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                               enum obd_notify_event ev);
 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                 CDEBUG(D_CONFIG, "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                               enum obd_notify_event ev)
 388 {
 389         struct lov_obd *lov = &obd->u.lov;
 390         struct lov_tgt_desc *tgt;
 391         int index, activate, active;
 392         ENTRY;
 393
 394         CDEBUG(D_INFO, "Searching in lov %p for uuid %s event(%d)\n",
 395                lov, uuid->uuid, ev);
 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 (ev == OBD_NOTIFY_DEACTIVATE || ev == OBD_NOTIFY_ACTIVATE) {
 414                 activate = (ev == OBD_NOTIFY_ACTIVATE) ? 1 : 0;
 415
 416                 if (lov->lov_tgts[index]->ltd_activate == activate) {
 417                         CDEBUG(D_INFO, "OSC %s already %sactivate!\n",
 418                                uuid->uuid, activate ? "" : "de");
 419                 } else {
 420                         lov->lov_tgts[index]->ltd_activate = activate;
 421                         CDEBUG(D_CONFIG, "%sactivate OSC %s\n",
 422                                activate ? "" : "de", obd_uuid2str(uuid));
 423                 }
 424
 425         } else if (ev == OBD_NOTIFY_INACTIVE || ev == OBD_NOTIFY_ACTIVE) {
 426                 active = (ev == OBD_NOTIFY_ACTIVE) ? 1 : 0;
 427
 428                 if (lov->lov_tgts[index]->ltd_active == active) {
 429                         CDEBUG(D_INFO, "OSC %s already %sactive!\n",
 430                                uuid->uuid, active ? "" : "in");
 431                         GOTO(out, index);
 432                 } else {
 433                         CDEBUG(D_CONFIG, "Marking OSC %s %sactive\n",
 434                                obd_uuid2str(uuid), active ? "" : "in");
 435                 }
 436
 437                 lov->lov_tgts[index]->ltd_active = active;
 438                 if (active) {
 439                         lov->desc.ld_active_tgt_count++;
 440                         lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 0;
 441                 } else {
 442                         lov->desc.ld_active_tgt_count--;
 443                         lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 1;
 444                 }
 445                 /* remove any old qos penalty */
 446                 lov->lov_tgts[index]->ltd_qos.ltq_penalty = 0;
 447         } else {
 448                 CERROR("Unknown event(%d) for uuid %s", ev, uuid->uuid);
 449         }
 450
 451  out:
 452         obd_putref(obd);
 453         RETURN(index);
 454 }
 455
 456 static int lov_notify(struct obd_device *obd, struct obd_device *watched,
 457                       enum obd_notify_event ev, void *data)
 458 {
 459         int rc = 0;
 460         ENTRY;
 461
 462         if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE ||
 463             ev == OBD_NOTIFY_ACTIVATE || ev == OBD_NOTIFY_DEACTIVATE) {
 464                 struct obd_uuid *uuid;
 465
 466                 LASSERT(watched);
 467
 468                 if (strcmp(watched->obd_type->typ_name, LUSTRE_OSC_NAME)) {
 469                         CERROR("unexpected notification of %s %s!\n",
 470                                watched->obd_type->typ_name,
 471                                watched->obd_name);
 472                         RETURN(-EINVAL);
 473                 }
 474                 uuid = &watched->u.cli.cl_target_uuid;
 475
 476                 /* Set OSC as active before notifying the observer, so the
 477                  * observer can use the OSC normally.
 478                  */
 479                 rc = lov_set_osc_active(obd, uuid, ev);
 480                 if (rc < 0) {
 481                         CERROR("event(%d) of %s failed: %d\n", ev,
 482                                obd_uuid2str(uuid), rc);
 483                         RETURN(rc);
 484                 }
 485                 /* active event should be pass lov target index as data */
 486                 data = &rc;
 487         }
 488
 489         /* Pass the notification up the chain. */
 490         if (watched) {
 491                 rc = obd_notify_observer(obd, watched, ev, data);
 492         } else {
 493                 /* NULL watched means all osc's in the lov (only for syncs) */
 494                 /* sync event should be send lov idx as data */
 495                 struct lov_obd *lov = &obd->u.lov;
 496                 int i, is_sync;
 497
 498                 data = &i;
 499                 is_sync = (ev == OBD_NOTIFY_SYNC) ||
 500                           (ev == OBD_NOTIFY_SYNC_NONBLOCK);
 501
 502                 obd_getref(obd);
 503                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
 504                         if (!lov->lov_tgts[i])
 505                                 continue;
 506
 507                         /* don't send sync event if target not
 508                          * connected/activated */
 509                         if (is_sync &&  !lov->lov_tgts[i]->ltd_active)
 510                                 continue;
 511
 512                         rc = obd_notify_observer(obd, lov->lov_tgts[i]->ltd_obd,
 513                                                  ev, data);
 514                         if (rc) {
 515                                 CERROR("%s: notify %s of %s failed %d\n",
 516                                        obd->obd_name,
 517                                        obd->obd_observer->obd_name,
 518                                        lov->lov_tgts[i]->ltd_obd->obd_name,
 519                                        rc);
 520                         }
 521                 }
 522                 obd_putref(obd);
 523         }
 524
 525         RETURN(rc);
 526 }
 527
 528 static int lov_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
 529                           __u32 index, int gen, int active)
 530 {
 531         struct lov_obd *lov = &obd->u.lov;
 532         struct lov_tgt_desc *tgt;
 533         struct obd_device *tgt_obd;
 534         int rc;
 535         ENTRY;
 536
 537         CDEBUG(D_CONFIG, "uuid:%s idx:%d gen:%d active:%d\n",
 538                uuidp->uuid, index, gen, active);
 539
 540         if (gen <= 0) {
 541                 CERROR("request to add OBD %s with invalid generation: %d\n",
 542                        uuidp->uuid, gen);
 543                 RETURN(-EINVAL);
 544         }
 545
 546         tgt_obd = class_find_client_obd(uuidp, LUSTRE_OSC_NAME,
 547                                         &obd->obd_uuid);
 548         if (tgt_obd == NULL)
 549                 RETURN(-EINVAL);
 550
 551         cfs_mutex_down(&lov->lov_lock);
 552
 553         if ((index < lov->lov_tgt_size) && (lov->lov_tgts[index] != NULL)) {
 554                 tgt = lov->lov_tgts[index];
 555                 CERROR("UUID %s already assigned at LOV target index %d\n",
 556                        obd_uuid2str(&tgt->ltd_uuid), index);
 557                 cfs_mutex_up(&lov->lov_lock);
 558                 RETURN(-EEXIST);
 559         }
 560
 561         if (index >= lov->lov_tgt_size) {
 562                 /* We need to reallocate the lov target array. */
 563                 struct lov_tgt_desc **newtgts, **old = NULL;
 564                 __u32 newsize, oldsize = 0;
 565
 566                 newsize = max(lov->lov_tgt_size, (__u32)2);
 567                 while (newsize < index + 1)
 568                         newsize = newsize << 1;
 569                 OBD_ALLOC(newtgts, sizeof(*newtgts) * newsize);
 570                 if (newtgts == NULL) {
 571                         cfs_mutex_up(&lov->lov_lock);
 572                         RETURN(-ENOMEM);
 573                 }
 574
 575                 if (lov->lov_tgt_size) {
 576                         memcpy(newtgts, lov->lov_tgts, sizeof(*newtgts) *
 577                                lov->lov_tgt_size);
 578                         old = lov->lov_tgts;
 579                         oldsize = lov->lov_tgt_size;
 580                 }
 581
 582                 lov->lov_tgts = newtgts;
 583                 lov->lov_tgt_size = newsize;
 584 #ifdef __KERNEL__
 585                 smp_rmb();
 586 #endif
 587                 if (old)
 588                         OBD_FREE(old, sizeof(*old) * oldsize);
 589
 590                 CDEBUG(D_CONFIG, "tgts: %p size: %d\n",
 591                        lov->lov_tgts, lov->lov_tgt_size);
 592         }
 593
 594         OBD_ALLOC_PTR(tgt);
 595         if (!tgt) {
 596                 cfs_mutex_up(&lov->lov_lock);
 597                 RETURN(-ENOMEM);
 598         }
 599
 600         rc = lov_ost_pool_add(&lov->lov_packed, index, lov->lov_tgt_size);
 601         if (rc) {
 602                 cfs_mutex_up(&lov->lov_lock);
 603                 OBD_FREE_PTR(tgt);
 604                 RETURN(rc);
 605         }
 606
 607         tgt->ltd_uuid = *uuidp;
 608         tgt->ltd_obd = tgt_obd;
 609         /* XXX - add a sanity check on the generation number. */
 610         tgt->ltd_gen = gen;
 611         tgt->ltd_index = index;
 612         tgt->ltd_activate = active;
 613         lov->lov_tgts[index] = tgt;
 614         if (index >= lov->desc.ld_tgt_count)
 615                 lov->desc.ld_tgt_count = index + 1;
 616
 617         cfs_mutex_up(&lov->lov_lock);
 618
 619         CDEBUG(D_CONFIG, "idx=%d ltd_gen=%d ld_tgt_count=%d\n",
 620                 index, tgt->ltd_gen, lov->desc.ld_tgt_count);
 621
 622         rc = obd_notify(obd, tgt_obd, OBD_NOTIFY_CREATE, &index);
 623
 624         if (lov->lov_connects == 0) {
 625                 /* lov_connect hasn't been called yet. We'll do the
 626                    lov_connect_obd on this target when that fn first runs,
 627                    because we don't know the connect flags yet. */
 628                 RETURN(0);
 629         }
 630
 631         obd_getref(obd);
 632
 633         rc = lov_connect_obd(obd, index, active, &lov->lov_ocd);
 634         if (rc)
 635                 GOTO(out, rc);
 636
 637         /* connect to administrative disabled ost */
 638         if (!tgt->ltd_exp)
 639                 GOTO(out, rc = 0);
 640
 641         rc = lov_notify(obd, tgt->ltd_exp->exp_obd,
 642                         active ? OBD_NOTIFY_CONNECT : OBD_NOTIFY_INACTIVE,
 643                         (void *)&index);
 644
 645 out:
 646         if (rc) {
 647                 CERROR("add failed (%d), deleting %s\n", rc,
 648                        obd_uuid2str(&tgt->ltd_uuid));
 649                 lov_del_target(obd, index, 0, 0);
 650         }
 651         obd_putref(obd);
 652         RETURN(rc);
 653 }
 654
 655 /* Schedule a target for deletion */
 656 int lov_del_target(struct obd_device *obd, __u32 index,
 657                    struct obd_uuid *uuidp, int gen)
 658 {
 659         struct lov_obd *lov = &obd->u.lov;
 660         int count = lov->desc.ld_tgt_count;
 661         int rc = 0;
 662         ENTRY;
 663
 664         if (index >= count) {
 665                 CERROR("LOV target index %d >= number of LOV OBDs %d.\n",
 666                        index, count);
 667                 RETURN(-EINVAL);
 668         }
 669
 670         obd_getref(obd);
 671
 672         if (!lov->lov_tgts[index]) {
 673                 CERROR("LOV target at index %d is not setup.\n", index);
 674                 GOTO(out, rc = -EINVAL);
 675         }
 676
 677         if (uuidp && !obd_uuid_equals(uuidp, &lov->lov_tgts[index]->ltd_uuid)) {
 678                 CERROR("LOV target UUID %s at index %d doesn't match %s.\n",
 679                        lov_uuid2str(lov, index), index,
 680                        obd_uuid2str(uuidp));
 681                 GOTO(out, rc = -EINVAL);
 682         }
 683
 684         CDEBUG(D_CONFIG, "uuid: %s idx: %d gen: %d exp: %p active: %d\n",
 685                lov_uuid2str(lov, index), index,
 686                lov->lov_tgts[index]->ltd_gen, lov->lov_tgts[index]->ltd_exp,
 687                lov->lov_tgts[index]->ltd_active);
 688
 689         lov->lov_tgts[index]->ltd_reap = 1;
 690         lov->lov_death_row++;
 691         /* we really delete it from obd_putref */
 692 out:
 693         obd_putref(obd);
 694
 695         RETURN(rc);
 696 }
 697
 698 static void __lov_del_obd(struct obd_device *obd, struct lov_tgt_desc *tgt)
 699 {
 700         struct obd_device *osc_obd;
 701
 702         LASSERT(tgt);
 703         LASSERT(tgt->ltd_reap);
 704
 705         osc_obd = class_exp2obd(tgt->ltd_exp);
 706
 707         CDEBUG(D_CONFIG, "Removing tgt %s : %s\n",
 708                tgt->ltd_uuid.uuid,
 709                osc_obd ? osc_obd->obd_name : "<no obd>");
 710
 711         if (tgt->ltd_exp)
 712                 lov_disconnect_obd(obd, tgt);
 713
 714         OBD_FREE_PTR(tgt);
 715
 716         /* Manual cleanup - no cleanup logs to clean up the osc's.  We must
 717            do it ourselves. And we can't do it from lov_cleanup,
 718            because we just lost our only reference to it. */
 719         if (osc_obd)
 720                 class_manual_cleanup(osc_obd);
