lcl_fm_ext[0].fe_length = obd_object_end -
2469                                                                       lun_start;
2470                                 lcl_fm_ext[0].fe_flags |= FIEMAP_EXTENT_UNKNOWN;
2471
2472                                 goto inactive_tgt;
2473                         }
2474
2475                         fm_local->fm_start = lun_start;
2476                         fm_local->fm_flags &= ~FIEMAP_FLAG_DEVICE_ORDER;
2477                         memcpy(&fm_key->fiemap, fm_local, sizeof(*fm_local));
2478                         *vallen=fiemap_count_to_size(fm_local->fm_extent_count);
2479                         rc = obd_get_info(lov->lov_tgts[ost_index]->ltd_exp,
2480                                           keylen, key, vallen, fm_local, lsm);
2481                         if (rc != 0)
2482                                 GOTO(out, rc);
2483
2484 inactive_tgt:
2485                         ext_count = fm_local->fm_mapped_extents;
2486                         if (ext_count == 0) {
2487                                 ost_done = 1;
2488                                 /* If last stripe has hole at the end,
2489                                  * then we need to return */
2490                                 if (cur_stripe_wrap == last_stripe) {
2491                                         fiemap->fm_mapped_extents = 0;
2492                                         goto finish;
2493                                 }
2494                                 break;
2495                         }
2496
2497                         /* If we just need num of extents then go to next device */
2498                         if (get_num_extents) {
2499                                 current_extent += ext_count;
2500                                 break;
2501                         }
2502
2503                         len_mapped_single_call = lcl_fm_ext[ext_count-1].fe_logical -
2504                                   lun_start + lcl_fm_ext[ext_count - 1].fe_length;
2505
2506                         /* Have we finished mapping on this device? */
2507                         if (req_fm_len <= len_mapped_single_call)
2508                                 ost_done = 1;
2509
2510                         /* Clear the EXTENT_LAST flag which can be present on
2511                          * last extent */
2512                         if (lcl_fm_ext[ext_count-1].fe_flags & FIEMAP_EXTENT_LAST)
2513                                 lcl_fm_ext[ext_count - 1].fe_flags &=
2514                                                             ~FIEMAP_EXTENT_LAST;
2515
2516                         curr_loc = lov_stripe_size(lsm,
2517                                            lcl_fm_ext[ext_count - 1].fe_logical+
2518                                            lcl_fm_ext[ext_count - 1].fe_length,
2519                                            cur_stripe);
2520                         if (curr_loc >= fm_key->oa.o_size)
2521                                 ost_eof = 1;
2522
2523                         fiemap_prepare_and_copy_exts(fiemap, lcl_fm_ext,
2524                                                      ost_index, ext_count,
2525                                                      current_extent);
2526
2527                         current_extent += ext_count;
2528
2529                         /* Ran out of available extents? */
2530                         if (current_extent >= fiemap->fm_extent_count)
2531                                 goto finish;
2532                 } while (ost_done == 0 && ost_eof == 0);
2533
2534                 if (cur_stripe_wrap == last_stripe)
2535                         goto finish;
2536         }
2537
2538 finish:
2539         /* Indicate that we are returning device offsets unless file just has
2540          * single stripe */
2541         if (lsm->lsm_stripe_count > 1)
2542                 fiemap->fm_flags |= FIEMAP_FLAG_DEVICE_ORDER;
2543
