lustre/lov/lov_obd.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * GPL HEADER START
   5  *
   6  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License version 2 only,
  10  * as published by the Free Software Foundation.
  11  *
  12  * This program is distributed in the hope that it will be useful, but
  13  * WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * General Public License version 2 for more details (a copy is included
  16  * in the LICENSE file that accompanied this code).
  17  *
  18  * You should have received a copy of the GNU General Public License
  19  * version 2 along with this program; If not, see
  20  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  21  *
  22  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  23  * CA 95054 USA or visit www.sun.com if you need additional information or
  24  * have any questions.
  25  *
  26  * GPL HEADER END
  27  */
  28 /*
  29  * Copyright  2008 Sun Microsystems, Inc. All rights reserved
  30  * Use is subject to license terms.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/lov/lov_obd.c
  37  *
  38  * Author: Phil Schwan <phil@clusterfs.com>
  39  * Author: Peter Braam <braam@clusterfs.com>
  40  * Author: Mike Shaver <shaver@clusterfs.com>
  41  * Author: Nathan Rutman <nathan@clusterfs.com>
  42  */
  43
  44 #ifndef EXPORT_SYMTAB
  45 # define EXPORT_SYMTAB
  46 #endif
  47 #define DEBUG_SUBSYSTEM S_LOV
  48 #ifdef __KERNEL__
  49 #include <libcfs/libcfs.h>
  50 #else
  51 #include <liblustre.h>
  52 #endif
  53
  54 #include <obd_support.h>
  55 #include <lustre_lib.h>
  56 #include <lustre_net.h>
  57 #include <lustre/lustre_idl.h>
  58 #include <lustre_dlm.h>
  59 #include <lustre_mds.h>
  60 #include <lustre_debug.h>
  61 #include <obd_class.h>
  62 #include <obd_lov.h>
  63 #include <obd_ost.h>
  64 #include <lprocfs_status.h>
  65 #include <lustre_param.h>
  66 #include <cl_object.h>
  67 #include <lustre/ll_fiemap.h>
  68
  69 #include "lov_internal.h"
  70
  71
  72 /* Keep a refcount of lov->tgt usage to prevent racing with addition/deletion.
  73    Any function that expects lov_tgts to remain stationary must take a ref. */
  74 void lov_getref(struct obd_device *obd)
  75 {
  76         struct lov_obd *lov = &obd->u.lov;
  77
  78         /* nobody gets through here until lov_putref is done */
  79         mutex_down(&lov->lov_lock);
  80         atomic_inc(&lov->lov_refcount);
  81         mutex_up(&lov->lov_lock);
  82         return;
  83 }
  84
  85 static void __lov_del_obd(struct obd_device *obd, struct lov_tgt_desc *tgt);
  86
  87 void lov_putref(struct obd_device *obd)
  88 {
  89         struct lov_obd *lov = &obd->u.lov;
  90
  91         mutex_down(&lov->lov_lock);
  92         /* ok to dec to 0 more than once -- ltd_exp's will be null */
  93         if (atomic_dec_and_test(&lov->lov_refcount) && lov->lov_death_row) {
  94                 CFS_LIST_HEAD(kill);
  95                 int i;
  96                 struct lov_tgt_desc *tgt, *n;
  97                 CDEBUG(D_CONFIG, "destroying %d lov targets\n",
  98                        lov->lov_death_row);
  99                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
 100                         tgt = lov->lov_tgts[i];
 101
 102                         if (!tgt || !tgt->ltd_reap)
 103                                 continue;
 104                         list_add(&tgt->ltd_kill, &kill);
 105                         /* XXX - right now there is a dependency on ld_tgt_count
 106                          * being the maximum tgt index for computing the
 107                          * mds_max_easize. So we can't shrink it. */
 108                         lov_ost_pool_remove(&lov->lov_packed, i);
 109                         lov->lov_tgts[i] = NULL;
 110                         lov->lov_death_row--;
 111                 }
 112                 mutex_up(&lov->lov_lock);
 113
 114                 list_for_each_entry_safe(tgt, n, &kill, ltd_kill) {
 115                         list_del(&tgt->ltd_kill);
 116                         /* Disconnect */
 117                         __lov_del_obd(obd, tgt);
 118                 }
 119         } else {
 120                 mutex_up(&lov->lov_lock);
 121         }
 122 }
 123
 124 static int lov_set_osc_active(struct obd_device *obd, struct obd_uuid *uuid,
 125                               int activate);
 126 static int lov_notify(struct obd_device *obd, struct obd_device *watched,
 127                       enum obd_notify_event ev, void *data);
 128
 129
 130 #define MAX_STRING_SIZE 128
 131 int lov_connect_obd(struct obd_device *obd, __u32 index, int activate,
 132                     struct obd_connect_data *data)
 133 {
 134         struct lov_obd *lov = &obd->u.lov;
 135         struct obd_uuid tgt_uuid;
 136         struct obd_device *tgt_obd;
