lustre/lov/lov_obd.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2012, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/lov/lov_obd.c
  37  *
  38  * Author: Phil Schwan <phil@clusterfs.com>
  39  * Author: Peter Braam <braam@clusterfs.com>
  40  * Author: Mike Shaver <shaver@clusterfs.com>
  41  * Author: Nathan Rutman <nathan@clusterfs.com>
  42  */
  43
  44 #define DEBUG_SUBSYSTEM S_LOV
  45 #ifdef __KERNEL__
  46 #include <libcfs/libcfs.h>
  47 #else
  48 #include <liblustre.h>
  49 #endif
  50
  51 #include <obd_support.h>
  52 #include <lustre_lib.h>
  53 #include <lustre_net.h>
  54 #include <lustre/lustre_idl.h>
  55 #include <lustre_dlm.h>
  56 #include <lustre_mds.h>
  57 #include <lustre_debug.h>
  58 #include <obd_class.h>
  59 #include <obd_lov.h>
  60 #include <obd_ost.h>
  61 #include <lprocfs_status.h>
  62 #include <lustre_param.h>
  63 #include <cl_object.h>
  64 #include <lclient.h> /* for cl_client_lru */
  65 #include <lustre/ll_fiemap.h>
  66 #include <lustre_log.h>
  67 #include <lustre_fid.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         mutex_lock(&lov->lov_lock);
  79         cfs_atomic_inc(&lov->lov_refcount);
  80         mutex_unlock(&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         mutex_lock(&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                 mutex_unlock(&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                 mutex_unlock(&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         RETURN(0);
 226 }
 227
 228 static int lov_connect(const struct lu_env *env,
 229                        struct obd_export **exp, struct obd_device *obd,
 230                        struct obd_uuid *cluuid, struct obd_connect_data *data,
 231                        void *localdata)
 232 {
 233         struct lov_obd *lov = &obd->u.lov;
 234         struct lov_tgt_desc *tgt;
 235         struct lustre_handle conn;
 236         int i, rc;
 237         ENTRY;
 238
 239         CDEBUG(D_CONFIG, "connect #%d\n", lov->lov_connects);
 240
 241         rc = class_connect(&conn, obd, cluuid);
 242         if (rc)
 243                 RETURN(rc);
 244
 245         *exp = class_conn2export(&conn);
 246
 247         /* Why should there ever be more than 1 connect? */
 248         lov->lov_connects++;
 249         LASSERT(lov->lov_connects == 1);
 250
 251         memset(&lov->lov_ocd, 0, sizeof(lov->lov_ocd));
 252         if (data)
 253                 lov->lov_ocd = *data;
 254
 255         obd_getref(obd);
 256         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
 257                 tgt = lov->lov_tgts[i];
 258                 if (!tgt || obd_uuid_empty(&tgt->ltd_uuid))
 259                         continue;
 260                 /* Flags will be lowest common denominator */
 261                 rc = lov_connect_obd(obd, i, tgt->ltd_activate, &lov->lov_ocd);
 262                 if (rc) {
 263                         CERROR("%s: lov connect tgt %d failed: %d\n",
 264                                obd->obd_name, i, rc);
 265                         continue;
 266                 }
 267                 /* connect to administrative disabled ost */
 268                 if (!lov->lov_tgts[i]->ltd_exp)
 269                         continue;
 270
 271                 rc = lov_notify(obd, lov->lov_tgts[i]->ltd_exp->exp_obd,
 272                                 OBD_NOTIFY_CONNECT, (void *)&i);
 273                 if (rc) {
 274                         CERROR("%s error sending notify %d\n",
 275                                obd->obd_name, rc);
 276                 }
 277         }
 278         obd_putref(obd);
 279
 280         RETURN(0);
 281 }
 282
 283 static int lov_disconnect_obd(struct obd_device *obd, struct lov_tgt_desc *tgt)
 284 {
 285         cfs_proc_dir_entry_t *lov_proc_dir;
 286         struct lov_obd *lov = &obd->u.lov;
 287         struct obd_device *osc_obd;
 288         int rc;
 289         ENTRY;
 290
 291         osc_obd = class_exp2obd(tgt->ltd_exp);
 292         CDEBUG(D_CONFIG, "%s: disconnecting target %s\n",
 293                obd->obd_name, osc_obd->obd_name);
 294
 295         if (tgt->ltd_active) {
 296                 tgt->ltd_active = 0;
 297                 lov->desc.ld_active_tgt_count--;
 298                 tgt->ltd_exp->exp_obd->obd_inactive = 1;
 299         }
 300
 301         lov_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
 302         if (lov_proc_dir) {
 303                 cfs_proc_dir_entry_t *osc_symlink;
 304
 305                 osc_symlink = lprocfs_srch(lov_proc_dir, osc_obd->obd_name);
 306                 if (osc_symlink) {
 307                         lprocfs_remove(&osc_symlink);
 308                 } else {
 309                         CERROR("/proc/fs/lustre/%s/%s/target_obds/%s missing.",
 310                                obd->obd_type->typ_name, obd->obd_name,
 311                                osc_obd->obd_name);
 312                 }
 313         }
 314
 315         if (osc_obd) {
 316                 /* Pass it on to our clients.
 317                  * XXX This should be an argument to disconnect,
 318                  * XXX not a back-door flag on the OBD.  Ah well.
 319                  */
 320                 osc_obd->obd_force = obd->obd_force;
 321                 osc_obd->obd_fail = obd->obd_fail;
 322                 osc_obd->obd_no_recov = obd->obd_no_recov;
 323         }
 324
 325         obd_register_observer(osc_obd, NULL);
 326
 327         rc = obd_disconnect(tgt->ltd_exp);
 328         if (rc) {
 329                 CERROR("Target %s disconnect error %d\n",
 330                        tgt->ltd_uuid.uuid, rc);
 331                 rc = 0;
 332         }
 333
 334         tgt->ltd_exp = NULL;
 335         RETURN(0);
 336 }
 337
 338 static int lov_disconnect(struct obd_export *exp)
 339 {
 340         struct obd_device *obd = class_exp2obd(exp);
 341         struct lov_obd *lov = &obd->u.lov;
 342         int i, rc;
 343         ENTRY;
 344
 345         if (!lov->lov_tgts)
 346                 goto out;
 347
 348         /* Only disconnect the underlying layers on the final disconnect. */
 349         lov->lov_connects--;
 350         if (lov->lov_connects != 0) {
 351                 /* why should there be more than 1 connect? */
 352                 CERROR("disconnect #%d\n", lov->lov_connects);
 353                 goto out;
 354         }
 355
 356         /* Let's hold another reference so lov_del_obd doesn't spin through
 357            putref every time */
 358         obd_getref(obd);
 359
 360         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
 361                 if (lov->lov_tgts[i] && lov->lov_tgts[i]->ltd_exp) {
 362                         /* Disconnection is the last we know about an obd */
 363                         lov_del_target(obd, i, 0, lov->lov_tgts[i]->ltd_gen);
 364                 }
 365         }
 366         obd_putref(obd);
 367
 368 out:
 369         rc = class_disconnect(exp); /* bz 9811 */
 370         RETURN(rc);
 371 }
 372
 373 /* Error codes:
 374  *
 375  *  -EINVAL  : UUID can't be found in the LOV's target list
 376  *  -ENOTCONN: The UUID is found, but the target connection is bad (!)
 377  *  -EBADF   : The UUID is found, but the OBD is the wrong type (!)
 378  *  any >= 0 : is log target index
 379  */
 380 static int lov_set_osc_active(struct obd_device *obd, struct obd_uuid *uuid,
 381                               enum obd_notify_event ev)
 382 {
 383         struct lov_obd *lov = &obd->u.lov;
 384         struct lov_tgt_desc *tgt;
 385         int index, activate, active;
 386         ENTRY;
 387
 388         CDEBUG(D_INFO, "Searching in lov %p for uuid %s event(%d)\n",
 389                lov, uuid->uuid, ev);
 390
 391         obd_getref(obd);
 392         for (index = 0; index < lov->desc.ld_tgt_count; index++) {
 393                 tgt = lov->lov_tgts[index];
 394                 if (!tgt)
 395                         continue;
 396                 /*
 397                  * LU-642, initially inactive OSC could miss the obd_connect,
 398                  * we make up for it here.
 399                  */
 400                 if (ev == OBD_NOTIFY_ACTIVATE && tgt->ltd_exp == NULL &&
 401                     obd_uuid_equals(uuid, &tgt->ltd_uuid)) {
 402                         struct obd_uuid lov_osc_uuid = {"LOV_OSC_UUID"};
 403
 404                         obd_connect(NULL, &tgt->ltd_exp, tgt->ltd_obd,
 405                                     &lov_osc_uuid, &lov->lov_ocd, NULL);
 406                 }
 407                 if (!tgt->ltd_exp)
 408                         continue;
 409
 410                 CDEBUG(D_INFO, "lov idx %d is %s conn "LPX64"\n",
 411                        index, obd_uuid2str(&tgt->ltd_uuid),
 412                        tgt->ltd_exp->exp_handle.h_cookie);
 413                 if (obd_uuid_equals(uuid, &tgt->ltd_uuid))
 414                         break;
 415         }
 416
 417         if (index == lov->desc.ld_tgt_count)
 418                 GOTO(out, index = -EINVAL);
 419
 420         if (ev == OBD_NOTIFY_DEACTIVATE || ev == OBD_NOTIFY_ACTIVATE) {
 421                 activate = (ev == OBD_NOTIFY_ACTIVATE) ? 1 : 0;
 422
 423                 if (lov->lov_tgts[index]->ltd_activate == activate) {
 424                         CDEBUG(D_INFO, "OSC %s already %sactivate!\n",
 425                                uuid->uuid, activate ? "" : "de");
 426                 } else {
 427                         lov->lov_tgts[index]->ltd_activate = activate;
 428                         CDEBUG(D_CONFIG, "%sactivate OSC %s\n",
 429                                activate ? "" : "de", obd_uuid2str(uuid));
 430                 }
 431
 432         } else if (ev == OBD_NOTIFY_INACTIVE || ev == OBD_NOTIFY_ACTIVE) {
 433                 active = (ev == OBD_NOTIFY_ACTIVE) ? 1 : 0;
 434
 435                 if (lov->lov_tgts[index]->ltd_active == active) {
 436                         CDEBUG(D_INFO, "OSC %s already %sactive!\n",
 437                                uuid->uuid, active ? "" : "in");
 438                         GOTO(out, index);
 439                 } else {
 440                         CDEBUG(D_CONFIG, "Marking OSC %s %sactive\n",
 441                                obd_uuid2str(uuid), active ? "" : "in");
 442                 }
 443
 444                 lov->lov_tgts[index]->ltd_active = active;
 445                 if (active) {
 446                         lov->desc.ld_active_tgt_count++;
 447                         lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 0;
 448                 } else {
 449                         lov->desc.ld_active_tgt_count--;
 450                         lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 1;
 451                 }
 452         } else {
 453                 CERROR("Unknown event(%d) for uuid %s", ev, uuid->uuid);
 454         }
 455
 456  out:
 457         obd_putref(obd);
 458         RETURN(index);
 459 }
 460
 461 static int lov_notify(struct obd_device *obd, struct obd_device *watched,
 462                       enum obd_notify_event ev, void *data)
 463 {
 464         int rc = 0;
 465         ENTRY;
 466
 467         if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE ||
 468             ev == OBD_NOTIFY_ACTIVATE || ev == OBD_NOTIFY_DEACTIVATE) {
 469                 struct obd_uuid *uuid;
 470
 471                 LASSERT(watched);
 472
 473                 if (strcmp(watched->obd_type->typ_name, LUSTRE_OSC_NAME)) {
 474                         CERROR("unexpected notification of %s %s!\n",
 475                                watched->obd_type->typ_name,
 476                                watched->obd_name);
 477                         RETURN(-EINVAL);
 478                 }
 479                 uuid = &watched->u.cli.cl_target_uuid;
 480
 481                 /* Set OSC as active before notifying the observer, so the
 482                  * observer can use the OSC normally.
