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         struct lov_obd *lov = &obd->u.lov;
 466         ENTRY;
 467
 468         down_read(&lov->lov_notify_lock);
 469         if (!lov->lov_connects) {
 470                 up_read(&lov->lov_notify_lock);
 471                 RETURN(rc);
 472         }
 473
 474         if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE ||
 475             ev == OBD_NOTIFY_ACTIVATE || ev == OBD_NOTIFY_DEACTIVATE) {
 476                 struct obd_uuid *uuid;
 477
 478                 LASSERT(watched);
 479
 480                 if (strcmp(watched->obd_type->typ_name, LUSTRE_OSC_NAME)) {
 481                         up_read(&lov->lov_notify_lock);
 482                         CERROR("unexpected notification of %s %s!\n",
 483                                watched->obd_type->typ_name,
 484                                watched->obd_name);
 485                         RETURN(-EINVAL);
 486                 }
 487                 uuid = &watched->u.cli.cl_target_uuid;
 488
 489                 /* Set OSC as active before notifying the observer, so the
 490                  * observer can use the OSC normally.
 491                  */
 492                 rc = lov_set_osc_active(obd, uuid, ev);
 493                 if (rc < 0) {
 494                         up_read(&lov->lov_notify_lock);
 495                         CERROR("event(%d) of %s failed: %d\n", ev,
 496                                obd_uuid2str(uuid), rc);
 497                         RETURN(rc);
 498                 }
 499                 /* active event should be pass lov target index as data */
 500                 data = &rc;
 501         }
 502
 503         /* Pass the notification up the chain. */
 504         if (watched) {
 505                 rc = obd_notify_observer(obd, watched, ev, data);
 506         } else {
 507                 /* NULL watched means all osc's in the lov (only for syncs) */
 508                 /* sync event should be send lov idx as data */
 509                 struct lov_obd *lov = &obd->u.lov;
 510                 int i, is_sync;
 511
 512                 data = &i;
 513                 is_sync = (ev == OBD_NOTIFY_SYNC) ||
 514                           (ev == OBD_NOTIFY_SYNC_NONBLOCK);
 515
 516                 obd_getref(obd);
 517                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
 518                         if (!lov->lov_tgts[i])
 519                                 continue;
 520
 521                         /* don't send sync event if target not
 522                          * connected/activated */
 523                         if (is_sync &&  !lov->lov_tgts[i]->ltd_active)
 524                                 continue;
 525
 526                         rc = obd_notify_observer(obd, lov->lov_tgts[i]->ltd_obd,
 527                                                  ev, data);
 528                         if (rc) {
 529                                 CERROR("%s: notify %s of %s failed %d\n",
 530                                        obd->obd_name,
 531                                        obd->obd_observer->obd_name,
 532                                        lov->lov_tgts[i]->ltd_obd->obd_name,
 533                                        rc);
 534                         }
 535                 }
 536                 obd_putref(obd);
 537         }
 538
 539         up_read(&lov->lov_notify_lock);
 540         RETURN(rc);
 541 }
 542
 543 static int lov_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
 544                           __u32 index, int gen, int active)
 545 {
 546         struct lov_obd *lov = &obd->u.lov;
 547         struct lov_tgt_desc *tgt;
 548         struct obd_device *tgt_obd;
 549         int rc;
 550         ENTRY;
 551
 552         CDEBUG(D_CONFIG, "uuid:%s idx:%d gen:%d active:%d\n",
 553                uuidp->uuid, index, gen, active);
 554
 555         if (gen <= 0) {
 556                 CERROR("request to add OBD %s with invalid generation: %d\n",
 557                        uuidp->uuid, gen);
 558                 RETURN(-EINVAL);
 559         }
 560
 561         tgt_obd = class_find_client_obd(uuidp, LUSTRE_OSC_NAME,
 562                                         &obd->obd_uuid);
 563         if (tgt_obd == NULL)
 564                 RETURN(-EINVAL);
 565
 566         mutex_lock(&lov->lov_lock);
 567
 568         if ((index < lov->lov_tgt_size) && (lov->lov_tgts[index] != NULL)) {
 569                 tgt = lov->lov_tgts[index];
 570                 CERROR("UUID %s already assigned at LOV target index %d\n",
 571                        obd_uuid2str(&tgt->ltd_uuid), index);
 572                 mutex_unlock(&lov->lov_lock);
 573                 RETURN(-EEXIST);
 574         }
 575
 576         if (index >= lov->lov_tgt_size) {
 577                 /* We need to reallocate the lov target array. */
 578                 struct lov_tgt_desc **newtgts, **old = NULL;
 579                 __u32 newsize, oldsize = 0;
 580
 581                 newsize = max(lov->lov_tgt_size, (__u32)2);
 582                 while (newsize < index + 1)
 583                         newsize = newsize << 1;
 584                 OBD_ALLOC(newtgts, sizeof(*newtgts) * newsize);
 585                 if (newtgts == NULL) {
 586                         mutex_unlock(&lov->lov_lock);
 587                         RETURN(-ENOMEM);
 588                 }
 589
 590                 if (lov->lov_tgt_size) {
 591                         memcpy(newtgts, lov->lov_tgts, sizeof(*newtgts) *
 592                                lov->lov_tgt_size);
 593                         old = lov->lov_tgts;
 594                         oldsize = lov->lov_tgt_size;
 595                 }
 596
 597                 lov->lov_tgts = newtgts;
 598                 lov->lov_tgt_size = newsize;
 599                 smp_rmb();
 600                 if (old)
 601                         OBD_FREE(old, sizeof(*old) * oldsize);
 602
 603                 CDEBUG(D_CONFIG, "tgts: %p size: %d\n",
 604                        lov->lov_tgts, lov->lov_tgt_size);
 605         }
 606
 607         OBD_ALLOC_PTR(tgt);
 608         if (!tgt) {
 609                 mutex_unlock(&lov->lov_lock);
 610                 RETURN(-ENOMEM);
 611         }
 612
 613         rc = lov_ost_pool_add(&lov->lov_packed, index, lov->lov_tgt_size);
 614         if (rc) {
 615                 mutex_unlock(&lov->lov_lock);
 616                 OBD_FREE_PTR(tgt);
 617                 RETURN(rc);
 618         }
 619
 620         tgt->ltd_uuid = *uuidp;
 621         tgt->ltd_obd = tgt_obd;
 622         /* XXX - add a sanity check on the generation number. */
 623         tgt->ltd_gen = gen;
 624         tgt->ltd_index = index;
 625         tgt->ltd_activate = active;
 626         lov->lov_tgts[index] = tgt;
 627         if (index >= lov->desc.ld_tgt_count)
 628                 lov->desc.ld_tgt_count = index + 1;
 629
 630         mutex_unlock(&lov->lov_lock);
 631
 632         CDEBUG(D_CONFIG, "idx=%d ltd_gen=%d ld_tgt_count=%d\n",
 633                 index, tgt->ltd_gen, lov->desc.ld_tgt_count);
 634
 635         rc = obd_notify(obd, tgt_obd, OBD_NOTIFY_CREATE, &index);
 636
 637         if (lov->lov_connects == 0) {
 638                 /* lov_connect hasn't been called yet. We'll do the
 639                    lov_connect_obd on this target when that fn first runs,
 640                    because we don't know the connect flags yet. */
 641                 RETURN(0);
 642         }
 643
 644         obd_getref(obd);
 645
 646         rc = lov_connect_obd(obd, index, active, &lov->lov_ocd);
 647         if (rc)
 648                 GOTO(out, rc);
 649
 650         /* connect to administrative disabled ost */
 651         if (!tgt->ltd_exp)
 652                 GOTO(out, rc = 0);
 653
 654         if (lov->lov_cache != NULL) {
 655                 rc = obd_set_info_async(NULL, tgt->ltd_exp,
 656                                 sizeof(KEY_CACHE_SET), KEY_CACHE_SET,
 657                                 sizeof(struct cl_client_cache), lov->lov_cache,
 658                                 NULL);
 659                 if (rc < 0)
 660                         GOTO(out, rc);
 661         }
 662
 663         rc = lov_notify(obd, tgt->ltd_exp->exp_obd,
 664                         active ? OBD_NOTIFY_CONNECT : OBD_NOTIFY_INACTIVE,
 665                         (void *)&index);
 666
 667 out:
 668         if (rc) {
 669                 CERROR("add failed (%d), deleting %s\n", rc,
 670                        obd_uuid2str(&tgt->ltd_uuid));
 671                 lov_del_target(obd, index, 0, 0);
 672         }
 673         obd_putref(obd);
 674         RETURN(rc);
 675 }
 676
 677 /* Schedule a target for deletion */
 678 int lov_del_target(struct obd_device *obd, __u32 index,
 679                    struct obd_uuid *uuidp, int gen)
 680 {
 681         struct lov_obd *lov = &obd->u.lov;
 682         int count = lov->desc.ld_tgt_count;
 683         int rc = 0;
 684         ENTRY;
 685
 686         if (index >= count) {
 687                 CERROR("LOV target index %d >= number of LOV OBDs %d.\n",
 688                        index, count);
 689                 RETURN(-EINVAL);
 690         }
 691
 692         /* to make sure there's no ongoing lov_notify() now */
 693         down_write(&lov->lov_notify_lock);
 694         obd_getref(obd);
 695
 696         if (!lov->lov_tgts[index]) {
 697                 CERROR("LOV target at index %d is not setup.\n", index);
 698                 GOTO(out, rc = -EINVAL);
 699         }
 700
 701         if (uuidp && !obd_uuid_equals(uuidp, &lov->lov_tgts[index]->ltd_uuid)) {
 702                 CERROR("LOV target UUID %s at index %d doesn't match %s.\n",
 703                        lov_uuid2str(lov, index), index,
 704                        obd_uuid2str(uuidp));
 705                 GOTO(out, rc = -EINVAL);
 706         }
 707
 708         CDEBUG(D_CONFIG, "uuid: %s idx: %d gen: %d exp: %p active: %d\n",
 709                lov_uuid2str(lov, index), index,
 710                lov->lov_tgts[index]->ltd_gen, lov->lov_tgts[index]->ltd_exp,
 711                lov->lov_tgts[index]->ltd_active);
 712
 713         lov->lov_tgts[index]->ltd_reap = 1;
 714         lov->lov_death_row++;
 715         /* we really delete it from obd_putref */
 716 out:
 717         obd_putref(obd);
 718         up_write(&lov->lov_notify_lock);
 719
 720         RETURN(rc);
 721 }
 722
 723 static void __lov_del_obd(struct obd_device *obd, struct lov_tgt_desc *tgt)
 724 {
 725         struct obd_device *osc_obd;
 726
 727         LASSERT(tgt);
 728         LASSERT(tgt->ltd_reap);
 729
 730         osc_obd = class_exp2obd(tgt->ltd_exp);
 731
 732         CDEBUG(D_CONFIG, "Removing tgt %s : %s\n",
 733                tgt->ltd_uuid.uuid,
 734                osc_obd ? osc_obd->obd_name : "<no obd>");
 735
 736         if (tgt->ltd_exp)
 737                 lov_disconnect_obd(obd, tgt);
 738
 739         OBD_FREE_PTR(tgt);
 740
 741         /* Manual cleanup - no cleanup logs to clean up the osc's.  We must
 742            do it ourselves. And we can't do it from lov_cleanup,
 743            because we just lost our only reference to it. */
 744         if (osc_obd)
 745                 class_manual_cleanup(osc_obd);
 746 }
 747
 748 void lov_fix_desc_stripe_size(__u64 *val)
 749 {
 750         if (*val < LOV_DEFAULT_STRIPE_SIZE) {
 751                 LCONSOLE_WARN("Increasing default stripe size to min %u\n",
 752                               LOV_DEFAULT_STRIPE_SIZE);
 753                 *val = LOV_DEFAULT_STRIPE_SIZE;
 754         } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) {
 755                 *val &= ~(LOV_MIN_STRIPE_SIZE - 1);
 756                 LCONSOLE_WARN("Changing default stripe size to "LPU64" (a "
 757                               "multiple of %u)\n",
 758                               *val, LOV_MIN_STRIPE_SIZE);
 759         }
 760 }
 761
 762 void lov_fix_desc_stripe_count(__u32 *val)
 763 {
 764         if (*val == 0)
 765                 *val = 1;
 766 }
 767
 768 void lov_fix_desc_pattern(__u32 *val)
 769 {
 770         /* from lov_setstripe */
 771         if ((*val != 0) && (*val != LOV_PATTERN_RAID0)) {
 772                 LCONSOLE_WARN("Unknown stripe pattern: %#x\n", *val);
 773                 *val = 0;
 774         }
 775 }
 776
 777 void lov_fix_desc_qos_maxage(__u32 *val)
 778 {
 779         /* fix qos_maxage */
 780         if (*val == 0)
 781                 *val = QOS_DEFAULT_MAXAGE;
 782 }
 783
 784 void lov_fix_desc(struct lov_desc *desc)
 785 {
 786         lov_fix_desc_stripe_size(&desc->ld_default_stripe_size);
 787         lov_fix_desc_stripe_count(&desc->ld_default_stripe_count);
 788         lov_fix_desc_pattern(&desc->ld_pattern);
 789         lov_fix_desc_qos_maxage(&desc->ld_qos_maxage);
 790 }
 791
 792 int lov_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
 793 {