 137         struct obd_uuid lov_osc_uuid = { "LOV_OSC_UUID" };
 138         struct obd_import *imp;
 139
 140 #ifdef __KERNEL__
 141         cfs_proc_dir_entry_t *lov_proc_dir;
 142 #endif
 143         int rc;
 144         ENTRY;
 145
 146         if (!lov->lov_tgts[index])
 147                 RETURN(-EINVAL);
 148
 149         tgt_uuid = lov->lov_tgts[index]->ltd_uuid;
 150
 151         tgt_obd = class_find_client_obd(&tgt_uuid, LUSTRE_OSC_NAME,
 152                                         &obd->obd_uuid);
 153
 154         if (!tgt_obd) {
 155                 CERROR("Target %s not attached\n", obd_uuid2str(&tgt_uuid));
 156                 RETURN(-EINVAL);
 157         }
 158         if (!tgt_obd->obd_set_up) {
 159                 CERROR("Target %s not set up\n", obd_uuid2str(&tgt_uuid));
 160                 RETURN(-EINVAL);
 161         }
 162
 163         /* override the sp_me from lov */
 164         tgt_obd->u.cli.cl_sp_me = lov->lov_sp_me;
 165
 166         if (data && (data->ocd_connect_flags & OBD_CONNECT_INDEX))
 167                 data->ocd_index = index;
 168
 169         /*
 170          * Divine LOV knows that OBDs under it are OSCs.
 171          */
 172         imp = tgt_obd->u.cli.cl_import;
 173
 174         if (activate) {
 175                 tgt_obd->obd_no_recov = 0;
 176                 /* FIXME this is probably supposed to be
 177                    ptlrpc_set_import_active.  Horrible naming. */
 178                 ptlrpc_activate_import(imp);
 179         }
 180
 181         rc = obd_register_observer(tgt_obd, obd);
 182         if (rc) {
 183                 CERROR("Target %s register_observer error %d\n",
 184                        obd_uuid2str(&tgt_uuid), rc);
 185                 RETURN(rc);
 186         }
 187
 188
 189         if (imp->imp_invalid) {
 190                 CERROR("not connecting OSC %s; administratively "
 191                        "disabled\n", obd_uuid2str(&tgt_uuid));
 192                 RETURN(0);
 193         }
 194
 195         rc = obd_connect(NULL, &lov->lov_tgts[index]->ltd_exp, tgt_obd,
 196                          &lov_osc_uuid, data, NULL);
 197         if (rc || !lov->lov_tgts[index]->ltd_exp) {
 198                 CERROR("Target %s connect error %d\n",
 199                        obd_uuid2str(&tgt_uuid), rc);
 200                 RETURN(-ENODEV);
 201         }
 202
 203         lov->lov_tgts[index]->ltd_reap = 0;
 204         CDEBUG(D_CONFIG, "Connected tgt idx %d %s (%s) %sactive\n", index,
 205                obd_uuid2str(&tgt_uuid), tgt_obd->obd_name, activate ? "":"in");
 206
 207 #ifdef __KERNEL__
 208         lov_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
 209         if (lov_proc_dir) {
 210                 struct obd_device *osc_obd = lov->lov_tgts[index]->ltd_exp->exp_obd;
 211                 cfs_proc_dir_entry_t *osc_symlink;
 212                 char name[MAX_STRING_SIZE];
 213
 214                 LASSERT(osc_obd != NULL);
 215                 LASSERT(osc_obd->obd_magic == OBD_DEVICE_MAGIC);
 216                 LASSERT(osc_obd->obd_type->typ_name != NULL);
 217                 snprintf(name, MAX_STRING_SIZE, "../../../%s/%s",
 218                          osc_obd->obd_type->typ_name,
 219                          osc_obd->obd_name);
 220                 osc_symlink = lprocfs_add_symlink(osc_obd->obd_name, lov_proc_dir,
 221                                                   name);
 222                 if (osc_symlink == NULL) {
 223                         CERROR("could not register LOV target "
 224                                "/proc/fs/lustre/%s/%s/target_obds/%s.",
 225                                obd->obd_type->typ_name, obd->obd_name,
 226                                osc_obd->obd_name);
 227                         lprocfs_remove(&lov_proc_dir);
 228                 }
 229         }
 230 #endif
 231
 232         rc = qos_add_tgt(obd, index);
 233         if (rc)
 234                 CERROR("qos_add_tgt failed %d\n", rc);
 235
 236         RETURN(0);
 237 }
 238
 239 static int lov_connect(const struct lu_env *env,
 240                        struct obd_export **exp, struct obd_device *obd,
 241                        struct obd_uuid *cluuid, struct obd_connect_data *data,
 242                        void *localdata)
 243 {
 244         struct lov_obd *lov = &obd->u.lov;
 245         struct lov_tgt_desc *tgt;
 246         struct lustre_handle conn;
 247         int i, rc;
 248         ENTRY;
 249
 250         CDEBUG(D_CONFIG, "connect #%d\n", lov->lov_connects);
 251
 252         rc = class_connect(&conn, obd, cluuid);
 253         if (rc)
 254                 RETURN(rc);
 255
 256         *exp = class_conn2export(&conn);
 257
 258         /* Why should there ever be more than 1 connect? */
 259         lov->lov_connects++;
 260         LASSERT(lov->lov_connects == 1);
 261
 262         memset(&lov->lov_ocd, 0, sizeof(lov->lov_ocd));