2544         if (get_num_extents)
2545                 goto skip_last_device_calc;
2546
2547         /* Check if we have reached the last stripe and whether mapping for that
2548          * stripe is done. */
2549         if (cur_stripe_wrap == last_stripe) {
2550                 if (ost_done || ost_eof)
2551                         fiemap->fm_extents[current_extent - 1].fe_flags |=
2552                                                              FIEMAP_EXTENT_LAST;
2553         }
2554
2555 skip_last_device_calc:
2556         fiemap->fm_mapped_extents = current_extent;
2557
2558 out:
2559         OBD_FREE(fm_local, buffer_size);
2560         return rc;
2561 }
2562
2563 static int lov_get_info(struct obd_export *exp, __u32 keylen,
2564                         void *key, __u32 *vallen, void *val,
2565                         struct lov_stripe_md *lsm)
2566 {
2567         struct obd_device *obddev = class_exp2obd(exp);
2568         struct lov_obd *lov = &obddev->u.lov;
2569         int i, rc;
2570         ENTRY;
2571
2572         if (!vallen || !val)
2573                 RETURN(-EFAULT);
2574
2575         obd_getref(obddev);
2576
2577         if (KEY_IS(KEY_LOCK_TO_STRIPE)) {
2578                 struct {
2579                         char name[16];
2580                         struct ldlm_lock *lock;
2581                 } *data = key;
2582                 struct ldlm_res_id *res_id = &data->lock->l_resource->lr_name;
2583                 struct lov_oinfo *loi;
2584                 __u32 *stripe = val;
2585
2586                 if (*vallen < sizeof(*stripe))
2587                         GOTO(out, rc = -EFAULT);
2588                 *vallen = sizeof(*stripe);
2589
2590                 /* XXX This is another one of those bits that will need to
2591                  * change if we ever actually support nested LOVs.  It uses
2592                  * the lock's export to find out which stripe it is. */
2593                 /* XXX - it's assumed all the locks for deleted OSTs have
2594                  * been cancelled. Also, the export for deleted OSTs will
2595                  * be NULL and won't match the lock's export. */
2596                 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2597                         loi = lsm->lsm_oinfo[i];
2598                         if (!lov->lov_tgts[loi->loi_ost_idx])
2599                                 continue;
2600                         if (lov->lov_tgts[loi->loi_ost_idx]->ltd_exp ==
2601                             data->lock->l_conn_export &&
2602                             osc_res_name_eq(loi->loi_id, loi->loi_seq, res_id)) {
2603                                 *stripe = i;
2604                                 GOTO(out, rc = 0);
2605                         }
2606                 }
2607                 LDLM_ERROR(data->lock, "lock on inode without such object");
2608                 dump_lsm(D_ERROR, lsm);
2609                 GOTO(out, rc = -ENXIO);
2610         } else if (KEY_IS(KEY_LAST_ID)) {
2611                 struct obd_id_info *info = val;
2612                 __u32 size = sizeof(obd_id);
2613                 struct lov_tgt_desc *tgt;
2614
2615                 LASSERT(*vallen == sizeof(struct obd_id_info));
2616                 tgt = lov->lov_tgts[info->idx];
2617
2618                 if (!tgt || !tgt->ltd_active)
2619                         GOTO(out, rc = -ESRCH);
2620
2621                 rc = obd_get_info(tgt->ltd_exp, keylen, key, &size, info->data, NULL);
2622                 GOTO(out, rc = 0);
2623         } else if (KEY_IS(KEY_LOVDESC)) {
2624                 struct lov_desc *desc_ret = val;
2625                 *desc_ret = lov->desc;
2626
2627                 GOTO(out, rc = 0);
2628         } else if (KEY_IS(KEY_FIEMAP)) {
2629                 rc = lov_fiemap(lov, keylen, key, vallen, val, lsm);
2630                 GOTO(out, rc);
2631         } else if (KEY_IS(KEY_CONNECT_FLAG)) {
2632                 struct lov_tgt_desc *tgt;