 794         struct lprocfs_static_vars lvars = { 0 };
 795         struct lov_desc *desc;
 796         struct lov_obd *lov = &obd->u.lov;
 797         int rc;
 798         ENTRY;
 799
 800         if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
 801                 CERROR("LOV setup requires a descriptor\n");
 802                 RETURN(-EINVAL);
 803         }
 804
 805         desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
 806
 807         if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
 808                 CERROR("descriptor size wrong: %d > %d\n",
 809                        (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
 810                 RETURN(-EINVAL);
 811         }
 812
 813         if (desc->ld_magic != LOV_DESC_MAGIC) {
 814                 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
 815                             CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
 816                                    obd->obd_name, desc);
 817                             lustre_swab_lov_desc(desc);
 818                 } else {
 819                         CERROR("%s: Bad lov desc magic: %#x\n",
 820                                obd->obd_name, desc->ld_magic);
 821                         RETURN(-EINVAL);
 822                 }
 823         }
 824
 825         lov_fix_desc(desc);
 826
 827         desc->ld_active_tgt_count = 0;
 828         lov->desc = *desc;
 829         lov->lov_tgt_size = 0;
 830
 831         mutex_init(&lov->lov_lock);
 832         cfs_atomic_set(&lov->lov_refcount, 0);
 833         lov->lov_sp_me = LUSTRE_SP_CLI;
 834
 835         init_rwsem(&lov->lov_notify_lock);
 836
 837         lov->lov_pools_hash_body = cfs_hash_create("POOLS", HASH_POOLS_CUR_BITS,
 838                                                    HASH_POOLS_MAX_BITS,
 839                                                    HASH_POOLS_BKT_BITS, 0,
 840                                                    CFS_HASH_MIN_THETA,
 841                                                    CFS_HASH_MAX_THETA,
 842                                                    &pool_hash_operations,
 843                                                    CFS_HASH_DEFAULT);
 844         CFS_INIT_LIST_HEAD(&lov->lov_pool_list);
 845         lov->lov_pool_count = 0;
 846         rc = lov_ost_pool_init(&lov->lov_packed, 0);
 847         if (rc)
 848                 GOTO(out, rc);
 849
 850         lprocfs_lov_init_vars(&lvars);
 851         lprocfs_obd_setup(obd, lvars.obd_vars);
 852 #ifdef LPROCFS
 853         {
 854                 int rc;
 855
 856                 rc = lprocfs_seq_create(obd->obd_proc_entry, "target_obd",
 857                                         0444, &lov_proc_target_fops, obd);
 858                 if (rc)
 859                         CWARN("Error adding the target_obd file\n");
 860         }
 861 #endif
 862         lov->lov_pool_proc_entry = lprocfs_register("pools",
 863                                                     obd->obd_proc_entry,
 864                                                     NULL, NULL);
 865
 866         RETURN(0);
 867
 868 out:
 869         return rc;
 870 }
 871
 872 static int lov_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
 873 {
 874         int rc = 0;
 875         struct lov_obd *lov = &obd->u.lov;
 876
 877         ENTRY;
 878
 879         switch (stage) {
 880         case OBD_CLEANUP_EARLY: {
 881                 int i;
 882                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
 883                         if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_active)
 884                                 continue;
 885                         obd_precleanup(class_exp2obd(lov->lov_tgts[i]->ltd_exp),
 886                                        OBD_CLEANUP_EARLY);
 887                 }
 888                 break;
 889         }
 890         case OBD_CLEANUP_EXPORTS:
 891                 rc = obd_llog_finish(obd, 0);
 892                 if (rc != 0)
 893                         CERROR("failed to cleanup llogging subsystems\n");
 894                 break;
 895         }
 896         RETURN(rc);
 897 }
 898
 899 static int lov_cleanup(struct obd_device *obd)
 900 {
 901         struct lov_obd *lov = &obd->u.lov;
 902         cfs_list_t *pos, *tmp;
 903         struct pool_desc *pool;
 904         ENTRY;
 905
 906         cfs_list_for_each_safe(pos, tmp, &lov->lov_pool_list) {
 907                 pool = cfs_list_entry(pos, struct pool_desc, pool_list);
 908                 /* free pool structs */
 909                 CDEBUG(D_INFO, "delete pool %p\n", pool);
 910                 /* In the function below, .hs_keycmp resolves to
 911                  * pool_hashkey_keycmp() */
 912                 /* coverity[overrun-buffer-val] */
 913                 lov_pool_del(obd, pool->pool_name);
 914         }
 915         cfs_hash_putref(lov->lov_pools_hash_body);
 916         lov_ost_pool_free(&lov->lov_packed);
 917
 918         lprocfs_obd_cleanup(obd);
 919         if (lov->lov_tgts) {
 920                 int i;
 921                 obd_getref(obd);
 922                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
 923                         if (!lov->lov_tgts[i])
 924                                 continue;
 925
 926                         /* Inactive targets may never have connected */
 927                         if (lov->lov_tgts[i]->ltd_active ||
 928                             cfs_atomic_read(&lov->lov_refcount))
 929                             /* We should never get here - these
 930                                should have been removed in the
 931                              disconnect. */
 932                                 CERROR("lov tgt %d not cleaned!"
 933                                        " deathrow=%d, lovrc=%d\n",
 934                                        i, lov->lov_death_row,
 935                                        cfs_atomic_read(&lov->lov_refcount));
 936                         lov_del_target(obd, i, 0, 0);
 937                 }
 938                 obd_putref(obd);
 939                 OBD_FREE(lov->lov_tgts, sizeof(*lov->lov_tgts) *
 940                          lov->lov_tgt_size);
 941                 lov->lov_tgt_size = 0;
 942         }
 943         RETURN(0);
 944 }
 945
 946 int lov_process_config_base(struct obd_device *obd, struct lustre_cfg *lcfg,
 947                             __u32 *indexp, int *genp)
 948 {
 949         struct obd_uuid obd_uuid;
 950         int cmd;
 951         int rc = 0;
 952         ENTRY;
 953
 954         switch(cmd = lcfg->lcfg_command) {
 955         case LCFG_LOV_ADD_OBD:
 956         case LCFG_LOV_ADD_INA:
 957         case LCFG_LOV_DEL_OBD: {
 958                 __u32 index;
 959                 int gen;
 960                 /* lov_modify_tgts add  0:lov_mdsA  1:ost1_UUID  2:0  3:1 */
 961                 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
 962                         GOTO(out, rc = -EINVAL);
 963
 964                 obd_str2uuid(&obd_uuid,  lustre_cfg_buf(lcfg, 1));
 965
 966                 if (sscanf(lustre_cfg_buf(lcfg, 2), "%d", indexp) != 1)
 967                         GOTO(out, rc = -EINVAL);
 968                 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", genp) != 1)
 969                         GOTO(out, rc = -EINVAL);
 970                 index = *indexp;
 971                 gen = *genp;
 972                 if (cmd == LCFG_LOV_ADD_OBD)
 973                         rc = lov_add_target(obd, &obd_uuid, index, gen, 1);
 974                 else if (cmd == LCFG_LOV_ADD_INA)
 975                         rc = lov_add_target(obd, &obd_uuid, index, gen, 0);
 976                 else
 977                         rc = lov_del_target(obd, index, &obd_uuid, gen);
 978                 GOTO(out, rc);
 979         }
 980         case LCFG_PARAM: {
 981                 struct lprocfs_static_vars lvars = { 0 };
 982                 struct lov_desc *desc = &(obd->u.lov.desc);
 983
 984                 if (!desc)
 985                         GOTO(out, rc = -EINVAL);
 986
 987                 lprocfs_lov_init_vars(&lvars);
 988
 989                 rc = class_process_proc_param(PARAM_LOV, lvars.obd_vars,
 990                                               lcfg, obd);
 991                 if (rc > 0)
 992                         rc = 0;
 993                 GOTO(out, rc);
 994         }
 995         case LCFG_POOL_NEW:
 996         case LCFG_POOL_ADD:
 997         case LCFG_POOL_DEL:
 998         case LCFG_POOL_REM:
 999                 GOTO(out, rc);
1000
1001         default: {
1002                 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
1003                 GOTO(out, rc = -EINVAL);
1004
1005         }
1006         }
1007 out:
1008         RETURN(rc);
1009 }
1010
1011 static int lov_recreate(struct obd_export *exp, struct obdo *src_oa,
1012                         struct lov_stripe_md **ea, struct obd_trans_info *oti)
1013 {
1014         struct lov_stripe_md *obj_mdp, *lsm;
1015         struct lov_obd *lov = &exp->exp_obd->u.lov;
1016         unsigned ost_idx;
1017         int rc, i;
1018         ENTRY;
1019
1020         LASSERT(src_oa->o_valid & OBD_MD_FLFLAGS &&
1021                 src_oa->o_flags & OBD_FL_RECREATE_OBJS);
1022
1023         OBD_ALLOC(obj_mdp, sizeof(*obj_mdp));
1024         if (obj_mdp == NULL)
1025                 RETURN(-ENOMEM);
1026
1027         ost_idx = src_oa->o_nlink;
1028         lsm = *ea;
1029         if (lsm == NULL)
1030                 GOTO(out, rc = -EINVAL);
1031         if (ost_idx >= lov->desc.ld_tgt_count ||
1032             !lov->lov_tgts[ost_idx])
1033                 GOTO(out, rc = -EINVAL);
1034
1035         for (i = 0; i < lsm->lsm_stripe_count; i++) {
1036                 if (lsm->lsm_oinfo[i]->loi_ost_idx == ost_idx) {
1037                         if (lsm->lsm_oinfo[i]->loi_id != src_oa->o_id)
1038                                 GOTO(out, rc = -EINVAL);
1039                         break;
1040                 }
1041         }
1042         if (i == lsm->lsm_stripe_count)
1043                 GOTO(out, rc = -EINVAL);
1044
1045         rc = obd_create(NULL, lov->lov_tgts[ost_idx]->ltd_exp,
1046                         src_oa, &obj_mdp, oti);
1047 out:
1048         OBD_FREE(obj_mdp, sizeof(*obj_mdp));
1049         RETURN(rc);
1050 }
1051
1052 /* the LOV expects oa->o_id to be set to the LOV object id */
1053 static int lov_create(const struct lu_env *env, struct obd_export *exp,
1054                       struct obdo *src_oa, struct lov_stripe_md **ea,
1055                       struct obd_trans_info *oti)
1056 {
1057         struct lov_obd *lov;
1058         int rc = 0;
1059         ENTRY;
1060
1061         LASSERT(ea != NULL);
1062         if (exp == NULL)
1063                 RETURN(-EINVAL);
1064
1065         if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
1066             src_oa->o_flags == OBD_FL_DELORPHAN) {
1067                 /* should be used with LOV anymore */
1068                 LBUG();
1069         }
1070
1071         lov = &exp->exp_obd->u.lov;
1072         if (!lov->desc.ld_active_tgt_count)
1073                 RETURN(-EIO);
1074
1075         obd_getref(exp->exp_obd);
1076         /* Recreate a specific object id at the given OST index */
1077         if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
1078             (src_oa->o_flags & OBD_FL_RECREATE_OBJS)) {
1079                  rc = lov_recreate(exp, src_oa, ea, oti);
1080         }
1081
1082         obd_putref(exp->exp_obd);
1083         RETURN(rc);
1084 }
1085
1086 #define ASSERT_LSM_MAGIC(lsmp)                                                  \
1087 do {                                                                            \
1088         LASSERT((lsmp) != NULL);                                                \
1089         LASSERTF(((lsmp)->lsm_magic == LOV_MAGIC_V1 ||                          \
1090                  (lsmp)->lsm_magic == LOV_MAGIC_V3),                            \
1091                  "%p->lsm_magic=%x\n", (lsmp), (lsmp)->lsm_magic);              \
1092 } while (0)
1093
1094 static int lov_destroy(const struct lu_env *env, struct obd_export *exp,
1095                        struct obdo *oa, struct lov_stripe_md *lsm,
1096                        struct obd_trans_info *oti, struct obd_export *md_exp,
1097                        void *capa)
1098 {
1099         struct lov_request_set *set;
1100         struct obd_info oinfo;
1101         struct lov_request *req;
1102         cfs_list_t *pos;
1103         struct lov_obd *lov;
1104         int rc = 0, err = 0;
1105         ENTRY;
1106
1107         ASSERT_LSM_MAGIC(lsm);
1108
1109         if (!exp || !exp->exp_obd)
1110                 RETURN(-ENODEV);
1111
1112         if (oa->o_valid & OBD_MD_FLCOOKIE) {
1113                 LASSERT(oti);
1114                 LASSERT(oti->oti_logcookies);