 263         if (data)
 264                 lov->lov_ocd = *data;
 265
 266         lov_getref(obd);
 267         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
 268                 tgt = lov->lov_tgts[i];
 269                 if (!tgt || obd_uuid_empty(&tgt->ltd_uuid))
 270                         continue;
 271                 /* Flags will be lowest common denominator */
 272                 rc = lov_connect_obd(obd, i, tgt->ltd_activate, &lov->lov_ocd);
 273                 if (rc) {
 274                         CERROR("%s: lov connect tgt %d failed: %d\n",
 275                                obd->obd_name, i, rc);
 276                         continue;
 277                 }
 278                 /* connect to administrative disabled ost */
 279                 if (!lov->lov_tgts[i]->ltd_exp)
 280                         continue;
 281
 282                 rc = lov_notify(obd, lov->lov_tgts[i]->ltd_exp->exp_obd,
 283                                 OBD_NOTIFY_CONNECT, (void *)&i);
 284                 if (rc) {
 285                         CERROR("%s error sending notify %d\n",
 286                                obd->obd_name, rc);
 287                 }
 288         }
 289         lov_putref(obd);
 290
 291         RETURN(0);
 292 }
 293
 294 static int lov_disconnect_obd(struct obd_device *obd, struct lov_tgt_desc *tgt)
 295 {
 296         cfs_proc_dir_entry_t *lov_proc_dir;
 297         struct lov_obd *lov = &obd->u.lov;
 298         struct obd_device *osc_obd;
 299         int rc;
 300         ENTRY;
 301
 302         osc_obd = class_exp2obd(tgt->ltd_exp);
 303         CDEBUG(D_CONFIG, "%s: disconnecting target %s\n",
 304                obd->obd_name, osc_obd->obd_name);
 305
 306         if (tgt->ltd_active) {
 307                 tgt->ltd_active = 0;
 308                 lov->desc.ld_active_tgt_count--;
 309                 tgt->ltd_exp->exp_obd->obd_inactive = 1;
 310         }
 311
 312         lov_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
 313         if (lov_proc_dir) {
 314                 cfs_proc_dir_entry_t *osc_symlink;
 315
 316                 osc_symlink = lprocfs_srch(lov_proc_dir, osc_obd->obd_name);
 317                 if (osc_symlink) {
 318                         lprocfs_remove(&osc_symlink);
 319                 } else {
 320                         CERROR("/proc/fs/lustre/%s/%s/target_obds/%s missing.",
 321                                obd->obd_type->typ_name, obd->obd_name,
 322                                osc_obd->obd_name);
 323                 }
 324         }
 325
 326         if (osc_obd) {
 327                 /* Pass it on to our clients.
 328                  * XXX This should be an argument to disconnect,
 329                  * XXX not a back-door flag on the OBD.  Ah well.
 330                  */
 331                 osc_obd->obd_force = obd->obd_force;
 332                 osc_obd->obd_fail = obd->obd_fail;
 333                 osc_obd->obd_no_recov = obd->obd_no_recov;
 334         }
 335
 336         obd_register_observer(osc_obd, NULL);
 337
 338         rc = obd_disconnect(tgt->ltd_exp);
 339         if (rc) {
 340                 CERROR("Target %s disconnect error %d\n",
 341                        tgt->ltd_uuid.uuid, rc);
 342                 rc = 0;
 343         }
 344
 345         qos_del_tgt(obd, tgt);
 346
 347         tgt->ltd_exp = NULL;
 348         RETURN(0);
 349 }
 350
 351 static int lov_disconnect(struct obd_export *exp)
 352 {
 353         struct obd_device *obd = class_exp2obd(exp);
 354         struct lov_obd *lov = &obd->u.lov;
 355         int i, rc;
 356         ENTRY;
 357
 358         if (!lov->lov_tgts)
 359                 goto out;
 360
 361         /* Only disconnect the underlying layers on the final disconnect. */
 362         lov->lov_connects--;
 363         if (lov->lov_connects != 0) {
 364                 /* why should there be more than 1 connect? */
 365                 CERROR("disconnect #%d\n", lov->lov_connects);
 366                 goto out;
 367         }
 368
 369         /* Let's hold another reference so lov_del_obd doesn't spin through
 370            putref every time */
 371         lov_getref(obd);
 372
 373         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
 374                 if (lov->lov_tgts[i] && lov->lov_tgts[i]->ltd_exp) {
 375                         /* Disconnection is the last we know about an obd */
 376                         lov_del_target(obd, i, 0, lov->lov_tgts[i]->ltd_gen);
 377                 }
 378         }
 379         lov_putref(obd);
 380
 381 out:
 382         rc = class_disconnect(exp); /* bz 9811 */
 383         RETURN(rc);
 384 }
 385
 386 /* Error codes:
 387  *
 388  *  -EINVAL  : UUID can't be found in the LOV's target list
 389  *  -ENOTCONN: The UUID is found, but the target connection is bad (!)