 721 }
 722
 723 void lov_fix_desc_stripe_size(__u64 *val)
 724 {
 725         if (*val < PTLRPC_MAX_BRW_SIZE) {
 726                 LCONSOLE_WARN("Increasing default stripe size to min %u\n",
 727                               PTLRPC_MAX_BRW_SIZE);
 728                 *val = PTLRPC_MAX_BRW_SIZE;
 729         } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) {
 730                 *val &= ~(LOV_MIN_STRIPE_SIZE - 1);
 731                 LCONSOLE_WARN("Changing default stripe size to "LPU64" (a "
 732                               "multiple of %u)\n",
 733                               *val, LOV_MIN_STRIPE_SIZE);
 734         }
 735 }
 736
 737 void lov_fix_desc_stripe_count(__u32 *val)
 738 {
 739         if (*val == 0)
 740                 *val = 1;
 741 }
 742
 743 void lov_fix_desc_pattern(__u32 *val)
 744 {
 745         /* from lov_setstripe */
 746         if ((*val != 0) && (*val != LOV_PATTERN_RAID0)) {
 747                 LCONSOLE_WARN("Unknown stripe pattern: %#x\n", *val);
 748                 *val = 0;
 749         }
 750 }
 751
 752 void lov_fix_desc_qos_maxage(__u32 *val)
 753 {
 754         /* fix qos_maxage */
 755         if (*val == 0)
 756                 *val = QOS_DEFAULT_MAXAGE;
 757 }
 758
 759 void lov_fix_desc(struct lov_desc *desc)
 760 {
 761         lov_fix_desc_stripe_size(&desc->ld_default_stripe_size);
 762         lov_fix_desc_stripe_count(&desc->ld_default_stripe_count);
 763         lov_fix_desc_pattern(&desc->ld_pattern);
 764         lov_fix_desc_qos_maxage(&desc->ld_qos_maxage);
 765 }
 766
 767 int lov_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
 768 {
 769         struct lprocfs_static_vars lvars = { 0 };
 770         struct lov_desc *desc;
 771         struct lov_obd *lov = &obd->u.lov;
 772         int rc;
 773         ENTRY;
 774
 775         if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
 776                 CERROR("LOV setup requires a descriptor\n");
 777                 RETURN(-EINVAL);
 778         }
 779
 780         desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
 781
 782         if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
 783                 CERROR("descriptor size wrong: %d > %d\n",
 784                        (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
 785                 RETURN(-EINVAL);
 786         }
 787
 788         if (desc->ld_magic != LOV_DESC_MAGIC) {
 789                 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
 790                             CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
 791                                    obd->obd_name, desc);
 792                             lustre_swab_lov_desc(desc);
 793                 } else {
 794                         CERROR("%s: Bad lov desc magic: %#x\n",
 795                                obd->obd_name, desc->ld_magic);
 796                         RETURN(-EINVAL);
 797                 }
 798         }
 799
 800         lov_fix_desc(desc);
 801
 802         desc->ld_active_tgt_count = 0;
 803         lov->desc = *desc;
 804         lov->lov_tgt_size = 0;
 805
 806         cfs_sema_init(&lov->lov_lock, 1);
 807         cfs_atomic_set(&lov->lov_refcount, 0);
 808         CFS_INIT_LIST_HEAD(&lov->lov_qos.lq_oss_list);
 809         cfs_init_rwsem(&lov->lov_qos.lq_rw_sem);
 810         lov->lov_sp_me = LUSTRE_SP_CLI;
 811         lov->lov_qos.lq_dirty = 1;
 812         lov->lov_qos.lq_rr.lqr_dirty = 1;
 813         lov->lov_qos.lq_reset = 1;
 814         /* Default priority is toward free space balance */
 815         lov->lov_qos.lq_prio_free = 232;
 816         /* Default threshold for rr (roughly 17%) */
 817         lov->lov_qos.lq_threshold_rr = 43;
 818         /* Init statfs fields */
 819         OBD_ALLOC_PTR(lov->lov_qos.lq_statfs_data);
 820         if (NULL == lov->lov_qos.lq_statfs_data)
 821                 RETURN(-ENOMEM);
 822         cfs_waitq_init(&lov->lov_qos.lq_statfs_waitq);
 823
 824         lov->lov_pools_hash_body = cfs_hash_create("POOLS", HASH_POOLS_CUR_BITS,
 825                                                    HASH_POOLS_MAX_BITS,
 826                                                    HASH_POOLS_BKT_BITS, 0,
 827                                                    CFS_HASH_MIN_THETA,
 828                                                    CFS_HASH_MAX_THETA,
 829                                                    &pool_hash_operations,
 830                                                    CFS_HASH_DEFAULT);
 831         CFS_INIT_LIST_HEAD(&lov->lov_pool_list);
 832         lov->lov_pool_count = 0;
 833         rc = lov_ost_pool_init(&lov->lov_packed, 0);
 834         if (rc)
 835                 GOTO(out_free_statfs, rc);
 836         rc = lov_ost_pool_init(&lov->lov_qos.lq_rr.lqr_pool, 0);
 837         if (rc)
 838                 GOTO(out_free_lov_packed, rc);
 839
 840         lprocfs_lov_init_vars(&lvars);
 841         lprocfs_obd_setup(obd, lvars.obd_vars);
 842 #ifdef LPROCFS
 843         {
 844                 int rc;
 845
 846                 rc = lprocfs_seq_create(obd->obd_proc_entry, "target_obd",
 847                                         0444, &lov_proc_target_fops, obd);
 848                 if (rc)
 849                         CWARN("Error adding the target_obd file\n");
 850         }
 851 #endif
 852         lov->lov_pool_proc_entry = lprocfs_register("pools",
 853                                                     obd->obd_proc_entry,
 854                                                     NULL, NULL);
 855
 856         RETURN(0);
 857
 858 out_free_lov_packed:
 859         lov_ost_pool_free(&lov->lov_packed);
 860 out_free_statfs:
 861         OBD_FREE_PTR(lov->lov_qos.lq_statfs_data);
 862         return rc;
 863 }
 864
 865 static int lov_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
 866 {
 867         int rc = 0;
 868         struct lov_obd *lov = &obd->u.lov;
 869
 870         ENTRY;
 871
 872         switch (stage) {
 873         case OBD_CLEANUP_EARLY: {
 874                 int i;
 875                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
 876                         if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_active)
 877                                 continue;
 878                         obd_precleanup(class_exp2obd(lov->lov_tgts[i]->ltd_exp),
 879                                        OBD_CLEANUP_EARLY);
 880                 }
 881                 break;
 882         }
 883         case OBD_CLEANUP_EXPORTS:
 884                 rc = obd_llog_finish(obd, 0);
 885                 if (rc != 0)
 886                         CERROR("failed to cleanup llogging subsystems\n");
 887                 break;
 888         }
 889         RETURN(rc);
 890 }
 891
 892 static int lov_cleanup(struct obd_device *obd)
 893 {
 894         struct lov_obd *lov = &obd->u.lov;
 895         cfs_list_t *pos, *tmp;
 896         struct pool_desc *pool;
 897
 898         cfs_list_for_each_safe(pos, tmp, &lov->lov_pool_list) {
 899                 pool = cfs_list_entry(pos, struct pool_desc, pool_list);
 900                 /* free pool structs */
 901                 CDEBUG(D_INFO, "delete pool %p\n", pool);
 902                 lov_pool_del(obd, pool->pool_name);
 903         }
 904         cfs_hash_putref(lov->lov_pools_hash_body);
 905         lov_ost_pool_free(&(lov->lov_qos.lq_rr.lqr_pool));
 906         lov_ost_pool_free(&lov->lov_packed);
 907
 908         if (lov->lov_tgts) {
 909                 int i;
 910                 obd_getref(obd);
 911                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
 912                         if (!lov->lov_tgts[i])
 913                                 continue;
 914
 915                         /* Inactive targets may never have connected */
 916                         if (lov->lov_tgts[i]->ltd_active ||
 917                             cfs_atomic_read(&lov->lov_refcount))
 918                             /* We should never get here - these
 919                                should have been removed in the
 920                              disconnect. */
 921                                 CERROR("lov tgt %d not cleaned!"
 922                                        " deathrow=%d, lovrc=%d\n",
 923                                        i, lov->lov_death_row,
 924                                        cfs_atomic_read(&lov->lov_refcount));
 925                         lov_del_target(obd, i, 0, 0);
 926                 }
 927                 obd_putref(obd);
 928                 OBD_FREE(lov->lov_tgts, sizeof(*lov->lov_tgts) *
 929                          lov->lov_tgt_size);
 930                 lov->lov_tgt_size = 0;
 931         }
 932
 933         /* clear pools parent proc entry only after all pools is killed */
 934         lprocfs_obd_cleanup(obd);
 935
 936         OBD_FREE_PTR(lov->lov_qos.lq_statfs_data);
 937         RETURN(0);
 938 }
 939
 940 int lov_process_config_base(struct obd_device *obd, struct lustre_cfg *lcfg,
 941                             __u32 *indexp, int *genp)
 942 {
 943         struct obd_uuid obd_uuid;
 944         int cmd;
 945         int rc = 0;
 946         ENTRY;
 947
 948         switch(cmd = lcfg->lcfg_command) {
 949         case LCFG_LOV_ADD_OBD:
 950         case LCFG_LOV_ADD_INA:
 951         case LCFG_LOV_DEL_OBD: {
 952                 __u32 index;
 953                 int gen;
 954                 /* lov_modify_tgts add  0:lov_mdsA  1:ost1_UUID  2:0  3:1 */
 955                 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
 956                         GOTO(out, rc = -EINVAL);
 957
 958                 obd_str2uuid(&obd_uuid,  lustre_cfg_buf(lcfg, 1));
 959
 960                 if (sscanf(lustre_cfg_buf(lcfg, 2), "%d", indexp) != 1)
 961                         GOTO(out, rc = -EINVAL);
 962                 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", genp) != 1)
 963                         GOTO(out, rc = -EINVAL);
 964                 index = *indexp;
 965                 gen = *genp;
 966                 if (cmd == LCFG_LOV_ADD_OBD)
 967                         rc = lov_add_target(obd, &obd_uuid, index, gen, 1);
 968                 else if (cmd == LCFG_LOV_ADD_INA)
 969                         rc = lov_add_target(obd, &obd_uuid, index, gen, 0);
 970                 else
 971                         rc = lov_del_target(obd, index, &obd_uuid, gen);
 972                 GOTO(out, rc);
 973         }
 974         case LCFG_PARAM: {
 975                 struct lprocfs_static_vars lvars = { 0 };
 976                 struct lov_desc *desc = &(obd->u.lov.desc);
 977
 978                 if (!desc)
 979                         GOTO(out, rc = -EINVAL);
 980
 981                 lprocfs_lov_init_vars(&lvars);
 982
 983                 rc = class_process_proc_param(PARAM_LOV, lvars.obd_vars,
 984                                               lcfg, obd);
 985                 if (rc > 0)
 986                         rc = 0;
 987                 GOTO(out, rc);
 988         }
 989         case LCFG_POOL_NEW:
 990         case LCFG_POOL_ADD:
 991         case LCFG_POOL_DEL:
 992         case LCFG_POOL_REM:
 993                 GOTO(out, rc);
 994
 995         default: {
 996                 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
 997                 GOTO(out, rc = -EINVAL);
 998
 999         }
1000         }
1001 out:
1002         RETURN(rc);
1003 }
1004
1005 #ifndef log2
1006 #define log2(n) cfs_ffz(~(n))
1007 #endif
1008
1009 static int lov_clear_orphans(struct obd_export *export, struct obdo *src_oa,
1010                              struct lov_stripe_md **ea,
1011                              struct obd_trans_info *oti)
1012 {
1013         struct lov_obd *lov;
1014         struct obdo *tmp_oa;
1015         struct obd_uuid *ost_uuid = NULL;
1016         int rc = 0, i;
1017         ENTRY;
1018
1019         LASSERT(src_oa->o_valid & OBD_MD_FLFLAGS &&
1020                 src_oa->o_flags == OBD_FL_DELORPHAN);
1021
1022         lov = &export->exp_obd->u.lov;
1023
1024         OBDO_ALLOC(tmp_oa);
1025         if (tmp_oa == NULL)
1026                 RETURN(-ENOMEM);
1027
1028         if (oti->oti_ost_uuid) {
1029                 ost_uuid = oti->oti_ost_uuid;
1030                 CDEBUG(D_HA, "clearing orphans only for %s\n",
1031                        ost_uuid->uuid);
1032         }
1033
1034         obd_getref(export->exp_obd);
1035         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
1036                 struct lov_stripe_md obj_md;
1037                 struct lov_stripe_md *obj_mdp = &obj_md;
1038                 struct lov_tgt_desc *tgt;
1039                 int err;
1040
1041                 tgt = lov->lov_tgts[i];
1042                 if (!tgt)
1043                         continue;
1044
1045                 /* if called for a specific target, we don't
1046                    care if it is not active. */
1047                 if (!lov->lov_tgts[i]->ltd_active && ost_uuid == NULL) {
1048                         CDEBUG(D_HA, "lov idx %d inactive\n", i);
1049                         continue;