 483                  */
 484                 rc = lov_set_osc_active(obd, uuid, ev);
 485                 if (rc < 0) {
 486                         CERROR("event(%d) of %s failed: %d\n", ev,
 487                                obd_uuid2str(uuid), rc);
 488                         RETURN(rc);
 489                 }
 490                 /* active event should be pass lov target index as data */
 491                 data = &rc;
 492         }
 493
 494         /* Pass the notification up the chain. */
 495         if (watched) {
 496                 rc = obd_notify_observer(obd, watched, ev, data);
 497         } else {
 498                 /* NULL watched means all osc's in the lov (only for syncs) */
 499                 /* sync event should be send lov idx as data */
 500                 struct lov_obd *lov = &obd->u.lov;
 501                 int i, is_sync;
 502
 503                 data = &i;
 504                 is_sync = (ev == OBD_NOTIFY_SYNC) ||
 505                           (ev == OBD_NOTIFY_SYNC_NONBLOCK);
 506
 507                 obd_getref(obd);
 508                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
 509                         if (!lov->lov_tgts[i])
 510                                 continue;
 511
 512                         /* don't send sync event if target not
 513                          * connected/activated */
 514                         if (is_sync &&  !lov->lov_tgts[i]->ltd_active)
 515                                 continue;
 516
 517                         rc = obd_notify_observer(obd, lov->lov_tgts[i]->ltd_obd,
 518                                                  ev, data);
 519                         if (rc) {
 520                                 CERROR("%s: notify %s of %s failed %d\n",
 521                                        obd->obd_name,
 522                                        obd->obd_observer->obd_name,
 523                                        lov->lov_tgts[i]->ltd_obd->obd_name,
 524                                        rc);
 525                         }
 526                 }
 527                 obd_putref(obd);
 528         }
 529
 530         RETURN(rc);
 531 }
 532
 533 static int lov_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
 534                           __u32 index, int gen, int active)
 535 {
 536         struct lov_obd *lov = &obd->u.lov;
 537         struct lov_tgt_desc *tgt;
 538         struct obd_device *tgt_obd;
 539         int rc;
 540         ENTRY;
 541
 542         CDEBUG(D_CONFIG, "uuid:%s idx:%d gen:%d active:%d\n",
 543                uuidp->uuid, index, gen, active);
 544
 545         if (gen <= 0) {
 546                 CERROR("request to add OBD %s with invalid generation: %d\n",
 547                        uuidp->uuid, gen);
 548                 RETURN(-EINVAL);
 549         }
 550
 551         tgt_obd = class_find_client_obd(uuidp, LUSTRE_OSC_NAME,
 552                                         &obd->obd_uuid);
 553         if (tgt_obd == NULL)
 554                 RETURN(-EINVAL);
 555
 556         mutex_lock(&lov->lov_lock);
 557
 558         if ((index < lov->lov_tgt_size) && (lov->lov_tgts[index] != NULL)) {
 559                 tgt = lov->lov_tgts[index];
 560                 CERROR("UUID %s already assigned at LOV target index %d\n",
 561                        obd_uuid2str(&tgt->ltd_uuid), index);
 562                 mutex_unlock(&lov->lov_lock);
 563                 RETURN(-EEXIST);
 564         }
 565
 566         if (index >= lov->lov_tgt_size) {
 567                 /* We need to reallocate the lov target array. */
 568                 struct lov_tgt_desc **newtgts, **old = NULL;
 569                 __u32 newsize, oldsize = 0;
 570
 571                 newsize = max(lov->lov_tgt_size, (__u32)2);
 572                 while (newsize < index + 1)
 573                         newsize = newsize << 1;
 574                 OBD_ALLOC(newtgts, sizeof(*newtgts) * newsize);
 575                 if (newtgts == NULL) {
 576                         mutex_unlock(&lov->lov_lock);
 577                         RETURN(-ENOMEM);
 578                 }
 579
 580                 if (lov->lov_tgt_size) {
 581                         memcpy(newtgts, lov->lov_tgts, sizeof(*newtgts) *
 582                                lov->lov_tgt_size);
 583                         old = lov->lov_tgts;
 584                         oldsize = lov->lov_tgt_size;
 585                 }
 586
 587                 lov->lov_tgts = newtgts;
 588                 lov->lov_tgt_size = newsize;
 589                 smp_rmb();
 590                 if (old)
 591                         OBD_FREE(old, sizeof(*old) * oldsize);
 592
 593                 CDEBUG(D_CONFIG, "tgts: %p size: %d\n",
 594                        lov->lov_tgts, lov->lov_tgt_size);
 595         }
 596
 597         OBD_ALLOC_PTR(tgt);
 598         if (!tgt) {
 599                 mutex_unlock(&lov->lov_lock);
 600                 RETURN(-ENOMEM);
 601         }
 602
 603         rc = lov_ost_pool_add(&lov->lov_packed, index, lov->lov_tgt_size);
 604         if (rc) {
 605                 mutex_unlock(&lov->lov_lock);
 606                 OBD_FREE_PTR(tgt);
 607                 RETURN(rc);
 608         }
 609
 610         tgt->ltd_uuid = *uuidp;
 611         tgt->ltd_obd = tgt_obd;
 612         /* XXX - add a sanity check on the generation number. */
 613         tgt->ltd_gen = gen;
 614         tgt->ltd_index = index;
 615         tgt->ltd_activate = active;
 616         lov->lov_tgts[index] = tgt;
 617         if (index >= lov->desc.ld_tgt_count)
 618                 lov->desc.ld_tgt_count = index + 1;
 619
 620         mutex_unlock(&lov->lov_lock);
 621
 622         CDEBUG(D_CONFIG, "idx=%d ltd_gen=%d ld_tgt_count=%d\n",
 623                 index, tgt->ltd_gen, lov->desc.ld_tgt_count);
 624
 625         rc = obd_notify(obd, tgt_obd, OBD_NOTIFY_CREATE, &index);
 626
 627         if (lov->lov_connects == 0) {
 628                 /* lov_connect hasn't been called yet. We'll do the
 629                    lov_connect_obd on this target when that fn first runs,
 630                    because we don't know the connect flags yet. */
 631                 RETURN(0);
 632         }
 633
 634         obd_getref(obd);
 635
 636         rc = lov_connect_obd(obd, index, active, &lov->lov_ocd);
 637         if (rc)
 638                 GOTO(out, rc);
 639
 640         /* connect to administrative disabled ost */
 641         if (!tgt->ltd_exp)
 642                 GOTO(out, rc = 0);
 643
 644         if (lov->lov_cache != NULL) {
 645                 rc = obd_set_info_async(NULL, tgt->ltd_exp,
 646                                 sizeof(KEY_CACHE_SET), KEY_CACHE_SET,
 647                                 sizeof(struct cl_client_cache), lov->lov_cache,
 648                                 NULL);
 649                 if (rc < 0)
 650                         GOTO(out, rc);
 651         }
 652
 653         rc = lov_notify(obd, tgt->ltd_exp->exp_obd,
 654                         active ? OBD_NOTIFY_CONNECT : OBD_NOTIFY_INACTIVE,
 655                         (void *)&index);
 656
 657 out:
 658         if (rc) {
 659                 CERROR("add failed (%d), deleting %s\n", rc,
 660                        obd_uuid2str(&tgt->ltd_uuid));
 661                 lov_del_target(obd, index, 0, 0);
 662         }
 663         obd_putref(obd);
 664         RETURN(rc);
 665 }
 666
 667 /* Schedule a target for deletion */
 668 int lov_del_target(struct obd_device *obd, __u32 index,
 669                    struct obd_uuid *uuidp, int gen)
 670 {
 671         struct lov_obd *lov = &obd->u.lov;
 672         int count = lov->desc.ld_tgt_count;
 673         int rc = 0;
 674         ENTRY;
 675
 676         if (index >= count) {
 677                 CERROR("LOV target index %d >= number of LOV OBDs %d.\n",
 678                        index, count);
 679                 RETURN(-EINVAL);
 680         }
 681
 682         obd_getref(obd);
 683
 684         if (!lov->lov_tgts[index]) {
 685                 CERROR("LOV target at index %d is not setup.\n", index);
 686                 GOTO(out, rc = -EINVAL);
 687         }
 688
 689         if (uuidp && !obd_uuid_equals(uuidp, &lov->lov_tgts[index]->ltd_uuid)) {
 690                 CERROR("LOV target UUID %s at index %d doesn't match %s.\n",
 691                        lov_uuid2str(lov, index), index,
 692                        obd_uuid2str(uuidp));
 693                 GOTO(out, rc = -EINVAL);
 694         }
 695
 696         CDEBUG(D_CONFIG, "uuid: %s idx: %d gen: %d exp: %p active: %d\n",
 697                lov_uuid2str(lov, index), index,
 698                lov->lov_tgts[index]->ltd_gen, lov->lov_tgts[index]->ltd_exp,
 699                lov->lov_tgts[index]->ltd_active);
 700
 701         lov->lov_tgts[index]->ltd_reap = 1;
 702         lov->lov_death_row++;
 703         /* we really delete it from obd_putref */
 704 out:
 705         obd_putref(obd);
 706
 707         RETURN(rc);
 708 }
 709
 710 static void __lov_del_obd(struct obd_device *obd, struct lov_tgt_desc *tgt)
 711 {
 712         struct obd_device *osc_obd;
 713
 714         LASSERT(tgt);
 715         LASSERT(tgt->ltd_reap);
 716
 717         osc_obd = class_exp2obd(tgt->ltd_exp);
 718
 719         CDEBUG(D_CONFIG, "Removing tgt %s : %s\n",
 720                tgt->ltd_uuid.uuid,
 721                osc_obd ? osc_obd->obd_name : "<no obd>");
 722
 723         if (tgt->ltd_exp)
 724                 lov_disconnect_obd(obd, tgt);
 725
 726         OBD_FREE_PTR(tgt);
 727
 728         /* Manual cleanup - no cleanup logs to clean up the osc's.  We must
 729            do it ourselves. And we can't do it from lov_cleanup,
 730            because we just lost our only reference to it. */
 731         if (osc_obd)
 732                 class_manual_cleanup(osc_obd);
 733 }
 734
 735 void lov_fix_desc_stripe_size(__u64 *val)
 736 {
 737         if (*val < LOV_DEFAULT_STRIPE_SIZE) {
 738                 LCONSOLE_WARN("Increasing default stripe size to min %u\n",
 739                               LOV_DEFAULT_STRIPE_SIZE);
 740                 *val = LOV_DEFAULT_STRIPE_SIZE;
 741         } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) {
 742                 *val &= ~(LOV_MIN_STRIPE_SIZE - 1);
 743                 LCONSOLE_WARN("Changing default stripe size to "LPU64" (a "
 744                               "multiple of %u)\n",
 745                               *val, LOV_MIN_STRIPE_SIZE);
 746         }
 747 }
 748
 749 void lov_fix_desc_stripe_count(__u32 *val)
 750 {
 751         if (*val == 0)
 752                 *val = 1;
 753 }
 754
 755 void lov_fix_desc_pattern(__u32 *val)
 756 {
 757         /* from lov_setstripe */
 758         if ((*val != 0) && (*val != LOV_PATTERN_RAID0)) {
 759                 LCONSOLE_WARN("Unknown stripe pattern: %#x\n", *val);
 760                 *val = 0;
 761         }
 762 }
 763
 764 void lov_fix_desc_qos_maxage(__u32 *val)
 765 {
 766         /* fix qos_maxage */
 767         if (*val == 0)
 768                 *val = QOS_DEFAULT_MAXAGE;
 769 }
 770
 771 void lov_fix_desc(struct lov_desc *desc)
 772 {
 773         lov_fix_desc_stripe_size(&desc->ld_default_stripe_size);
 774         lov_fix_desc_stripe_count(&desc->ld_default_stripe_count);
 775         lov_fix_desc_pattern(&desc->ld_pattern);
 776         lov_fix_desc_qos_maxage(&desc->ld_qos_maxage);
 777 }
 778
 779 int lov_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
 780 {
 781         struct lprocfs_static_vars lvars = { 0 };
 782         struct lov_desc *desc;
 783         struct lov_obd *lov = &obd->u.lov;
 784         int rc;
 785         ENTRY;
 786
 787         if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
 788                 CERROR("LOV setup requires a descriptor\n");
 789                 RETURN(-EINVAL);
 790         }
 791
 792         desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
 793
 794         if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
 795                 CERROR("descriptor size wrong: %d > %d\n",
 796                        (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
 797                 RETURN(-EINVAL);
 798         }
 799
 800         if (desc->ld_magic != LOV_DESC_MAGIC) {
 801                 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
 802                             CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
 803                                    obd->obd_name, desc);
 804                             lustre_swab_lov_desc(desc);
 805                 } else {
 806                         CERROR("%s: Bad lov desc magic: %#x\n",
 807                                obd->obd_name, desc->ld_magic);
 808                         RETURN(-EINVAL);
 809                 }
 810         }
 811
 812         lov_fix_desc(desc);
 813
 814         desc->ld_active_tgt_count = 0;
 815         lov->desc = *desc;
 816         lov->lov_tgt_size = 0;
 817
 818         mutex_init(&lov->lov_lock);
 819         cfs_atomic_set(&lov->lov_refcount, 0);
 820         lov->lov_sp_me = LUSTRE_SP_CLI;
 821
 822         lov->lov_pools_hash_body = cfs_hash_create("POOLS", HASH_POOLS_CUR_BITS,
 823                                                    HASH_POOLS_MAX_BITS,
 824                                                    HASH_POOLS_BKT_BITS, 0,
 825                                                    CFS_HASH_MIN_THETA,
 826                                                    CFS_HASH_MAX_THETA,
 827                                                    &pool_hash_operations,
 828                                                    CFS_HASH_DEFAULT);
 829         CFS_INIT_LIST_HEAD(&lov->lov_pool_list);
 830         lov->lov_pool_count = 0;
 831         rc = lov_ost_pool_init(&lov->lov_packed, 0);
 832         if (rc)
 833                 GOTO(out, rc);
 834
 835         lprocfs_lov_init_vars(&lvars);
 836         lprocfs_obd_setup(obd, lvars.obd_vars);
 837 #ifdef LPROCFS
 838         {
 839                 int rc;
 840
 841                 rc = lprocfs_seq_create(obd->obd_proc_entry, "target_obd",
 842                                         0444, &lov_proc_target_fops, obd);
 843                 if (rc)
 844                         CWARN("Error adding the target_obd file\n");
 845         }
 846 #endif
 847         lov->lov_pool_proc_entry = lprocfs_register("pools",
 848                                                     obd->obd_proc_entry,
 849                                                     NULL, NULL);
 850
 851         RETURN(0);
 852
 853 out:
 854         return rc;
 855 }
 856
 857 static int lov_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
 858 {
 859         int rc = 0;
 860         struct lov_obd *lov = &obd->u.lov;
 861
 862         ENTRY;
 863
 864         switch (stage) {
 865         case OBD_CLEANUP_EARLY: {
 866                 int i;
 867                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
 868                         if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_active)
 869                                 continue;
 870                         obd_precleanup(class_exp2obd(lov->lov_tgts[i]->ltd_exp),
 871                                        OBD_CLEANUP_EARLY);
 872                 }
 873                 break;
 874         }
 875         case OBD_CLEANUP_EXPORTS:
 876                 rc = obd_llog_finish(obd, 0);
 877                 if (rc != 0)
 878                         CERROR("failed to cleanup llogging subsystems\n");
 879                 break;
 880         }
 881         RETURN(rc);
 882 }
 883
 884 static int lov_cleanup(struct obd_device *obd)
 885 {
 886         struct lov_obd *lov = &obd->u.lov;
 887         cfs_list_t *pos, *tmp;
 888         struct pool_desc *pool;
 889         ENTRY;
 890
 891         cfs_list_for_each_safe(pos, tmp, &lov->lov_pool_list) {
 892                 pool = cfs_list_entry(pos, struct pool_desc, pool_list);
 893                 /* free pool structs */
 894                 CDEBUG(D_INFO, "delete pool %p\n", pool);
 895                 /* In the function below, .hs_keycmp resolves to
 896                  * pool_hashkey_keycmp() */
 897                 /* coverity[overrun-buffer-val] */
 898                 lov_pool_del(obd, pool->pool_name);
 899         }
 900         cfs_hash_putref(lov->lov_pools_hash_body);
 901         lov_ost_pool_free(&lov->lov_packed);
 902
 903         lprocfs_obd_cleanup(obd);
 904         if (lov->lov_tgts) {
 905                 int i;
 906                 obd_getref(obd);
 907                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
 908                         if (!lov->lov_tgts[i])
 909                                 continue;
 910
 911                         /* Inactive targets may never have connected */
 912                         if (lov->lov_tgts[i]->ltd_active ||
 913                             cfs_atomic_read(&lov->lov_refcount))