 390  *  -EBADF   : The UUID is found, but the OBD is the wrong type (!)
 391  *  any >= 0 : is log target index
 392  */
 393 static int lov_set_osc_active(struct obd_device *obd, struct obd_uuid *uuid,
 394                               int activate)
 395 {
 396         struct lov_obd *lov = &obd->u.lov;
 397         struct lov_tgt_desc *tgt;
 398         int index;
 399         ENTRY;
 400
 401         CDEBUG(D_INFO, "Searching in lov %p for uuid %s (activate=%d)\n",
 402                lov, uuid->uuid, activate);
 403
 404         lov_getref(obd);
 405         for (index = 0; index < lov->desc.ld_tgt_count; index++) {
 406                 tgt = lov->lov_tgts[index];
 407                 if (!tgt || !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 (lov->lov_tgts[index]->ltd_active == activate) {
 421                 CDEBUG(D_INFO, "OSC %s already %sactive!\n", uuid->uuid,
 422                        activate ? "" : "in");
 423                 GOTO(out, index);
 424         }
 425
 426         CDEBUG(D_CONFIG, "Marking OSC %s %sactive\n", obd_uuid2str(uuid),
 427                activate ? "" : "in");
 428
 429         lov->lov_tgts[index]->ltd_active = activate;
 430
 431         if (activate) {
 432                 lov->desc.ld_active_tgt_count++;
 433                 lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 0;
 434         } else {
 435                 lov->desc.ld_active_tgt_count--;
 436                 lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 1;
 437         }
 438         /* remove any old qos penalty */
 439         lov->lov_tgts[index]->ltd_qos.ltq_penalty = 0;
 440
 441  out:
 442         lov_putref(obd);
 443         RETURN(index);
 444 }
 445
 446 static int lov_notify(struct obd_device *obd, struct obd_device *watched,
 447                       enum obd_notify_event ev, void *data)
 448 {
 449         int rc = 0;
 450         ENTRY;
 451
 452         if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE) {
 453                 struct obd_uuid *uuid;
 454
 455                 LASSERT(watched);
 456
 457                 if (strcmp(watched->obd_type->typ_name, LUSTRE_OSC_NAME)) {
 458                         CERROR("unexpected notification of %s %s!\n",
 459                                watched->obd_type->typ_name,
 460                                watched->obd_name);
 461                         RETURN(-EINVAL);
 462                 }
 463                 uuid = &watched->u.cli.cl_target_uuid;
 464
 465                 /* Set OSC as active before notifying the observer, so the
 466                  * observer can use the OSC normally.