2633                 __u64 ost_idx = *((__u64*)val);
2634
2635                 LASSERT(*vallen == sizeof(__u64));
2636                 LASSERT(ost_idx < lov->desc.ld_tgt_count);
2637                 tgt = lov->lov_tgts[ost_idx];
2638
2639                 if (!tgt || !tgt->ltd_exp)
2640                         GOTO(out, rc = -ESRCH);
2641
2642                 *((__u64*)val) = tgt->ltd_exp->exp_connect_flags;
2643                 GOTO(out, rc = 0);
2644         }
2645
2646         rc = -EINVAL;
2647
2648 out:
2649         obd_putref(obddev);
2650         RETURN(rc);
2651 }
2652
2653 static int lov_set_info_async(struct obd_export *exp, obd_count keylen,
2654                               void *key, obd_count vallen, void *val,
2655                               struct ptlrpc_request_set *set)
2656 {
2657         struct obd_device *obddev = class_exp2obd(exp);
2658         struct lov_obd *lov = &obddev->u.lov;
2659         obd_count count;
2660         int i, rc = 0, err;
2661         struct lov_tgt_desc *tgt;
2662         unsigned incr, check_uuid,
2663                  do_inactive, no_set;
2664         unsigned next_id = 0,  mds_con = 0, capa = 0;
2665         ENTRY;
2666
2667         incr = check_uuid = do_inactive = no_set = 0;
2668         if (set == NULL) {
2669                 no_set = 1;
2670                 set = ptlrpc_prep_set();
2671                 if (!set)
2672                         RETURN(-ENOMEM);
2673         }
2674
2675         obd_getref(obddev);
2676         count = lov->desc.ld_tgt_count;
2677
2678         if (KEY_IS(KEY_NEXT_ID)) {
2679                 count = vallen / sizeof(struct obd_id_info);
2680                 vallen = sizeof(obd_id);
2681                 incr = sizeof(struct obd_id_info);
2682                 do_inactive = 1;
2683                 next_id = 1;
2684         } else if (KEY_IS(KEY_CHECKSUM)) {
2685                 do_inactive = 1;
2686         } else if (KEY_IS(KEY_EVICT_BY_NID)) {
2687                 /* use defaults:  do_inactive = incr = 0; */
2688         } else if (KEY_IS(KEY_MDS_CONN)) {
2689                 mds_con = 1;
2690         } else if (KEY_IS(KEY_CAPA_KEY)) {
2691                 capa = 1;
2692         }
2693
2694         for (i = 0; i < count; i++, val = (char *)val + incr) {
2695                 if (next_id) {
2696                         tgt = lov->lov_tgts[((struct obd_id_info*)val)->idx];
2697                 } else {
2698                         tgt = lov->lov_tgts[i];
2699                 }
2700                 /* OST was disconnected */
2701                 if (!tgt || !tgt->ltd_exp)
2702                         continue;
2703
2704                 /* OST is inactive and we don't want inactive OSCs */
2705                 if (!tgt->ltd_active && !do_inactive)
2706                         continue;
2707
2708                 if (mds_con) {
2709                         struct mds_group_info *mgi;
2710
2711                         LASSERT(vallen == sizeof(*mgi));
2712                         mgi = (struct mds_group_info *)val;
2713
2714                         /* Only want a specific OSC */
2715                         if (mgi->uuid && !obd_uuid_equals(mgi->uuid,
2716                                                 &tgt->ltd_uuid))
2717                                 continue;
2718
2719                         err = obd_set_info_async(tgt->ltd_exp,
2720                                          keylen, key, sizeof(int),
2721                                          &mgi->group, set);
2722                 } else if (next_id) {
2723                         err = obd_set_info_async(tgt->ltd_exp,
2724                                          keylen, key, vallen,
2725                                          ((struct obd_id_info*)val)->data, set);
2726                 } else if (capa) {
2727                         struct mds_capa_info *info = (struct mds_capa_info*)val;
2728
2729                         LASSERT(vallen == sizeof(*info));