 914                             /* We should never get here - these
 915                                should have been removed in the
 916                              disconnect. */
 917                                 CERROR("lov tgt %d not cleaned!"
 918                                        " deathrow=%d, lovrc=%d\n",
 919                                        i, lov->lov_death_row,
 920                                        cfs_atomic_read(&lov->lov_refcount));
 921                         lov_del_target(obd, i, 0, 0);
 922                 }
 923                 obd_putref(obd);
 924                 OBD_FREE(lov->lov_tgts, sizeof(*lov->lov_tgts) *
 925                          lov->lov_tgt_size);
 926                 lov->lov_tgt_size = 0;
 927         }
 928         RETURN(0);
 929 }
 930
 931 int lov_process_config_base(struct obd_device *obd, struct lustre_cfg *lcfg,
 932                             __u32 *indexp, int *genp)
 933 {
 934         struct obd_uuid obd_uuid;
 935         int cmd;
 936         int rc = 0;
 937         ENTRY;
 938
 939         switch(cmd = lcfg->lcfg_command) {
 940         case LCFG_LOV_ADD_OBD:
 941         case LCFG_LOV_ADD_INA:
 942         case LCFG_LOV_DEL_OBD: {
 943                 __u32 index;
 944                 int gen;
 945                 /* lov_modify_tgts add  0:lov_mdsA  1:ost1_UUID  2:0  3:1 */
 946                 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
 947                         GOTO(out, rc = -EINVAL);
 948
 949                 obd_str2uuid(&obd_uuid,  lustre_cfg_buf(lcfg, 1));
 950
 951                 if (sscanf(lustre_cfg_buf(lcfg, 2), "%d", indexp) != 1)
 952                         GOTO(out, rc = -EINVAL);
 953                 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", genp) != 1)
 954                         GOTO(out, rc = -EINVAL);
 955                 index = *indexp;
 956                 gen = *genp;
 957                 if (cmd == LCFG_LOV_ADD_OBD)
 958                         rc = lov_add_target(obd, &obd_uuid, index, gen, 1);
 959                 else if (cmd == LCFG_LOV_ADD_INA)
 960                         rc = lov_add_target(obd, &obd_uuid, index, gen, 0);
 961                 else
 962                         rc = lov_del_target(obd, index, &obd_uuid, gen);
 963                 GOTO(out, rc);
 964         }
 965         case LCFG_PARAM: {
 966                 struct lprocfs_static_vars lvars = { 0 };
 967                 struct lov_desc *desc = &(obd->u.lov.desc);
 968
 969                 if (!desc)
 970                         GOTO(out, rc = -EINVAL);
 971
 972                 lprocfs_lov_init_vars(&lvars);
 973
 974                 rc = class_process_proc_param(PARAM_LOV, lvars.obd_vars,
 975                                               lcfg, obd);
 976                 if (rc > 0)
 977                         rc = 0;
 978                 GOTO(out, rc);
 979         }
 980         case LCFG_POOL_NEW:
 981         case LCFG_POOL_ADD:
 982         case LCFG_POOL_DEL:
 983         case LCFG_POOL_REM:
 984                 GOTO(out, rc);
 985
 986         default: {
 987                 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
 988                 GOTO(out, rc = -EINVAL);
 989
 990         }
 991         }
 992 out:
 993         RETURN(rc);
 994 }
 995
 996 static int lov_recreate(struct obd_export *exp, struct obdo *src_oa,
 997                         struct lov_stripe_md **ea, struct obd_trans_info *oti)
 998 {
 999         struct lov_stripe_md *obj_mdp, *lsm;
1000         struct lov_obd *lov = &exp->exp_obd->u.lov;
1001         unsigned ost_idx;
1002         int rc, i;
1003         ENTRY;
1004
1005         LASSERT(src_oa->o_valid & OBD_MD_FLFLAGS &&
1006                 src_oa->o_flags & OBD_FL_RECREATE_OBJS);
1007
1008         OBD_ALLOC(obj_mdp, sizeof(*obj_mdp));
1009         if (obj_mdp == NULL)
1010                 RETURN(-ENOMEM);
1011
1012         ost_idx = src_oa->o_nlink;
1013         lsm = *ea;
1014         if (lsm == NULL)
1015                 GOTO(out, rc = -EINVAL);
1016         if (ost_idx >= lov->desc.ld_tgt_count ||
1017             !lov->lov_tgts[ost_idx])
1018                 GOTO(out, rc = -EINVAL);
1019
1020         for (i = 0; i < lsm->lsm_stripe_count; i++) {
1021                 if (lsm->lsm_oinfo[i]->loi_ost_idx == ost_idx) {
1022                         if (lsm->lsm_oinfo[i]->loi_id != src_oa->o_id)
1023                                 GOTO(out, rc = -EINVAL);
1024                         break;
1025                 }
1026         }
1027         if (i == lsm->lsm_stripe_count)
1028                 GOTO(out, rc = -EINVAL);
1029
1030         rc = obd_create(NULL, lov->lov_tgts[ost_idx]->ltd_exp,
1031                         src_oa, &obj_mdp, oti);
1032 out:
1033         OBD_FREE(obj_mdp, sizeof(*obj_mdp));
1034         RETURN(rc);
1035 }
1036
1037 /* the LOV expects oa->o_id to be set to the LOV object id */
1038 static int lov_create(const struct lu_env *env, struct obd_export *exp,
1039                       struct obdo *src_oa, struct lov_stripe_md **ea,
1040                       struct obd_trans_info *oti)
1041 {
1042         struct lov_obd *lov;
1043         int rc = 0;
1044         ENTRY;
1045
1046         LASSERT(ea != NULL);
1047         if (exp == NULL)
1048                 RETURN(-EINVAL);
1049
1050         if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
1051             src_oa->o_flags == OBD_FL_DELORPHAN) {
1052                 /* should be used with LOV anymore */
1053                 LBUG();
1054         }
1055
1056         lov = &exp->exp_obd->u.lov;
1057         if (!lov->desc.ld_active_tgt_count)
1058                 RETURN(-EIO);
1059
1060         obd_getref(exp->exp_obd);
1061         /* Recreate a specific object id at the given OST index */
1062         if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
1063             (src_oa->o_flags & OBD_FL_RECREATE_OBJS)) {
1064                  rc = lov_recreate(exp, src_oa, ea, oti);
1065         }
1066
1067         obd_putref(exp->exp_obd);
1068         RETURN(rc);
1069 }
1070
1071 #define ASSERT_LSM_MAGIC(lsmp)                                                  \
1072 do {                                                                            \
1073         LASSERT((lsmp) != NULL);                                                \
1074         LASSERTF(((lsmp)->lsm_magic == LOV_MAGIC_V1 ||                          \
1075                  (lsmp)->lsm_magic == LOV_MAGIC_V3),                            \
1076                  "%p->lsm_magic=%x\n", (lsmp), (lsmp)->lsm_magic);              \
1077 } while (0)
1078
1079 static int lov_destroy(const struct lu_env *env, struct obd_export *exp,
1080                        struct obdo *oa, struct lov_stripe_md *lsm,
1081                        struct obd_trans_info *oti, struct obd_export *md_exp,
1082                        void *capa)
1083 {
1084         struct lov_request_set *set;
1085         struct obd_info oinfo;
1086         struct lov_request *req;
1087         cfs_list_t *pos;
1088         struct lov_obd *lov;
1089         int rc = 0, err = 0;
1090         ENTRY;
1091
1092         ASSERT_LSM_MAGIC(lsm);
1093
1094         if (!exp || !exp->exp_obd)
1095                 RETURN(-ENODEV);
1096
1097         if (oa->o_valid & OBD_MD_FLCOOKIE) {
1098                 LASSERT(oti);
1099                 LASSERT(oti->oti_logcookies);
1100         }
1101
1102         lov = &exp->exp_obd->u.lov;
1103         obd_getref(exp->exp_obd);
1104         rc = lov_prep_destroy_set(exp, &oinfo, oa, lsm, oti, &set);
1105         if (rc)
1106                 GOTO(out, rc);
1107
1108         cfs_list_for_each (pos, &set->set_list) {
1109                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1110
1111                 if (oa->o_valid & OBD_MD_FLCOOKIE)
1112                         oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1113
1114                 err = obd_destroy(env, lov->lov_tgts[req->rq_idx]->ltd_exp,
1115                                   req->rq_oi.oi_oa, NULL, oti, NULL, capa);
1116                 err = lov_update_common_set(set, req, err);
1117                 if (err) {
1118                         CERROR("error: destroying objid "LPX64" subobj "
1119                                LPX64" on OST idx %d: rc = %d\n",
1120                                oa->o_id, req->rq_oi.oi_oa->o_id,
1121                                req->rq_idx, err);
1122                         if (!rc)
1123                                 rc = err;
1124                 }
1125         }
1126
1127         if (rc == 0) {
1128                 LASSERT(lsm_op_find(lsm->lsm_magic) != NULL);
1129                 rc = lsm_op_find(lsm->lsm_magic)->lsm_destroy(lsm, oa, md_exp);
1130         }
1131         err = lov_fini_destroy_set(set);
1132 out:
1133         obd_putref(exp->exp_obd);
1134         RETURN(rc ? rc : err);
1135 }
1136
1137 static int lov_getattr(const struct lu_env *env, struct obd_export *exp,
1138                        struct obd_info *oinfo)
1139 {
1140         struct lov_request_set *set;
1141         struct lov_request *req;
1142         cfs_list_t *pos;
1143         struct lov_obd *lov;
1144         int err = 0, rc = 0;
1145         ENTRY;
1146
1147         LASSERT(oinfo);
1148         ASSERT_LSM_MAGIC(oinfo->oi_md);
1149
1150         if (!exp || !exp->exp_obd)
1151                 RETURN(-ENODEV);
1152
1153         lov = &exp->exp_obd->u.lov;
1154
1155         rc = lov_prep_getattr_set(exp, oinfo, &set);
1156         if (rc)
1157                 RETURN(rc);
1158
1159         cfs_list_for_each (pos, &set->set_list) {
1160                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1161
1162                 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
1163                        "%u\n", oinfo->oi_oa->o_id, req->rq_stripe,
1164                        req->rq_oi.oi_oa->o_id, req->rq_idx);
1165
1166                 rc = obd_getattr(env, lov->lov_tgts[req->rq_idx]->ltd_exp,
1167                                  &req->rq_oi);
1168                 err = lov_update_common_set(set, req, rc);
1169                 if (err) {
1170                         CERROR("error: getattr objid "LPX64" subobj "
1171                                LPX64" on OST idx %d: rc = %d\n",
1172                                oinfo->oi_oa->o_id, req->rq_oi.oi_oa->o_id,
1173                                req->rq_idx, err);
1174                         break;
1175                 }
1176         }
1177
1178         rc = lov_fini_getattr_set(set);
1179         if (err)
1180                 rc = err;
1181         RETURN(rc);
1182 }
1183
1184 static int lov_getattr_interpret(struct ptlrpc_request_set *rqset,
1185                                  void *data, int rc)
1186 {
1187         struct lov_request_set *lovset = (struct lov_request_set *)data;
1188         int err;
1189         ENTRY;
1190
1191         /* don't do attribute merge if this aysnc op failed */
1192         if (rc)
1193                 cfs_atomic_set(&lovset->set_completes, 0);
1194         err = lov_fini_getattr_set(lovset);
1195         RETURN(rc ? rc : err);
1196 }
1197
1198 static int lov_getattr_async(struct obd_export *exp, struct obd_info *oinfo,
1199                               struct ptlrpc_request_set *rqset)
1200 {
1201         struct lov_request_set *lovset;
1202         struct lov_obd *lov;
1203         cfs_list_t *pos;
1204         struct lov_request *req;
1205         int rc = 0, err;
1206         ENTRY;
1207
1208         LASSERT(oinfo);
1209         ASSERT_LSM_MAGIC(oinfo->oi_md);
1210
1211         if (!exp || !exp->exp_obd)
1212                 RETURN(-ENODEV);
1213
1214         lov = &exp->exp_obd->u.lov;
1215
1216         rc = lov_prep_getattr_set(exp, oinfo, &lovset);
1217         if (rc)
1218                 RETURN(rc);
1219
1220         CDEBUG(D_INFO, "objid "LPX64": %ux%u byte stripes\n",
1221                oinfo->oi_md->lsm_object_id, oinfo->oi_md->lsm_stripe_count,
1222                oinfo->oi_md->lsm_stripe_size);
1223
1224         cfs_list_for_each (pos, &lovset->set_list) {
1225                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1226
1227                 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
1228                        "%u\n", oinfo->oi_oa->o_id, req->rq_stripe,
1229                        req->rq_oi.oi_oa->o_id, req->rq_idx);
1230                 rc = obd_getattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1231                                        &req->rq_oi, rqset);
1232                 if (rc) {
1233                         CERROR("error: getattr objid "LPX64" subobj "
1234                                LPX64" on OST idx %d: rc = %d\n",
1235                                oinfo->oi_oa->o_id, req->rq_oi.oi_oa->o_id,
1236                                req->rq_idx, rc);
1237                         GOTO(out, rc);
1238                 }
1239         }
1240
1241         if (!cfs_list_empty(&rqset->set_requests)) {
1242                 LASSERT(rc == 0);
1243                 LASSERT (rqset->set_interpret == NULL);
1244                 rqset->set_interpret = lov_getattr_interpret;
1245                 rqset->set_arg = (void *)lovset;
1246                 RETURN(rc);
1247         }
1248 out:
1249         if (rc)
1250                 cfs_atomic_set(&lovset->set_completes, 0);
1251         err = lov_fini_getattr_set(lovset);