 467                  */
 468                 rc = lov_set_osc_active(obd, uuid, ev == OBD_NOTIFY_ACTIVE);
 469                 if (rc < 0) {
 470                         CERROR("%sactivation of %s failed: %d\n",
 471                                (ev == OBD_NOTIFY_ACTIVE) ? "" : "de",
 472                                obd_uuid2str(uuid), rc);
 473                         RETURN(rc);
 474                 }
 475                 /* active event should be pass lov target index as data */
 476                 data = &rc;
 477         }
 478
 479         /* Pass the notification up the chain. */
 480         if (watched) {
 481                 rc = obd_notify_observer(obd, watched, ev, data);
 482         } else {
 483                 /* NULL watched means all osc's in the lov (only for syncs) */
 484                 /* sync event should be send lov idx as data */
 485                 struct lov_obd *lov = &obd->u.lov;
 486                 struct obd_device *tgt_obd;
 487                 int i;
 488                 lov_getref(obd);
 489                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
 490                         /* don't send sync event if target not
 491                          * connected/activated */
 492                         if (!lov->lov_tgts[i] ||
 493                             !lov->lov_tgts[i]->ltd_active)
 494                                  continue;
 495
 496                         if ((ev == OBD_NOTIFY_SYNC) ||
 497                             (ev == OBD_NOTIFY_SYNC_NONBLOCK))
 498                                 data = &i;
 499
 500                         tgt_obd = class_exp2obd(lov->lov_tgts[i]->ltd_exp);
 501
 502                         rc = obd_notify_observer(obd, tgt_obd, ev, data);
 503                         if (rc) {
 504                                 CERROR("%s: notify %s of %s failed %d\n",
 505                                        obd->obd_name,
 506                                        obd->obd_observer->obd_name,
 507                                        tgt_obd->obd_name, rc);
 508                                 break;
 509                         }
 510                 }
 511                 lov_putref(obd);
 512         }
 513
 514         RETURN(rc);
 515 }
 516
 517 int lov_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
 518                    __u32 index, int gen, int active)
 519 {
 520         struct lov_obd *lov = &obd->u.lov;
 521         struct lov_tgt_desc *tgt;
 522         int rc;
 523         ENTRY;
 524
 525         CDEBUG(D_CONFIG, "uuid:%s idx:%d gen:%d active:%d\n",
 526                uuidp->uuid, index, gen, active);
 527
 528         if (gen <= 0) {
 529                 CERROR("request to add OBD %s with invalid generation: %d\n",
 530                        uuidp->uuid, gen);
 531                 RETURN(-EINVAL);
 532         }
 533
 534         mutex_down(&lov->lov_lock);
 535
 536         if ((index < lov->lov_tgt_size) && (lov->lov_tgts[index] != NULL)) {
 537                 tgt = lov->lov_tgts[index];
 538                 CERROR("UUID %s already assigned at LOV target index %d\n",
 539                        obd_uuid2str(&tgt->ltd_uuid), index);
 540                 mutex_up(&lov->lov_lock);
 541                 RETURN(-EEXIST);
 542         }
 543
 544         if (index >= lov->lov_tgt_size) {
 545                 /* We need to reallocate the lov target array. */
 546                 struct lov_tgt_desc **newtgts, **old = NULL;
 547                 __u32 newsize, oldsize = 0;
 548
 549                 newsize = max(lov->lov_tgt_size, (__u32)2);
 550                 while (newsize < index + 1)
 551                         newsize = newsize << 1;
 552                 OBD_ALLOC(newtgts, sizeof(*newtgts) * newsize);
 553                 if (newtgts == NULL) {
 554                         mutex_up(&lov->lov_lock);
 555                         RETURN(-ENOMEM);
 556                 }
 557
 558                 if (lov->lov_tgt_size) {
 559                         memcpy(newtgts, lov->lov_tgts, sizeof(*newtgts) *
 560                                lov->lov_tgt_size);
 561                         old = lov->lov_tgts;
 562                         oldsize = lov->lov_tgt_size;
 563                 }
 564
 565                 lov->lov_tgts = newtgts;
 566                 lov->lov_tgt_size = newsize;
 567 #ifdef __KERNEL__
 568                 smp_rmb();
 569 #endif
 570                 if (old)
 571                         OBD_FREE(old, sizeof(*old) * oldsize);
 572
 573                 CDEBUG(D_CONFIG, "tgts: %p size: %d\n",
 574                        lov->lov_tgts, lov->lov_tgt_size);
 575         }
 576
 577         OBD_ALLOC_PTR(tgt);
 578         if (!tgt) {
 579                 mutex_up(&lov->lov_lock);
 580                 RETURN(-ENOMEM);
 581         }
 582
 583         memset(tgt, 0, sizeof(*tgt));
 584         tgt->ltd_uuid = *uuidp;