1050                 }
1051
1052                 if (ost_uuid && !obd_uuid_equals(ost_uuid, &tgt->ltd_uuid))
1053                         continue;
1054
1055                 CDEBUG(D_CONFIG,"Clear orphans for %d:%s\n", i,
1056                        obd_uuid2str(ost_uuid));
1057
1058                 memcpy(tmp_oa, src_oa, sizeof(*tmp_oa));
1059
1060                 LASSERT(lov->lov_tgts[i]->ltd_exp);
1061                 /* XXX: LOV STACKING: use real "obj_mdp" sub-data */
1062                 err = obd_create(lov->lov_tgts[i]->ltd_exp,
1063                                  tmp_oa, &obj_mdp, oti);
1064                 if (err) {
1065                         /* This export will be disabled until it is recovered,
1066                            and then orphan recovery will be completed. */
1067                         CERROR("error in orphan recovery on OST idx %d/%d: "
1068                                "rc = %d\n", i, lov->desc.ld_tgt_count, err);
1069                         rc = err;
1070                 }
1071
1072                 if (ost_uuid)
1073                         break;
1074         }
1075         obd_putref(export->exp_obd);
1076
1077         OBDO_FREE(tmp_oa);
1078         RETURN(rc);
1079 }
1080
1081 static int lov_recreate(struct obd_export *exp, struct obdo *src_oa,
1082                         struct lov_stripe_md **ea, struct obd_trans_info *oti)
1083 {
1084         struct lov_stripe_md *obj_mdp, *lsm;
1085         struct lov_obd *lov = &exp->exp_obd->u.lov;
1086         unsigned ost_idx;
1087         int rc, i;
1088         ENTRY;
1089
1090         LASSERT(src_oa->o_valid & OBD_MD_FLFLAGS &&
1091                 src_oa->o_flags & OBD_FL_RECREATE_OBJS);
1092
1093         OBD_ALLOC(obj_mdp, sizeof(*obj_mdp));
1094         if (obj_mdp == NULL)
1095                 RETURN(-ENOMEM);
1096
1097         ost_idx = src_oa->o_nlink;
1098         lsm = *ea;
1099         if (lsm == NULL)
1100                 GOTO(out, rc = -EINVAL);
1101         if (ost_idx >= lov->desc.ld_tgt_count ||
1102             !lov->lov_tgts[ost_idx])
1103                 GOTO(out, rc = -EINVAL);
1104
1105         for (i = 0; i < lsm->lsm_stripe_count; i++) {
1106                 if (lsm->lsm_oinfo[i]->loi_ost_idx == ost_idx) {
1107                         if (lsm->lsm_oinfo[i]->loi_id != src_oa->o_id)
1108                                 GOTO(out, rc = -EINVAL);
1109                         break;
1110                 }
1111         }
1112         if (i == lsm->lsm_stripe_count)
1113                 GOTO(out, rc = -EINVAL);
1114
1115         rc = obd_create(lov->lov_tgts[ost_idx]->ltd_exp, src_oa, &obj_mdp, oti);
1116 out:
1117         OBD_FREE(obj_mdp, sizeof(*obj_mdp));
1118         RETURN(rc);
1119 }
1120
1121 /* the LOV expects oa->o_id to be set to the LOV object id */
1122 static int lov_create(struct obd_export *exp, struct obdo *src_oa,
1123                       struct lov_stripe_md **ea, struct obd_trans_info *oti)
1124 {
1125         struct lov_obd *lov;
1126         struct obd_info oinfo;
1127         struct lov_request_set *set = NULL;
1128         struct lov_request *req;
1129         struct l_wait_info  lwi = { 0 };
1130         int rc = 0;
1131         ENTRY;
1132
1133         LASSERT(ea != NULL);
1134         if (exp == NULL)
1135                 RETURN(-EINVAL);
1136
1137         if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
1138             src_oa->o_flags == OBD_FL_DELORPHAN) {
1139                 rc = lov_clear_orphans(exp, src_oa, ea, oti);
1140                 RETURN(rc);
1141         }
1142
1143         lov = &exp->exp_obd->u.lov;
1144         if (!lov->desc.ld_active_tgt_count)
1145                 RETURN(-EIO);
1146
1147         obd_getref(exp->exp_obd);
1148         /* Recreate a specific object id at the given OST index */
1149         if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
1150             (src_oa->o_flags & OBD_FL_RECREATE_OBJS)) {
1151                  rc = lov_recreate(exp, src_oa, ea, oti);
1152                  GOTO(out, rc);
1153         }
1154
1155         /* issue statfs rpcs if the osfs data is older than qos_maxage - 1s,
1156          * later in alloc_qos(), we will wait for those rpcs to complete if
1157          * the osfs age is older than 2 * qos_maxage */
1158         qos_statfs_update(exp->exp_obd,
1159                           cfs_time_shift_64(-lov->desc.ld_qos_maxage +
1160                                             OBD_STATFS_CACHE_SECONDS),
1161                           0);
1162
1163         rc = lov_prep_create_set(exp, &oinfo, ea, src_oa, oti, &set);
1164         if (rc)
1165                 GOTO(out, rc);
1166
1167         cfs_list_for_each_entry(req, &set->set_list, rq_link) {
1168                 /* XXX: LOV STACKING: use real "obj_mdp" sub-data */
1169                 rc = obd_create_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1170                                       &req->rq_oi, &req->rq_oi.oi_md, oti);
1171         }
1172
1173         /* osc_create have timeout equ obd_timeout/2 so waiting don't be
1174          * longer then this */
1175         l_wait_event(set->set_waitq, lov_finished_set(set), &lwi);
1176
1177         /* we not have ptlrpc set for assign set->interpret and should
1178          * be call interpret function himself. calling from cb_create_update
1179          * not permited because lov_fini_create_set can sleep for long time,
1180          * but we must avoid sleeping in ptlrpcd interpret function. */
1181         rc = lov_fini_create_set(set, ea);
1182 out:
1183         obd_putref(exp->exp_obd);
1184         RETURN(rc);
1185 }
1186
1187 #define ASSERT_LSM_MAGIC(lsmp)                                                  \
1188 do {                                                                            \
1189         LASSERT((lsmp) != NULL);                                                \
1190         LASSERTF(((lsmp)->lsm_magic == LOV_MAGIC_V1 ||                          \
1191                  (lsmp)->lsm_magic == LOV_MAGIC_V3),                            \
1192                  "%p->lsm_magic=%x\n", (lsmp), (lsmp)->lsm_magic);              \
1193 } while (0)
1194
1195 static int lov_destroy(struct obd_export *exp, struct obdo *oa,
1196                        struct lov_stripe_md *lsm, struct obd_trans_info *oti,
1197                        struct obd_export *md_exp, void *capa)
1198 {
1199         struct lov_request_set *set;
1200         struct obd_info oinfo;
1201         struct lov_request *req;
1202         cfs_list_t *pos;
1203         struct lov_obd *lov;
1204         int rc = 0, err = 0;
1205         ENTRY;
1206
1207         ASSERT_LSM_MAGIC(lsm);
1208
1209         if (!exp || !exp->exp_obd)
1210                 RETURN(-ENODEV);
1211
1212         if (oa->o_valid & OBD_MD_FLCOOKIE) {
1213                 LASSERT(oti);
1214                 LASSERT(oti->oti_logcookies);
1215         }
1216
1217         lov = &exp->exp_obd->u.lov;
1218         obd_getref(exp->exp_obd);
1219         rc = lov_prep_destroy_set(exp, &oinfo, oa, lsm, oti, &set);
1220         if (rc)
1221                 GOTO(out, rc);
1222
1223         cfs_list_for_each (pos, &set->set_list) {
1224                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1225
1226                 if (oa->o_valid & OBD_MD_FLCOOKIE)
1227                         oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1228
1229                 err = obd_destroy(lov->lov_tgts[req->rq_idx]->ltd_exp,
1230                                   req->rq_oi.oi_oa, NULL, oti, NULL, capa);
1231                 err = lov_update_common_set(set, req, err);
1232                 if (err) {
1233                         CERROR("error: destroying objid "LPX64" subobj "
1234                                LPX64" on OST idx %d: rc = %d\n",
1235                                oa->o_id, req->rq_oi.oi_oa->o_id,
1236                                req->rq_idx, err);
1237                         if (!rc)
1238                                 rc = err;
1239                 }
1240         }
1241
1242         if (rc == 0) {
1243                 LASSERT(lsm_op_find(lsm->lsm_magic) != NULL);
1244                 rc = lsm_op_find(lsm->lsm_magic)->lsm_destroy(lsm, oa, md_exp);
1245         }
1246         err = lov_fini_destroy_set(set);
1247 out:
1248         obd_putref(exp->exp_obd);
1249         RETURN(rc ? rc : err);
1250 }
1251
1252 static int lov_getattr(struct obd_export *exp, struct obd_info *oinfo)
1253 {
1254         struct lov_request_set *set;
1255         struct lov_request *req;
1256         cfs_list_t *pos;
1257         struct lov_obd *lov;
1258         int err = 0, rc = 0;
1259         ENTRY;
1260
1261         LASSERT(oinfo);
1262         ASSERT_LSM_MAGIC(oinfo->oi_md);
1263
1264         if (!exp || !exp->exp_obd)
1265                 RETURN(-ENODEV);
1266
1267         lov = &exp->exp_obd->u.lov;
1268
1269         rc = lov_prep_getattr_set(exp, oinfo, &set);
1270         if (rc)
1271                 RETURN(rc);
1272
1273         cfs_list_for_each (pos, &set->set_list) {
1274                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1275
1276                 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
1277                        "%u\n", oinfo->oi_oa->o_id, req->rq_stripe,
1278                        req->rq_oi.oi_oa->o_id, req->rq_idx);
1279
1280                 rc = obd_getattr(lov->lov_tgts[req->rq_idx]->ltd_exp,
1281                                  &req->rq_oi);
1282                 err = lov_update_common_set(set, req, rc);
1283                 if (err) {
1284                         CERROR("error: getattr objid "LPX64" subobj "
1285                                LPX64" on OST idx %d: rc = %d\n",
1286                                oinfo->oi_oa->o_id, req->rq_oi.oi_oa->o_id,
1287                                req->rq_idx, err);
1288                         break;
1289                 }
1290         }
1291
1292         rc = lov_fini_getattr_set(set);
1293         if (err)
1294                 rc = err;
1295         RETURN(rc);
1296 }
1297
1298 static int lov_getattr_interpret(struct ptlrpc_request_set *rqset,
1299                                  void *data, int rc)
1300 {
1301         struct lov_request_set *lovset = (struct lov_request_set *)data;
1302         int err;
1303         ENTRY;
1304
1305         /* don't do attribute merge if this aysnc op failed */
1306         if (rc)
1307                 lovset->set_completes = 0;
1308         err = lov_fini_getattr_set(lovset);
1309         RETURN(rc ? rc : err);
1310 }
1311
1312 static int lov_getattr_async(struct obd_export *exp, struct obd_info *oinfo,
1313                               struct ptlrpc_request_set *rqset)
1314 {
1315         struct lov_request_set *lovset;
1316         struct lov_obd *lov;
1317         cfs_list_t *pos;
1318         struct lov_request *req;
1319         int rc = 0, err;
1320         ENTRY;
1321
1322         LASSERT(oinfo);
1323         ASSERT_LSM_MAGIC(oinfo->oi_md);
1324
1325         if (!exp || !exp->exp_obd)
1326                 RETURN(-ENODEV);
1327
1328         lov = &exp->exp_obd->u.lov;
1329
1330         rc = lov_prep_getattr_set(exp, oinfo, &lovset);
1331         if (rc)
1332                 RETURN(rc);
1333
1334         CDEBUG(D_INFO, "objid "LPX64": %ux%u byte stripes\n",
1335                oinfo->oi_md->lsm_object_id, oinfo->oi_md->lsm_stripe_count,
1336                oinfo->oi_md->lsm_stripe_size);
1337
1338         cfs_list_for_each (pos, &lovset->set_list) {
1339                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1340
1341                 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
1342                        "%u\n", oinfo->oi_oa->o_id, req->rq_stripe,
1343                        req->rq_oi.oi_oa->o_id, req->rq_idx);
1344                 rc = obd_getattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1345                                        &req->rq_oi, rqset);
1346                 if (rc) {
1347                         CERROR("error: getattr objid "LPX64" subobj "
1348                                LPX64" on OST idx %d: rc = %d\n",
1349                                oinfo->oi_oa->o_id, req->rq_oi.oi_oa->o_id,
1350                                req->rq_idx, rc);
1351                         GOTO(out, rc);
1352                 }
1353         }
1354
1355         if (!cfs_list_empty(&rqset->set_requests)) {
1356                 LASSERT(rc == 0);
1357                 LASSERT (rqset->set_interpret == NULL);
1358                 rqset->set_interpret = lov_getattr_interpret;
1359                 rqset->set_arg = (void *)lovset;
1360                 RETURN(rc);
1361         }
1362 out:
1363         if (rc)
1364                 lovset->set_completes = 0;
1365         err = lov_fini_getattr_set(lovset);
1366         RETURN(rc ? rc : err);
1367 }
1368
1369 static int lov_setattr(struct obd_export *exp, struct obd_info *oinfo,
1370                        struct obd_trans_info *oti)
1371 {
1372         struct lov_request_set *set;
1373         struct lov_obd *lov;
1374         cfs_list_t *pos;
1375         struct lov_request *req;