1115         }
1116
1117         lov = &exp->exp_obd->u.lov;
1118         obd_getref(exp->exp_obd);
1119         rc = lov_prep_destroy_set(exp, &oinfo, oa, lsm, oti, &set);
1120         if (rc)
1121                 GOTO(out, rc);
1122
1123         cfs_list_for_each (pos, &set->set_list) {
1124                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1125
1126                 if (oa->o_valid & OBD_MD_FLCOOKIE)
1127                         oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1128
1129                 err = obd_destroy(env, lov->lov_tgts[req->rq_idx]->ltd_exp,
1130                                   req->rq_oi.oi_oa, NULL, oti, NULL, capa);
1131                 err = lov_update_common_set(set, req, err);
1132                 if (err) {
1133                         CERROR("error: destroying objid "LPX64" subobj "
1134                                LPX64" on OST idx %d: rc = %d\n",
1135                                oa->o_id, req->rq_oi.oi_oa->o_id,
1136                                req->rq_idx, err);
1137                         if (!rc)
1138                                 rc = err;
1139                 }
1140         }
1141
1142         if (rc == 0) {
1143                 LASSERT(lsm_op_find(lsm->lsm_magic) != NULL);
1144                 rc = lsm_op_find(lsm->lsm_magic)->lsm_destroy(lsm, oa, md_exp);
1145         }
1146         err = lov_fini_destroy_set(set);
1147 out:
1148         obd_putref(exp->exp_obd);
1149         RETURN(rc ? rc : err);
1150 }
1151
1152 static int lov_getattr(const struct lu_env *env, struct obd_export *exp,
1153                        struct obd_info *oinfo)
1154 {
1155         struct lov_request_set *set;
1156         struct lov_request *req;
1157         cfs_list_t *pos;
1158         struct lov_obd *lov;
1159         int err = 0, rc = 0;
1160         ENTRY;
1161
1162         LASSERT(oinfo);
1163         ASSERT_LSM_MAGIC(oinfo->oi_md);
1164
1165         if (!exp || !exp->exp_obd)
1166                 RETURN(-ENODEV);
1167
1168         lov = &exp->exp_obd->u.lov;
1169
1170         rc = lov_prep_getattr_set(exp, oinfo, &set);
1171         if (rc)
1172                 RETURN(rc);
1173
1174         cfs_list_for_each (pos, &set->set_list) {
1175                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1176
1177                 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
1178                        "%u\n", oinfo->oi_oa->o_id, req->rq_stripe,
1179                        req->rq_oi.oi_oa->o_id, req->rq_idx);
1180
1181                 rc = obd_getattr(env, lov->lov_tgts[req->rq_idx]->ltd_exp,
1182                                  &req->rq_oi);
1183                 err = lov_update_common_set(set, req, rc);
1184                 if (err) {
1185                         CERROR("error: getattr objid "LPX64" subobj "
1186                                LPX64" on OST idx %d: rc = %d\n",
1187                                oinfo->oi_oa->o_id, req->rq_oi.oi_oa->o_id,
1188                                req->rq_idx, err);
1189                         break;
1190                 }
1191         }
1192
1193         rc = lov_fini_getattr_set(set);
1194         if (err)
1195                 rc = err;
1196         RETURN(rc);
1197 }
1198
1199 static int lov_getattr_interpret(struct ptlrpc_request_set *rqset,
1200                                  void *data, int rc)
1201 {
1202         struct lov_request_set *lovset = (struct lov_request_set *)data;
1203         int err;
1204         ENTRY;
1205
1206         /* don't do attribute merge if this aysnc op failed */
1207         if (rc)
1208                 cfs_atomic_set(&lovset->set_completes, 0);
1209         err = lov_fini_getattr_set(lovset);
1210         RETURN(rc ? rc : err);
1211 }
1212
1213 static int lov_getattr_async(struct obd_export *exp, struct obd_info *oinfo,
1214                               struct ptlrpc_request_set *rqset)
1215 {
1216         struct lov_request_set *lovset;
1217         struct lov_obd *lov;
1218         cfs_list_t *pos;
1219         struct lov_request *req;
1220         int rc = 0, err;
1221         ENTRY;
1222
1223         LASSERT(oinfo);
1224         ASSERT_LSM_MAGIC(oinfo->oi_md);
1225
1226         if (!exp || !exp->exp_obd)
1227                 RETURN(-ENODEV);
1228
1229         lov = &exp->exp_obd->u.lov;
1230
1231         rc = lov_prep_getattr_set(exp, oinfo, &lovset);
1232         if (rc)
1233                 RETURN(rc);
1234
1235         CDEBUG(D_INFO, "objid "LPX64": %ux%u byte stripes\n",
1236                oinfo->oi_md->lsm_object_id, oinfo->oi_md->lsm_stripe_count,
1237                oinfo->oi_md->lsm_stripe_size);
1238
1239         cfs_list_for_each (pos, &lovset->set_list) {
1240                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1241
1242                 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
1243                        "%u\n", oinfo->oi_oa->o_id, req->rq_stripe,
1244                        req->rq_oi.oi_oa->o_id, req->rq_idx);
1245                 rc = obd_getattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1246                                        &req->rq_oi, rqset);
1247                 if (rc) {
1248                         CERROR("error: getattr objid "LPX64" subobj "
1249                                LPX64" on OST idx %d: rc = %d\n",
1250                                oinfo->oi_oa->o_id, req->rq_oi.oi_oa->o_id,
1251                                req->rq_idx, rc);
1252                         GOTO(out, rc);
1253                 }
1254         }
1255
1256         if (!cfs_list_empty(&rqset->set_requests)) {
1257                 LASSERT(rc == 0);
1258                 LASSERT (rqset->set_interpret == NULL);
1259                 rqset->set_interpret = lov_getattr_interpret;
1260                 rqset->set_arg = (void *)lovset;
1261                 RETURN(rc);
1262         }
1263 out:
1264         if (rc)
1265                 cfs_atomic_set(&lovset->set_completes, 0);
1266         err = lov_fini_getattr_set(lovset);
1267         RETURN(rc ? rc : err);
1268 }
1269
1270 static int lov_setattr(const struct lu_env *env, struct obd_export *exp,
1271                        struct obd_info *oinfo, struct obd_trans_info *oti)
1272 {
1273         struct lov_request_set *set;
1274         struct lov_obd *lov;
1275         cfs_list_t *pos;
1276         struct lov_request *req;
1277         int err = 0, rc = 0;
1278         ENTRY;
1279
1280         LASSERT(oinfo);
1281         ASSERT_LSM_MAGIC(oinfo->oi_md);
1282
1283         if (!exp || !exp->exp_obd)
1284                 RETURN(-ENODEV);
1285
1286         /* for now, we only expect the following updates here */
1287         LASSERT(!(oinfo->oi_oa->o_valid & ~(OBD_MD_FLID | OBD_MD_FLTYPE |
1288                                             OBD_MD_FLMODE | OBD_MD_FLATIME |
1289                                             OBD_MD_FLMTIME | OBD_MD_FLCTIME |
1290                                             OBD_MD_FLFLAGS | OBD_MD_FLSIZE |
1291                                             OBD_MD_FLGROUP | OBD_MD_FLUID |
1292                                             OBD_MD_FLGID | OBD_MD_FLFID |
1293                                             OBD_MD_FLGENER)));
1294         lov = &exp->exp_obd->u.lov;
1295         rc = lov_prep_setattr_set(exp, oinfo, oti, &set);
1296         if (rc)
1297                 RETURN(rc);
1298
1299         cfs_list_for_each (pos, &set->set_list) {
1300                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1301
1302                 rc = obd_setattr(env, lov->lov_tgts[req->rq_idx]->ltd_exp,
1303                                  &req->rq_oi, NULL);
1304                 err = lov_update_setattr_set(set, req, rc);
1305                 if (err) {
1306                         CERROR("error: setattr objid "LPX64" subobj "
1307                                LPX64" on OST idx %d: rc = %d\n",
1308                                set->set_oi->oi_oa->o_id,
1309                                req->rq_oi.oi_oa->o_id, req->rq_idx, err);
1310                         if (!rc)
1311                                 rc = err;
1312                 }
1313         }
1314         err = lov_fini_setattr_set(set);
1315         if (!rc)
1316                 rc = err;
1317         RETURN(rc);
1318 }
1319
1320 static int lov_setattr_interpret(struct ptlrpc_request_set *rqset,
1321                                  void *data, int rc)
1322 {
1323         struct lov_request_set *lovset = (struct lov_request_set *)data;
1324         int err;
1325         ENTRY;
1326
1327         if (rc)
1328                 cfs_atomic_set(&lovset->set_completes, 0);
1329         err = lov_fini_setattr_set(lovset);
1330         RETURN(rc ? rc : err);
1331 }
1332
1333 /* If @oti is given, the request goes from MDS and responses from OSTs are not
1334    needed. Otherwise, a client is waiting for responses. */
1335 static int lov_setattr_async(struct obd_export *exp, struct obd_info *oinfo,
1336                              struct obd_trans_info *oti,
1337                              struct ptlrpc_request_set *rqset)
1338 {
1339         struct lov_request_set *set;
1340         struct lov_request *req;
1341         cfs_list_t *pos;
1342         struct lov_obd *lov;
1343         int rc = 0;
1344         ENTRY;
1345
1346         LASSERT(oinfo);
1347         ASSERT_LSM_MAGIC(oinfo->oi_md);
1348         if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE) {
1349                 LASSERT(oti);
1350                 LASSERT(oti->oti_logcookies);
1351         }
1352
1353         if (!exp || !exp->exp_obd)
1354                 RETURN(-ENODEV);
1355
1356         lov = &exp->exp_obd->u.lov;
1357         rc = lov_prep_setattr_set(exp, oinfo, oti, &set);
1358         if (rc)
1359                 RETURN(rc);
1360
1361         CDEBUG(D_INFO, "objid "LPX64": %ux%u byte stripes\n",
1362                oinfo->oi_md->lsm_object_id, oinfo->oi_md->lsm_stripe_count,
1363                oinfo->oi_md->lsm_stripe_size);
1364
1365         cfs_list_for_each (pos, &set->set_list) {
1366                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1367
1368                 if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE)
1369                         oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1370
1371                 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
1372                        "%u\n", oinfo->oi_oa->o_id, req->rq_stripe,
1373                        req->rq_oi.oi_oa->o_id, req->rq_idx);
1374
1375                 rc = obd_setattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1376                                        &req->rq_oi, oti, rqset);
1377                 if (rc) {
1378                         CERROR("error: setattr objid "LPX64" subobj "
1379                                LPX64" on OST idx %d: rc = %d\n",
1380                                set->set_oi->oi_oa->o_id,
1381                                req->rq_oi.oi_oa->o_id,
1382                                req->rq_idx, rc);
1383                         break;
1384                 }
1385         }
1386
1387         /* If we are not waiting for responses on async requests, return. */
1388         if (rc || !rqset || cfs_list_empty(&rqset->set_requests)) {
1389                 int err;
1390                 if (rc)
1391                         cfs_atomic_set(&set->set_completes, 0);
1392                 err = lov_fini_setattr_set(set);
1393                 RETURN(rc ? rc : err);
1394         }
1395
1396         LASSERT(rqset->set_interpret == NULL);
1397         rqset->set_interpret = lov_setattr_interpret;
1398         rqset->set_arg = (void *)set;
1399
1400         RETURN(0);
1401 }
1402
1403 static int lov_punch_interpret(struct ptlrpc_request_set *rqset,
1404                                void *data, int rc)
1405 {
1406         struct lov_request_set *lovset = (struct lov_request_set *)data;
1407         int err;
1408         ENTRY;
1409
1410         if (rc)
1411                 cfs_atomic_set(&lovset->set_completes, 0);
1412         err = lov_fini_punch_set(lovset);
1413         RETURN(rc ? rc : err);
1414 }
1415
1416 /* FIXME: maybe we'll just make one node the authoritative attribute node, then
1417  * we can send this 'punch' to just the authoritative node and the nodes
1418  * that the punch will affect. */
1419 static int lov_punch(const struct lu_env *env, struct obd_export *exp,
1420                      struct obd_info *oinfo, struct obd_trans_info *oti,
1421                      struct ptlrpc_request_set *rqset)
1422 {
1423         struct lov_request_set *set;
1424         struct lov_obd *lov;
1425         cfs_list_t *pos;
1426         struct lov_request *req;
1427         int rc = 0;
1428         ENTRY;
1429
1430         LASSERT(oinfo);
1431         ASSERT_LSM_MAGIC(oinfo->oi_md);
1432
1433         if (!exp || !exp->exp_obd)
1434                 RETURN(-ENODEV);
1435
1436         lov = &exp->exp_obd->u.lov;