1252         RETURN(rc ? rc : err);
1253 }
1254
1255 static int lov_setattr(const struct lu_env *env, struct obd_export *exp,
1256                        struct obd_info *oinfo, struct obd_trans_info *oti)
1257 {
1258         struct lov_request_set *set;
1259         struct lov_obd *lov;
1260         cfs_list_t *pos;
1261         struct lov_request *req;
1262         int err = 0, rc = 0;
1263         ENTRY;
1264
1265         LASSERT(oinfo);
1266         ASSERT_LSM_MAGIC(oinfo->oi_md);
1267
1268         if (!exp || !exp->exp_obd)
1269                 RETURN(-ENODEV);
1270
1271         /* for now, we only expect the following updates here */
1272         LASSERT(!(oinfo->oi_oa->o_valid & ~(OBD_MD_FLID | OBD_MD_FLTYPE |
1273                                             OBD_MD_FLMODE | OBD_MD_FLATIME |
1274                                             OBD_MD_FLMTIME | OBD_MD_FLCTIME |
1275                                             OBD_MD_FLFLAGS | OBD_MD_FLSIZE |
1276                                             OBD_MD_FLGROUP | OBD_MD_FLUID |
1277                                             OBD_MD_FLGID | OBD_MD_FLFID |
1278                                             OBD_MD_FLGENER)));
1279         lov = &exp->exp_obd->u.lov;
1280         rc = lov_prep_setattr_set(exp, oinfo, oti, &set);
1281         if (rc)
1282                 RETURN(rc);
1283
1284         cfs_list_for_each (pos, &set->set_list) {
1285                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1286
1287                 rc = obd_setattr(env, lov->lov_tgts[req->rq_idx]->ltd_exp,
1288                                  &req->rq_oi, NULL);
1289                 err = lov_update_setattr_set(set, req, rc);
1290                 if (err) {
1291                         CERROR("error: setattr objid "LPX64" subobj "
1292                                LPX64" on OST idx %d: rc = %d\n",
1293                                set->set_oi->oi_oa->o_id,
1294                                req->rq_oi.oi_oa->o_id, req->rq_idx, err);
1295                         if (!rc)
1296                                 rc = err;
1297                 }
1298         }
1299         err = lov_fini_setattr_set(set);
1300         if (!rc)
1301                 rc = err;
1302         RETURN(rc);
1303 }
1304
1305 static int lov_setattr_interpret(struct ptlrpc_request_set *rqset,
1306                                  void *data, int rc)
1307 {
1308         struct lov_request_set *lovset = (struct lov_request_set *)data;
1309         int err;
1310         ENTRY;
1311
1312         if (rc)
1313                 cfs_atomic_set(&lovset->set_completes, 0);
1314         err = lov_fini_setattr_set(lovset);
1315         RETURN(rc ? rc : err);
1316 }
1317
1318 /* If @oti is given, the request goes from MDS and responses from OSTs are not
1319    needed. Otherwise, a client is waiting for responses. */
1320 static int lov_setattr_async(struct obd_export *exp, struct obd_info *oinfo,
1321                              struct obd_trans_info *oti,
1322                              struct ptlrpc_request_set *rqset)
1323 {
1324         struct lov_request_set *set;
1325         struct lov_request *req;
1326         cfs_list_t *pos;
1327         struct lov_obd *lov;
1328         int rc = 0;
1329         ENTRY;
1330
1331         LASSERT(oinfo);
1332         ASSERT_LSM_MAGIC(oinfo->oi_md);
1333         if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE) {
1334                 LASSERT(oti);
1335                 LASSERT(oti->oti_logcookies);
1336         }
1337
1338         if (!exp || !exp->exp_obd)
1339                 RETURN(-ENODEV);
1340
1341         lov = &exp->exp_obd->u.lov;
1342         rc = lov_prep_setattr_set(exp, oinfo, oti, &set);
1343         if (rc)
1344                 RETURN(rc);
1345
1346         CDEBUG(D_INFO, "objid "LPX64": %ux%u byte stripes\n",
1347                oinfo->oi_md->lsm_object_id, oinfo->oi_md->lsm_stripe_count,
1348                oinfo->oi_md->lsm_stripe_size);
1349
1350         cfs_list_for_each (pos, &set->set_list) {
1351                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1352
1353                 if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE)
1354                         oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1355
1356                 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
1357                        "%u\n", oinfo->oi_oa->o_id, req->rq_stripe,
1358                        req->rq_oi.oi_oa->o_id, req->rq_idx);
1359
1360                 rc = obd_setattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1361                                        &req->rq_oi, oti, rqset);
1362                 if (rc) {
1363                         CERROR("error: setattr objid "LPX64" subobj "
1364                                LPX64" on OST idx %d: rc = %d\n",
1365                                set->set_oi->oi_oa->o_id,
1366                                req->rq_oi.oi_oa->o_id,
1367                                req->rq_idx, rc);
1368                         break;
1369                 }
1370         }
1371
1372         /* If we are not waiting for responses on async requests, return. */
1373         if (rc || !rqset || cfs_list_empty(&rqset->set_requests)) {
1374                 int err;
1375                 if (rc)
1376                         cfs_atomic_set(&set->set_completes, 0);
1377                 err = lov_fini_setattr_set(set);
1378                 RETURN(rc ? rc : err);
1379         }
1380
1381         LASSERT(rqset->set_interpret == NULL);
1382         rqset->set_interpret = lov_setattr_interpret;
1383         rqset->set_arg = (void *)set;
1384
1385         RETURN(0);
1386 }
1387
1388 static int lov_punch_interpret(struct ptlrpc_request_set *rqset,
1389                                void *data, int rc)
1390 {
1391         struct lov_request_set *lovset = (struct lov_request_set *)data;
1392         int err;
1393         ENTRY;
1394
1395         if (rc)
1396                 cfs_atomic_set(&lovset->set_completes, 0);
1397         err = lov_fini_punch_set(lovset);
1398         RETURN(rc ? rc : err);
1399 }
1400
1401 /* FIXME: maybe we'll just make one node the authoritative attribute node, then
1402  * we can send this 'punch' to just the authoritative node and the nodes
1403  * that the punch will affect. */
1404 static int lov_punch(const struct lu_env *env, struct obd_export *exp,
1405                      struct obd_info *oinfo, struct obd_trans_info *oti,
1406                      struct ptlrpc_request_set *rqset)
1407 {
1408         struct lov_request_set *set;
1409         struct lov_obd *lov;
1410         cfs_list_t *pos;
1411         struct lov_request *req;
1412         int rc = 0;
1413         ENTRY;
1414
1415         LASSERT(oinfo);
1416         ASSERT_LSM_MAGIC(oinfo->oi_md);
1417
1418         if (!exp || !exp->exp_obd)
1419                 RETURN(-ENODEV);
1420
1421         lov = &exp->exp_obd->u.lov;
1422         rc = lov_prep_punch_set(exp, oinfo, oti, &set);
1423         if (rc)
1424                 RETURN(rc);
1425
1426         cfs_list_for_each (pos, &set->set_list) {
1427                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1428
1429                 rc = obd_punch(env, lov->lov_tgts[req->rq_idx]->ltd_exp,
1430                                &req->rq_oi, NULL, rqset);
1431                 if (rc) {
1432                         CERROR("error: punch objid "LPX64" subobj "LPX64
1433                                " on OST idx %d: rc = %d\n",
1434                                set->set_oi->oi_oa->o_id,
1435                                req->rq_oi.oi_oa->o_id, req->rq_idx, rc);
1436                         break;
1437                 }
1438         }
1439
1440         if (rc || cfs_list_empty(&rqset->set_requests)) {
1441                 int err;
1442                 err = lov_fini_punch_set(set);
1443                 RETURN(rc ? rc : err);
1444         }
1445
1446         LASSERT(rqset->set_interpret == NULL);
1447         rqset->set_interpret = lov_punch_interpret;
1448         rqset->set_arg = (void *)set;
1449
1450         RETURN(0);
1451 }
1452
1453 static int lov_sync_interpret(struct ptlrpc_request_set *rqset,
1454                               void *data, int rc)
1455 {
1456         struct lov_request_set *lovset = data;
1457         int err;
1458         ENTRY;
1459
1460         if (rc)
1461                 cfs_atomic_set(&lovset->set_completes, 0);
1462         err = lov_fini_sync_set(lovset);
1463         RETURN(rc ?: err);
1464 }
1465
1466 static int lov_sync(const struct lu_env *env, struct obd_export *exp,
1467                     struct obd_info *oinfo, obd_off start, obd_off end,
1468                     struct ptlrpc_request_set *rqset)
1469 {
1470         struct lov_request_set *set = NULL;
1471         struct lov_obd *lov;
1472         cfs_list_t *pos;
1473         struct lov_request *req;
1474         int rc = 0;
1475         ENTRY;
1476
1477         ASSERT_LSM_MAGIC(oinfo->oi_md);
1478         LASSERT(rqset != NULL);
1479
1480         if (!exp->exp_obd)
1481                 RETURN(-ENODEV);
1482
1483         lov = &exp->exp_obd->u.lov;
1484         rc = lov_prep_sync_set(exp, oinfo, start, end, &set);
1485         if (rc)
1486                 RETURN(rc);
1487
1488         CDEBUG(D_INFO, "fsync objid "LPX64" ["LPX64", "LPX64"]\n",
1489                set->set_oi->oi_oa->o_id, start, end);
1490
1491         cfs_list_for_each (pos, &set->set_list) {
1492                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1493
1494                 rc = obd_sync(env, lov->lov_tgts[req->rq_idx]->ltd_exp,
1495                               &req->rq_oi, req->rq_oi.oi_policy.l_extent.start,
1496                               req->rq_oi.oi_policy.l_extent.end, rqset);
1497                 if (rc) {
1498                         CERROR("error: fsync objid "LPX64" subobj "LPX64
1499                                " on OST idx %d: rc = %d\n",
1500                                set->set_oi->oi_oa->o_id,
1501                                req->rq_oi.oi_oa->o_id, req->rq_idx, rc);
1502                         break;
1503                 }
1504         }
1505
1506         /* If we are not waiting for responses on async requests, return. */
1507         if (rc || cfs_list_empty(&rqset->set_requests)) {
1508                 int err = lov_fini_sync_set(set);
1509
1510                 RETURN(rc ?: err);
1511         }
1512
1513         LASSERT(rqset->set_interpret == NULL);
1514         rqset->set_interpret = lov_sync_interpret;
1515         rqset->set_arg = (void *)set;
1516
1517         RETURN(0);
1518 }
1519
1520 static int lov_brw_check(struct lov_obd *lov, struct obd_info *lov_oinfo,
1521                          obd_count oa_bufs, struct brw_page *pga)
1522 {
1523         struct obd_info oinfo = { { { 0 } } };
1524         int i, rc = 0;
1525
1526         oinfo.oi_oa = lov_oinfo->oi_oa;
1527
1528         /* The caller just wants to know if there's a chance that this
1529          * I/O can succeed */
1530         for (i = 0; i < oa_bufs; i++) {
1531                 int stripe = lov_stripe_number(lov_oinfo->oi_md, pga[i].off);
1532                 int ost = lov_oinfo->oi_md->lsm_oinfo[stripe]->loi_ost_idx;
1533                 obd_off start, end;
1534
1535                 if (!lov_stripe_intersects(lov_oinfo->oi_md, i, pga[i].off,
1536                                            pga[i].off + pga[i].count - 1,
1537                                            &start, &end))
1538                         continue;
1539
1540                 if (!lov->lov_tgts[ost] || !lov->lov_tgts[ost]->ltd_active) {
1541                         CDEBUG(D_HA, "lov idx %d inactive\n", ost);
1542                         return -EIO;
1543                 }
1544
1545                 rc = obd_brw(OBD_BRW_CHECK, lov->lov_tgts[ost]->ltd_exp, &oinfo,
1546                              1, &pga[i], NULL);
1547                 if (rc)
1548                         break;
1549         }
1550         return rc;
1551 }
1552
1553 static int lov_brw(int cmd, struct obd_export *exp, struct obd_info *oinfo,
1554                    obd_count oa_bufs, struct brw_page *pga,
1555                    struct obd_trans_info *oti)
1556 {
1557         struct lov_request_set *set;
1558         struct lov_request *req;
1559         cfs_list_t *pos;
1560         struct lov_obd *lov = &exp->exp_obd->u.lov;
1561         int err, rc = 0;
1562         ENTRY;
1563
1564         ASSERT_LSM_MAGIC(oinfo->oi_md);
1565
1566         if (cmd == OBD_BRW_CHECK) {
1567                 rc = lov_brw_check(lov, oinfo, oa_bufs, pga);
1568                 RETURN(rc);
1569         }
1570
1571         rc = lov_prep_brw_set(exp, oinfo, oa_bufs, pga, oti, &set);
1572         if (rc)
1573                 RETURN(rc);
1574
1575         cfs_list_for_each (pos, &set->set_list) {
1576                 struct obd_export *sub_exp;
1577                 struct brw_page *sub_pga;
1578                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1579
1580                 sub_exp = lov->lov_tgts[req->rq_idx]->ltd_exp;
1581                 sub_pga = set->set_pga + req->rq_pgaidx;
1582                 rc = obd_brw(cmd, sub_exp, &req->rq_oi, req->rq_oabufs,
1583                              sub_pga, oti);
1584                 if (rc)
1585                         break;
1586                 lov_update_common_set(set, req, rc);
1587         }