2730
2731                          /* Only want a specific OSC */
2732                         if (info->uuid &&
2733                             !obd_uuid_equals(info->uuid, &tgt->ltd_uuid))
2734                                 continue;
2735
2736                         err = obd_set_info_async(tgt->ltd_exp, keylen, key,
2737                                                  sizeof(*info->capa),
2738                                                  info->capa, set);
2739
2740                 } else {
2741                         /* Only want a specific OSC */
2742                         if (check_uuid &&
2743                             !obd_uuid_equals(val, &tgt->ltd_uuid))
2744                                 continue;
2745
2746                         err = obd_set_info_async(tgt->ltd_exp,
2747                                          keylen, key, vallen, val, set);
2748                 }
2749
2750                 if (!rc)
2751                         rc = err;
2752         }
2753
2754         obd_putref(obddev);
2755         if (no_set) {
2756                 err = ptlrpc_set_wait(set);
2757                 if (!rc)
2758                         rc = err;
2759                 ptlrpc_set_destroy(set);
2760         }
2761         RETURN(rc);
2762 }
2763
2764 int lov_test_and_clear_async_rc(struct lov_stripe_md *lsm)
2765 {
2766         int i, rc = 0;
2767         ENTRY;
2768
2769         for (i = 0; i < lsm->lsm_stripe_count; i++) {
2770                 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
2771                 if (loi->loi_ar.ar_rc && !rc)
2772                         rc = loi->loi_ar.ar_rc;
2773                 loi->loi_ar.ar_rc = 0;
2774         }
2775         RETURN(rc);
2776 }
2777 EXPORT_SYMBOL(lov_test_and_clear_async_rc);
2778
2779
2780 static int lov_extent_calc(struct obd_export *exp, struct lov_stripe_md *lsm,
2781                            int cmd, __u64 *offset)
2782 {
2783         __u32 ssize = lsm->lsm_stripe_size;
2784         __u64 start;
2785
2786         start = *offset;
2787         do_div(start, ssize);
2788         start = start * ssize;
2789
2790         CDEBUG(D_DLMTRACE, "offset "LPU64", stripe %u, start "LPU64
2791                            ", end "LPU64"\n", *offset, ssize, start,
2792                            start + ssize - 1);
2793         if (cmd == OBD_CALC_STRIPE_END) {
2794                 *offset = start + ssize - 1;
2795         } else if (cmd == OBD_CALC_STRIPE_START) {
2796                 *offset = start;
2797         } else {
2798                 LBUG();
2799         }
2800
2801         RETURN(0);
2802 }
2803
2804 void lov_stripe_lock(struct lov_stripe_md *md)
2805 {
2806         LASSERT(md->lsm_lock_owner != cfs_curproc_pid());
2807         cfs_spin_lock(&md->lsm_lock);
2808         LASSERT(md->lsm_lock_owner == 0);
2809         md->lsm_lock_owner = cfs_curproc_pid();
2810 }
2811 EXPORT_SYMBOL(lov_stripe_lock);
2812
2813 void lov_stripe_unlock(struct lov_stripe_md *md)
2814 {
2815         LASSERT(md->lsm_lock_owner == cfs_curproc_pid());
2816         md->lsm_lock_owner = 0;
2817         cfs_spin_unlock(&md->lsm_lock);
2818 }
2819 EXPORT_SYMBOL(lov_stripe_unlock);
2820
2821
2822 struct obd_ops lov_obd_ops = {
2823         .o_owner               = THIS_MODULE,
2824         .o_setup               = lov_setup,
2825         .o_precleanup          = lov_precleanup,
2826         .o_cleanup             = lov_cleanup,
2827         //.o_process_config      = lov_process_config,
2828         .o_connect             = lov_connect,
2829         .o_disconnect          = lov_disconnect,
2830         .o_statfs              = lov_statfs,
2831         .o_statfs_async        = lov_statfs_async,
2832         .o_packmd              = lov_packmd,
2833         .o_unpackmd            = lov_unpackmd,
2834         .o_create              = lov_create,
2835         .o_destroy             = lov_destroy,
2836         .o_getattr             = lov_getattr,