1437         rc = lov_prep_punch_set(exp, oinfo, oti, &set);
1438         if (rc)
1439                 RETURN(rc);
1440
1441         cfs_list_for_each (pos, &set->set_list) {
1442                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1443
1444                 rc = obd_punch(env, lov->lov_tgts[req->rq_idx]->ltd_exp,
1445                                &req->rq_oi, NULL, rqset);
1446                 if (rc) {
1447                         CERROR("error: punch objid "LPX64" subobj "LPX64
1448                                " on OST idx %d: rc = %d\n",
1449                                set->set_oi->oi_oa->o_id,
1450                                req->rq_oi.oi_oa->o_id, req->rq_idx, rc);
1451                         break;
1452                 }
1453         }
1454
1455         if (rc || cfs_list_empty(&rqset->set_requests)) {
1456                 int err;
1457                 err = lov_fini_punch_set(set);
1458                 RETURN(rc ? rc : err);
1459         }
1460
1461         LASSERT(rqset->set_interpret == NULL);
1462         rqset->set_interpret = lov_punch_interpret;
1463         rqset->set_arg = (void *)set;
1464
1465         RETURN(0);
1466 }
1467
1468 static int lov_sync_interpret(struct ptlrpc_request_set *rqset,
1469                               void *data, int rc)
1470 {
1471         struct lov_request_set *lovset = data;
1472         int err;
1473         ENTRY;
1474
1475         if (rc)
1476                 cfs_atomic_set(&lovset->set_completes, 0);
1477         err = lov_fini_sync_set(lovset);
1478         RETURN(rc ?: err);
1479 }
1480
1481 static int lov_sync(const struct lu_env *env, struct obd_export *exp,
1482                     struct obd_info *oinfo, obd_off start, obd_off end,
1483                     struct ptlrpc_request_set *rqset)
1484 {
1485         struct lov_request_set *set = NULL;
1486         struct lov_obd *lov;
1487         cfs_list_t *pos;
1488         struct lov_request *req;
1489         int rc = 0;
1490         ENTRY;
1491
1492         ASSERT_LSM_MAGIC(oinfo->oi_md);
1493         LASSERT(rqset != NULL);
1494
1495         if (!exp->exp_obd)
1496                 RETURN(-ENODEV);
1497
1498         lov = &exp->exp_obd->u.lov;
1499         rc = lov_prep_sync_set(exp, oinfo, start, end, &set);
1500         if (rc)
1501                 RETURN(rc);
1502
1503         CDEBUG(D_INFO, "fsync objid "LPX64" ["LPX64", "LPX64"]\n",
1504                set->set_oi->oi_oa->o_id, start, end);
1505
1506         cfs_list_for_each (pos, &set->set_list) {
1507                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1508
1509                 rc = obd_sync(env, lov->lov_tgts[req->rq_idx]->ltd_exp,
1510                               &req->rq_oi, req->rq_oi.oi_policy.l_extent.start,
1511                               req->rq_oi.oi_policy.l_extent.end, rqset);
1512                 if (rc) {
1513                         CERROR("error: fsync objid "LPX64" subobj "LPX64
1514                                " on OST idx %d: rc = %d\n",
1515                                set->set_oi->oi_oa->o_id,
1516                                req->rq_oi.oi_oa->o_id, req->rq_idx, rc);
1517                         break;
1518                 }
1519         }
1520
1521         /* If we are not waiting for responses on async requests, return. */
1522         if (rc || cfs_list_empty(&rqset->set_requests)) {
1523                 int err = lov_fini_sync_set(set);
1524
1525                 RETURN(rc ?: err);
1526         }
1527
1528         LASSERT(rqset->set_interpret == NULL);
1529         rqset->set_interpret = lov_sync_interpret;
1530         rqset->set_arg = (void *)set;
1531
1532         RETURN(0);
1533 }
1534
1535 static int lov_brw_check(struct lov_obd *lov, struct obd_info *lov_oinfo,
1536                          obd_count oa_bufs, struct brw_page *pga)
1537 {
1538         struct obd_info oinfo = { { { 0 } } };
1539         int i, rc = 0;
1540
1541         oinfo.oi_oa = lov_oinfo->oi_oa;
1542
1543         /* The caller just wants to know if there's a chance that this
1544          * I/O can succeed */
1545         for (i = 0; i < oa_bufs; i++) {
1546                 int stripe = lov_stripe_number(lov_oinfo->oi_md, pga[i].off);
1547                 int ost = lov_oinfo->oi_md->lsm_oinfo[stripe]->loi_ost_idx;
1548                 obd_off start, end;
1549
1550                 if (!lov_stripe_intersects(lov_oinfo->oi_md, i, pga[i].off,
1551                                            pga[i].off + pga[i].count - 1,
1552                                            &start, &end))
1553                         continue;
1554
1555                 if (!lov->lov_tgts[ost] || !lov->lov_tgts[ost]->ltd_active) {
1556                         CDEBUG(D_HA, "lov idx %d inactive\n", ost);
1557                         return -EIO;
1558                 }
1559
1560                 rc = obd_brw(OBD_BRW_CHECK, lov->lov_tgts[ost]->ltd_exp, &oinfo,
1561                              1, &pga[i], NULL);
1562                 if (rc)
1563                         break;
1564         }
1565         return rc;
1566 }
1567
1568 static int lov_brw(int cmd, struct obd_export *exp, struct obd_info *oinfo,
1569                    obd_count oa_bufs, struct brw_page *pga,
1570                    struct obd_trans_info *oti)
1571 {
1572         struct lov_request_set *set;
1573         struct lov_request *req;
1574         cfs_list_t *pos;
1575         struct lov_obd *lov = &exp->exp_obd->u.lov;
1576         int err, rc = 0;
1577         ENTRY;
1578
1579         ASSERT_LSM_MAGIC(oinfo->oi_md);
1580
1581         if (cmd == OBD_BRW_CHECK) {
1582                 rc = lov_brw_check(lov, oinfo, oa_bufs, pga);
1583                 RETURN(rc);
1584         }
1585
1586         rc = lov_prep_brw_set(exp, oinfo, oa_bufs, pga, oti, &set);
1587         if (rc)
1588                 RETURN(rc);
1589
1590         cfs_list_for_each (pos, &set->set_list) {
1591                 struct obd_export *sub_exp;
1592                 struct brw_page *sub_pga;
1593                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1594
1595                 sub_exp = lov->lov_tgts[req->rq_idx]->ltd_exp;
1596                 sub_pga = set->set_pga + req->rq_pgaidx;
1597                 rc = obd_brw(cmd, sub_exp, &req->rq_oi, req->rq_oabufs,
1598                              sub_pga, oti);
1599                 if (rc)
1600                         break;
1601                 lov_update_common_set(set, req, rc);
1602         }
1603
1604         err = lov_fini_brw_set(set);
1605         if (!rc)
1606                 rc = err;
1607         RETURN(rc);
1608 }
1609
1610 static int lov_enqueue_interpret(struct ptlrpc_request_set *rqset,
1611                                  void *data, int rc)
1612 {
1613         struct lov_request_set *lovset = (struct lov_request_set *)data;
1614         ENTRY;
1615         rc = lov_fini_enqueue_set(lovset, lovset->set_ei->ei_mode, rc, rqset);
1616         RETURN(rc);
1617 }
1618
1619 static int lov_enqueue(struct obd_export *exp, struct obd_info *oinfo,
1620                        struct ldlm_enqueue_info *einfo,
1621                        struct ptlrpc_request_set *rqset)
1622 {
1623         ldlm_mode_t mode = einfo->ei_mode;
1624         struct lov_request_set *set;
1625         struct lov_request *req;
1626         cfs_list_t *pos;
1627         struct lov_obd *lov;
1628         ldlm_error_t rc;
1629         ENTRY;
1630
1631         LASSERT(oinfo);
1632         ASSERT_LSM_MAGIC(oinfo->oi_md);
1633         LASSERT(mode == (mode & -mode));
1634
1635         /* we should never be asked to replay a lock this way. */
1636         LASSERT((oinfo->oi_flags & LDLM_FL_REPLAY) == 0);
1637
1638         if (!exp || !exp->exp_obd)
1639                 RETURN(-ENODEV);
1640
1641         lov = &exp->exp_obd->u.lov;
1642         rc = lov_prep_enqueue_set(exp, oinfo, einfo, &set);
1643         if (rc)
1644                 RETURN(rc);
1645
1646         cfs_list_for_each (pos, &set->set_list) {
1647                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1648
1649                 rc = obd_enqueue(lov->lov_tgts[req->rq_idx]->ltd_exp,
1650                                  &req->rq_oi, einfo, rqset);
1651                 if (rc != ELDLM_OK)
1652                         GOTO(out, rc);
1653         }
1654
1655         if (rqset && !cfs_list_empty(&rqset->set_requests)) {
1656                 LASSERT(rc == 0);
1657                 LASSERT(rqset->set_interpret == NULL);
1658                 rqset->set_interpret = lov_enqueue_interpret;
1659                 rqset->set_arg = (void *)set;
1660                 RETURN(rc);
1661         }
1662 out:
1663         rc = lov_fini_enqueue_set(set, mode, rc, rqset);
1664         RETURN(rc);
1665 }
1666
1667 static int lov_change_cbdata(struct obd_export *exp,
1668                              struct lov_stripe_md *lsm, ldlm_iterator_t it,
1669                              void *data)
1670 {
1671         struct lov_obd *lov;
1672         int rc = 0, i;
1673         ENTRY;
1674
1675         ASSERT_LSM_MAGIC(lsm);
1676
1677         if (!exp || !exp->exp_obd)
1678                 RETURN(-ENODEV);
1679
1680         lov = &exp->exp_obd->u.lov;
1681         for (i = 0; i < lsm->lsm_stripe_count; i++) {
1682                 struct lov_stripe_md submd;
1683                 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
1684
1685                 if (!lov->lov_tgts[loi->loi_ost_idx]) {
1686                         CDEBUG(D_HA, "lov idx %d NULL \n", loi->loi_ost_idx);
1687                         continue;
1688                 }
1689
1690                 submd.lsm_object_id = loi->loi_id;
1691                 submd.lsm_object_seq = loi->loi_seq;
1692                 submd.lsm_stripe_count = 0;
1693                 rc = obd_change_cbdata(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1694                                        &submd, it, data);
1695         }
1696         RETURN(rc);
1697 }
1698
1699 /* find any ldlm lock of the inode in lov
1700  * return 0    not find
1701  *        1    find one
1702  *      < 0    error */
1703 static int lov_find_cbdata(struct obd_export *exp,
1704                            struct lov_stripe_md *lsm, ldlm_iterator_t it,
1705                            void *data)
1706 {
1707         struct lov_obd *lov;
1708         int rc = 0, i;
1709         ENTRY;
1710
1711         ASSERT_LSM_MAGIC(lsm);
1712
1713         if (!exp || !exp->exp_obd)
1714                 RETURN(-ENODEV);
1715
1716         lov = &exp->exp_obd->u.lov;
1717         for (i = 0; i < lsm->lsm_stripe_count; i++) {
1718                 struct lov_stripe_md submd;
1719                 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
1720
1721                 if (!lov->lov_tgts[loi->loi_ost_idx]) {
1722                         CDEBUG(D_HA, "lov idx %d NULL \n", loi->loi_ost_idx);
1723                         continue;
1724                 }
1725
1726                 submd.lsm_object_id = loi->loi_id;
1727                 submd.lsm_object_seq = loi->loi_seq;
1728                 submd.lsm_stripe_count = 0;
1729                 rc = obd_find_cbdata(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1730                                      &submd, it, data);
1731                 if (rc != 0)
1732                         RETURN(rc);
1733         }
1734         RETURN(rc);
1735 }
1736
1737 static int lov_cancel(struct obd_export *exp, struct lov_stripe_md *lsm,
1738                       __u32 mode, struct lustre_handle *lockh)
1739 {
1740         struct lov_request_set *set;
1741         struct obd_info oinfo;
1742         struct lov_request *req;
1743         cfs_list_t *pos;
1744         struct lov_obd *lov;
1745         struct lustre_handle *lov_lockhp;
1746         int err = 0, rc = 0;
1747         ENTRY;
1748
1749         ASSERT_LSM_MAGIC(lsm);
1750
1751         if (!exp || !exp->exp_obd)
1752                 RETURN(-ENODEV);
1753
1754         LASSERT(lockh);
1755         lov = &exp->exp_obd->u.lov;
1756         rc = lov_prep_cancel_set(exp, &oinfo, lsm, mode, lockh, &set);
1757         if (rc)
1758                 RETURN(rc);
1759
1760         cfs_list_for_each (pos, &set->set_list) {
1761                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1762                 lov_lockhp = set->set_lockh->llh_handles + req->rq_stripe;
1763
1764                 rc = obd_cancel(lov->lov_tgts[req->rq_idx]->ltd_exp,
1765                                 req->rq_oi.oi_md, mode, lov_lockhp);
1766                 rc = lov_update_common_set(set, req, rc);
1767                 if (rc) {
1768                         CERROR("error: cancel objid "LPX64" subobj "
1769                                LPX64" on OST idx %d: rc = %d\n",
1770                                lsm->lsm_object_id,
1771                                req->rq_oi.oi_md->lsm_object_id,
1772                                req->rq_idx, rc);