1376         int err = 0, rc = 0;
1377         ENTRY;
1378
1379         LASSERT(oinfo);
1380         ASSERT_LSM_MAGIC(oinfo->oi_md);
1381
1382         if (!exp || !exp->exp_obd)
1383                 RETURN(-ENODEV);
1384
1385         /* for now, we only expect the following updates here */
1386         LASSERT(!(oinfo->oi_oa->o_valid & ~(OBD_MD_FLID | OBD_MD_FLTYPE |
1387                                             OBD_MD_FLMODE | OBD_MD_FLATIME |
1388                                             OBD_MD_FLMTIME | OBD_MD_FLCTIME |
1389                                             OBD_MD_FLFLAGS | OBD_MD_FLSIZE |
1390                                             OBD_MD_FLGROUP | OBD_MD_FLUID |
1391                                             OBD_MD_FLGID | OBD_MD_FLFID |
1392                                             OBD_MD_FLGENER)));
1393         lov = &exp->exp_obd->u.lov;
1394         rc = lov_prep_setattr_set(exp, oinfo, oti, &set);
1395         if (rc)
1396                 RETURN(rc);
1397
1398         cfs_list_for_each (pos, &set->set_list) {
1399                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1400
1401                 rc = obd_setattr(lov->lov_tgts[req->rq_idx]->ltd_exp,
1402                                  &req->rq_oi, NULL);
1403                 err = lov_update_setattr_set(set, req, rc);
1404                 if (err) {
1405                         CERROR("error: setattr objid "LPX64" subobj "
1406                                LPX64" on OST idx %d: rc = %d\n",
1407                                set->set_oi->oi_oa->o_id,
1408                                req->rq_oi.oi_oa->o_id, req->rq_idx, err);
1409                         if (!rc)
1410                                 rc = err;
1411                 }
1412         }
1413         err = lov_fini_setattr_set(set);
1414         if (!rc)
1415                 rc = err;
1416         RETURN(rc);
1417 }
1418
1419 static int lov_setattr_interpret(struct ptlrpc_request_set *rqset,
1420                                  void *data, int rc)
1421 {
1422         struct lov_request_set *lovset = (struct lov_request_set *)data;
1423         int err;
1424         ENTRY;
1425
1426         if (rc)
1427                 lovset->set_completes = 0;
1428         err = lov_fini_setattr_set(lovset);
1429         RETURN(rc ? rc : err);
1430 }
1431
1432 /* If @oti is given, the request goes from MDS and responses from OSTs are not
1433    needed. Otherwise, a client is waiting for responses. */
1434 static int lov_setattr_async(struct obd_export *exp, struct obd_info *oinfo,
1435                              struct obd_trans_info *oti,
1436                              struct ptlrpc_request_set *rqset)
1437 {
1438         struct lov_request_set *set;
1439         struct lov_request *req;
1440         cfs_list_t *pos;
1441         struct lov_obd *lov;
1442         int rc = 0;
1443         ENTRY;
1444
1445         LASSERT(oinfo);
1446         ASSERT_LSM_MAGIC(oinfo->oi_md);
1447         if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE) {
1448                 LASSERT(oti);
1449                 LASSERT(oti->oti_logcookies);
1450         }
1451
1452         if (!exp || !exp->exp_obd)
1453                 RETURN(-ENODEV);
1454
1455         lov = &exp->exp_obd->u.lov;
1456         rc = lov_prep_setattr_set(exp, oinfo, oti, &set);
1457         if (rc)
1458                 RETURN(rc);
1459
1460         CDEBUG(D_INFO, "objid "LPX64": %ux%u byte stripes\n",
1461                oinfo->oi_md->lsm_object_id, oinfo->oi_md->lsm_stripe_count,
1462                oinfo->oi_md->lsm_stripe_size);
1463
1464         cfs_list_for_each (pos, &set->set_list) {
1465                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1466
1467                 if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE)
1468                         oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1469
1470                 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
1471                        "%u\n", oinfo->oi_oa->o_id, req->rq_stripe,
1472                        req->rq_oi.oi_oa->o_id, req->rq_idx);
1473
1474                 rc = obd_setattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1475                                        &req->rq_oi, oti, rqset);
1476                 if (rc) {
1477                         CERROR("error: setattr objid "LPX64" subobj "
1478                                LPX64" on OST idx %d: rc = %d\n",
1479                                set->set_oi->oi_oa->o_id,
1480                                req->rq_oi.oi_oa->o_id,
1481                                req->rq_idx, rc);
1482                         break;
1483                 }
1484         }
1485
1486         /* If we are not waiting for responses on async requests, return. */
1487         if (rc || !rqset || cfs_list_empty(&rqset->set_requests)) {
1488                 int err;
1489                 if (rc)
1490                         set->set_completes = 0;
1491                 err = lov_fini_setattr_set(set);
1492                 RETURN(rc ? rc : err);
1493         }
1494
1495         LASSERT(rqset->set_interpret == NULL);
1496         rqset->set_interpret = lov_setattr_interpret;
1497         rqset->set_arg = (void *)set;
1498
1499         RETURN(0);
1500 }
1501
1502 static int lov_punch_interpret(struct ptlrpc_request_set *rqset,
1503                                void *data, int rc)
1504 {
1505         struct lov_request_set *lovset = (struct lov_request_set *)data;
1506         int err;
1507         ENTRY;
1508
1509         if (rc)
1510                 lovset->set_completes = 0;
1511         err = lov_fini_punch_set(lovset);
1512         RETURN(rc ? rc : err);
1513 }
1514
1515 /* FIXME: maybe we'll just make one node the authoritative attribute node, then
1516  * we can send this 'punch' to just the authoritative node and the nodes
1517  * that the punch will affect. */
1518 static int lov_punch(struct obd_export *exp, struct obd_info *oinfo,
1519                      struct obd_trans_info *oti,
1520                      struct ptlrpc_request_set *rqset)
1521 {
1522         struct lov_request_set *set;
1523         struct lov_obd *lov;
1524         cfs_list_t *pos;
1525         struct lov_request *req;
1526         int rc = 0;
1527         ENTRY;
1528
1529         LASSERT(oinfo);
1530         ASSERT_LSM_MAGIC(oinfo->oi_md);
1531
1532         if (!exp || !exp->exp_obd)
1533                 RETURN(-ENODEV);
1534
1535         lov = &exp->exp_obd->u.lov;
1536         rc = lov_prep_punch_set(exp, oinfo, oti, &set);
1537         if (rc)
1538                 RETURN(rc);
1539
1540         cfs_list_for_each (pos, &set->set_list) {
1541                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1542
1543                 rc = obd_punch(lov->lov_tgts[req->rq_idx]->ltd_exp,
1544                                &req->rq_oi, NULL, rqset);
1545                 if (rc) {
1546                         CERROR("error: punch objid "LPX64" subobj "LPX64
1547                                " on OST idx %d: rc = %d\n",
1548                                set->set_oi->oi_oa->o_id,
1549                                req->rq_oi.oi_oa->o_id, req->rq_idx, rc);
1550                         break;
1551                 }
1552         }
1553
1554         if (rc || cfs_list_empty(&rqset->set_requests)) {
1555                 int err;
1556                 err = lov_fini_punch_set(set);
1557                 RETURN(rc ? rc : err);
1558         }
1559
1560         LASSERT(rqset->set_interpret == NULL);
1561         rqset->set_interpret = lov_punch_interpret;
1562         rqset->set_arg = (void *)set;
1563
1564         RETURN(0);
1565 }
1566
1567 static int lov_sync_interpret(struct ptlrpc_request_set *rqset,
1568                               void *data, int rc)
1569 {
1570         struct lov_request_set *lovset = data;
1571         int err;
1572         ENTRY;
1573
1574         if (rc)
1575                 lovset->set_completes = 0;
1576         err = lov_fini_sync_set(lovset);
1577         RETURN(rc ?: err);
1578 }
1579
1580 static int lov_sync(struct obd_export *exp, struct obd_info *oinfo,
1581                     obd_off start, obd_off end,
1582                     struct ptlrpc_request_set *rqset)
1583 {
1584         struct lov_request_set *set = NULL;
1585         struct lov_obd *lov;
1586         cfs_list_t *pos;
1587         struct lov_request *req;
1588         int rc = 0;
1589         ENTRY;
1590
1591         ASSERT_LSM_MAGIC(oinfo->oi_md);
1592         LASSERT(rqset != NULL);
1593
1594         if (!exp->exp_obd)
1595                 RETURN(-ENODEV);
1596
1597         lov = &exp->exp_obd->u.lov;
1598         rc = lov_prep_sync_set(exp, oinfo, start, end, &set);
1599         if (rc)
1600                 RETURN(rc);
1601
1602         CDEBUG(D_INFO, "fsync objid "LPX64" ["LPX64", "LPX64"]\n",
1603                set->set_oi->oi_oa->o_id, start, end);
1604
1605         cfs_list_for_each (pos, &set->set_list) {
1606                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1607
1608                 rc = obd_sync(lov->lov_tgts[req->rq_idx]->ltd_exp, &req->rq_oi,
1609                               req->rq_oi.oi_policy.l_extent.start,
1610                               req->rq_oi.oi_policy.l_extent.end, rqset);
1611                 if (rc) {
1612                         CERROR("error: fsync objid "LPX64" subobj "LPX64
1613                                " on OST idx %d: rc = %d\n",
1614                                set->set_oi->oi_oa->o_id,
1615                                req->rq_oi.oi_oa->o_id, req->rq_idx, rc);
1616                         break;
1617                 }
1618         }
1619
1620         /* If we are not waiting for responses on async requests, return. */
1621         if (rc || cfs_list_empty(&rqset->set_requests)) {
1622                 int err = lov_fini_sync_set(set);
1623
1624                 RETURN(rc ?: err);
1625         }
1626
1627         LASSERT(rqset->set_interpret == NULL);
1628         rqset->set_interpret = lov_sync_interpret;
1629         rqset->set_arg = (void *)set;
1630
1631         RETURN(0);
1632 }
1633
1634 static int lov_brw_check(struct lov_obd *lov, struct obd_info *lov_oinfo,
1635                          obd_count oa_bufs, struct brw_page *pga)
1636 {
1637         struct obd_info oinfo = { { { 0 } } };
1638         int i, rc = 0;
1639
1640         oinfo.oi_oa = lov_oinfo->oi_oa;
1641
1642         /* The caller just wants to know if there's a chance that this
1643          * I/O can succeed */
1644         for (i = 0; i < oa_bufs; i++) {
1645                 int stripe = lov_stripe_number(lov_oinfo->oi_md, pga[i].off);
1646                 int ost = lov_oinfo->oi_md->lsm_oinfo[stripe]->loi_ost_idx;
1647                 obd_off start, end;
1648
1649                 if (!lov_stripe_intersects(lov_oinfo->oi_md, i, pga[i].off,
1650                                            pga[i].off + pga[i].count - 1,
1651                                            &start, &end))
1652                         continue;
1653
1654                 if (!lov->lov_tgts[ost] || !lov->lov_tgts[ost]->ltd_active) {
1655                         CDEBUG(D_HA, "lov idx %d inactive\n", ost);
1656                         return -EIO;
1657                 }
1658
1659                 rc = obd_brw(OBD_BRW_CHECK, lov->lov_tgts[ost]->ltd_exp, &oinfo,
1660                              1, &pga[i], NULL);
1661                 if (rc)
1662                         break;
1663         }
1664         return rc;
1665 }
1666
1667 static int lov_brw(int cmd, struct obd_export *exp, struct obd_info *oinfo,
1668                    obd_count oa_bufs, struct brw_page *pga,
1669                    struct obd_trans_info *oti)
1670 {
1671         struct lov_request_set *set;
1672         struct lov_request *req;
1673         cfs_list_t *pos;
1674         struct lov_obd *lov = &exp->exp_obd->u.lov;
1675         int err, rc = 0;
1676         ENTRY;
1677
1678         ASSERT_LSM_MAGIC(oinfo->oi_md);
1679
1680         if (cmd == OBD_BRW_CHECK) {
1681                 rc = lov_brw_check(lov, oinfo, oa_bufs, pga);
1682                 RETURN(rc);
1683         }
1684
1685         rc = lov_prep_brw_set(exp, oinfo, oa_bufs, pga, oti, &set);
1686         if (rc)
1687                 RETURN(rc);
1688
1689         cfs_list_for_each (pos, &set->set_list) {
1690                 struct obd_export *sub_exp;
1691                 struct brw_page *sub_pga;
1692                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1693
1694                 sub_exp = lov->lov_tgts[req->rq_idx]->ltd_exp;
1695                 sub_pga = set->set_pga + req->rq_pgaidx;
1696                 rc = obd_brw(cmd, sub_exp, &req->rq_oi, req->rq_oabufs,
1697                              sub_pga, oti);
1698                 if (rc)
1699                         break;
1700                 lov_update_common_set(set, req, rc);
1701         }
1702
1703         err = lov_fini_brw_set(set);
1704         if (!rc)
1705                 rc = err;
1706         RETURN(rc);
1707 }
1708
1709 static int lov_enqueue_interpret(struct ptlrpc_request_set *rqset,