2837         .o_getattr_async       = lov_getattr_async,
2838         .o_setattr             = lov_setattr,
2839         .o_setattr_async       = lov_setattr_async,
2840         .o_brw                 = lov_brw,
2841         .o_merge_lvb           = lov_merge_lvb,
2842         .o_adjust_kms          = lov_adjust_kms,
2843         .o_punch               = lov_punch,
2844         .o_sync                = lov_sync,
2845         .o_enqueue             = lov_enqueue,
2846         .o_change_cbdata       = lov_change_cbdata,
2847         .o_find_cbdata         = lov_find_cbdata,
2848         .o_cancel              = lov_cancel,
2849         .o_cancel_unused       = lov_cancel_unused,
2850         .o_iocontrol           = lov_iocontrol,
2851         .o_get_info            = lov_get_info,
2852         .o_set_info_async      = lov_set_info_async,
2853         .o_extent_calc         = lov_extent_calc,
2854         .o_llog_init           = lov_llog_init,
2855         .o_llog_finish         = lov_llog_finish,
2856         .o_notify              = lov_notify,
2857         .o_pool_new            = lov_pool_new,
2858         .o_pool_rem            = lov_pool_remove,
2859         .o_pool_add            = lov_pool_add,
2860         .o_pool_del            = lov_pool_del,
2861         .o_getref              = lov_getref,
2862         .o_putref              = lov_putref,
2863         .o_sync_fs             = lov_sync_fs,
2864 };
2865
2866 static quota_interface_t *quota_interface;
2867 extern quota_interface_t lov_quota_interface;
2868
2869 cfs_mem_cache_t *lov_oinfo_slab;
2870
2871 extern struct lu_kmem_descr lov_caches[];
2872
2873 int __init lov_init(void)
2874 {
2875         struct lprocfs_static_vars lvars = { 0 };
2876         int rc, rc2;
2877         ENTRY;
2878
2879         /* print an address of _any_ initialized kernel symbol from this
2880          * module, to allow debugging with gdb that doesn't support data
2881          * symbols from modules.*/
2882         CDEBUG(D_CONSOLE, "Lustre LOV module (%p).\n", &lov_caches);
2883
2884         rc = lu_kmem_init(lov_caches);
2885         if (rc)
2886                 return rc;
2887
2888         lov_oinfo_slab = cfs_mem_cache_create("lov_oinfo",
2889                                               sizeof(struct lov_oinfo),
2890                                               0, CFS_SLAB_HWCACHE_ALIGN);
2891         if (lov_oinfo_slab == NULL) {
2892                 lu_kmem_fini(lov_caches);
2893                 return -ENOMEM;
2894         }
2895         lprocfs_lov_init_vars(&lvars);
2896
2897         cfs_request_module("lquota");
2898         quota_interface = PORTAL_SYMBOL_GET(lov_quota_interface);
2899         init_obd_quota_ops(quota_interface, &lov_obd_ops);
2900
2901         rc = class_register_type(&lov_obd_ops, NULL, lvars.module_vars,
2902                                  LUSTRE_LOV_NAME, &lov_device_type);
2903
2904         if (rc) {
2905                 if (quota_interface)
2906                         PORTAL_SYMBOL_PUT(lov_quota_interface);
2907                 rc2 = cfs_mem_cache_destroy(lov_oinfo_slab);
2908                 LASSERT(rc2 == 0);
2909                 lu_kmem_fini(lov_caches);
2910         }
2911
2912         RETURN(rc);
2913 }
2914
2915 #ifdef __KERNEL__
2916 static void /*__exit*/ lov_exit(void)
2917 {
2918         int rc;
2919
2920         lu_device_type_fini(&lov_device_type);
2921         lu_kmem_fini(lov_caches);
2922
2923         if (quota_interface)
2924                 PORTAL_SYMBOL_PUT(lov_quota_interface);
2925
2926         class_unregister_type(LUSTRE_LOV_NAME);
2927         rc = cfs_mem_cache_destroy(lov_oinfo_slab);
2928         LASSERT(rc == 0);
2929 }
2930
2931 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2932 MODULE_DESCRIPTION("Lustre Logical Object Volume OBD driver");
2933 MODULE_LICENSE("GPL");
2934
2935 cfs_module(lov, LUSTRE_VERSION_STRING, lov_init, lov_exit);
2936 #endif