1773                         err = rc;
1774                 }
1775
1776         }
1777         lov_fini_cancel_set(set);
1778         RETURN(err);
1779 }
1780
1781 static int lov_cancel_unused(struct obd_export *exp,
1782                              struct lov_stripe_md *lsm,
1783                              ldlm_cancel_flags_t flags, void *opaque)
1784 {
1785         struct lov_obd *lov;
1786         int rc = 0, i;
1787         ENTRY;
1788
1789         if (!exp || !exp->exp_obd)
1790                 RETURN(-ENODEV);
1791
1792         lov = &exp->exp_obd->u.lov;
1793         if (lsm == NULL) {
1794                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
1795                         int err;
1796                         if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
1797                                 continue;
1798
1799                         err = obd_cancel_unused(lov->lov_tgts[i]->ltd_exp, NULL,
1800                                                 flags, opaque);
1801                         if (!rc)
1802                                 rc = err;
1803                 }
1804                 RETURN(rc);
1805         }
1806
1807         ASSERT_LSM_MAGIC(lsm);
1808
1809         for (i = 0; i < lsm->lsm_stripe_count; i++) {
1810                 struct lov_stripe_md submd;
1811                 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
1812                 int err;
1813
1814                 if (!lov->lov_tgts[loi->loi_ost_idx]) {
1815                         CDEBUG(D_HA, "lov idx %d NULL\n", loi->loi_ost_idx);
1816                         continue;
1817                 }
1818
1819                 if (!lov->lov_tgts[loi->loi_ost_idx]->ltd_active)
1820                         CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
1821
1822                 submd.lsm_object_id = loi->loi_id;
1823                 submd.lsm_object_seq = loi->loi_seq;
1824                 submd.lsm_stripe_count = 0;
1825                 err = obd_cancel_unused(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1826                                         &submd, flags, opaque);
1827                 if (err && lov->lov_tgts[loi->loi_ost_idx]->ltd_active) {
1828                         CERROR("error: cancel unused objid "LPX64" subobj "LPX64
1829                                " on OST idx %d: rc = %d\n", lsm->lsm_object_id,
1830                                loi->loi_id, loi->loi_ost_idx, err);
1831                         if (!rc)
1832                                 rc = err;
1833                 }
1834         }
1835         RETURN(rc);
1836 }
1837
1838 int lov_statfs_interpret(struct ptlrpc_request_set *rqset, void *data, int rc)
1839 {
1840         struct lov_request_set *lovset = (struct lov_request_set *)data;
1841         int err;
1842         ENTRY;
1843
1844         if (rc)
1845                 cfs_atomic_set(&lovset->set_completes, 0);
1846
1847         err = lov_fini_statfs_set(lovset);
1848         RETURN(rc ? rc : err);
1849 }
1850
1851 static int lov_statfs_async(struct obd_export *exp, struct obd_info *oinfo,
1852                             __u64 max_age, struct ptlrpc_request_set *rqset)
1853 {
1854         struct obd_device      *obd = class_exp2obd(exp);
1855         struct lov_request_set *set;
1856         struct lov_request *req;
1857         cfs_list_t *pos;
1858         struct lov_obd *lov;
1859         int rc = 0;
1860         ENTRY;
1861
1862         LASSERT(oinfo != NULL);
1863         LASSERT(oinfo->oi_osfs != NULL);
1864
1865         lov = &obd->u.lov;
1866         rc = lov_prep_statfs_set(obd, oinfo, &set);
1867         if (rc)
1868                 RETURN(rc);
1869
1870         cfs_list_for_each (pos, &set->set_list) {
1871                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1872                 rc = obd_statfs_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1873                                       &req->rq_oi, max_age, rqset);
1874                 if (rc)
1875                         break;
1876         }
1877
1878         if (rc || cfs_list_empty(&rqset->set_requests)) {
1879                 int err;
1880                 if (rc)
1881                         cfs_atomic_set(&set->set_completes, 0);
1882                 err = lov_fini_statfs_set(set);
1883                 RETURN(rc ? rc : err);
1884         }
1885
1886         LASSERT(rqset->set_interpret == NULL);
1887         rqset->set_interpret = lov_statfs_interpret;
1888         rqset->set_arg = (void *)set;
1889         RETURN(0);
1890 }
1891
1892 static int lov_statfs(const struct lu_env *env, struct obd_export *exp,
1893                       struct obd_statfs *osfs, __u64 max_age, __u32 flags)
1894 {
1895         struct ptlrpc_request_set *set = NULL;
1896         struct obd_info oinfo = { { { 0 } } };
1897         int rc = 0;
1898         ENTRY;
1899
1900
1901         /* for obdclass we forbid using obd_statfs_rqset, but prefer using async
1902          * statfs requests */
1903         set = ptlrpc_prep_set();
1904         if (set == NULL)
1905                 RETURN(-ENOMEM);
1906
1907         oinfo.oi_osfs = osfs;
1908         oinfo.oi_flags = flags;
1909         rc = lov_statfs_async(exp, &oinfo, max_age, set);
1910         if (rc == 0)
1911                 rc = ptlrpc_set_wait(set);
1912         ptlrpc_set_destroy(set);
1913
1914         RETURN(rc);
1915 }
1916
1917 static int lov_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1918                          void *karg, void *uarg)
1919 {
1920         struct obd_device *obddev = class_exp2obd(exp);
1921         struct lov_obd *lov = &obddev->u.lov;
1922         int i = 0, rc = 0, count = lov->desc.ld_tgt_count;
1923         struct obd_uuid *uuidp;
1924         ENTRY;
1925
1926         switch (cmd) {
1927         case IOC_OBD_STATFS: {
1928                 struct obd_ioctl_data *data = karg;
1929                 struct obd_device *osc_obd;
1930                 struct obd_statfs stat_buf = {0};
1931                 __u32 index;
1932                 __u32 flags;
1933
1934                 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
1935                 if ((index >= count))
1936                         RETURN(-ENODEV);
1937
1938                 if (!lov->lov_tgts[index])
1939                         /* Try again with the next index */
1940                         RETURN(-EAGAIN);
1941                 if (!lov->lov_tgts[index]->ltd_active)
1942                         RETURN(-ENODATA);
1943
1944                 osc_obd = class_exp2obd(lov->lov_tgts[index]->ltd_exp);
1945                 if (!osc_obd)
1946                         RETURN(-EINVAL);
1947
1948                 /* copy UUID */
1949                 if (cfs_copy_to_user(data->ioc_pbuf2, obd2cli_tgt(osc_obd),
1950                                      min((int) data->ioc_plen2,
1951                                          (int) sizeof(struct obd_uuid))))
1952                         RETURN(-EFAULT);
1953
1954                 flags = uarg ? *(__u32*)uarg : 0;
1955                 /* got statfs data */
1956                 rc = obd_statfs(NULL, lov->lov_tgts[index]->ltd_exp, &stat_buf,
1957                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1958                                 flags);
1959                 if (rc)
1960                         RETURN(rc);
1961                 if (cfs_copy_to_user(data->ioc_pbuf1, &stat_buf,
1962                                      min((int) data->ioc_plen1,
1963                                          (int) sizeof(stat_buf))))
1964                         RETURN(-EFAULT);
1965                 break;
1966         }
1967         case OBD_IOC_LOV_GET_CONFIG: {
1968                 struct obd_ioctl_data *data;
1969                 struct lov_desc *desc;
1970                 char *buf = NULL;
1971                 __u32 *genp;
1972
1973                 len = 0;
1974                 if (obd_ioctl_getdata(&buf, &len, (void *)uarg))
1975                         RETURN(-EINVAL);
1976
1977                 data = (struct obd_ioctl_data *)buf;
1978
1979                 if (sizeof(*desc) > data->ioc_inllen1) {
1980                         obd_ioctl_freedata(buf, len);
1981                         RETURN(-EINVAL);
1982                 }
1983
1984                 if (sizeof(uuidp->uuid) * count > data->ioc_inllen2) {
1985                         obd_ioctl_freedata(buf, len);
1986                         RETURN(-EINVAL);
1987                 }
1988
1989                 if (sizeof(__u32) * count > data->ioc_inllen3) {
1990                         obd_ioctl_freedata(buf, len);
1991                         RETURN(-EINVAL);
1992                 }
1993
1994                 desc = (struct lov_desc *)data->ioc_inlbuf1;
1995                 memcpy(desc, &(lov->desc), sizeof(*desc));
1996
1997                 uuidp = (struct obd_uuid *)data->ioc_inlbuf2;
1998                 genp = (__u32 *)data->ioc_inlbuf3;
1999                 /* the uuid will be empty for deleted OSTs */
2000                 for (i = 0; i < count; i++, uuidp++, genp++) {
2001                         if (!lov->lov_tgts[i])
2002                                 continue;
2003                         *uuidp = lov->lov_tgts[i]->ltd_uuid;
2004                         *genp = lov->lov_tgts[i]->ltd_gen;
2005                 }
2006
2007                 if (cfs_copy_to_user((void *)uarg, buf, len))
2008                         rc = -EFAULT;
2009                 obd_ioctl_freedata(buf, len);
2010                 break;
2011         }
2012         case LL_IOC_LOV_SETSTRIPE:
2013                 rc = lov_setstripe(exp, len, karg, uarg);
2014                 break;
2015         case LL_IOC_LOV_GETSTRIPE:
2016                 rc = lov_getstripe(exp, karg, uarg);
2017                 break;
2018         case LL_IOC_LOV_SETEA:
2019                 rc = lov_setea(exp, karg, uarg);
2020                 break;
2021         case OBD_IOC_QUOTACTL: {
2022                 struct if_quotactl *qctl = karg;
2023                 struct lov_tgt_desc *tgt = NULL;
2024                 struct obd_quotactl *oqctl;
2025
2026                 if (qctl->qc_valid == QC_OSTIDX) {
2027                         if (qctl->qc_idx < 0 || count <= qctl->qc_idx)
2028                                 RETURN(-EINVAL);
2029
2030                         tgt = lov->lov_tgts[qctl->qc_idx];
2031                         if (!tgt || !tgt->ltd_exp)
2032                                 RETURN(-EINVAL);
2033                 } else if (qctl->qc_valid == QC_UUID) {
2034                         for (i = 0; i < count; i++) {
2035                                 tgt = lov->lov_tgts[i];
2036                                 if (!tgt ||
2037                                     !obd_uuid_equals(&tgt->ltd_uuid,
2038                                                      &qctl->obd_uuid))
2039                                         continue;
2040
2041                                 if (tgt->ltd_exp == NULL)
2042                                         RETURN(-EINVAL);
2043
2044                                 break;
2045                         }
2046                 } else {
2047                         RETURN(-EINVAL);
2048                 }
2049
2050                 if (i >= count)
2051                         RETURN(-EAGAIN);
2052
2053                 LASSERT(tgt && tgt->ltd_exp);
2054                 OBD_ALLOC_PTR(oqctl);
2055                 if (!oqctl)
2056                         RETURN(-ENOMEM);
2057
2058                 QCTL_COPY(oqctl, qctl);
2059                 rc = obd_quotactl(tgt->ltd_exp, oqctl);
2060                 if (rc == 0) {
2061                         QCTL_COPY(qctl, oqctl);
2062                         qctl->qc_valid = QC_OSTIDX;
2063                         qctl->obd_uuid = tgt->ltd_uuid;
2064                 }
2065                 OBD_FREE_PTR(oqctl);
2066                 break;
2067         }
2068         default: {
2069                 int set = 0;
2070
2071                 if (count == 0)
2072                         RETURN(-ENOTTY);
2073
2074                 for (i = 0; i < count; i++) {
2075                         int err;
2076                         struct obd_device *osc_obd;
2077
2078                         /* OST was disconnected */
2079                         if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
2080                                 continue;
2081
2082                         /* ll_umount_begin() sets force flag but for lov, not
2083                          * osc. Let's pass it through */
2084                         osc_obd = class_exp2obd(lov->lov_tgts[i]->ltd_exp);
2085                         osc_obd->obd_force = obddev->obd_force;
2086                         err = obd_iocontrol(cmd, lov->lov_tgts[i]->ltd_exp,
2087                                             len, karg, uarg);
2088                         if (err == -ENODATA && cmd == OBD_IOC_POLL_QUOTACHECK) {