1710                                  void *data, int rc)
1711 {
1712         struct lov_request_set *lovset = (struct lov_request_set *)data;
1713         ENTRY;
1714         rc = lov_fini_enqueue_set(lovset, lovset->set_ei->ei_mode, rc, rqset);
1715         RETURN(rc);
1716 }
1717
1718 static int lov_enqueue(struct obd_export *exp, struct obd_info *oinfo,
1719                        struct ldlm_enqueue_info *einfo,
1720                        struct ptlrpc_request_set *rqset)
1721 {
1722         ldlm_mode_t mode = einfo->ei_mode;
1723         struct lov_request_set *set;
1724         struct lov_request *req;
1725         cfs_list_t *pos;
1726         struct lov_obd *lov;
1727         ldlm_error_t rc;
1728         ENTRY;
1729
1730         LASSERT(oinfo);
1731         ASSERT_LSM_MAGIC(oinfo->oi_md);
1732         LASSERT(mode == (mode & -mode));
1733
1734         /* we should never be asked to replay a lock this way. */
1735         LASSERT((oinfo->oi_flags & LDLM_FL_REPLAY) == 0);
1736
1737         if (!exp || !exp->exp_obd)
1738                 RETURN(-ENODEV);
1739
1740         lov = &exp->exp_obd->u.lov;
1741         rc = lov_prep_enqueue_set(exp, oinfo, einfo, &set);
1742         if (rc)
1743                 RETURN(rc);
1744
1745         cfs_list_for_each (pos, &set->set_list) {
1746                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1747
1748                 rc = obd_enqueue(lov->lov_tgts[req->rq_idx]->ltd_exp,
1749                                  &req->rq_oi, einfo, rqset);
1750                 if (rc != ELDLM_OK)
1751                         GOTO(out, rc);
1752         }
1753
1754         if (rqset && !cfs_list_empty(&rqset->set_requests)) {
1755                 LASSERT(rc == 0);
1756                 LASSERT(rqset->set_interpret == NULL);
1757                 rqset->set_interpret = lov_enqueue_interpret;
1758                 rqset->set_arg = (void *)set;
1759                 RETURN(rc);
1760         }
1761 out:
1762         rc = lov_fini_enqueue_set(set, mode, rc, rqset);
1763         RETURN(rc);
1764 }
1765
1766 static int lov_change_cbdata(struct obd_export *exp,
1767                              struct lov_stripe_md *lsm, ldlm_iterator_t it,
1768                              void *data)
1769 {
1770         struct lov_obd *lov;
1771         int rc = 0, i;
1772         ENTRY;
1773
1774         ASSERT_LSM_MAGIC(lsm);
1775
1776         if (!exp || !exp->exp_obd)
1777                 RETURN(-ENODEV);
1778
1779         lov = &exp->exp_obd->u.lov;
1780         for (i = 0; i < lsm->lsm_stripe_count; i++) {
1781                 struct lov_stripe_md submd;
1782                 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
1783
1784                 if (!lov->lov_tgts[loi->loi_ost_idx]) {
1785                         CDEBUG(D_HA, "lov idx %d NULL \n", loi->loi_ost_idx);
1786                         continue;
1787                 }
1788
1789                 LASSERT_SEQ_IS_MDT(loi->loi_seq);
1790                 submd.lsm_object_id = loi->loi_id;
1791                 submd.lsm_object_seq = loi->loi_seq;
1792                 submd.lsm_stripe_count = 0;
1793                 rc = obd_change_cbdata(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1794                                        &submd, it, data);
1795         }
1796         RETURN(rc);
1797 }
1798
1799 /* find any ldlm lock of the inode in lov
1800  * return 0    not find
1801  *        1    find one
1802  *      < 0    error */
1803 static int lov_find_cbdata(struct obd_export *exp,
1804                            struct lov_stripe_md *lsm, ldlm_iterator_t it,
1805                            void *data)
1806 {
1807         struct lov_obd *lov;
1808         int rc = 0, i;
1809         ENTRY;
1810
1811         ASSERT_LSM_MAGIC(lsm);
1812
1813         if (!exp || !exp->exp_obd)
1814                 RETURN(-ENODEV);
1815
1816         lov = &exp->exp_obd->u.lov;
1817         for (i = 0; i < lsm->lsm_stripe_count; i++) {
1818                 struct lov_stripe_md submd;
1819                 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
1820
1821                 if (!lov->lov_tgts[loi->loi_ost_idx]) {
1822                         CDEBUG(D_HA, "lov idx %d NULL \n", loi->loi_ost_idx);
1823                         continue;
1824                 }
1825
1826                 LASSERT_SEQ_IS_MDT(loi->loi_seq);
1827                 submd.lsm_object_id = loi->loi_id;
1828                 submd.lsm_object_seq = loi->loi_seq;
1829                 submd.lsm_stripe_count = 0;
1830                 rc = obd_find_cbdata(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1831                                      &submd, it, data);
1832                 if (rc != 0)
1833                         RETURN(rc);
1834         }
1835         RETURN(rc);
1836 }
1837
1838 static int lov_cancel(struct obd_export *exp, struct lov_stripe_md *lsm,
1839                       __u32 mode, struct lustre_handle *lockh)
1840 {
1841         struct lov_request_set *set;
1842         struct obd_info oinfo;
1843         struct lov_request *req;
1844         cfs_list_t *pos;
1845         struct lov_obd *lov;
1846         struct lustre_handle *lov_lockhp;
1847         int err = 0, rc = 0;
1848         ENTRY;
1849
1850         ASSERT_LSM_MAGIC(lsm);
1851
1852         if (!exp || !exp->exp_obd)
1853                 RETURN(-ENODEV);
1854
1855         LASSERT_SEQ_IS_MDT(lsm->lsm_object_seq);
1856         LASSERT(lockh);
1857         lov = &exp->exp_obd->u.lov;
1858         rc = lov_prep_cancel_set(exp, &oinfo, lsm, mode, lockh, &set);
1859         if (rc)
1860                 RETURN(rc);
1861
1862         cfs_list_for_each (pos, &set->set_list) {
1863                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1864                 lov_lockhp = set->set_lockh->llh_handles + req->rq_stripe;
1865
1866                 rc = obd_cancel(lov->lov_tgts[req->rq_idx]->ltd_exp,
1867                                 req->rq_oi.oi_md, mode, lov_lockhp);
1868                 rc = lov_update_common_set(set, req, rc);
1869                 if (rc) {
1870                         CERROR("error: cancel objid "LPX64" subobj "
1871                                LPX64" on OST idx %d: rc = %d\n",
1872                                lsm->lsm_object_id,
1873                                req->rq_oi.oi_md->lsm_object_id,
1874                                req->rq_idx, rc);
1875                         err = rc;
1876                 }
1877
1878         }
1879         lov_fini_cancel_set(set);
1880         RETURN(err);
1881 }
1882
1883 static int lov_cancel_unused(struct obd_export *exp,
1884                              struct lov_stripe_md *lsm,
1885                              ldlm_cancel_flags_t flags, void *opaque)
1886 {
1887         struct lov_obd *lov;
1888         int rc = 0, i;
1889         ENTRY;
1890
1891         if (!exp || !exp->exp_obd)
1892                 RETURN(-ENODEV);
1893
1894         lov = &exp->exp_obd->u.lov;
1895         if (lsm == NULL) {
1896                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
1897                         int err;
1898                         if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
1899                                 continue;
1900
1901                         err = obd_cancel_unused(lov->lov_tgts[i]->ltd_exp, NULL,
1902                                                 flags, opaque);
1903                         if (!rc)
1904                                 rc = err;
1905                 }
1906                 RETURN(rc);
1907         }
1908
1909         ASSERT_LSM_MAGIC(lsm);
1910
1911         LASSERT_SEQ_IS_MDT(lsm->lsm_object_seq);
1912         for (i = 0; i < lsm->lsm_stripe_count; i++) {
1913                 struct lov_stripe_md submd;
1914                 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
1915                 int err;
1916
1917                 if (!lov->lov_tgts[loi->loi_ost_idx]) {
1918                         CDEBUG(D_HA, "lov idx %d NULL\n", loi->loi_ost_idx);
1919                         continue;
1920                 }
1921
1922                 if (!lov->lov_tgts[loi->loi_ost_idx]->ltd_active)
1923                         CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
1924
1925                 submd.lsm_object_id = loi->loi_id;
1926                 submd.lsm_object_seq = loi->loi_seq;
1927                 submd.lsm_stripe_count = 0;
1928                 err = obd_cancel_unused(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1929                                         &submd, flags, opaque);
1930                 if (err && lov->lov_tgts[loi->loi_ost_idx]->ltd_active) {
1931                         CERROR("error: cancel unused objid "LPX64" subobj "LPX64
1932                                " on OST idx %d: rc = %d\n", lsm->lsm_object_id,
1933                                loi->loi_id, loi->loi_ost_idx, err);
1934                         if (!rc)
1935                                 rc = err;
1936                 }
1937         }
1938         RETURN(rc);
1939 }
1940
1941 int lov_statfs_interpret(struct ptlrpc_request_set *rqset, void *data, int rc)
1942 {
1943         struct lov_request_set *lovset = (struct lov_request_set *)data;
1944         int err;
1945         ENTRY;
1946
1947         if (rc)
1948                 lovset->set_completes = 0;
1949
1950         err = lov_fini_statfs_set(lovset);
1951         RETURN(rc ? rc : err);
1952 }
1953
1954 static int lov_statfs_async(struct obd_device *obd, struct obd_info *oinfo,
1955                             __u64 max_age, struct ptlrpc_request_set *rqset)
1956 {
1957         struct lov_request_set *set;
1958         struct lov_request *req;
1959         cfs_list_t *pos;
1960         struct lov_obd *lov;
1961         int rc = 0;
1962         ENTRY;
1963
1964         LASSERT(oinfo != NULL);
1965         LASSERT(oinfo->oi_osfs != NULL);
1966
1967         lov = &obd->u.lov;
1968         rc = lov_prep_statfs_set(obd, oinfo, &set);
1969         if (rc)
1970                 RETURN(rc);
1971
1972         cfs_list_for_each (pos, &set->set_list) {
1973                 struct obd_device *osc_obd;
1974
1975                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1976
1977                 osc_obd = class_exp2obd(lov->lov_tgts[req->rq_idx]->ltd_exp);
1978                 rc = obd_statfs_async(osc_obd, &req->rq_oi, max_age, rqset);
1979                 if (rc)
1980                         break;
1981         }
1982
1983         if (rc || cfs_list_empty(&rqset->set_requests)) {
1984                 int err;
1985                 if (rc)
1986                         set->set_completes = 0;
1987                 err = lov_fini_statfs_set(set);
1988                 RETURN(rc ? rc : err);
1989         }
1990
1991         LASSERT(rqset->set_interpret == NULL);
1992         rqset->set_interpret = lov_statfs_interpret;
1993         rqset->set_arg = (void *)set;
1994         RETURN(0);
1995 }
1996
1997 static int lov_statfs(struct obd_device *obd, struct obd_statfs *osfs,
1998                       __u64 max_age, __u32 flags)
1999 {
2000         struct ptlrpc_request_set *set = NULL;
2001         struct obd_info oinfo = { { { 0 } } };
2002         int rc = 0;
2003         ENTRY;
2004
2005
2006         /* for obdclass we forbid using obd_statfs_rqset, but prefer using async
2007          * statfs requests */
2008         set = ptlrpc_prep_set();
2009         if (set == NULL)
2010                 RETURN(-ENOMEM);
2011
2012         oinfo.oi_osfs = osfs;
2013         oinfo.oi_flags = flags;
2014         rc = lov_statfs_async(obd, &oinfo, max_age, set);
2015         if (rc == 0)
2016                 rc = ptlrpc_set_wait(set);
2017         ptlrpc_set_destroy(set);
2018
2019         RETURN(rc);
2020 }
2021
2022 static int lov_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
2023                          void *karg, void *uarg)
2024 {
2025         struct obd_device *obddev = class_exp2obd(exp);
2026         struct lov_obd *lov = &obddev->u.lov;
2027         int i = 0, rc = 0, count = lov->desc.ld_tgt_count;
2028         struct obd_uuid *uuidp;
2029         ENTRY;
2030
2031         switch (cmd) {
2032         case IOC_OBD_STATFS: {
2033                 struct obd_ioctl_data *data = karg;
2034                 struct obd_device *osc_obd;
2035                 struct obd_statfs stat_buf = {0};
2036                 __u32 index;
2037
2038                 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
2039                 if ((index >= count))
2040                         RETURN(-ENODEV);
2041
2042                 if (!lov->lov_tgts[index])
2043                         /* Try again with the next index */
2044                         RETURN(-EAGAIN);
2045                 if (!lov->lov_tgts[index]->ltd_active)
2046                         RETURN(-ENODATA);
2047
2048                 osc_obd = class_exp2obd(lov->lov_tgts[index]->ltd_exp);
2049                 if (!osc_obd)
2050                         RETURN(-EINVAL);
2051
2052                 /* copy UUID */
2053                 if (cfs_copy_to_user(data->ioc_pbuf2, obd2cli_tgt(osc_obd),