 585         /* XXX - add a sanity check on the generation number. */
 586         tgt->ltd_gen = gen;
 587         tgt->ltd_index = index;
 588         tgt->ltd_activate = active;
 589         lov->lov_tgts[index] = tgt;
 590         if (index >= lov->desc.ld_tgt_count)
 591                 lov->desc.ld_tgt_count = index + 1;
 592
 593         rc = lov_ost_pool_add(&lov->lov_packed, index, lov->lov_tgt_size);
 594         if (rc)
 595                 RETURN(rc);
 596
 597         mutex_up(&lov->lov_lock);
 598
 599         CDEBUG(D_CONFIG, "idx=%d ltd_gen=%d ld_tgt_count=%d\n",
 600                 index, tgt->ltd_gen, lov->desc.ld_tgt_count);
 601
 602         if (lov->lov_connects == 0) {
 603                 /* lov_connect hasn't been called yet. We'll do the
 604                    lov_connect_obd on this target when that fn first runs,
 605                    because we don't know the connect flags yet. */
 606                 RETURN(0);
 607         }
 608
 609         lov_getref(obd);
 610
 611         rc = lov_connect_obd(obd, index, active, &lov->lov_ocd);
 612         if (rc)
 613                 GOTO(out, rc);
 614
 615         /* connect to administrative disabled ost */
 616         if (!tgt->ltd_exp)
 617                 GOTO(out, rc = 0);
 618
 619         rc = lov_notify(obd, tgt->ltd_exp->exp_obd,
 620                         active ? OBD_NOTIFY_CONNECT : OBD_NOTIFY_INACTIVE,
 621                         (void *)&index);
 622
 623 out:
 624         if (rc) {
 625                 CERROR("add failed (%d), deleting %s\n", rc,
 626                        obd_uuid2str(&tgt->ltd_uuid));
 627                 lov_del_target(obd, index, 0, 0);
 628         }
 629         lov_putref(obd);
 630         RETURN(rc);
 631 }
 632
 633 /* Schedule a target for deletion */
 634 int lov_del_target(struct obd_device *obd, __u32 index,
 635                    struct obd_uuid *uuidp, int gen)
 636 {
 637         struct lov_obd *lov = &obd->u.lov;
 638         int count = lov->desc.ld_tgt_count;
 639         int rc = 0;
 640         ENTRY;
 641
 642         if (index >= count) {
 643                 CERROR("LOV target index %d >= number of LOV OBDs %d.\n",
 644                        index, count);
 645                 RETURN(-EINVAL);
 646         }
 647
 648         lov_getref(obd);
 649
 650         if (!lov->lov_tgts[index]) {
 651                 CERROR("LOV target at index %d is not setup.\n", index);
 652                 GOTO(out, rc = -EINVAL);
 653         }
 654
 655         if (uuidp && !obd_uuid_equals(uuidp, &lov->lov_tgts[index]->ltd_uuid)) {
 656                 CERROR("LOV target UUID %s at index %d doesn't match %s.\n",
 657                        lov_uuid2str(lov, index), index,
 658                        obd_uuid2str(uuidp));
 659                 GOTO(out, rc = -EINVAL);
 660         }
 661
 662         CDEBUG(D_CONFIG, "uuid: %s idx: %d gen: %d exp: %p active: %d\n",
 663                lov_uuid2str(lov, index), index,
 664                lov->lov_tgts[index]->ltd_gen, lov->lov_tgts[index]->ltd_exp,
 665                lov->lov_tgts[index]->ltd_active);
 666
 667         lov->lov_tgts[index]->ltd_reap = 1;
 668         lov->lov_death_row++;
 669         /* we really delete it from lov_putref */
 670 out:
 671         lov_putref(obd);
 672
 673         RETURN(rc);
 674 }
 675
 676 static void __lov_del_obd(struct obd_device *obd, struct lov_tgt_desc *tgt)
 677 {
 678         struct obd_device *osc_obd;
 679
 680         LASSERT(tgt);
 681         LASSERT(tgt->ltd_reap);
 682
 683         osc_obd = class_exp2obd(tgt->ltd_exp);
 684
 685         CDEBUG(D_CONFIG, "Removing tgt %s : %s\n",
 686                tgt->ltd_uuid.uuid,
 687                osc_obd ? osc_obd->obd_name : "<no obd>");
 688
 689         if (tgt->ltd_exp)
 690                 lov_disconnect_obd(obd, tgt);
 691
 692         OBD_FREE_PTR(tgt);
 693
 694         /* Manual cleanup - no cleanup logs to clean up the osc's.  We must
 695            do it ourselves. And we can't do it from lov_cleanup,
 696            because we just lost our only reference to it. */
 697         if (osc_obd)
 698                 class_manual_cleanup(osc_obd);
 699 }
 700
 701 void lov_fix_desc_stripe_size(__u64 *val)
 702 {
 703         if (*val < PTLRPC_MAX_BRW_SIZE) {
 704                 LCONSOLE_WARN("Increasing default stripe size to min %u\n",
 705                               PTLRPC_MAX_BRW_SIZE);
 706                 *val = PTLRPC_MAX_BRW_SIZE;
 707         } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) {
 708                 *val &= ~(LOV_MIN_STRIPE_SIZE - 1);
 709                 LCONSOLE_WARN("Changing default stripe size to "LPU64" (a "
 710                               "multiple of %u)\n",
 711                               *val, LOV_MIN_STRIPE_SIZE);
 712         }
 713 }
 714
 715 void lov_fix_desc_stripe_count(__u32 *val)
 716 {
 717         if (*val == 0)
 718                 *val = 1;
 719 }
 720