2089                                 RETURN(err);
2090                         } else if (err) {
2091                                 if (lov->lov_tgts[i]->ltd_active) {
2092                                         CDEBUG(err == -ENOTTY ?
2093                                                D_IOCTL : D_WARNING,
2094                                                "iocontrol OSC %s on OST "
2095                                                "idx %d cmd %x: err = %d\n",
2096                                                lov_uuid2str(lov, i),
2097                                                i, cmd, err);
2098                                         if (!rc)
2099                                                 rc = err;
2100                                 }
2101                         } else {
2102                                 set = 1;
2103                         }
2104                 }
2105                 if (!set && !rc)
2106                         rc = -EIO;
2107         }
2108         }
2109
2110         RETURN(rc);
2111 }
2112
2113 #define FIEMAP_BUFFER_SIZE 4096
2114
2115 /**
2116  * Non-zero fe_logical indicates that this is a continuation FIEMAP
2117  * call. The local end offset and the device are sent in the first
2118  * fm_extent. This function calculates the stripe number from the index.
2119  * This function returns a stripe_no on which mapping is to be restarted.
2120  *
2121  * This function returns fm_end_offset which is the in-OST offset at which
2122  * mapping should be restarted. If fm_end_offset=0 is returned then caller
2123  * will re-calculate proper offset in next stripe.
2124  * Note that the first extent is passed to lov_get_info via the value field.
2125  *
2126  * \param fiemap fiemap request header
2127  * \param lsm striping information for the file
2128  * \param fm_start logical start of mapping
2129  * \param fm_end logical end of mapping
2130  * \param start_stripe starting stripe will be returned in this
2131  */
2132 obd_size fiemap_calc_fm_end_offset(struct ll_user_fiemap *fiemap,
2133                                    struct lov_stripe_md *lsm, obd_size fm_start,
2134                                    obd_size fm_end, int *start_stripe)
2135 {
2136         obd_size local_end = fiemap->fm_extents[0].fe_logical;
2137         obd_off lun_start, lun_end;
2138         obd_size fm_end_offset;
2139         int stripe_no = -1, i;
2140
2141         if (fiemap->fm_extent_count == 0 ||
2142             fiemap->fm_extents[0].fe_logical == 0)
2143                 return 0;
2144
2145         /* Find out stripe_no from ost_index saved in the fe_device */
2146         for (i = 0; i < lsm->lsm_stripe_count; i++) {
2147                 if (lsm->lsm_oinfo[i]->loi_ost_idx ==
2148                                         fiemap->fm_extents[0].fe_device) {
2149                         stripe_no = i;
2150                         break;
2151                 }
2152         }
2153         if (stripe_no == -1)
2154                 return -EINVAL;
2155
2156         /* If we have finished mapping on previous device, shift logical
2157          * offset to start of next device */
2158         if ((lov_stripe_intersects(lsm, stripe_no, fm_start, fm_end,
2159                                    &lun_start, &lun_end)) != 0 &&
2160                                    local_end < lun_end) {
2161                 fm_end_offset = local_end;
2162                 *start_stripe = stripe_no;
2163         } else {
2164                 /* This is a special value to indicate that caller should
2165                  * calculate offset in next stripe. */
2166                 fm_end_offset = 0;
2167                 *start_stripe = (stripe_no + 1) % lsm->lsm_stripe_count;
2168         }
2169
2170         return fm_end_offset;
2171 }
2172
2173 /**
2174  * We calculate on which OST the mapping will end. If the length of mapping
2175  * is greater than (stripe_size * stripe_count) then the last_stripe will
2176  * will be one just before start_stripe. Else we check if the mapping
2177  * intersects each OST and find last_stripe.
2178  * This function returns the last_stripe and also sets the stripe_count
2179  * over which the mapping is spread
2180  *
2181  * \param lsm striping information for the file
2182  * \param fm_start logical start of mapping
2183  * \param fm_end logical end of mapping
2184  * \param start_stripe starting stripe of the mapping
2185  * \param stripe_count the number of stripes across which to map is returned
2186  *
2187  * \retval last_stripe return the last stripe of the mapping
2188  */
2189 int fiemap_calc_last_stripe(struct lov_stripe_md *lsm, obd_size fm_start,
2190                             obd_size fm_end, int start_stripe,
2191                             int *stripe_count)
2192 {
2193         int last_stripe;
2194         obd_off obd_start, obd_end;
2195         int i, j;
2196
2197         if (fm_end - fm_start > lsm->lsm_stripe_size * lsm->lsm_stripe_count) {
2198                 last_stripe = (start_stripe < 1 ? lsm->lsm_stripe_count - 1 :
2199                                                               start_stripe - 1);
2200                 *stripe_count = lsm->lsm_stripe_count;
2201         } else {
2202                 for (j = 0, i = start_stripe; j < lsm->lsm_stripe_count;
2203                      i = (i + 1) % lsm->lsm_stripe_count, j++) {
2204                         if ((lov_stripe_intersects(lsm, i, fm_start, fm_end,
2205                                                    &obd_start, &obd_end)) == 0)
2206                                 break;
2207                 }
2208                 *stripe_count = j;
2209                 last_stripe = (start_stripe + j - 1) %lsm->lsm_stripe_count;
2210         }
2211
2212         return last_stripe;
2213 }
2214
2215 /**
2216  * Set fe_device and copy extents from local buffer into main return buffer.
2217  *
2218  * \param fiemap fiemap request header
2219  * \param lcl_fm_ext array of local fiemap extents to be copied
2220  * \param ost_index OST index to be written into the fm_device field for each
2221                     extent
2222  * \param ext_count number of extents to be copied
2223  * \param current_extent where to start copying in main extent array
2224  */
2225 void fiemap_prepare_and_copy_exts(struct ll_user_fiemap *fiemap,
2226                                   struct ll_fiemap_extent *lcl_fm_ext,
2227                                   int ost_index, unsigned int ext_count,
2228                                   int current_extent)
2229 {
2230         char *to;
2231         int ext;
2232
2233         for (ext = 0; ext < ext_count; ext++) {
2234                 lcl_fm_ext[ext].fe_device = ost_index;
2235                 lcl_fm_ext[ext].fe_flags |= FIEMAP_EXTENT_NET;
2236         }
2237
2238         /* Copy fm_extent's from fm_local to return buffer */
2239         to = (char *)fiemap + fiemap_count_to_size(current_extent);
2240         memcpy(to, lcl_fm_ext, ext_count * sizeof(struct ll_fiemap_extent));
2241 }
2242
2243 /**
2244  * Break down the FIEMAP request and send appropriate calls to individual OSTs.
2245  * This also handles the restarting of FIEMAP calls in case mapping overflows
2246  * the available number of extents in single call.
2247  */
2248 static int lov_fiemap(struct lov_obd *lov, __u32 keylen, void *key,
2249                       __u32 *vallen, void *val, struct lov_stripe_md *lsm)
2250 {
2251         struct ll_fiemap_info_key *fm_key = key;
2252         struct ll_user_fiemap *fiemap = val;
2253         struct ll_user_fiemap *fm_local = NULL;
2254         struct ll_fiemap_extent *lcl_fm_ext;
2255         int count_local;
2256         unsigned int get_num_extents = 0;
2257         int ost_index = 0, actual_start_stripe, start_stripe;
2258         obd_size fm_start, fm_end, fm_length, fm_end_offset;
2259         obd_size curr_loc;
2260         int current_extent = 0, rc = 0, i;
2261         int ost_eof = 0; /* EOF for object */
2262         int ost_done = 0; /* done with required mapping for this OST? */
2263         int last_stripe;
2264         int cur_stripe = 0, cur_stripe_wrap = 0, stripe_count;
2265         unsigned int buffer_size = FIEMAP_BUFFER_SIZE;
2266
2267         if (lsm == NULL)
2268                 GOTO(out, rc = 0);
2269
2270         if (fiemap_count_to_size(fm_key->fiemap.fm_extent_count) < buffer_size)
2271                 buffer_size = fiemap_count_to_size(fm_key->fiemap.fm_extent_count);
2272
2273         OBD_ALLOC_LARGE(fm_local, buffer_size);
2274         if (fm_local == NULL)
2275                 GOTO(out, rc = -ENOMEM);
2276         lcl_fm_ext = &fm_local->fm_extents[0];
2277
2278         count_local = fiemap_size_to_count(buffer_size);
2279
2280         memcpy(fiemap, &fm_key->fiemap, sizeof(*fiemap));
2281         fm_start = fiemap->fm_start;
2282         fm_length = fiemap->fm_length;
2283         /* Calculate start stripe, last stripe and length of mapping */
2284         actual_start_stripe = start_stripe = lov_stripe_number(lsm, fm_start);
2285         fm_end = (fm_length == ~0ULL ? fm_key->oa.o_size :
2286                                                 fm_start + fm_length - 1);
2287         /* If fm_length != ~0ULL but fm_start+fm_length-1 exceeds file size */
2288         if (fm_end > fm_key->oa.o_size)
2289                 fm_end = fm_key->oa.o_size;
2290
2291         last_stripe = fiemap_calc_last_stripe(lsm, fm_start, fm_end,
2292                                             actual_start_stripe, &stripe_count);
2293
2294         fm_end_offset = fiemap_calc_fm_end_offset(fiemap, lsm, fm_start,
2295                                                   fm_end, &start_stripe);
2296         if (fm_end_offset == -EINVAL)
2297                 GOTO(out, rc = -EINVAL);
2298
2299         if (fiemap->fm_extent_count == 0) {
2300                 get_num_extents = 1;
2301                 count_local = 0;
2302         }
2303
2304         /* Check each stripe */
2305         for (cur_stripe = start_stripe, i = 0; i < stripe_count;
2306              i++, cur_stripe = (cur_stripe + 1) % lsm->lsm_stripe_count) {
2307                 obd_size req_fm_len; /* Stores length of required mapping */
2308                 obd_size len_mapped_single_call;
2309                 obd_off lun_start, lun_end, obd_object_end;
2310                 unsigned int ext_count;
2311
2312                 cur_stripe_wrap = cur_stripe;
2313
2314                 /* Find out range of mapping on this stripe */
2315                 if ((lov_stripe_intersects(lsm, cur_stripe, fm_start, fm_end,
2316                                            &lun_start, &obd_object_end)) == 0)
2317                         continue;
2318
2319                 /* If this is a continuation FIEMAP call and we are on
2320                  * starting stripe then lun_start needs to be set to
2321                  * fm_end_offset */
2322                 if (fm_end_offset != 0 && cur_stripe == start_stripe)
2323                         lun_start = fm_end_offset;
2324
2325                 if (fm_length != ~0ULL) {
2326                         /* Handle fm_start + fm_length overflow */
2327                         if (fm_start + fm_length < fm_start)
2328                                 fm_length = ~0ULL - fm_start;
2329                         lun_end = lov_size_to_stripe(lsm, fm_start + fm_length,
2330                                                      cur_stripe);
2331                 } else {
2332                         lun_end = ~0ULL;
2333                 }
2334
2335                 if (lun_start == lun_end)
2336                         continue;
2337
2338                 req_fm_len = obd_object_end - lun_start;
2339                 fm_local->fm_length = 0;
2340                 len_mapped_single_call = 0;
2341
2342                 /* If the output buffer is very large and the objects have many
2343                  * extents we may need to loop on a single OST repeatedly */
2344                 ost_eof = 0;
2345                 ost_done = 0;
2346                 do {
2347                         if (get_num_extents == 0) {
2348                                 /* Don't get too many extents. */
2349                                 if (current_extent + count_local >
2350                                     fiemap->fm_extent_count)