1588
1589         err = lov_fini_brw_set(set);
1590         if (!rc)
1591                 rc = err;
1592         RETURN(rc);
1593 }
1594
1595 static int lov_enqueue_interpret(struct ptlrpc_request_set *rqset,
1596                                  void *data, int rc)
1597 {
1598         struct lov_request_set *lovset = (struct lov_request_set *)data;
1599         ENTRY;
1600         rc = lov_fini_enqueue_set(lovset, lovset->set_ei->ei_mode, rc, rqset);
1601         RETURN(rc);
1602 }
1603
1604 static int lov_enqueue(struct obd_export *exp, struct obd_info *oinfo,
1605                        struct ldlm_enqueue_info *einfo,
1606                        struct ptlrpc_request_set *rqset)
1607 {
1608         ldlm_mode_t mode = einfo->ei_mode;
1609         struct lov_request_set *set;
1610         struct lov_request *req;
1611         cfs_list_t *pos;
1612         struct lov_obd *lov;
1613         ldlm_error_t rc;
1614         ENTRY;
1615
1616         LASSERT(oinfo);
1617         ASSERT_LSM_MAGIC(oinfo->oi_md);
1618         LASSERT(mode == (mode & -mode));
1619
1620         /* we should never be asked to replay a lock this way. */
1621         LASSERT((oinfo->oi_flags & LDLM_FL_REPLAY) == 0);
1622
1623         if (!exp || !exp->exp_obd)
1624                 RETURN(-ENODEV);
1625
1626         lov = &exp->exp_obd->u.lov;
1627         rc = lov_prep_enqueue_set(exp, oinfo, einfo, &set);
1628         if (rc)
1629                 RETURN(rc);
1630
1631         cfs_list_for_each (pos, &set->set_list) {
1632                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1633
1634                 rc = obd_enqueue(lov->lov_tgts[req->rq_idx]->ltd_exp,
1635                                  &req->rq_oi, einfo, rqset);
1636                 if (rc != ELDLM_OK)
1637                         GOTO(out, rc);
1638         }
1639
1640         if (rqset && !cfs_list_empty(&rqset->set_requests)) {
1641                 LASSERT(rc == 0);
1642                 LASSERT(rqset->set_interpret == NULL);
1643                 rqset->set_interpret = lov_enqueue_interpret;
1644                 rqset->set_arg = (void *)set;
1645                 RETURN(rc);
1646         }
1647 out:
1648         rc = lov_fini_enqueue_set(set, mode, rc, rqset);
1649         RETURN(rc);
1650 }
1651
1652 static int lov_change_cbdata(struct obd_export *exp,
1653                              struct lov_stripe_md *lsm, ldlm_iterator_t it,
1654                              void *data)
1655 {
1656         struct lov_obd *lov;
1657         int rc = 0, i;
1658         ENTRY;
1659
1660         ASSERT_LSM_MAGIC(lsm);
1661
1662         if (!exp || !exp->exp_obd)
1663                 RETURN(-ENODEV);
1664
1665         lov = &exp->exp_obd->u.lov;
1666         for (i = 0; i < lsm->lsm_stripe_count; i++) {
1667                 struct lov_stripe_md submd;
1668                 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
1669
1670                 if (!lov->lov_tgts[loi->loi_ost_idx]) {
1671                         CDEBUG(D_HA, "lov idx %d NULL \n", loi->loi_ost_idx);
1672                         continue;
1673                 }
1674
1675                 LASSERT_SEQ_IS_MDT(loi->loi_seq);
1676                 submd.lsm_object_id = loi->loi_id;
1677                 submd.lsm_object_seq = loi->loi_seq;
1678                 submd.lsm_stripe_count = 0;
1679                 rc = obd_change_cbdata(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1680                                        &submd, it, data);
1681         }
1682         RETURN(rc);
1683 }
1684
1685 /* find any ldlm lock of the inode in lov
1686  * return 0    not find
1687  *        1    find one
1688  *      < 0    error */
1689 static int lov_find_cbdata(struct obd_export *exp,
1690                            struct lov_stripe_md *lsm, ldlm_iterator_t it,
1691                            void *data)
1692 {
1693         struct lov_obd *lov;
1694         int rc = 0, i;
1695         ENTRY;
1696
1697         ASSERT_LSM_MAGIC(lsm);
1698
1699         if (!exp || !exp->exp_obd)
1700                 RETURN(-ENODEV);
1701
1702         lov = &exp->exp_obd->u.lov;
1703         for (i = 0; i < lsm->lsm_stripe_count; i++) {
1704                 struct lov_stripe_md submd;
1705                 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
1706
1707                 if (!lov->lov_tgts[loi->loi_ost_idx]) {
1708                         CDEBUG(D_HA, "lov idx %d NULL \n", loi->loi_ost_idx);
1709                         continue;
1710                 }
1711
1712                 LASSERT_SEQ_IS_MDT(loi->loi_seq);
1713                 submd.lsm_object_id = loi->loi_id;
1714                 submd.lsm_object_seq = loi->loi_seq;
1715                 submd.lsm_stripe_count = 0;
1716                 rc = obd_find_cbdata(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1717                                      &submd, it, data);
1718                 if (rc != 0)
1719                         RETURN(rc);
1720         }
1721         RETURN(rc);
1722 }
1723
1724 static int lov_cancel(struct obd_export *exp, struct lov_stripe_md *lsm,
1725                       __u32 mode, struct lustre_handle *lockh)
1726 {
1727         struct lov_request_set *set;
1728         struct obd_info oinfo;
1729         struct lov_request *req;
1730         cfs_list_t *pos;
1731         struct lov_obd *lov;
1732         struct lustre_handle *lov_lockhp;
1733         int err = 0, rc = 0;
1734         ENTRY;
1735
1736         ASSERT_LSM_MAGIC(lsm);
1737
1738         if (!exp || !exp->exp_obd)
1739                 RETURN(-ENODEV);
1740
1741         LASSERT_SEQ_IS_MDT(lsm->lsm_object_seq);
1742         LASSERT(lockh);
1743         lov = &exp->exp_obd->u.lov;
1744         rc = lov_prep_cancel_set(exp, &oinfo, lsm, mode, lockh, &set);
1745         if (rc)
1746                 RETURN(rc);
1747
1748         cfs_list_for_each (pos, &set->set_list) {
1749                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1750                 lov_lockhp = set->set_lockh->llh_handles + req->rq_stripe;
1751
1752                 rc = obd_cancel(lov->lov_tgts[req->rq_idx]->ltd_exp,
1753                                 req->rq_oi.oi_md, mode, lov_lockhp);
1754                 rc = lov_update_common_set(set, req, rc);
1755                 if (rc) {
1756                         CERROR("error: cancel objid "LPX64" subobj "
1757                                LPX64" on OST idx %d: rc = %d\n",
1758                                lsm->lsm_object_id,
1759                                req->rq_oi.oi_md->lsm_object_id,
1760                                req->rq_idx, rc);
1761                         err = rc;
1762                 }
1763
1764         }
1765         lov_fini_cancel_set(set);
1766         RETURN(err);
1767 }
1768
1769 static int lov_cancel_unused(struct obd_export *exp,
1770                              struct lov_stripe_md *lsm,
1771                              ldlm_cancel_flags_t flags, void *opaque)
1772 {
1773         struct lov_obd *lov;
1774         int rc = 0, i;
1775         ENTRY;
1776
1777         if (!exp || !exp->exp_obd)
1778                 RETURN(-ENODEV);
1779
1780         lov = &exp->exp_obd->u.lov;
1781         if (lsm == NULL) {
1782                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
1783                         int err;
1784                         if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
1785                                 continue;
1786
1787                         err = obd_cancel_unused(lov->lov_tgts[i]->ltd_exp, NULL,
1788                                                 flags, opaque);
1789                         if (!rc)
1790                                 rc = err;
1791                 }
1792                 RETURN(rc);
1793         }
1794
1795         ASSERT_LSM_MAGIC(lsm);
1796
1797         LASSERT_SEQ_IS_MDT(lsm->lsm_object_seq);
1798         for (i = 0; i < lsm->lsm_stripe_count; i++) {
1799                 struct lov_stripe_md submd;
1800                 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
1801                 int err;
1802
1803                 if (!lov->lov_tgts[loi->loi_ost_idx]) {
1804                         CDEBUG(D_HA, "lov idx %d NULL\n", loi->loi_ost_idx);
1805                         continue;
1806                 }
1807
1808                 if (!lov->lov_tgts[loi->loi_ost_idx]->ltd_active)
1809                         CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
1810
1811                 submd.lsm_object_id = loi->loi_id;
1812                 submd.lsm_object_seq = loi->loi_seq;
1813                 submd.lsm_stripe_count = 0;
1814                 err = obd_cancel_unused(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1815                                         &submd, flags, opaque);
1816                 if (err && lov->lov_tgts[loi->loi_ost_idx]->ltd_active) {
1817                         CERROR("error: cancel unused objid "LPX64" subobj "LPX64
1818                                " on OST idx %d: rc = %d\n", lsm->lsm_object_id,
1819                                loi->loi_id, loi->loi_ost_idx, err);
1820                         if (!rc)
1821                                 rc = err;
1822                 }
1823         }
1824         RETURN(rc);
1825 }
1826
1827 int lov_statfs_interpret(struct ptlrpc_request_set *rqset, void *data, int rc)
1828 {
1829         struct lov_request_set *lovset = (struct lov_request_set *)data;
1830         int err;
1831         ENTRY;
1832
1833         if (rc)
1834                 cfs_atomic_set(&lovset->set_completes, 0);
1835
1836         err = lov_fini_statfs_set(lovset);
1837         RETURN(rc ? rc : err);
1838 }
1839
1840 static int lov_statfs_async(struct obd_export *exp, struct obd_info *oinfo,
1841                             __u64 max_age, struct ptlrpc_request_set *rqset)
1842 {
1843         struct obd_device      *obd = class_exp2obd(exp);
1844         struct lov_request_set *set;
1845         struct lov_request *req;
1846         cfs_list_t *pos;
1847         struct lov_obd *lov;
1848         int rc = 0;
1849         ENTRY;
1850
1851         LASSERT(oinfo != NULL);
1852         LASSERT(oinfo->oi_osfs != NULL);
1853
1854         lov = &obd->u.lov;
1855         rc = lov_prep_statfs_set(obd, oinfo, &set);
1856         if (rc)
1857                 RETURN(rc);
1858
1859         cfs_list_for_each (pos, &set->set_list) {
1860                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1861                 rc = obd_statfs_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1862                                       &req->rq_oi, max_age, rqset);
1863                 if (rc)
1864                         break;
1865         }
1866
1867         if (rc || cfs_list_empty(&rqset->set_requests)) {
1868                 int err;
1869                 if (rc)
1870                         cfs_atomic_set(&set->set_completes, 0);
1871                 err = lov_fini_statfs_set(set);
1872                 RETURN(rc ? rc : err);
1873         }
1874
1875         LASSERT(rqset->set_interpret == NULL);
1876         rqset->set_interpret = lov_statfs_interpret;
1877         rqset->set_arg = (void *)set;
1878         RETURN(0);
1879 }
1880
1881 static int lov_statfs(const struct lu_env *env, struct obd_export *exp,
1882                       struct obd_statfs *osfs, __u64 max_age, __u32 flags)
1883 {
1884         struct ptlrpc_request_set *set = NULL;
1885         struct obd_info oinfo = { { { 0 } } };
1886         int rc = 0;
1887         ENTRY;
1888
1889
1890         /* for obdclass we forbid using obd_statfs_rqset, but prefer using async
1891          * statfs requests */
1892         set = ptlrpc_prep_set();
1893         if (set == NULL)
1894                 RETURN(-ENOMEM);
1895
1896         oinfo.oi_osfs = osfs;
1897         oinfo.oi_flags = flags;
1898         rc = lov_statfs_async(exp, &oinfo, max_age, set);
1899         if (rc == 0)
1900                 rc = ptlrpc_set_wait(set);
1901         ptlrpc_set_destroy(set);
1902
1903         RETURN(rc);
1904 }
1905
1906 static int lov_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1907                          void *karg, void *uarg)
1908 {
1909         struct obd_device *obddev = class_exp2obd(exp);
1910         struct lov_obd *lov = &obddev->u.lov;
1911         int i = 0, rc = 0, count = lov->desc.ld_tgt_count;
1912         struct obd_uuid *uuidp;
1913         ENTRY;
1914
1915         switch (cmd) {
1916         case IOC_OBD_STATFS: {
1917                 struct obd_ioctl_data *data = karg;
1918                 struct obd_device *osc_obd;
1919                 struct obd_statfs stat_buf = {0};
1920                 __u32 index;
1921                 __u32 flags;
1922
1923                 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
1924                 if ((index >= count))
1925                         RETURN(-ENODEV);
1926
1927                 if (!lov->lov_tgts[index])
1928                         /* Try again with the next index */
1929                         RETURN(-EAGAIN);
1930                 if (!lov->lov_tgts[index]->ltd_active)
1931                         RETURN(-ENODATA);
1932
1933                 osc_obd = class_exp2obd(lov->lov_tgts[index]->ltd_exp);
1934                 if (!osc_obd)
1935                         RETURN(-EINVAL);
1936
1937                 /* copy UUID */
1938                 if (cfs_copy_to_user(data->ioc_pbuf2, obd2cli_tgt(osc_obd),