 721 void lov_fix_desc_pattern(__u32 *val)
 722 {
 723         /* from lov_setstripe */
 724         if ((*val != 0) && (*val != LOV_PATTERN_RAID0)) {
 725                 LCONSOLE_WARN("Unknown stripe pattern: %#x\n", *val);
 726                 *val = 0;
 727         }
 728 }
 729
 730 void lov_fix_desc_qos_maxage(__u32 *val)
 731 {
 732         /* fix qos_maxage */
 733         if (*val == 0)
 734                 *val = QOS_DEFAULT_MAXAGE;
 735 }
 736
 737 void lov_fix_desc(struct lov_desc *desc)
 738 {
 739         lov_fix_desc_stripe_size(&desc->ld_default_stripe_size);
 740         lov_fix_desc_stripe_count(&desc->ld_default_stripe_count);
 741         lov_fix_desc_pattern(&desc->ld_pattern);
 742         lov_fix_desc_qos_maxage(&desc->ld_qos_maxage);
 743 }
 744
 745 int lov_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
 746 {
 747         struct lprocfs_static_vars lvars = { 0 };
 748         struct lov_desc *desc;
 749         struct lov_obd *lov = &obd->u.lov;
 750         int count;
 751         int rc;
 752         ENTRY;
 753
 754         if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
 755                 CERROR("LOV setup requires a descriptor\n");
 756                 RETURN(-EINVAL);
 757         }
 758
 759         desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
 760
 761         if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
 762                 CERROR("descriptor size wrong: %d > %d\n",
 763                        (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
 764                 RETURN(-EINVAL);
 765         }
 766
 767         if (desc->ld_magic != LOV_DESC_MAGIC) {
 768                 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
 769                             CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
 770                                    obd->obd_name, desc);
 771                             lustre_swab_lov_desc(desc);
 772                 } else {
 773                         CERROR("%s: Bad lov desc magic: %#x\n",
 774                                obd->obd_name, desc->ld_magic);
 775                         RETURN(-EINVAL);
 776                 }
 777         }
 778
 779         lov_fix_desc(desc);
 780
 781         /* Because of 64-bit divide/mod operations only work with a 32-bit
 782          * divisor in a 32-bit kernel, we cannot support a stripe width
 783          * of 4GB or larger on 32-bit CPUs. */
 784         count = desc->ld_default_stripe_count;
 785         if ((count > 0 ? count : desc->ld_tgt_count) *
 786             desc->ld_default_stripe_size > 0xffffffff) {
 787                 CERROR("LOV: stripe width "LPU64"x%u > 4294967295 bytes\n",
 788                        desc->ld_default_stripe_size, count);
 789                 RETURN(-EINVAL);
 790         }
 791
 792         desc->ld_active_tgt_count = 0;
 793         lov->desc = *desc;
 794         lov->lov_tgt_size = 0;
 795
 796         sema_init(&lov->lov_lock, 1);
 797         atomic_set(&lov->lov_refcount, 0);
 798         CFS_INIT_LIST_HEAD(&lov->lov_qos.lq_oss_list);
 799         init_rwsem(&lov->lov_qos.lq_rw_sem);
 800         lov->lov_sp_me = LUSTRE_SP_CLI;
 801         lov->lov_qos.lq_dirty = 1;
 802         lov->lov_qos.lq_rr.lqr_dirty = 1;
 803         lov->lov_qos.lq_reset = 1;
 804         /* Default priority is toward free space balance */
 805         lov->lov_qos.lq_prio_free = 232;
 806         /* Default threshold for rr (roughly 17%) */
 807         lov->lov_qos.lq_threshold_rr = 43;
 808         /* Init statfs fields */
 809         OBD_ALLOC_PTR(lov->lov_qos.lq_statfs_data);
 810         if (NULL == lov->lov_qos.lq_statfs_data)
 811                 RETURN(-ENOMEM);
 812         cfs_waitq_init(&lov->lov_qos.lq_statfs_waitq);
 813
 814         lov->lov_pools_hash_body = lustre_hash_init("POOLS", 7, 7,
 815                                                     &pool_hash_operations, 0);
 816         CFS_INIT_LIST_HEAD(&lov->lov_pool_list);
 817         lov->lov_pool_count = 0;
 818         rc = lov_ost_pool_init(&lov->lov_packed, 0);
 819         if (rc)
 820                 RETURN(rc);
 821         rc = lov_ost_pool_init(&lov->lov_qos.lq_rr.lqr_pool, 0);
 822         if (rc) {
 823                 lov_ost_pool_free(&lov->lov_packed);
 824                 RETURN(rc);
 825         }
 826
 827         lprocfs_lov_init_vars(&lvars);
 828         lprocfs_obd_setup(obd, lvars.obd_vars);
 829 #ifdef LPROCFS
 830         {
 831                 int rc;
 832
 833                 rc = lprocfs_seq_create(obd->obd_proc_entry, "target_obd",
 834                                         0444, &lov_proc_target_fops, obd);
 835                 if (rc)
 836                         CWARN("Error adding the target_obd file\n");
 837         }
 838 #endif
 839         lov->lov_pool_proc_entry = lprocfs_register("pools",
 840                                                     obd->obd_proc_entry,