2351                                         count_local = fiemap->fm_extent_count -
2352                                                                  current_extent;
2353                         }
2354
2355                         lun_start += len_mapped_single_call;
2356                         fm_local->fm_length = req_fm_len - len_mapped_single_call;
2357                         req_fm_len = fm_local->fm_length;
2358                         fm_local->fm_extent_count = count_local;
2359                         fm_local->fm_mapped_extents = 0;
2360                         fm_local->fm_flags = fiemap->fm_flags;
2361
2362                         fm_key->oa.o_id = lsm->lsm_oinfo[cur_stripe]->loi_id;
2363                         fm_key->oa.o_seq = lsm->lsm_oinfo[cur_stripe]->loi_seq;
2364                         ost_index = lsm->lsm_oinfo[cur_stripe]->loi_ost_idx;
2365
2366                         if (ost_index < 0 || ost_index >=lov->desc.ld_tgt_count)
2367                                 GOTO(out, rc = -EINVAL);
2368
2369                         /* If OST is inactive, return extent with UNKNOWN flag */
2370                         if (!lov->lov_tgts[ost_index]->ltd_active) {
2371                                 fm_local->fm_flags |= FIEMAP_EXTENT_LAST;
2372                                 fm_local->fm_mapped_extents = 1;
2373
2374                                 lcl_fm_ext[0].fe_logical = lun_start;
2375                                 lcl_fm_ext[0].fe_length = obd_object_end -
2376                                                                       lun_start;
2377                                 lcl_fm_ext[0].fe_flags |= FIEMAP_EXTENT_UNKNOWN;
2378
2379                                 goto inactive_tgt;
2380                         }
2381
2382                         fm_local->fm_start = lun_start;
2383                         fm_local->fm_flags &= ~FIEMAP_FLAG_DEVICE_ORDER;
2384                         memcpy(&fm_key->fiemap, fm_local, sizeof(*fm_local));
2385                         *vallen=fiemap_count_to_size(fm_local->fm_extent_count);
2386                         rc = obd_get_info(NULL,
2387                                           lov->lov_tgts[ost_index]->ltd_exp,
2388                                           keylen, key, vallen, fm_local, lsm);
2389                         if (rc != 0)
2390                                 GOTO(out, rc);
2391
2392 inactive_tgt:
2393                         ext_count = fm_local->fm_mapped_extents;
2394                         if (ext_count == 0) {
2395                                 ost_done = 1;
2396                                 /* If last stripe has hole at the end,
2397                                  * then we need to return */
2398                                 if (cur_stripe_wrap == last_stripe) {
2399                                         fiemap->fm_mapped_extents = 0;
2400                                         goto finish;
2401                                 }
2402                                 break;
2403                         }
2404
2405                         /* If we just need num of extents then go to next device */
2406                         if (get_num_extents) {
2407                                 current_extent += ext_count;
2408                                 break;
2409                         }
2410
2411                         len_mapped_single_call = lcl_fm_ext[ext_count-1].fe_logical -
2412                                   lun_start + lcl_fm_ext[ext_count - 1].fe_length;
2413
2414                         /* Have we finished mapping on this device? */
2415                         if (req_fm_len <= len_mapped_single_call)
2416                                 ost_done = 1;
2417
2418                         /* Clear the EXTENT_LAST flag which can be present on
2419                          * last extent */
2420                         if (lcl_fm_ext[ext_count-1].fe_flags & FIEMAP_EXTENT_LAST)
2421                                 lcl_fm_ext[ext_count - 1].fe_flags &=
2422                                                             ~FIEMAP_EXTENT_LAST;
2423
2424                         curr_loc = lov_stripe_size(lsm,
2425                                            lcl_fm_ext[ext_count - 1].fe_logical+
2426                                            lcl_fm_ext[ext_count - 1].fe_length,
2427                                            cur_stripe);
2428                         if (curr_loc >= fm_key->oa.o_size)
2429                                 ost_eof = 1;
2430
2431                         fiemap_prepare_and_copy_exts(fiemap, lcl_fm_ext,
2432                                                      ost_index, ext_count,
2433                                                      current_extent);
2434
2435                         current_extent += ext_count;
2436
2437                         /* Ran out of available extents? */
2438                         if (current_extent >= fiemap->fm_extent_count)
2439                                 goto finish;
2440                 } while (ost_done == 0 && ost_eof == 0);
2441
2442                 if (cur_stripe_wrap == last_stripe)
2443                         goto finish;
2444         }
2445
2446 finish:
2447         /* Indicate that we are returning device offsets unless file just has
2448          * single stripe */
2449         if (lsm->lsm_stripe_count > 1)
2450                 fiemap->fm_flags |= FIEMAP_FLAG_DEVICE_ORDER;
2451
2452         if (get_num_extents)
2453                 goto skip_last_device_calc;
2454
2455         /* Check if we have reached the last stripe and whether mapping for that
2456          * stripe is done. */
2457         if (cur_stripe_wrap == last_stripe) {
2458                 if (ost_done || ost_eof)
2459                         fiemap->fm_extents[current_extent - 1].fe_flags |=
2460                                                              FIEMAP_EXTENT_LAST;
2461         }
2462
2463 skip_last_device_calc:
2464         fiemap->fm_mapped_extents = current_extent;
2465
2466 out:
2467         OBD_FREE_LARGE(fm_local, buffer_size);
2468         return rc;
2469 }
2470
2471 static int lov_get_info(const struct lu_env *env, struct obd_export *exp,
2472                         __u32 keylen, void *key, __u32 *vallen, void *val,
2473                         struct lov_stripe_md *lsm)
2474 {
2475         struct obd_device *obddev = class_exp2obd(exp);
2476         struct lov_obd *lov = &obddev->u.lov;
2477         int i, rc;
2478         ENTRY;
2479
2480         if (!vallen || !val)
2481                 RETURN(-EFAULT);
2482
2483         obd_getref(obddev);
2484
2485         if (KEY_IS(KEY_LOCK_TO_STRIPE)) {
2486                 struct {
2487                         char name[16];
2488                         struct ldlm_lock *lock;
2489                 } *data = key;
2490                 struct ldlm_res_id *res_id = &data->lock->l_resource->lr_name;
2491                 struct lov_oinfo *loi;
2492                 __u32 *stripe = val;
2493
2494                 if (*vallen < sizeof(*stripe))
2495                         GOTO(out, rc = -EFAULT);
2496                 *vallen = sizeof(*stripe);
2497
2498                 /* XXX This is another one of those bits that will need to
2499                  * change if we ever actually support nested LOVs.  It uses
2500                  * the lock's export to find out which stripe it is. */
2501                 /* XXX - it's assumed all the locks for deleted OSTs have
2502                  * been cancelled. Also, the export for deleted OSTs will
2503                  * be NULL and won't match the lock's export. */
2504                 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2505                         loi = lsm->lsm_oinfo[i];
2506                         if (!lov->lov_tgts[loi->loi_ost_idx])
2507                                 continue;
2508                         if (lov->lov_tgts[loi->loi_ost_idx]->ltd_exp ==
2509                             data->lock->l_conn_export &&
2510                             ostid_res_name_eq(&loi->loi_oi, res_id)) {
2511                                 *stripe = i;
2512                                 GOTO(out, rc = 0);
2513                         }
2514                 }
2515                 LDLM_ERROR(data->lock, "lock on inode without such object");
2516                 dump_lsm(D_ERROR, lsm);
2517                 GOTO(out, rc = -ENXIO);
2518         } else if (KEY_IS(KEY_LAST_ID)) {
2519                 struct obd_id_info *info = val;
2520                 __u32 size = sizeof(obd_id);
2521                 struct lov_tgt_desc *tgt;
2522
2523                 LASSERT(*vallen == sizeof(struct obd_id_info));
2524                 tgt = lov->lov_tgts[info->idx];
2525
2526                 if (!tgt || !tgt->ltd_active)
2527                         GOTO(out, rc = -ESRCH);
2528
2529                 rc = obd_get_info(env, tgt->ltd_exp, keylen, key,
2530                                   &size, info->data, NULL);
2531                 GOTO(out, rc = 0);
2532         } else if (KEY_IS(KEY_LOVDESC)) {
2533                 struct lov_desc *desc_ret = val;
2534                 *desc_ret = lov->desc;
2535
2536                 GOTO(out, rc = 0);
2537         } else if (KEY_IS(KEY_FIEMAP)) {
2538                 rc = lov_fiemap(lov, keylen, key, vallen, val, lsm);
2539                 GOTO(out, rc);
2540         } else if (KEY_IS(KEY_CONNECT_FLAG)) {
2541                 struct lov_tgt_desc *tgt;
2542                 __u64 ost_idx = *((__u64*)val);
2543
2544                 LASSERT(*vallen == sizeof(__u64));
2545                 LASSERT(ost_idx < lov->desc.ld_tgt_count);
2546                 tgt = lov->lov_tgts[ost_idx];
2547
2548                 if (!tgt || !tgt->ltd_exp)
2549                         GOTO(out, rc = -ESRCH);
2550
2551                 *((__u64 *)val) = exp_connect_flags(tgt->ltd_exp);
2552                 GOTO(out, rc = 0);
2553         } else if (KEY_IS(KEY_TGT_COUNT)) {
2554                 *((int *)val) = lov->desc.ld_tgt_count;
2555                 GOTO(out, rc = 0);
2556         }
2557
2558         rc = -EINVAL;
2559
2560 out:
2561         obd_putref(obddev);
2562         RETURN(rc);
2563 }
2564
2565 static int lov_set_info_async(const struct lu_env *env, struct obd_export *exp,
2566                               obd_count keylen, void *key, obd_count vallen,
2567                               void *val, struct ptlrpc_request_set *set)
2568 {
2569         struct obd_device *obddev = class_exp2obd(exp);
2570         struct lov_obd *lov = &obddev->u.lov;
2571         obd_count count;
2572         int i, rc = 0, err;
2573         struct lov_tgt_desc *tgt;
2574         unsigned incr, check_uuid,
2575                  do_inactive, no_set;
2576         unsigned next_id = 0,  mds_con = 0, capa = 0;
2577         ENTRY;
2578
2579         incr = check_uuid = do_inactive = no_set = 0;
2580         if (set == NULL) {
2581                 no_set = 1;
2582                 set = ptlrpc_prep_set();
2583                 if (!set)
2584                         RETURN(-ENOMEM);
2585         }
2586
2587         obd_getref(obddev);
2588         count = lov->desc.ld_tgt_count;
2589
2590         if (KEY_IS(KEY_NEXT_ID)) {
2591                 count = vallen / sizeof(struct obd_id_info);
2592                 vallen = sizeof(obd_id);
2593                 incr = sizeof(struct obd_id_info);
2594                 do_inactive = 1;
2595                 next_id = 1;
2596         } else if (KEY_IS(KEY_CHECKSUM)) {
2597                 do_inactive = 1;
2598         } else if (KEY_IS(KEY_EVICT_BY_NID)) {
2599                 /* use defaults:  do_inactive = incr = 0; */
2600         } else if (KEY_IS(KEY_MDS_CONN)) {
2601                 mds_con = 1;
2602         } else if (KEY_IS(KEY_CAPA_KEY)) {
2603                 capa = 1;
2604         } else if (KEY_IS(KEY_CACHE_SET)) {
2605                 LASSERT(lov->lov_cache == NULL);
2606                 lov->lov_cache = val;
2607                 do_inactive = 1;
2608         }
2609
2610         for (i = 0; i < count; i++, val = (char *)val + incr) {
2611                 if (next_id) {
2612                         tgt = lov->lov_tgts[((struct obd_id_info*)val)->idx];
2613                 } else {
2614                         tgt = lov->lov_tgts[i];
2615                 }
