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