1939                                      min((int) data->ioc_plen2,
1940                                          (int) sizeof(struct obd_uuid))))
1941                         RETURN(-EFAULT);
1942
1943                 flags = uarg ? *(__u32*)uarg : 0;
1944                 /* got statfs data */
1945                 rc = obd_statfs(NULL, lov->lov_tgts[index]->ltd_exp, &stat_buf,
1946                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1947                                 flags);
1948                 if (rc)
1949                         RETURN(rc);
1950                 if (cfs_copy_to_user(data->ioc_pbuf1, &stat_buf,
1951                                      min((int) data->ioc_plen1,
1952                                          (int) sizeof(stat_buf))))
1953                         RETURN(-EFAULT);
1954                 break;
1955         }
1956         case OBD_IOC_LOV_GET_CONFIG: {
1957                 struct obd_ioctl_data *data;
1958                 struct lov_desc *desc;
1959                 char *buf = NULL;
1960                 __u32 *genp;
1961
1962                 len = 0;
1963                 if (obd_ioctl_getdata(&buf, &len, (void *)uarg))
1964                         RETURN(-EINVAL);
1965
1966                 data = (struct obd_ioctl_data *)buf;
1967
1968                 if (sizeof(*desc) > data->ioc_inllen1) {
1969                         obd_ioctl_freedata(buf, len);
1970                         RETURN(-EINVAL);
1971                 }
1972
1973                 if (sizeof(uuidp->uuid) * count > data->ioc_inllen2) {
1974                         obd_ioctl_freedata(buf, len);
1975                         RETURN(-EINVAL);
1976                 }
1977
1978                 if (sizeof(__u32) * count > data->ioc_inllen3) {
1979                         obd_ioctl_freedata(buf, len);
1980                         RETURN(-EINVAL);
1981                 }
1982
1983                 desc = (struct lov_desc *)data->ioc_inlbuf1;
1984                 memcpy(desc, &(lov->desc), sizeof(*desc));
1985
1986                 uuidp = (struct obd_uuid *)data->ioc_inlbuf2;
1987                 genp = (__u32 *)data->ioc_inlbuf3;
1988                 /* the uuid will be empty for deleted OSTs */
1989                 for (i = 0; i < count; i++, uuidp++, genp++) {
1990                         if (!lov->lov_tgts[i])
1991                                 continue;
1992                         *uuidp = lov->lov_tgts[i]->ltd_uuid;
1993                         *genp = lov->lov_tgts[i]->ltd_gen;
1994                 }
1995
1996                 if (cfs_copy_to_user((void *)uarg, buf, len))
1997                         rc = -EFAULT;
1998                 obd_ioctl_freedata(buf, len);
1999                 break;
2000         }
2001         case LL_IOC_LOV_SETSTRIPE:
2002                 rc = lov_setstripe(exp, len, karg, uarg);
2003                 break;
2004         case LL_IOC_LOV_GETSTRIPE:
2005                 rc = lov_getstripe(exp, karg, uarg);
2006                 break;
2007         case LL_IOC_LOV_SETEA:
2008                 rc = lov_setea(exp, karg, uarg);
2009                 break;
2010         case OBD_IOC_QUOTACTL: {
2011                 struct if_quotactl *qctl = karg;
2012                 struct lov_tgt_desc *tgt = NULL;
2013                 struct obd_quotactl *oqctl;
2014
2015                 if (qctl->qc_valid == QC_OSTIDX) {
2016                         if (qctl->qc_idx < 0 || count <= qctl->qc_idx)
2017                                 RETURN(-EINVAL);
2018
2019                         tgt = lov->lov_tgts[qctl->qc_idx];
2020                         if (!tgt || !tgt->ltd_exp)
2021                                 RETURN(-EINVAL);
2022                 } else if (qctl->qc_valid == QC_UUID) {
2023                         for (i = 0; i < count; i++) {
2024                                 tgt = lov->lov_tgts[i];
2025                                 if (!tgt ||
2026                                     !obd_uuid_equals(&tgt->ltd_uuid,
2027                                                      &qctl->obd_uuid))
2028                                         continue;
2029
2030                                 if (tgt->ltd_exp == NULL)
2031                                         RETURN(-EINVAL);
2032
2033                                 break;
2034                         }
2035                 } else {
2036                         RETURN(-EINVAL);
2037                 }
2038
2039                 if (i >= count)
2040                         RETURN(-EAGAIN);
2041
2042                 LASSERT(tgt && tgt->ltd_exp);
2043                 OBD_ALLOC_PTR(oqctl);
2044                 if (!oqctl)
2045                         RETURN(-ENOMEM);
2046
2047                 QCTL_COPY(oqctl, qctl);
2048                 rc = obd_quotactl(tgt->ltd_exp, oqctl);
2049                 if (rc == 0) {
2050                         QCTL_COPY(qctl, oqctl);
2051                         qctl->qc_valid = QC_OSTIDX;
2052                         qctl->obd_uuid = tgt->ltd_uuid;
2053                 }
2054                 OBD_FREE_PTR(oqctl);
2055                 break;
2056         }
2057         default: {
2058                 int set = 0;
2059
2060                 if (count == 0)
2061                         RETURN(-ENOTTY);
2062
2063                 for (i = 0; i < count; i++) {
2064                         int err;
2065                         struct obd_device *osc_obd;
2066
2067                         /* OST was disconnected */
2068                         if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
2069                                 continue;
2070
2071                         /* ll_umount_begin() sets force flag but for lov, not
2072                          * osc. Let's pass it through */
2073                         osc_obd = class_exp2obd(lov->lov_tgts[i]->ltd_exp);
2074                         osc_obd->obd_force = obddev->obd_force;
2075                         err = obd_iocontrol(cmd, lov->lov_tgts[i]->ltd_exp,
2076                                             len, karg, uarg);
2077                         if (err == -ENODATA && cmd == OBD_IOC_POLL_QUOTACHECK) {
2078                                 RETURN(err);
2079                         } else if (err) {
2080                                 if (lov->lov_tgts[i]->ltd_active) {
2081                                         CDEBUG(err == -ENOTTY ?
2082                                                D_IOCTL : D_WARNING,
2083                                                "iocontrol OSC %s on OST "
2084                                                "idx %d cmd %x: err = %d\n",
2085                                                lov_uuid2str(lov, i),
2086                                                i, cmd, err);
2087                                         if (!rc)
2088                                                 rc = err;
2089                                 }
2090                         } else {
2091                                 set = 1;
2092                         }
2093                 }
2094                 if (!set && !rc)
2095                         rc = -EIO;
2096         }
2097         }
2098
2099         RETURN(rc);
2100 }
2101
2102 #define FIEMAP_BUFFER_SIZE 4096
2103
2104 /**
2105  * Non-zero fe_logical indicates that this is a continuation FIEMAP
2106  * call. The local end offset and the device are sent in the first
2107  * fm_extent. This function calculates the stripe number from the index.
2108  * This function returns a stripe_no on which mapping is to be restarted.
2109  *
2110  * This function returns fm_end_offset which is the in-OST offset at which
2111  * mapping should be restarted. If fm_end_offset=0 is returned then caller
2112  * will re-calculate proper offset in next stripe.
2113  * Note that the first extent is passed to lov_get_info via the value field.
2114  *
2115  * \param fiemap fiemap request header
2116  * \param lsm striping information for the file
2117  * \param fm_start logical start of mapping
2118  * \param fm_end logical end of mapping
2119  * \param start_stripe starting stripe will be returned in this
2120  */
2121 obd_size fiemap_calc_fm_end_offset(struct ll_user_fiemap *fiemap,
2122                                    struct lov_stripe_md *lsm, obd_size fm_start,
2123                                    obd_size fm_end, int *start_stripe)
2124 {
2125         obd_size local_end = fiemap->fm_extents[0].fe_logical;
2126         obd_off lun_start, lun_end;
2127         obd_size fm_end_offset;
2128         int stripe_no = -1, i;
2129
2130         if (fiemap->fm_extent_count == 0 ||
2131             fiemap->fm_extents[0].fe_logical == 0)
2132                 return 0;
2133
2134         /* Find out stripe_no from ost_index saved in the fe_device */
2135         for (i = 0; i < lsm->lsm_stripe_count; i++) {
2136                 if (lsm->lsm_oinfo[i]->loi_ost_idx ==
2137                                         fiemap->fm_extents[0].fe_device) {
2138                         stripe_no = i;
2139                         break;
2140                 }
2141         }
2142         if (stripe_no == -1)
2143                 return -EINVAL;
2144
2145         /* If we have finished mapping on previous device, shift logical
2146          * offset to start of next device */
2147         if ((lov_stripe_intersects(lsm, stripe_no, fm_start, fm_end,
2148                                    &lun_start, &lun_end)) != 0 &&
2149                                    local_end < lun_end) {
2150                 fm_end_offset = local_end;
2151                 *start_stripe = stripe_no;
2152         } else {
2153                 /* This is a special value to indicate that caller should
2154                  * calculate offset in next stripe. */
2155                 fm_end_offset = 0;
2156                 *start_stripe = (stripe_no + 1) % lsm->lsm_stripe_count;
2157         }
2158
2159         return fm_end_offset;
2160 }
2161
2162 /**
2163  * We calculate on which OST the mapping will end. If the length of mapping
2164  * is greater than (stripe_size * stripe_count) then the last_stripe will
2165  * will be one just before start_stripe. Else we check if the mapping
2166  * intersects each OST and find last_stripe.
2167  * This function returns the last_stripe and also sets the stripe_count
2168  * over which the mapping is spread
2169  *
2170  * \param lsm striping information for the file
2171  * \param fm_start logical start of mapping
2172  * \param fm_end logical end of mapping
2173  * \param start_stripe starting stripe of the mapping
2174  * \param stripe_count the number of stripes across which to map is returned
2175  *
2176  * \retval last_stripe return the last stripe of the mapping
2177  */
2178 int fiemap_calc_last_stripe(struct lov_stripe_md *lsm, obd_size fm_start,
2179                             obd_size fm_end, int start_stripe,
2180                             int *stripe_count)
2181 {
2182         int last_stripe;
2183         obd_off obd_start, obd_end;
2184         int i, j;
2185
2186         if (fm_end - fm_start > lsm->lsm_stripe_size * lsm->lsm_stripe_count) {
2187                 last_stripe = (start_stripe < 1 ? lsm->lsm_stripe_count - 1 :
2188                                                               start_stripe - 1);
2189                 *stripe_count = lsm->lsm_stripe_count;
2190         } else {
2191                 for (j = 0, i = start_stripe; j < lsm->lsm_stripe_count;
2192                      i = (i + 1) % lsm->lsm_stripe_count, j++) {
2193                         if ((lov_stripe_intersects(lsm, i, fm_start, fm_end,
2194                                                    &obd_start, &obd_end)) == 0)
2195                                 break;
2196                 }
2197                 *stripe_count = j;
2198                 last_stripe = (start_stripe + j - 1) %lsm->lsm_stripe_count;
2199         }
2200
2201         return last_stripe;
2202 }
2203
2204 /**
2205  * Set fe_device and copy extents from local buffer into main return buffer.
2206  *
2207  * \param fiemap fiemap request header
2208  * \param lcl_fm_ext array of local fiemap extents to be copied
2209  * \param ost_index OST index to be written into the fm_device field for each
2210                     extent
2211  * \param ext_count number of extents to be copied
2212  * \param current_extent where to start copying in main extent array
2213  */
2214 void fiemap_prepare_and_copy_exts(struct ll_user_fiemap *fiemap,
2215                                   struct ll_fiemap_extent *lcl_fm_ext,
2216                                   int ost_index, unsigned int ext_count,
2217                                   int current_extent)
2218 {
2219         char *to;
2220         int ext;
2221
2222         for (ext = 0; ext < ext_count; ext++) {
2223                 lcl_fm_ext[ext].fe_device = ost_index;
2224                 lcl_fm_ext[ext].fe_flags |= FIEMAP_EXTENT_NET;
2225         }
2226
2227         /* Copy fm_extent's from fm_local to return buffer */
2228         to = (char *)fiemap + fiemap_count_to_size(current_extent);
2229         memcpy(to, lcl_fm_ext, ext_count * sizeof(struct ll_fiemap_extent));
2230 }
2231
2232 /**
2233  * Break down the FIEMAP request and send appropriate calls to individual OSTs.
2234  * This also handles the restarting of FIEMAP calls in case mapping overflows
2235  * the available number of extents in single call.