2054                                      min((int) data->ioc_plen2,
2055                                          (int) sizeof(struct obd_uuid))))
2056                         RETURN(-EFAULT);
2057
2058                 /* got statfs data */
2059                 rc = obd_statfs(osc_obd, &stat_buf,
2060                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
2061                                 0);
2062                 if (rc)
2063                         RETURN(rc);
2064                 if (cfs_copy_to_user(data->ioc_pbuf1, &stat_buf,
2065                                      min((int) data->ioc_plen1,
2066                                          (int) sizeof(stat_buf))))
2067                         RETURN(-EFAULT);
2068                 break;
2069         }
2070         case OBD_IOC_LOV_GET_CONFIG: {
2071                 struct obd_ioctl_data *data;
2072                 struct lov_desc *desc;
2073                 char *buf = NULL;
2074                 __u32 *genp;
2075
2076                 len = 0;
2077                 if (obd_ioctl_getdata(&buf, &len, (void *)uarg))
2078                         RETURN(-EINVAL);
2079
2080                 data = (struct obd_ioctl_data *)buf;
2081
2082                 if (sizeof(*desc) > data->ioc_inllen1) {
2083                         obd_ioctl_freedata(buf, len);
2084                         RETURN(-EINVAL);
2085                 }
2086
2087                 if (sizeof(uuidp->uuid) * count > data->ioc_inllen2) {
2088                         obd_ioctl_freedata(buf, len);
2089                         RETURN(-EINVAL);
2090                 }
2091
2092                 if (sizeof(__u32) * count > data->ioc_inllen3) {
2093                         obd_ioctl_freedata(buf, len);
2094                         RETURN(-EINVAL);
2095                 }
2096
2097                 desc = (struct lov_desc *)data->ioc_inlbuf1;
2098                 memcpy(desc, &(lov->desc), sizeof(*desc));
2099
2100                 uuidp = (struct obd_uuid *)data->ioc_inlbuf2;
2101                 genp = (__u32 *)data->ioc_inlbuf3;
2102                 /* the uuid will be empty for deleted OSTs */
2103                 for (i = 0; i < count; i++, uuidp++, genp++) {
2104                         if (!lov->lov_tgts[i])
2105                                 continue;
2106                         *uuidp = lov->lov_tgts[i]->ltd_uuid;
2107                         *genp = lov->lov_tgts[i]->ltd_gen;
2108                 }
2109
2110                 if (cfs_copy_to_user((void *)uarg, buf, len))
2111                         rc = -EFAULT;
2112                 obd_ioctl_freedata(buf, len);
2113                 break;
2114         }
2115         case LL_IOC_LOV_SETSTRIPE:
2116                 rc = lov_setstripe(exp, len, karg, uarg);
2117                 break;
2118         case LL_IOC_LOV_GETSTRIPE:
2119                 rc = lov_getstripe(exp, karg, uarg);
2120                 break;
2121         case LL_IOC_LOV_SETEA:
2122                 rc = lov_setea(exp, karg, uarg);
2123                 break;
2124         case OBD_IOC_QUOTACTL: {
2125                 struct if_quotactl *qctl = karg;
2126                 struct lov_tgt_desc *tgt = NULL;
2127                 struct obd_quotactl *oqctl;
2128
2129                 if (qctl->qc_valid == QC_OSTIDX) {
2130                         if (qctl->qc_idx < 0 || count <= qctl->qc_idx)
2131                                 RETURN(-EINVAL);
2132
2133                         tgt = lov->lov_tgts[qctl->qc_idx];
2134                         if (!tgt || !tgt->ltd_exp)
2135                                 RETURN(-EINVAL);
2136                 } else if (qctl->qc_valid == QC_UUID) {
2137                         for (i = 0; i < count; i++) {
2138                                 tgt = lov->lov_tgts[i];
2139                                 if (!tgt ||
2140                                     !obd_uuid_equals(&tgt->ltd_uuid,
2141                                                      &qctl->obd_uuid))
2142                                         continue;
2143
2144                                 if (tgt->ltd_exp == NULL)
2145                                         RETURN(-EINVAL);
2146
2147                                 break;
2148                         }
2149                 } else {
2150                         RETURN(-EINVAL);
2151                 }
2152
2153                 if (i >= count)
2154                         RETURN(-EAGAIN);
2155
2156                 LASSERT(tgt && tgt->ltd_exp);
2157                 OBD_ALLOC_PTR(oqctl);
2158                 if (!oqctl)
2159                         RETURN(-ENOMEM);
2160
2161                 QCTL_COPY(oqctl, qctl);
2162                 rc = obd_quotactl(tgt->ltd_exp, oqctl);
2163                 if (rc == 0) {
2164                         QCTL_COPY(qctl, oqctl);
2165                         qctl->qc_valid = QC_OSTIDX;
2166                         qctl->obd_uuid = tgt->ltd_uuid;
2167                 }
2168                 OBD_FREE_PTR(oqctl);
2169                 break;
2170         }
2171         default: {
2172                 int set = 0;
2173
2174                 if (count == 0)
2175                         RETURN(-ENOTTY);
2176
2177                 for (i = 0; i < count; i++) {
2178                         int err;
2179                         struct obd_device *osc_obd;
2180
2181                         /* OST was disconnected */
2182                         if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
2183                                 continue;
2184
2185                         /* ll_umount_begin() sets force flag but for lov, not
2186                          * osc. Let's pass it through */
2187                         osc_obd = class_exp2obd(lov->lov_tgts[i]->ltd_exp);
2188                         osc_obd->obd_force = obddev->obd_force;
2189                         err = obd_iocontrol(cmd, lov->lov_tgts[i]->ltd_exp,
2190                                             len, karg, uarg);
2191                         if (err == -ENODATA && cmd == OBD_IOC_POLL_QUOTACHECK) {
2192                                 RETURN(err);
2193                         } else if (err) {
2194                                 if (lov->lov_tgts[i]->ltd_active) {
2195                                         CDEBUG(err == -ENOTTY ?
2196                                                D_IOCTL : D_WARNING,
2197                                                "iocontrol OSC %s on OST "
2198                                                "idx %d cmd %x: err = %d\n",
2199                                                lov_uuid2str(lov, i),
2200                                                i, cmd, err);
2201                                         if (!rc)
2202                                                 rc = err;
2203                                 }
2204                         } else {
2205                                 set = 1;
2206                         }
2207                 }
2208                 if (!set && !rc)
2209                         rc = -EIO;
2210         }
2211         }
2212
2213         RETURN(rc);
2214 }
2215
2216 #define FIEMAP_BUFFER_SIZE 4096
2217
2218 /**
2219  * Non-zero fe_logical indicates that this is a continuation FIEMAP
2220  * call. The local end offset and the device are sent in the first
2221  * fm_extent. This function calculates the stripe number from the index.
2222  * This function returns a stripe_no on which mapping is to be restarted.
2223  *
2224  * This function returns fm_end_offset which is the in-OST offset at which
2225  * mapping should be restarted. If fm_end_offset=0 is returned then caller
2226  * will re-calculate proper offset in next stripe.
2227  * Note that the first extent is passed to lov_get_info via the value field.
2228  *
2229  * \param fiemap fiemap request header
2230  * \param lsm striping information for the file
2231  * \param fm_start logical start of mapping
2232  * \param fm_end logical end of mapping
2233  * \param start_stripe starting stripe will be returned in this
2234  */
2235 obd_size fiemap_calc_fm_end_offset(struct ll_user_fiemap *fiemap,
2236                                    struct lov_stripe_md *lsm, obd_size fm_start,
2237                                    obd_size fm_end, int *start_stripe)
2238 {
2239         obd_size local_end = fiemap->fm_extents[0].fe_logical;
2240         obd_off lun_start, lun_end;
2241         obd_size fm_end_offset;
2242         int stripe_no = -1, i;
2243
2244         if (fiemap->fm_extent_count == 0 ||
2245             fiemap->fm_extents[0].fe_logical == 0)
2246                 return 0;
2247
2248         /* Find out stripe_no from ost_index saved in the fe_device */
2249         for (i = 0; i < lsm->lsm_stripe_count; i++) {
2250                 if (lsm->lsm_oinfo[i]->loi_ost_idx ==
2251                                         fiemap->fm_extents[0].fe_device) {
2252                         stripe_no = i;
2253                         break;
2254                 }
2255         }
2256
2257         /* If we have finished mapping on previous device, shift logical
2258          * offset to start of next device */
2259         if ((lov_stripe_intersects(lsm, stripe_no, fm_start, fm_end,
2260                                    &lun_start, &lun_end)) != 0 &&
2261                                    local_end < lun_end) {
2262                 fm_end_offset = local_end;
2263                 *start_stripe = stripe_no;
2264         } else {
2265                 /* This is a special value to indicate that caller should
2266                  * calculate offset in next stripe. */
2267                 fm_end_offset = 0;
2268                 *start_stripe = (stripe_no + 1) % lsm->lsm_stripe_count;
2269         }
2270
2271         return fm_end_offset;
2272 }
2273
2274 /**
2275  * We calculate on which OST the mapping will end. If the length of mapping
2276  * is greater than (stripe_size * stripe_count) then the last_stripe will
2277  * will be one just before start_stripe. Else we check if the mapping
2278  * intersects each OST and find last_stripe.
2279  * This function returns the last_stripe and also sets the stripe_count
2280  * over which the mapping is spread
2281  *
2282  * \param lsm striping information for the file
2283  * \param fm_start logical start of mapping
2284  * \param fm_end logical end of mapping
2285  * \param start_stripe starting stripe of the mapping
2286  * \param stripe_count the number of stripes across which to map is returned
2287  *
2288  * \retval last_stripe return the last stripe of the mapping
2289  */
2290 int fiemap_calc_last_stripe(struct lov_stripe_md *lsm, obd_size fm_start,
2291                             obd_size fm_end, int start_stripe,
2292                             int *stripe_count)
2293 {
2294         int last_stripe;
2295         obd_off obd_start, obd_end;
2296         int i, j;
2297
2298         if (fm_end - fm_start > lsm->lsm_stripe_size * lsm->lsm_stripe_count) {
2299                 last_stripe = (start_stripe < 1 ? lsm->lsm_stripe_count - 1 :
2300                                                               start_stripe - 1);
2301                 *stripe_count = lsm->lsm_stripe_count;
2302         } else {
2303                 for (j = 0, i = start_stripe; j < lsm->lsm_stripe_count;
2304                      i = (i + 1) % lsm->lsm_stripe_count, j++) {
2305                         if ((lov_stripe_intersects(lsm, i, fm_start, fm_end,
2306                                                    &obd_start, &obd_end)) == 0)
2307                                 break;
2308                 }
2309                 *stripe_count = j;
2310                 last_stripe = (start_stripe + j - 1) %lsm->lsm_stripe_count;
2311         }
2312
2313         return last_stripe;
2314 }
2315
2316 /**
2317  * Set fe_device and copy extents from local buffer into main return buffer.
2318  *
2319  * \param fiemap fiemap request header
2320  * \param lcl_fm_ext array of local fiemap extents to be copied
2321  * \param ost_index OST index to be written into the fm_device field for each
2322                     extent
2323  * \param ext_count number of extents to be copied
2324  * \param current_extent where to start copying in main extent array
2325  */
2326 void fiemap_prepare_and_copy_exts(struct ll_user_fiemap *fiemap,
2327                                   struct ll_fiemap_extent *lcl_fm_ext,
2328                                   int ost_index, unsigned int ext_count,
2329                                   int current_extent)
2330 {
2331         char *to;
2332         int ext;
2333
2334         for (ext = 0; ext < ext_count; ext++) {
2335                 lcl_fm_ext[ext].fe_device = ost_index;
2336                 lcl_fm_ext[ext].fe_flags |= FIEMAP_EXTENT_NET;
2337         }
2338
2339         /* Copy fm_extent's from fm_local to return buffer */
2340         to = (char *)fiemap + fiemap_count_to_size(current_extent);
2341         memcpy(to, lcl_fm_ext, ext_count * sizeof(struct ll_fiemap_extent));
2342 }
2343
2344 /**
2345  * Break down the FIEMAP request and send appropriate calls to individual OSTs.