2616                 /* OST was disconnected */
2617                 if (!tgt || !tgt->ltd_exp)
2618                         continue;
2619
2620                 /* OST is inactive and we don't want inactive OSCs */
2621                 if (!tgt->ltd_active && !do_inactive)
2622                         continue;
2623
2624                 if (mds_con) {
2625                         struct mds_group_info *mgi;
2626
2627                         LASSERT(vallen == sizeof(*mgi));
2628                         mgi = (struct mds_group_info *)val;
2629
2630                         /* Only want a specific OSC */
2631                         if (mgi->uuid && !obd_uuid_equals(mgi->uuid,
2632                                                 &tgt->ltd_uuid))
2633                                 continue;
2634
2635                         err = obd_set_info_async(env, tgt->ltd_exp,
2636                                          keylen, key, sizeof(int),
2637                                          &mgi->group, set);
2638                 } else if (next_id) {
2639                         err = obd_set_info_async(env, tgt->ltd_exp,
2640                                          keylen, key, vallen,
2641                                          ((struct obd_id_info*)val)->data, set);
2642                 } else if (capa) {
2643                         struct mds_capa_info *info = (struct mds_capa_info*)val;
2644
2645                         LASSERT(vallen == sizeof(*info));
2646
2647                          /* Only want a specific OSC */
2648                         if (info->uuid &&
2649                             !obd_uuid_equals(info->uuid, &tgt->ltd_uuid))
2650                                 continue;
2651
2652                         err = obd_set_info_async(env, tgt->ltd_exp, keylen,
2653                                                  key, sizeof(*info->capa),
2654                                                  info->capa, set);
2655                 } else {
2656                         /* Only want a specific OSC */
2657                         if (check_uuid &&
2658                             !obd_uuid_equals(val, &tgt->ltd_uuid))
2659                                 continue;
2660
2661                         err = obd_set_info_async(env, tgt->ltd_exp,
2662                                          keylen, key, vallen, val, set);
2663                 }
2664
2665                 if (!rc)
2666                         rc = err;
2667         }
2668
2669         obd_putref(obddev);
2670         if (no_set) {
2671                 err = ptlrpc_set_wait(set);
2672                 if (!rc)
2673                         rc = err;
2674                 ptlrpc_set_destroy(set);
2675         }
2676         RETURN(rc);
2677 }
2678
2679 static int lov_extent_calc(struct obd_export *exp, struct lov_stripe_md *lsm,
2680                            int cmd, __u64 *offset)
2681 {
2682         __u32 ssize = lsm->lsm_stripe_size;
2683         __u64 start;
2684
2685         start = *offset;
2686         lov_do_div64(start, ssize);
2687         start = start * ssize;
2688
2689         CDEBUG(D_DLMTRACE, "offset "LPU64", stripe %u, start "LPU64
2690                            ", end "LPU64"\n", *offset, ssize, start,
2691                            start + ssize - 1);
2692         if (cmd == OBD_CALC_STRIPE_END) {
2693                 *offset = start + ssize - 1;
2694         } else if (cmd == OBD_CALC_STRIPE_START) {
2695                 *offset = start;
2696         } else {
2697                 LBUG();
2698         }
2699
2700         RETURN(0);
2701 }
2702
2703 void lov_stripe_lock(struct lov_stripe_md *md)
2704 {
2705         LASSERT(md->lsm_lock_owner != cfs_curproc_pid());
2706         spin_lock(&md->lsm_lock);
2707         LASSERT(md->lsm_lock_owner == 0);
2708         md->lsm_lock_owner = cfs_curproc_pid();
2709 }
2710 EXPORT_SYMBOL(lov_stripe_lock);
2711
2712 void lov_stripe_unlock(struct lov_stripe_md *md)
2713 {
2714         LASSERT(md->lsm_lock_owner == cfs_curproc_pid());
2715         md->lsm_lock_owner = 0;
2716         spin_unlock(&md->lsm_lock);
2717 }
2718 EXPORT_SYMBOL(lov_stripe_unlock);
2719
2720 static int lov_quotactl(struct obd_device *obd, struct obd_export *exp,
2721                         struct obd_quotactl *oqctl)
2722 {
2723         struct lov_obd      *lov = &obd->u.lov;
2724         struct lov_tgt_desc *tgt;
2725         __u64                curspace = 0;
2726         __u64                bhardlimit = 0;
2727         int                  i, rc = 0;
2728         ENTRY;
2729
2730         if (oqctl->qc_cmd != LUSTRE_Q_QUOTAON &&
2731             oqctl->qc_cmd != LUSTRE_Q_QUOTAOFF &&
2732             oqctl->qc_cmd != Q_GETOQUOTA &&
2733             oqctl->qc_cmd != Q_INITQUOTA &&
2734             oqctl->qc_cmd != LUSTRE_Q_SETQUOTA &&
2735             oqctl->qc_cmd != Q_FINVALIDATE) {
2736                 CERROR("bad quota opc %x for lov obd", oqctl->qc_cmd);
2737                 RETURN(-EFAULT);
2738         }
2739
2740         /* for lov tgt */
2741         obd_getref(obd);
2742         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2743                 int err;
2744
2745                 tgt = lov->lov_tgts[i];
2746
2747                 if (!tgt)
2748                         continue;
2749
2750                 if (!tgt->ltd_active || tgt->ltd_reap) {
2751                         if (oqctl->qc_cmd == Q_GETOQUOTA &&
2752                             lov->lov_tgts[i]->ltd_activate) {
2753                                 rc = -EREMOTEIO;
2754                                 CERROR("ost %d is inactive\n", i);
2755                         } else {
2756                                 CDEBUG(D_HA, "ost %d is inactive\n", i);
2757                         }
2758                         continue;
2759                 }
2760
2761                 err = obd_quotactl(tgt->ltd_exp, oqctl);
2762                 if (err) {
2763                         if (tgt->ltd_active && !rc)
2764                                 rc = err;
2765                         continue;
2766                 }
2767
2768                 if (oqctl->qc_cmd == Q_GETOQUOTA) {
2769                         curspace += oqctl->qc_dqblk.dqb_curspace;
2770                         bhardlimit += oqctl->qc_dqblk.dqb_bhardlimit;
2771                 }
2772         }
2773         obd_putref(obd);
2774
2775         if (oqctl->qc_cmd == Q_GETOQUOTA) {
2776                 oqctl->qc_dqblk.dqb_curspace = curspace;
2777                 oqctl->qc_dqblk.dqb_bhardlimit = bhardlimit;
2778         }
2779         RETURN(rc);
2780 }
2781
2782 static int lov_quotacheck(struct obd_device *obd, struct obd_export *exp,
2783                           struct obd_quotactl *oqctl)
2784 {
2785         struct lov_obd *lov = &obd->u.lov;
2786         int             i, rc = 0;
2787         ENTRY;
2788
2789         obd_getref(obd);
2790
2791         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2792                 if (!lov->lov_tgts[i])
2793                         continue;
2794
2795                 /* Skip quota check on the administratively disabled OSTs. */
2796                 if (!lov->lov_tgts[i]->ltd_activate) {
2797                         CWARN("lov idx %d was administratively disabled, "
2798                               "skip quotacheck on it.\n", i);
2799                         continue;
2800                 }
2801
2802                 if (!lov->lov_tgts[i]->ltd_active) {
2803                         CERROR("lov idx %d inactive\n", i);
2804                         rc = -EIO;
2805                         goto out;
2806                 }
2807         }
2808
2809         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2810                 int err;
2811
2812                 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_activate)
2813                         continue;
2814
2815                 err = obd_quotacheck(lov->lov_tgts[i]->ltd_exp, oqctl);
2816                 if (err && !rc)
2817                         rc = err;
2818         }
2819
2820 out:
2821         obd_putref(obd);
2822
2823         RETURN(rc);
2824 }
2825
2826 struct obd_ops lov_obd_ops = {
2827         .o_owner               = THIS_MODULE,
2828         .o_setup               = lov_setup,
2829         .o_precleanup          = lov_precleanup,
2830         .o_cleanup             = lov_cleanup,
2831         //.o_process_config      = lov_process_config,
2832         .o_connect             = lov_connect,
2833         .o_disconnect          = lov_disconnect,
2834         .o_statfs              = lov_statfs,
2835         .o_statfs_async        = lov_statfs_async,
2836         .o_packmd              = lov_packmd,
2837         .o_unpackmd            = lov_unpackmd,
2838         .o_create              = lov_create,
2839         .o_destroy             = lov_destroy,
2840         .o_getattr             = lov_getattr,
2841         .o_getattr_async       = lov_getattr_async,
2842         .o_setattr             = lov_setattr,
2843         .o_setattr_async       = lov_setattr_async,
2844         .o_brw                 = lov_brw,
2845         .o_merge_lvb           = lov_merge_lvb,
2846         .o_adjust_kms          = lov_adjust_kms,
2847         .o_punch               = lov_punch,
2848         .o_sync                = lov_sync,
2849         .o_enqueue             = lov_enqueue,
2850         .o_change_cbdata       = lov_change_cbdata,
2851         .o_find_cbdata         = lov_find_cbdata,
2852         .o_cancel              = lov_cancel,
2853         .o_cancel_unused       = lov_cancel_unused,
2854         .o_iocontrol           = lov_iocontrol,
2855         .o_get_info            = lov_get_info,
2856         .o_set_info_async      = lov_set_info_async,
2857         .o_extent_calc         = lov_extent_calc,
2858         .o_llog_init           = lov_llog_init,
2859         .o_llog_finish         = lov_llog_finish,
2860         .o_notify              = lov_notify,
2861         .o_pool_new            = lov_pool_new,
2862         .o_pool_rem            = lov_pool_remove,
2863         .o_pool_add            = lov_pool_add,
2864         .o_pool_del            = lov_pool_del,
2865         .o_getref              = lov_getref,
2866         .o_putref              = lov_putref,
2867         .o_quotactl            = lov_quotactl,
2868         .o_quotacheck          = lov_quotacheck,
2869 };
2870
2871 cfs_mem_cache_t *lov_oinfo_slab;
2872
2873 extern struct lu_kmem_descr lov_caches[];
2874
2875 int __init lov_init(void)
2876 {
2877         struct lprocfs_static_vars lvars = { 0 };
2878         int rc, rc2;
2879         ENTRY;
2880
2881         /* print an address of _any_ initialized kernel symbol from this
2882          * module, to allow debugging with gdb that doesn't support data
2883          * symbols from modules.*/
2884         CDEBUG(D_INFO, "Lustre LOV module (%p).\n", &lov_caches);
2885
2886         rc = lu_kmem_init(lov_caches);
2887         if (rc)
2888                 return rc;
2889
2890         lov_oinfo_slab = cfs_mem_cache_create("lov_oinfo",
2891                                               sizeof(struct lov_oinfo),
2892                                               0, CFS_SLAB_HWCACHE_ALIGN);
2893         if (lov_oinfo_slab == NULL) {
2894                 lu_kmem_fini(lov_caches);
2895                 return -ENOMEM;
2896         }
2897         lprocfs_lov_init_vars(&lvars);
2898
2899         rc = class_register_type(&lov_obd_ops, NULL, lvars.module_vars,
2900                                  LUSTRE_LOV_NAME, &lov_device_type);
2901
2902         if (rc) {
2903                 rc2 = cfs_mem_cache_destroy(lov_oinfo_slab);
2904                 LASSERT(rc2 == 0);
2905                 lu_kmem_fini(lov_caches);
2906         }
2907
2908         RETURN(rc);
2909 }
2910
2911 #ifdef __KERNEL__
2912 static void /*__exit*/ lov_exit(void)
2913 {
2914         int rc;
2915
2916         class_unregister_type(LUSTRE_LOV_NAME);
2917         rc = cfs_mem_cache_destroy(lov_oinfo_slab);
2918         LASSERT(rc == 0);
2919
2920         lu_kmem_fini(lov_caches);
2921 }
2922
2923 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2924 MODULE_DESCRIPTION("Lustre Logical Object Volume OBD driver");
2925 MODULE_LICENSE("GPL");
2926
2927 cfs_module(lov, LUSTRE_VERSION_STRING, lov_init, lov_exit);
2928 #endif