2236  */
2237 static int lov_fiemap(struct lov_obd *lov, __u32 keylen, void *key,
2238                       __u32 *vallen, void *val, struct lov_stripe_md *lsm)
2239 {
2240         struct ll_fiemap_info_key *fm_key = key;
2241         struct ll_user_fiemap *fiemap = val;
2242         struct ll_user_fiemap *fm_local = NULL;
2243         struct ll_fiemap_extent *lcl_fm_ext;
2244         int count_local;
2245         unsigned int get_num_extents = 0;
2246         int ost_index = 0, actual_start_stripe, start_stripe;
2247         obd_size fm_start, fm_end, fm_length, fm_end_offset;
2248         obd_size curr_loc;
2249         int current_extent = 0, rc = 0, i;
2250         int ost_eof = 0; /* EOF for object */
2251         int ost_done = 0; /* done with required mapping for this OST? */
2252         int last_stripe;
2253         int cur_stripe = 0, cur_stripe_wrap = 0, stripe_count;
2254         unsigned int buffer_size = FIEMAP_BUFFER_SIZE;
2255
2256         if (lsm == NULL)
2257                 GOTO(out, rc = 0);
2258
2259         if (fiemap_count_to_size(fm_key->fiemap.fm_extent_count) < buffer_size)
2260                 buffer_size = fiemap_count_to_size(fm_key->fiemap.fm_extent_count);
2261
2262         OBD_ALLOC_LARGE(fm_local, buffer_size);
2263         if (fm_local == NULL)
2264                 GOTO(out, rc = -ENOMEM);
2265         lcl_fm_ext = &fm_local->fm_extents[0];
2266
2267         count_local = fiemap_size_to_count(buffer_size);
2268
2269         memcpy(fiemap, &fm_key->fiemap, sizeof(*fiemap));
2270         fm_start = fiemap->fm_start;
2271         fm_length = fiemap->fm_length;
2272         /* Calculate start stripe, last stripe and length of mapping */
2273         actual_start_stripe = start_stripe = lov_stripe_number(lsm, fm_start);
2274         fm_end = (fm_length == ~0ULL ? fm_key->oa.o_size :
2275                                                 fm_start + fm_length - 1);
2276         /* If fm_length != ~0ULL but fm_start+fm_length-1 exceeds file size */
2277         if (fm_end > fm_key->oa.o_size)
2278                 fm_end = fm_key->oa.o_size;
2279
2280         last_stripe = fiemap_calc_last_stripe(lsm, fm_start, fm_end,
2281                                             actual_start_stripe, &stripe_count);
2282
2283         fm_end_offset = fiemap_calc_fm_end_offset(fiemap, lsm, fm_start,
2284                                                   fm_end, &start_stripe);
2285         if (fm_end_offset == -EINVAL)
2286                 GOTO(out, rc = -EINVAL);
2287
2288         if (fiemap->fm_extent_count == 0) {
2289                 get_num_extents = 1;
2290                 count_local = 0;
2291         }
2292
2293         /* Check each stripe */
2294         for (cur_stripe = start_stripe, i = 0; i < stripe_count;
2295              i++, cur_stripe = (cur_stripe + 1) % lsm->lsm_stripe_count) {
2296                 obd_size req_fm_len; /* Stores length of required mapping */
2297                 obd_size len_mapped_single_call;
2298                 obd_off lun_start, lun_end, obd_object_end;
2299                 unsigned int ext_count;
2300
2301                 cur_stripe_wrap = cur_stripe;
2302
2303                 /* Find out range of mapping on this stripe */
2304                 if ((lov_stripe_intersects(lsm, cur_stripe, fm_start, fm_end,
2305                                            &lun_start, &obd_object_end)) == 0)
2306                         continue;
2307
2308                 /* If this is a continuation FIEMAP call and we are on
2309                  * starting stripe then lun_start needs to be set to
2310                  * fm_end_offset */
2311                 if (fm_end_offset != 0 && cur_stripe == start_stripe)
2312                         lun_start = fm_end_offset;
2313
2314                 if (fm_length != ~0ULL) {
2315                         /* Handle fm_start + fm_length overflow */
2316                         if (fm_start + fm_length < fm_start)
2317                                 fm_length = ~0ULL - fm_start;
2318                         lun_end = lov_size_to_stripe(lsm, fm_start + fm_length,
2319                                                      cur_stripe);
2320                 } else {
2321                         lun_end = ~0ULL;
2322                 }
2323
2324                 if (lun_start == lun_end)
2325                         continue;
2326
2327                 req_fm_len = obd_object_end - lun_start;
2328                 fm_local->fm_length = 0;
2329                 len_mapped_single_call = 0;
2330
2331                 /* If the output buffer is very large and the objects have many
2332                  * extents we may need to loop on a single OST repeatedly */
2333                 ost_eof = 0;
2334                 ost_done = 0;
2335                 do {
2336                         if (get_num_extents == 0) {
2337                                 /* Don't get too many extents. */
2338                                 if (current_extent + count_local >
2339                                     fiemap->fm_extent_count)
2340                                         count_local = fiemap->fm_extent_count -
2341                                                                  current_extent;
2342                         }
2343
2344                         lun_start += len_mapped_single_call;
2345                         fm_local->fm_length = req_fm_len - len_mapped_single_call;
2346                         req_fm_len = fm_local->fm_length;
2347                         fm_local->fm_extent_count = count_local;
2348                         fm_local->fm_mapped_extents = 0;
2349                         fm_local->fm_flags = fiemap->fm_flags;
2350
2351                         fm_key->oa.o_id = lsm->lsm_oinfo[cur_stripe]->loi_id;
2352                         fm_key->oa.o_seq = lsm->lsm_oinfo[cur_stripe]->loi_seq;
2353                         ost_index = lsm->lsm_oinfo[cur_stripe]->loi_ost_idx;
2354
2355                         if (ost_index < 0 || ost_index >=lov->desc.ld_tgt_count)
2356                                 GOTO(out, rc = -EINVAL);
2357
2358                         /* If OST is inactive, return extent with UNKNOWN flag */
2359                         if (!lov->lov_tgts[ost_index]->ltd_active) {
2360                                 fm_local->fm_flags |= FIEMAP_EXTENT_LAST;
2361                                 fm_local->fm_mapped_extents = 1;
2362
2363                                 lcl_fm_ext[0].fe_logical = lun_start;
2364                                 lcl_fm_ext[0].fe_length = obd_object_end -
2365                                                                       lun_start;
2366                                 lcl_fm_ext[0].fe_flags |= FIEMAP_EXTENT_UNKNOWN;
2367
2368                                 goto inactive_tgt;
2369                         }
2370
2371                         fm_local->fm_start = lun_start;
2372                         fm_local->fm_flags &= ~FIEMAP_FLAG_DEVICE_ORDER;
2373                         memcpy(&fm_key->fiemap, fm_local, sizeof(*fm_local));
2374                         *vallen=fiemap_count_to_size(fm_local->fm_extent_count);
2375                         rc = obd_get_info(NULL,
2376                                           lov->lov_tgts[ost_index]->ltd_exp,
2377                                           keylen, key, vallen, fm_local, lsm);
2378                         if (rc != 0)
2379                                 GOTO(out, rc);
2380
2381 inactive_tgt:
2382                         ext_count = fm_local->fm_mapped_extents;
2383                         if (ext_count == 0) {
2384                                 ost_done = 1;
2385                                 /* If last stripe has hole at the end,
2386                                  * then we need to return */
2387                                 if (cur_stripe_wrap == last_stripe) {
2388                                         fiemap->fm_mapped_extents = 0;
2389                                         goto finish;
2390                                 }
2391                                 break;
2392                         }
2393
2394                         /* If we just need num of extents then go to next device */
2395                         if (get_num_extents) {
2396                                 current_extent += ext_count;
2397                                 break;
2398                         }
2399
2400                         len_mapped_single_call = lcl_fm_ext[ext_count-1].fe_logical -
2401                                   lun_start + lcl_fm_ext[ext_count - 1].fe_length;
2402
2403                         /* Have we finished mapping on this device? */
2404                         if (req_fm_len <= len_mapped_single_call)
2405                                 ost_done = 1;
2406
2407                         /* Clear the EXTENT_LAST flag which can be present on
2408                          * last extent */
2409                         if (lcl_fm_ext[ext_count-1].fe_flags & FIEMAP_EXTENT_LAST)
2410                                 lcl_fm_ext[ext_count - 1].fe_flags &=
2411                                                             ~FIEMAP_EXTENT_LAST;
2412
2413                         curr_loc = lov_stripe_size(lsm,
2414                                            lcl_fm_ext[ext_count - 1].fe_logical+
2415                                            lcl_fm_ext[ext_count - 1].fe_length,
2416                                            cur_stripe);
2417                         if (curr_loc >= fm_key->oa.o_size)
2418                                 ost_eof = 1;
2419
2420                         fiemap_prepare_and_copy_exts(fiemap, lcl_fm_ext,
2421                                                      ost_index, ext_count,
2422                                                      current_extent);
2423
2424                         current_extent += ext_count;
2425
2426                         /* Ran out of available extents? */
2427                         if (current_extent >= fiemap->fm_extent_count)
2428                                 goto finish;
2429                 } while (ost_done == 0 && ost_eof == 0);
2430
2431                 if (cur_stripe_wrap == last_stripe)
2432                         goto finish;
2433         }
2434
2435 finish:
2436         /* Indicate that we are returning device offsets unless file just has
2437          * single stripe */
2438         if (lsm->lsm_stripe_count > 1)
2439                 fiemap->fm_flags |= FIEMAP_FLAG_DEVICE_ORDER;
2440
2441         if (get_num_extents)
2442                 goto skip_last_device_calc;
2443
2444         /* Check if we have reached the last stripe and whether mapping for that
2445          * stripe is done. */
2446         if (cur_stripe_wrap == last_stripe) {
2447                 if (ost_done || ost_eof)
2448                         fiemap->fm_extents[current_extent - 1].fe_flags |=
2449                                                              FIEMAP_EXTENT_LAST;
2450         }
2451
2452 skip_last_device_calc:
2453         fiemap->fm_mapped_extents = current_extent;
2454
2455 out:
2456         OBD_FREE_LARGE(fm_local, buffer_size);
2457         return rc;
2458 }
2459
2460 static int lov_get_info(const struct lu_env *env, struct obd_export *exp,
2461                         __u32 keylen, void *key, __u32 *vallen, void *val,
2462                         struct lov_stripe_md *lsm)
2463 {
2464         struct obd_device *obddev = class_exp2obd(exp);
2465         struct lov_obd *lov = &obddev->u.lov;
2466         int i, rc;
2467         ENTRY;
2468
2469         if (!vallen || !val)
2470                 RETURN(-EFAULT);
2471
2472         obd_getref(obddev);
2473
2474         if (KEY_IS(KEY_LOCK_TO_STRIPE)) {
2475                 struct {
2476                         char name[16];
2477                         struct ldlm_lock *lock;
2478                 } *data = key;
2479                 struct ldlm_res_id *res_id = &data->lock->l_resource->lr_name;
2480                 struct lov_oinfo *loi;
2481                 __u32 *stripe = val;
2482
2483                 if (*vallen < sizeof(*stripe))
2484                         GOTO(out, rc = -EFAULT);
2485                 *vallen = sizeof(*stripe);
2486
2487                 /* XXX This is another one of those bits that will need to
2488                  * change if we ever actually support nested LOVs.  It uses
2489                  * the lock's export to find out which stripe it is. */
2490                 /* XXX - it's assumed all the locks for deleted OSTs have
2491                  * been cancelled. Also, the export for deleted OSTs will
2492                  * be NULL and won't match the lock's export. */
2493                 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2494                         loi = lsm->lsm_oinfo[i];
2495                         if (!lov->lov_tgts[loi->loi_ost_idx])
2496                                 continue;
2497                         if (lov->lov_tgts[loi->loi_ost_idx]->ltd_exp ==
2498                             data->lock->l_conn_export &&
2499                             ostid_res_name_eq(&loi->loi_oi, res_id)) {
2500                                 *stripe = i;
2501                                 GOTO(out, rc = 0);
2502                         }
2503                 }
2504                 LDLM_ERROR(data->lock, "lock on inode without such object");
2505                 dump_lsm(D_ERROR, lsm);
2506                 GOTO(out, rc = -ENXIO);
2507         } else if (KEY_IS(KEY_LAST_ID)) {
2508                 struct obd_id_info *info = val;
2509                 __u32 size = sizeof(obd_id);
2510                 struct lov_tgt_desc *tgt;
2511
2512                 LASSERT(*vallen == sizeof(struct obd_id_info));
2513                 tgt = lov->lov_tgts[info->idx];
2514
2515                 if (!tgt || !tgt->ltd_active)
2516                         GOTO(out, rc = -ESRCH);
2517
2518                 rc = obd_get_info(env, tgt->ltd_exp, keylen, key,
2519                                   &size, info->data, NULL);
2520                 GOTO(out, rc = 0);
2521         } else if (KEY_IS(KEY_LOVDESC)) {
2522                 struct lov_desc *desc_ret = val;
2523                 *desc_ret = lov->desc;
2524
2525                 GOTO(out, rc = 0);
2526         } else if (KEY_IS(KEY_FIEMAP)) {
2527                 rc = lov_fiemap(lov, keylen, key, vallen, val, lsm);
2528                 GOTO(out, rc);
2529         } else if (KEY_IS(KEY_CONNECT_FLAG)) {
2530                 struct lov_tgt_desc *tgt;
2531                 __u64 ost_idx = *((__u64*)val);
2532
2533                 LASSERT(*vallen == sizeof(__u64));
2534                 LASSERT(ost_idx < lov->desc.ld_tgt_count);
2535                 tgt = lov->lov_tgts[ost_idx];
2536
2537                 if (!tgt || !tgt->ltd_exp)
2538                         GOTO(out, rc = -ESRCH);