2346  * This also handles the restarting of FIEMAP calls in case mapping overflows
2347  * the available number of extents in single call.
2348  */
2349 static int lov_fiemap(struct lov_obd *lov, __u32 keylen, void *key,
2350                       __u32 *vallen, void *val, struct lov_stripe_md *lsm)
2351 {
2352         struct ll_fiemap_info_key *fm_key = key;
2353         struct ll_user_fiemap *fiemap = val;
2354         struct ll_user_fiemap *fm_local = NULL;
2355         struct ll_fiemap_extent *lcl_fm_ext;
2356         int count_local;
2357         unsigned int get_num_extents = 0;
2358         int ost_index = 0, actual_start_stripe, start_stripe;
2359         obd_size fm_start, fm_end, fm_length, fm_end_offset = 0;
2360         obd_size curr_loc;
2361         int current_extent = 0, rc = 0, i;
2362         int ost_eof = 0; /* EOF for object */
2363         int ost_done = 0; /* done with required mapping for this OST? */
2364         int last_stripe;
2365         int cur_stripe = 0, cur_stripe_wrap = 0, stripe_count;
2366         unsigned int buffer_size = FIEMAP_BUFFER_SIZE;
2367
2368         if (lsm == NULL)
2369                 GOTO(out, rc = 0);
2370
2371         if (fiemap_count_to_size(fm_key->fiemap.fm_extent_count) < buffer_size)
2372                 buffer_size = fiemap_count_to_size(fm_key->fiemap.fm_extent_count);
2373
2374         OBD_ALLOC_LARGE(fm_local, buffer_size);
2375         if (fm_local == NULL)
2376                 GOTO(out, rc = -ENOMEM);
2377         lcl_fm_ext = &fm_local->fm_extents[0];
2378
2379         count_local = fiemap_size_to_count(buffer_size);
2380
2381         memcpy(fiemap, &fm_key->fiemap, sizeof(*fiemap));
2382         fm_start = fiemap->fm_start;
2383         fm_length = fiemap->fm_length;
2384         /* Calculate start stripe, last stripe and length of mapping */
2385         actual_start_stripe = start_stripe = lov_stripe_number(lsm, fm_start);
2386         fm_end = (fm_length == ~0ULL ? fm_key->oa.o_size :
2387                                                 fm_start + fm_length - 1);
2388         /* If fm_length != ~0ULL but fm_start+fm_length-1 exceeds file size */
2389         if (fm_end > fm_key->oa.o_size)
2390                 fm_end = fm_key->oa.o_size;
2391
2392         last_stripe = fiemap_calc_last_stripe(lsm, fm_start, fm_end,
2393                                             actual_start_stripe, &stripe_count);
2394
2395         fm_end_offset = fiemap_calc_fm_end_offset(fiemap, lsm, fm_start, fm_end,
2396                                                   &start_stripe);
2397
2398         if (fiemap->fm_extent_count == 0) {
2399                 get_num_extents = 1;
2400                 count_local = 0;
2401         }
2402
2403         /* Check each stripe */
2404         for (cur_stripe = start_stripe, i = 0; i < stripe_count;
2405              i++, cur_stripe = (cur_stripe + 1) % lsm->lsm_stripe_count) {
2406                 obd_size req_fm_len; /* Stores length of required mapping */
2407                 obd_size len_mapped_single_call;
2408                 obd_off lun_start, lun_end, obd_object_end;
2409                 unsigned int ext_count;
2410
2411                 cur_stripe_wrap = cur_stripe;
2412
2413                 /* Find out range of mapping on this stripe */
2414                 if ((lov_stripe_intersects(lsm, cur_stripe, fm_start, fm_end,
2415                                            &lun_start, &obd_object_end)) == 0)
2416                         continue;
2417
2418                 /* If this is a continuation FIEMAP call and we are on
2419                  * starting stripe then lun_start needs to be set to
2420                  * fm_end_offset */
2421                 if (fm_end_offset != 0 && cur_stripe == start_stripe)
2422                         lun_start = fm_end_offset;
2423
2424                 if (fm_length != ~0ULL) {
2425                         /* Handle fm_start + fm_length overflow */
2426                         if (fm_start + fm_length < fm_start)
2427                                 fm_length = ~0ULL - fm_start;
2428                         lun_end = lov_size_to_stripe(lsm, fm_start + fm_length,
2429                                                      cur_stripe);
2430                 } else {
2431                         lun_end = ~0ULL;
2432                 }
2433
2434                 if (lun_start == lun_end)
2435                         continue;
2436
2437                 req_fm_len = obd_object_end - lun_start;
2438                 fm_local->fm_length = 0;
2439                 len_mapped_single_call = 0;
2440
2441                 /* If the output buffer is very large and the objects have many
2442                  * extents we may need to loop on a single OST repeatedly */
2443                 ost_eof = 0;
2444                 ost_done = 0;
2445                 do {
2446                         if (get_num_extents == 0) {
2447                                 /* Don't get too many extents. */
2448                                 if (current_extent + count_local >
2449                                     fiemap->fm_extent_count)
2450                                         count_local = fiemap->fm_extent_count -
2451                                                                  current_extent;
2452                         }
2453
2454                         lun_start += len_mapped_single_call;
2455                         fm_local->fm_length = req_fm_len - len_mapped_single_call;
2456                         req_fm_len = fm_local->fm_length;
2457                         fm_local->fm_extent_count = count_local;
2458                         fm_local->fm_mapped_extents = 0;
2459                         fm_local->fm_flags = fiemap->fm_flags;
2460
2461                         fm_key->oa.o_id = lsm->lsm_oinfo[cur_stripe]->loi_id;
2462                         fm_key->oa.o_seq = lsm->lsm_oinfo[cur_stripe]->loi_seq;
2463                         ost_index = lsm->lsm_oinfo[cur_stripe]->loi_ost_idx;
2464
2465                         if (ost_index < 0 || ost_index >=lov->desc.ld_tgt_count)
2466                                 GOTO(out, rc = -EINVAL);
2467
2468                         /* If OST is inactive, return extent with UNKNOWN flag */
2469                         if (!lov->lov_tgts[ost_index]->ltd_active) {
2470                                 fm_local->fm_flags |= FIEMAP_EXTENT_LAST;
2471                                 fm_local->fm_mapped_extents = 1;
2472
2473                                 lcl_fm_ext[0].fe_logical = lun_start;
2474                                 lcl_fm_ext[0].fe_length = obd_object_end -
2475                                                                       lun_start;
2476                                 lcl_fm_ext[0].fe_flags |= FIEMAP_EXTENT_UNKNOWN;
2477
2478                                 goto inactive_tgt;
2479                         }
2480
2481                         fm_local->fm_start = lun_start;
2482                         fm_local->fm_flags &= ~FIEMAP_FLAG_DEVICE_ORDER;
2483                         memcpy(&fm_key->fiemap, fm_local, sizeof(*fm_local));
2484                         *vallen=fiemap_count_to_size(fm_local->fm_extent_count);
2485                         rc = obd_get_info(lov->lov_tgts[ost_index]->ltd_exp,
2486                                           keylen, key, vallen, fm_local, lsm);
2487                         if (rc != 0)
2488                                 GOTO(out, rc);
2489
2490 inactive_tgt:
2491                         ext_count = fm_local->fm_mapped_extents;
2492                         if (ext_count == 0) {
2493                                 ost_done = 1;
2494                                 /* If last stripe has hole at the end,
2495                                  * then we need to return */
2496                                 if (cur_stripe_wrap == last_stripe) {
2497                                         fiemap->fm_mapped_extents = 0;
2498                                         goto finish;
2499                                 }
2500                                 break;
2501                         }
2502
2503                         /* If we just need num of extents then go to next device */
2504                         if (get_num_extents) {
2505                                 current_extent += ext_count;
2506                                 break;
2507                         }
2508
2509                         len_mapped_single_call = lcl_fm_ext[ext_count-1].fe_logical -
2510                                   lun_start + lcl_fm_ext[ext_count - 1].fe_length;
2511
2512                         /* Have we finished mapping on this device? */
2513                         if (req_fm_len <= len_mapped_single_call)
2514                                 ost_done = 1;
2515
2516                         /* Clear the EXTENT_LAST flag which can be present on
2517                          * last extent */
2518                         if (lcl_fm_ext[ext_count-1].fe_flags & FIEMAP_EXTENT_LAST)
2519                                 lcl_fm_ext[ext_count - 1].fe_flags &=
2520                                                             ~FIEMAP_EXTENT_LAST;
2521
2522                         curr_loc = lov_stripe_size(lsm,
2523                                            lcl_fm_ext[ext_count - 1].fe_logical+
2524                                            lcl_fm_ext[ext_count - 1].fe_length,
2525                                            cur_stripe);
2526                         if (curr_loc >= fm_key->oa.o_size)
2527                                 ost_eof = 1;
2528
2529                         fiemap_prepare_and_copy_exts(fiemap, lcl_fm_ext,
2530                                                      ost_index, ext_count,
2531                                                      current_extent);
2532
2533                         current_extent += ext_count;
2534
2535                         /* Ran out of available extents? */
2536                         if (current_extent >= fiemap->fm_extent_count)
2537                                 goto finish;
2538                 } while (ost_done == 0 && ost_eof == 0);
2539
2540                 if (cur_stripe_wrap == last_stripe)
2541                         goto finish;
2542         }
2543
2544 finish:
2545         /* Indicate that we are returning device offsets unless file just has
2546          * single stripe */
2547         if (lsm->lsm_stripe_count > 1)
2548                 fiemap->fm_flags |= FIEMAP_FLAG_DEVICE_ORDER;
2549
2550         if (get_num_extents)
2551                 goto skip_last_device_calc;
2552
2553         /* Check if we have reached the last stripe and whether mapping for that
2554          * stripe is done. */
2555         if (cur_stripe_wrap == last_stripe) {
2556                 if (ost_done || ost_eof)
2557                         fiemap->fm_extents[current_extent - 1].fe_flags |=
2558                                                              FIEMAP_EXTENT_LAST;
2559         }
2560
2561 skip_last_device_calc:
2562         fiemap->fm_mapped_extents = current_extent;
2563
2564 out:
2565         OBD_FREE_LARGE(fm_local, buffer_size);
2566         return rc;
2567 }
2568
2569 static int lov_get_info(struct obd_export *exp, __u32 keylen,
2570                         void *key, __u32 *vallen, void *val,
2571                         struct lov_stripe_md *lsm)
2572 {
2573         struct obd_device *obddev = class_exp2obd(exp);
2574         struct lov_obd *lov = &obddev->u.lov;
2575         int i, rc;
2576         ENTRY;
2577
2578         if (!vallen || !val)
2579                 RETURN(-EFAULT);
2580
2581         obd_getref(obddev);
2582
2583         if (KEY_IS(KEY_LOCK_TO_STRIPE)) {
2584                 struct {
2585                         char name[16];
2586                         struct ldlm_lock *lock;
2587                 } *data = key;
2588                 struct ldlm_res_id *res_id = &data->lock->l_resource->lr_name;
2589                 struct lov_oinfo *loi;
2590                 __u32 *stripe = val;
2591
2592                 if (*vallen < sizeof(*stripe))
2593                         GOTO(out, rc = -EFAULT);
2594                 *vallen = sizeof(*stripe);
2595
2596                 /* XXX This is another one of those bits that will need to
2597                  * change if we ever actually support nested LOVs.  It uses
2598                  * the lock's export to find out which stripe it is. */
2599                 /* XXX - it's assumed all the locks for deleted OSTs have
2600                  * been cancelled. Also, the export for deleted OSTs will
2601                  * be NULL and won't match the lock's export. */
2602                 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2603                         loi = lsm->lsm_oinfo[i];
2604                         if (!lov->lov_tgts[loi->loi_ost_idx])
2605                                 continue;
2606                         if (lov->lov_tgts[loi->loi_ost_idx]->ltd_exp ==
2607                             data->lock->l_conn_export &&
2608                             osc_res_name_eq(loi->loi_id, loi->loi_seq, res_id)) {