2539
2540                 *((__u64 *)val) = exp_connect_flags(tgt->ltd_exp);
2541                 GOTO(out, rc = 0);
2542         } else if (KEY_IS(KEY_TGT_COUNT)) {
2543                 *((int *)val) = lov->desc.ld_tgt_count;
2544                 GOTO(out, rc = 0);
2545         }
2546
2547         rc = -EINVAL;
2548
2549 out:
2550         obd_putref(obddev);
2551         RETURN(rc);
2552 }
2553
2554 static int lov_set_info_async(const struct lu_env *env, struct obd_export *exp,
2555                               obd_count keylen, void *key, obd_count vallen,
2556                               void *val, struct ptlrpc_request_set *set)
2557 {
2558         struct obd_device *obddev = class_exp2obd(exp);
2559         struct lov_obd *lov = &obddev->u.lov;
2560         obd_count count;
2561         int i, rc = 0, err;
2562         struct lov_tgt_desc *tgt;
2563         unsigned incr, check_uuid,
2564                  do_inactive, no_set;
2565         unsigned next_id = 0,  mds_con = 0, capa = 0;
2566         ENTRY;
2567
2568         incr = check_uuid = do_inactive = no_set = 0;
2569         if (set == NULL) {
2570                 no_set = 1;
2571                 set = ptlrpc_prep_set();
2572                 if (!set)
2573                         RETURN(-ENOMEM);
2574         }
2575
2576         obd_getref(obddev);
2577         count = lov->desc.ld_tgt_count;
2578
2579         if (KEY_IS(KEY_NEXT_ID)) {
2580                 count = vallen / sizeof(struct obd_id_info);
2581                 vallen = sizeof(obd_id);
2582                 incr = sizeof(struct obd_id_info);
2583                 do_inactive = 1;
2584                 next_id = 1;
2585         } else if (KEY_IS(KEY_CHECKSUM)) {
2586                 do_inactive = 1;
2587         } else if (KEY_IS(KEY_EVICT_BY_NID)) {
2588                 /* use defaults:  do_inactive = incr = 0; */
2589         } else if (KEY_IS(KEY_MDS_CONN)) {
2590                 mds_con = 1;
2591         } else if (KEY_IS(KEY_CAPA_KEY)) {
2592                 capa = 1;
2593         } else if (KEY_IS(KEY_CACHE_SET)) {
2594                 LASSERT(lov->lov_cache == NULL);
2595                 lov->lov_cache = val;
2596                 do_inactive = 1;
2597         }
2598
2599         for (i = 0; i < count; i++, val = (char *)val + incr) {
2600                 if (next_id) {
2601                         tgt = lov->lov_tgts[((struct obd_id_info*)val)->idx];
2602                 } else {
2603                         tgt = lov->lov_tgts[i];
2604                 }
2605                 /* OST was disconnected */
2606                 if (!tgt || !tgt->ltd_exp)
2607                         continue;
2608
2609                 /* OST is inactive and we don't want inactive OSCs */
2610                 if (!tgt->ltd_active && !do_inactive)
2611                         continue;
2612
2613                 if (mds_con) {
2614                         struct mds_group_info *mgi;
2615
2616                         LASSERT(vallen == sizeof(*mgi));
2617                         mgi = (struct mds_group_info *)val;
2618
2619                         /* Only want a specific OSC */
2620                         if (mgi->uuid && !obd_uuid_equals(mgi->uuid,
2621                                                 &tgt->ltd_uuid))
2622                                 continue;
2623
2624                         err = obd_set_info_async(env, tgt->ltd_exp,
2625                                          keylen, key, sizeof(int),
2626                                          &mgi->group, set);
2627                 } else if (next_id) {
2628                         err = obd_set_info_async(env, tgt->ltd_exp,
2629                                          keylen, key, vallen,
2630                                          ((struct obd_id_info*)val)->data, set);
2631                 } else if (capa) {
2632                         struct mds_capa_info *info = (struct mds_capa_info*)val;
2633
2634                         LASSERT(vallen == sizeof(*info));
2635
2636                          /* Only want a specific OSC */
2637                         if (info->uuid &&
2638                             !obd_uuid_equals(info->uuid, &tgt->ltd_uuid))
2639                                 continue;
2640
2641                         err = obd_set_info_async(env, tgt->ltd_exp, keylen,
2642                                                  key, sizeof(*info->capa),
2643                                                  info->capa, set);
2644                 } else {
2645                         /* Only want a specific OSC */
2646                         if (check_uuid &&
2647                             !obd_uuid_equals(val, &tgt->ltd_uuid))
2648                                 continue;
2649
2650                         err = obd_set_info_async(env, tgt->ltd_exp,
2651                                          keylen, key, vallen, val, set);
2652                 }
2653
2654                 if (!rc)
2655                         rc = err;
2656         }
2657
2658         obd_putref(obddev);
2659         if (no_set) {
2660                 err = ptlrpc_set_wait(set);
2661                 if (!rc)
2662                         rc = err;
2663                 ptlrpc_set_destroy(set);
2664         }
2665         RETURN(rc);
2666 }
2667
2668 static int lov_extent_calc(struct obd_export *exp, struct lov_stripe_md *lsm,
2669                            int cmd, __u64 *offset)
2670 {
2671         __u32 ssize = lsm->lsm_stripe_size;
2672         __u64 start;
2673
2674         start = *offset;
2675         lov_do_div64(start, ssize);
2676         start = start * ssize;
2677
2678         CDEBUG(D_DLMTRACE, "offset "LPU64", stripe %u, start "LPU64
2679                            ", end "LPU64"\n", *offset, ssize, start,
2680                            start + ssize - 1);
2681         if (cmd == OBD_CALC_STRIPE_END) {
2682                 *offset = start + ssize - 1;
2683         } else if (cmd == OBD_CALC_STRIPE_START) {
2684                 *offset = start;
2685         } else {
2686                 LBUG();
2687         }
2688
2689         RETURN(0);
2690 }
2691
2692 void lov_stripe_lock(struct lov_stripe_md *md)
2693 {
2694         LASSERT(md->lsm_lock_owner != cfs_curproc_pid());
2695         spin_lock(&md->lsm_lock);
2696         LASSERT(md->lsm_lock_owner == 0);
2697         md->lsm_lock_owner = cfs_curproc_pid();
2698 }
2699 EXPORT_SYMBOL(lov_stripe_lock);
2700
2701 void lov_stripe_unlock(struct lov_stripe_md *md)
2702 {
2703         LASSERT(md->lsm_lock_owner == cfs_curproc_pid());
2704         md->lsm_lock_owner = 0;
2705         spin_unlock(&md->lsm_lock);
2706 }
2707 EXPORT_SYMBOL(lov_stripe_unlock);
2708
2709 static int lov_quotactl(struct obd_device *obd, struct obd_export *exp,
2710                         struct obd_quotactl *oqctl)
2711 {
2712         struct lov_obd      *lov = &obd->u.lov;
2713         struct lov_tgt_desc *tgt;
2714         __u64                curspace = 0;
2715         __u64                bhardlimit = 0;
2716         int                  i, rc = 0;
2717         ENTRY;
2718
2719         if (oqctl->qc_cmd != LUSTRE_Q_QUOTAON &&
2720             oqctl->qc_cmd != LUSTRE_Q_QUOTAOFF &&
2721             oqctl->qc_cmd != Q_GETOQUOTA &&
2722             oqctl->qc_cmd != Q_INITQUOTA &&
2723             oqctl->qc_cmd != LUSTRE_Q_SETQUOTA &&
2724             oqctl->qc_cmd != Q_FINVALIDATE) {
2725                 CERROR("bad quota opc %x for lov obd", oqctl->qc_cmd);
2726                 RETURN(-EFAULT);
2727         }
2728
2729         /* for lov tgt */
2730         obd_getref(obd);
2731         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2732                 int err;
2733
2734                 tgt = lov->lov_tgts[i];
2735
2736                 if (!tgt)
2737                         continue;
2738
2739                 if (!tgt->ltd_active || tgt->ltd_reap) {
2740                         if (oqctl->qc_cmd == Q_GETOQUOTA &&
2741                             lov->lov_tgts[i]->ltd_activate) {
2742                                 rc = -EREMOTEIO;
2743                                 CERROR("ost %d is inactive\n", i);
2744                         } else {
2745                                 CDEBUG(D_HA, "ost %d is inactive\n", i);
2746                         }
2747                         continue;
2748                 }
2749
2750                 err = obd_quotactl(tgt->ltd_exp, oqctl);
2751                 if (err) {
2752                         if (tgt->ltd_active && !rc)
2753                                 rc = err;
2754                         continue;
2755                 }
2756
2757                 if (oqctl->qc_cmd == Q_GETOQUOTA) {
2758                         curspace += oqctl->qc_dqblk.dqb_curspace;
2759                         bhardlimit += oqctl->qc_dqblk.dqb_bhardlimit;
2760                 }
2761         }
2762         obd_putref(obd);
2763
2764         if (oqctl->qc_cmd == Q_GETOQUOTA) {
2765                 oqctl->qc_dqblk.dqb_curspace = curspace;
2766                 oqctl->qc_dqblk.dqb_bhardlimit = bhardlimit;
2767         }
2768         RETURN(rc);
2769 }
2770
2771 static int lov_quotacheck(struct obd_device *obd, struct obd_export *exp,
2772                           struct obd_quotactl *oqctl)
2773 {
2774         struct lov_obd *lov = &obd->u.lov;
2775         int             i, rc = 0;
2776         ENTRY;
2777
2778         obd_getref(obd);
2779
2780         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2781                 if (!lov->lov_tgts[i])
2782                         continue;
2783
2784                 /* Skip quota check on the administratively disabled OSTs. */
2785                 if (!lov->lov_tgts[i]->ltd_activate) {
2786                         CWARN("lov idx %d was administratively disabled, "
2787                               "skip quotacheck on it.\n", i);
2788                         continue;
2789                 }
2790
2791                 if (!lov->lov_tgts[i]->ltd_active) {
2792                         CERROR("lov idx %d inactive\n", i);
2793                         rc = -EIO;
2794                         goto out;
2795                 }
2796         }
2797
2798         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2799                 int err;
2800
2801                 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_activate)
2802                         continue;
2803
2804                 err = obd_quotacheck(lov->lov_tgts[i]->ltd_exp, oqctl);
2805                 if (err && !rc)
2806                         rc = err;
2807         }
2808
2809 out:
2810         obd_putref(obd);
2811
2812         RETURN(rc);
2813 }
2814
2815 struct obd_ops lov_obd_ops = {
2816         .o_owner               = THIS_MODULE,
2817         .o_setup               = lov_setup,
2818         .o_precleanup          = lov_precleanup,
2819         .o_cleanup             = lov_cleanup,
2820         //.o_process_config      = lov_process_config,
2821         .o_connect             = lov_connect,
2822         .o_disconnect          = lov_disconnect,
2823         .o_statfs              = lov_statfs,
2824         .o_statfs_async        = lov_statfs_async,
2825         .o_packmd              = lov_packmd,
2826         .o_unpackmd            = lov_unpackmd,
2827         .o_create              = lov_create,
2828         .o_destroy             = lov_destroy,
2829         .o_getattr             = lov_getattr,
2830         .o_getattr_async       = lov_getattr_async,
2831         .o_setattr             = lov_setattr,
2832         .o_setattr_async       = lov_setattr_async,
2833         .o_brw                 = lov_brw,
2834         .o_merge_lvb           = lov_merge_lvb,
2835         .o_adjust_kms          = lov_adjust_kms,
2836         .o_punch               = lov_punch,
2837         .o_sync                = lov_sync,
2838         .o_enqueue             = lov_enqueue,
2839         .o_change_cbdata       = lov_change_cbdata,
2840         .o_find_cbdata         = lov_find_cbdata,
2841         .o_cancel              = lov_cancel,
2842         .o_cancel_unused       = lov_cancel_unused,
2843         .o_iocontrol           = lov_iocontrol,
2844         .o_get_info            = lov_get_info,
2845         .o_set_info_async      = lov_set_info_async,
2846         .o_extent_calc         = lov_extent_calc,
2847         .o_llog_init           = lov_llog_init,
2848         .o_llog_finish         = lov_llog_finish,
2849         .o_notify              = lov_notify,
2850         .o_pool_new            = lov_pool_new,
2851         .o_pool_rem            = lov_pool_remove,
2852         .o_pool_add            = lov_pool_add,
2853         .o_pool_del            = lov_pool_del,
2854         .o_getref              = lov_getref,
2855         .o_putref              = lov_putref,
2856         .o_quotactl            = lov_quotactl,
2857         .o_quotacheck          = lov_quotacheck,
2858 };
2859
2860 cfs_mem_cache_t *lov_oinfo_slab;
2861
2862 extern struct lu_kmem_descr lov_caches[];
2863
2864 int __init lov_init(void)
2865 {
2866         struct lprocfs_static_vars lvars = { 0 };
2867         int rc, rc2;
2868         ENTRY;
2869
2870         /* print an address of _any_ initialized kernel symbol from this
2871          * module, to allow debugging with gdb that doesn't support data
2872          * symbols from modules.*/
2873         CDEBUG(D_INFO, "Lustre LOV module (%p).\n", &lov_caches);
2874
2875         rc = lu_kmem_init(lov_caches);
2876         if (rc)
2877                 return rc;
2878
2879         lov_oinfo_slab = cfs_mem_cache_create("lov_oinfo",
2880                                               sizeof(struct lov_oinfo),
2881                                               0, CFS_SLAB_HWCACHE_ALIGN);
2882         if (lov_oinfo_slab == NULL) {
2883                 lu_kmem_fini(lov_caches);
2884                 return -ENOMEM;
2885         }
2886         lprocfs_lov_init_vars(&lvars);
2887
2888         rc = class_register_type(&lov_obd_ops, NULL, lvars.module_vars,
2889                                  LUSTRE_LOV_NAME, &lov_device_type);
2890
2891         if (rc) {
2892                 rc2 = cfs_mem_cache_destroy(lov_oinfo_slab);
2893                 LASSERT(rc2 == 0);
2894                 lu_kmem_fini(lov_caches);
2895         }
2896
2897         RETURN(rc);
2898 }
2899
2900 #ifdef __KERNEL__
2901 static void /*__exit*/ lov_exit(void)
2902 {
2903         int rc;
2904
2905         class_unregister_type(LUSTRE_LOV_NAME);
2906         rc = cfs_mem_cache_destroy(lov_oinfo_slab);
2907         LASSERT(rc == 0);
2908
2909         lu_kmem_fini(lov_caches);
2910 }
2911
2912 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2913 MODULE_DESCRIPTION("Lustre Logical Object Volume OBD driver");
2914 MODULE_LICENSE("GPL");
2915
2916 cfs_module(lov, LUSTRE_VERSION_STRING, lov_init, lov_exit);
2917 #endif