2609                                 *stripe = i;
2610                                 GOTO(out, rc = 0);
2611                         }
2612                 }
2613                 LDLM_ERROR(data->lock, "lock on inode without such object");
2614                 dump_lsm(D_ERROR, lsm);
2615                 GOTO(out, rc = -ENXIO);
2616         } else if (KEY_IS(KEY_LAST_ID)) {
2617                 struct obd_id_info *info = val;
2618                 __u32 size = sizeof(obd_id);
2619                 struct lov_tgt_desc *tgt;
2620
2621                 LASSERT(*vallen == sizeof(struct obd_id_info));
2622                 tgt = lov->lov_tgts[info->idx];
2623
2624                 if (!tgt || !tgt->ltd_active)
2625                         GOTO(out, rc = -ESRCH);
2626
2627                 rc = obd_get_info(tgt->ltd_exp, keylen, key, &size, info->data, NULL);
2628                 GOTO(out, rc = 0);
2629         } else if (KEY_IS(KEY_LOVDESC)) {
2630                 struct lov_desc *desc_ret = val;
2631                 *desc_ret = lov->desc;
2632
2633                 GOTO(out, rc = 0);
2634         } else if (KEY_IS(KEY_FIEMAP)) {
2635                 rc = lov_fiemap(lov, keylen, key, vallen, val, lsm);
2636                 GOTO(out, rc);
2637         } else if (KEY_IS(KEY_CONNECT_FLAG)) {
2638                 struct lov_tgt_desc *tgt;
2639                 __u64 ost_idx = *((__u64*)val);
2640
2641                 LASSERT(*vallen == sizeof(__u64));
2642                 LASSERT(ost_idx < lov->desc.ld_tgt_count);
2643                 tgt = lov->lov_tgts[ost_idx];
2644
2645                 if (!tgt || !tgt->ltd_exp)
2646                         GOTO(out, rc = -ESRCH);
2647
2648                 *((__u64*)val) = tgt->ltd_exp->exp_connect_flags;
2649                 GOTO(out, rc = 0);
2650         }
2651
2652         rc = -EINVAL;
2653
2654 out:
2655         obd_putref(obddev);
2656         RETURN(rc);
2657 }
2658
2659 static int lov_set_info_async(struct obd_export *exp, obd_count keylen,
2660                               void *key, obd_count vallen, void *val,
2661                               struct ptlrpc_request_set *set)
2662 {
2663         struct obd_device *obddev = class_exp2obd(exp);
2664         struct lov_obd *lov = &obddev->u.lov;
2665         obd_count count;
2666         int i, rc = 0, err;
2667         struct lov_tgt_desc *tgt;
2668         unsigned incr, check_uuid,
2669                  do_inactive, no_set;
2670         unsigned next_id = 0,  mds_con = 0, capa = 0;
2671         ENTRY;
2672
2673         incr = check_uuid = do_inactive = no_set = 0;
2674         if (set == NULL) {
2675                 no_set = 1;
2676                 set = ptlrpc_prep_set();
2677                 if (!set)
2678                         RETURN(-ENOMEM);
2679         }
2680
2681         obd_getref(obddev);
2682         count = lov->desc.ld_tgt_count;
2683
2684         if (KEY_IS(KEY_NEXT_ID)) {
2685                 count = vallen / sizeof(struct obd_id_info);
2686                 vallen = sizeof(obd_id);
2687                 incr = sizeof(struct obd_id_info);
2688                 do_inactive = 1;
2689                 next_id = 1;
2690         } else if (KEY_IS(KEY_CHECKSUM)) {
2691                 do_inactive = 1;
2692         } else if (KEY_IS(KEY_EVICT_BY_NID)) {
2693                 /* use defaults:  do_inactive = incr = 0; */
2694         } else if (KEY_IS(KEY_MDS_CONN)) {
2695                 mds_con = 1;
2696         } else if (KEY_IS(KEY_CAPA_KEY)) {
2697                 capa = 1;
2698         }
2699
2700         for (i = 0; i < count; i++, val = (char *)val + incr) {
2701                 if (next_id) {
2702                         tgt = lov->lov_tgts[((struct obd_id_info*)val)->idx];
2703                 } else {
2704                         tgt = lov->lov_tgts[i];
2705                 }
2706                 /* OST was disconnected */
2707                 if (!tgt || !tgt->ltd_exp)
2708                         continue;
2709
2710                 /* OST is inactive and we don't want inactive OSCs */
2711                 if (!tgt->ltd_active && !do_inactive)
2712                         continue;
2713
2714                 if (mds_con) {
2715                         struct mds_group_info *mgi;
2716
2717                         LASSERT(vallen == sizeof(*mgi));
2718                         mgi = (struct mds_group_info *)val;
2719
2720                         /* Only want a specific OSC */
2721                         if (mgi->uuid && !obd_uuid_equals(mgi->uuid,
2722                                                 &tgt->ltd_uuid))
2723                                 continue;
2724
2725                         err = obd_set_info_async(tgt->ltd_exp,
2726                                          keylen, key, sizeof(int),
2727                                          &mgi->group, set);
2728                 } else if (next_id) {
2729                         err = obd_set_info_async(tgt->ltd_exp,
2730                                          keylen, key, vallen,
2731                                          ((struct obd_id_info*)val)->data, set);
2732                 } else if (capa) {
2733                         struct mds_capa_info *info = (struct mds_capa_info*)val;
2734
2735                         LASSERT(vallen == sizeof(*info));
2736
2737                          /* Only want a specific OSC */
2738                         if (info->uuid &&
2739                             !obd_uuid_equals(info->uuid, &tgt->ltd_uuid))
2740                                 continue;
2741
2742                         err = obd_set_info_async(tgt->ltd_exp, keylen, key,
2743                                                  sizeof(*info->capa),
2744                                                  info->capa, set);
2745
2746                 } else {
2747                         /* Only want a specific OSC */
2748                         if (check_uuid &&
2749                             !obd_uuid_equals(val, &tgt->ltd_uuid))
2750                                 continue;
2751
2752                         err = obd_set_info_async(tgt->ltd_exp,
2753                                          keylen, key, vallen, val, set);
2754                 }
2755
2756                 if (!rc)
2757                         rc = err;
2758         }
2759
2760         obd_putref(obddev);
2761         if (no_set) {
2762                 err = ptlrpc_set_wait(set);
2763                 if (!rc)
2764                         rc = err;
2765                 ptlrpc_set_destroy(set);
2766         }
2767         RETURN(rc);
2768 }
2769
2770 int lov_test_and_clear_async_rc(struct lov_stripe_md *lsm)
2771 {
2772         int i, rc = 0;
2773         ENTRY;
2774
2775         for (i = 0; i < lsm->lsm_stripe_count; i++) {
2776                 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
2777                 if (loi->loi_ar.ar_rc && !rc)
2778                         rc = loi->loi_ar.ar_rc;
2779                 loi->loi_ar.ar_rc = 0;
2780         }
2781         RETURN(rc);
2782 }
2783 EXPORT_SYMBOL(lov_test_and_clear_async_rc);
2784
2785
2786 static int lov_extent_calc(struct obd_export *exp, struct lov_stripe_md *lsm,
2787                            int cmd, __u64 *offset)
2788 {
2789         __u32 ssize = lsm->lsm_stripe_size;
2790         __u64 start;
2791
2792         start = *offset;
2793         do_div(start, ssize);
2794         start = start * ssize;
2795
2796         CDEBUG(D_DLMTRACE, "offset "LPU64", stripe %u, start "LPU64
2797                            ", end "LPU64"\n", *offset, ssize, start,
2798                            start + ssize - 1);
2799         if (cmd == OBD_CALC_STRIPE_END) {
2800                 *offset = start + ssize - 1;
2801         } else if (cmd == OBD_CALC_STRIPE_START) {
2802                 *offset = start;
2803         } else {
2804                 LBUG();
2805         }
2806
2807         RETURN(0);
2808 }
2809
2810 void lov_stripe_lock(struct lov_stripe_md *md)
2811 {
2812         LASSERT(md->lsm_lock_owner != cfs_curproc_pid());
2813         cfs_spin_lock(&md->lsm_lock);
2814         LASSERT(md->lsm_lock_owner == 0);
2815         md->lsm_lock_owner = cfs_curproc_pid();
2816 }
2817 EXPORT_SYMBOL(lov_stripe_lock);
2818
2819 void lov_stripe_unlock(struct lov_stripe_md *md)
2820 {
2821         LASSERT(md->lsm_lock_owner == cfs_curproc_pid());
2822         md->lsm_lock_owner = 0;
2823         cfs_spin_unlock(&md->lsm_lock);
2824 }
2825 EXPORT_SYMBOL(lov_stripe_unlock);
2826
2827
2828 struct obd_ops lov_obd_ops = {
2829         .o_owner               = THIS_MODULE,
2830         .o_setup               = lov_setup,
2831         .o_precleanup          = lov_precleanup,
2832         .o_cleanup             = lov_cleanup,
2833         //.o_process_config      = lov_process_config,
2834         .o_connect             = lov_connect,
2835         .o_disconnect          = lov_disconnect,
2836         .o_statfs              = lov_statfs,
2837         .o_statfs_async        = lov_statfs_async,
2838         .o_packmd              = lov_packmd,
2839         .o_unpackmd            = lov_unpackmd,
2840         .o_create              = lov_create,
2841         .o_destroy             = lov_destroy,
2842         .o_getattr             = lov_getattr,
2843         .o_getattr_async       = lov_getattr_async,
2844         .o_setattr             = lov_setattr,
2845         .o_setattr_async       = lov_setattr_async,
2846         .o_brw                 = lov_brw,
2847         .o_merge_lvb           = lov_merge_lvb,
2848         .o_adjust_kms          = lov_adjust_kms,
2849         .o_punch               = lov_punch,
2850         .o_sync                = lov_sync,
2851         .o_enqueue             = lov_enqueue,
2852         .o_change_cbdata       = lov_change_cbdata,
2853         .o_find_cbdata         = lov_find_cbdata,
2854         .o_cancel              = lov_cancel,
2855         .o_cancel_unused       = lov_cancel_unused,
2856         .o_iocontrol           = lov_iocontrol,
2857         .o_get_info            = lov_get_info,
2858         .o_set_info_async      = lov_set_info_async,
2859         .o_extent_calc         = lov_extent_calc,
2860         .o_llog_init           = lov_llog_init,
2861         .o_llog_finish         = lov_llog_finish,
2862         .o_notify              = lov_notify,
2863         .o_pool_new            = lov_pool_new,
2864         .o_pool_rem            = lov_pool_remove,
2865         .o_pool_add            = lov_pool_add,
2866         .o_pool_del            = lov_pool_del,
2867         .o_getref              = lov_getref,
2868         .o_putref              = lov_putref,
2869 };
2870
2871 static quota_interface_t *quota_interface;
2872 extern quota_interface_t lov_quota_interface;
2873
2874 cfs_mem_cache_t *lov_oinfo_slab;
2875
2876 extern struct lu_kmem_descr lov_caches[];
2877
2878 int __init lov_init(void)
2879 {
2880         struct lprocfs_static_vars lvars = { 0 };
2881         int rc, rc2;
2882         ENTRY;
2883
2884         /* print an address of _any_ initialized kernel symbol from this
2885          * module, to allow debugging with gdb that doesn't support data
2886          * symbols from modules.*/
2887         CDEBUG(D_CONSOLE, "Lustre LOV module (%p).\n", &lov_caches);
2888
2889         rc = lu_kmem_init(lov_caches);
2890         if (rc)
2891                 return rc;
2892
2893         lov_oinfo_slab = cfs_mem_cache_create("lov_oinfo",
2894                                               sizeof(struct lov_oinfo),
2895                                               0, CFS_SLAB_HWCACHE_ALIGN);
2896         if (lov_oinfo_slab == NULL) {
2897                 lu_kmem_fini(lov_caches);
2898                 return -ENOMEM;
2899         }
2900         lprocfs_lov_init_vars(&lvars);
2901
2902         cfs_request_module("lquota");
2903         quota_interface = PORTAL_SYMBOL_GET(lov_quota_interface);
2904         init_obd_quota_ops(quota_interface, &lov_obd_ops);
2905
2906         rc = class_register_type(&lov_obd_ops, NULL, lvars.module_vars,
2907                                  LUSTRE_LOV_NAME, &lov_device_type);
2908
2909         if (rc) {
2910                 if (quota_interface)
2911                         PORTAL_SYMBOL_PUT(lov_quota_interface);
2912                 rc2 = cfs_mem_cache_destroy(lov_oinfo_slab);
2913                 LASSERT(rc2 == 0);
2914                 lu_kmem_fini(lov_caches);
2915         }
2916
2917         RETURN(rc);
2918 }
2919
2920 #ifdef __KERNEL__
2921 static void /*__exit*/ lov_exit(void)
2922 {
2923         int rc;
2924
2925         lu_device_type_fini(&lov_device_type);
2926         lu_kmem_fini(lov_caches);
2927
2928         if (quota_interface)
2929                 PORTAL_SYMBOL_PUT(lov_quota_interface);
2930
2931         class_unregister_type(LUSTRE_LOV_NAME);
2932         rc = cfs_mem_cache_destroy(lov_oinfo_slab);
2933         LASSERT(rc == 0);
2934 }
2935
2936 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2937 MODULE_DESCRIPTION("Lustre Logical Object Volume OBD driver");
2938 MODULE_LICENSE("GPL");
2939
2940 cfs_module(lov, LUSTRE_VERSION_STRING, lov_init, lov_exit);
2941 #endif