 841                                                     NULL, NULL);
 842
 843         RETURN(0);
 844 }
 845
 846 static int lov_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
 847 {
 848         int rc = 0;
 849         struct lov_obd *lov = &obd->u.lov;
 850
 851         ENTRY;
 852
 853         switch (stage) {
 854         case OBD_CLEANUP_EARLY: {
 855                 int i;
 856                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
 857                         if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_active)
 858                                 continue;
 859                         obd_precleanup(class_exp2obd(lov->lov_tgts[i]->ltd_exp),
 860                                        OBD_CLEANUP_EARLY);
 861                 }
 862                 break;
 863         }
 864         case OBD_CLEANUP_EXPORTS:
 865                 rc = obd_llog_finish(obd, 0);
 866                 if (rc != 0)
 867                         CERROR("failed to cleanup llogging subsystems\n");
 868                 break;
 869         }
 870         RETURN(rc);
 871 }
 872
 873 static int lov_cleanup(struct obd_device *obd)
 874 {
 875         struct lov_obd *lov = &obd->u.lov;
 876         struct list_head *pos, *tmp;
 877         struct pool_desc *pool;
 878
 879         list_for_each_safe(pos, tmp, &lov->lov_pool_list) {
 880                 pool = list_entry(pos, struct pool_desc, pool_list);
 881                 /* free pool structs */
 882                 CDEBUG(D_INFO, "delete pool %p\n", pool);
 883                 lov_pool_del(obd, pool->pool_name);
 884         }
 885         lustre_hash_exit(lov->lov_pools_hash_body);
 886         lov_ost_pool_free(&(lov->lov_qos.lq_rr.lqr_pool));
 887         lov_ost_pool_free(&lov->lov_packed);
 888
 889         if (lov->lov_tgts) {
 890                 int i;
 891                 lov_getref(obd);
 892                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
 893                         if (!lov->lov_tgts[i])
 894                                 continue;
 895
 896                         /* Inactive targets may never have connected */
 897                         if (lov->lov_tgts[i]->ltd_active ||
 898                             atomic_read(&lov->lov_refcount))
 899                             /* We should never get here - these
 900                                should have been removed in the
 901                              disconnect. */
 902                                 CERROR("lov tgt %d not cleaned!"
 903                                        " deathrow=%d, lovrc=%d\n",
 904                                        i, lov->lov_death_row,
 905                                        atomic_read(&lov->lov_refcount));
 906                         lov_del_target(obd, i, 0, 0);
 907                 }
 908                 lov_putref(obd);
 909                 OBD_FREE(lov->lov_tgts, sizeof(*lov->lov_tgts) *
 910                          lov->lov_tgt_size);
 911                 lov->lov_tgt_size = 0;
 912         }
 913
 914         /* clear pools parent proc entry only after all pools is killed */
 915         lprocfs_obd_cleanup(obd);
 916
 917         OBD_FREE_PTR(lov->lov_qos.lq_statfs_data);
 918         RETURN(0);
 919 }
 920
 921 int lov_process_config_base(struct obd_device *obd, struct lustre_cfg *lcfg,
 922                             __u32 *indexp, int *genp)
 923 {
 924         struct obd_uuid obd_uuid;
 925         int cmd;
 926         int rc = 0;
 927         ENTRY;
 928
 929         switch(cmd = lcfg->lcfg_command) {
 930         case LCFG_LOV_ADD_OBD:
 931         case LCFG_LOV_ADD_INA:
 932         case LCFG_LOV_DEL_OBD: {
 933                 __u32 index;
 934                 int gen;
 935                 /* lov_modify_tgts add  0:lov_mdsA  1:ost1_UUID  2:0  3:1 */
 936                 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
 937                         GOTO(out, rc = -EINVAL);
 938
 939                 obd_str2uuid(&obd_uuid,  lustre_cfg_buf(lcfg, 1));
 940
 941                 if (sscanf(lustre_cfg_buf(lcfg, 2), "%d", indexp) != 1)
 942                         GOTO(out, rc = -EINVAL);
 943                 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", genp) != 1)
 944                         GOTO(out, rc = -EINVAL);
 945                 index = *indexp;
 946                 gen = *genp;
 947                 if (cmd == LCFG_LOV_ADD_OBD)
 948                         rc = lov_add_target(obd, &obd_uuid, index, gen, 1);
 949                 else if (cmd == LCFG_LOV_ADD_INA)
 950                         rc = lov_add_target(obd, &obd_uuid, index, gen, 0);
 951                 else
 952                         rc = lov_del_target(obd, index, &obd_uuid, gen);
 953                 GOTO(out, rc);
 954         }
 955         case LCFG_PARAM: {
 956                 struct lprocfs_static_vars lvars = { 0 };
 957                 struct lov_desc *desc = &(obd->u.lov.desc);
 958
 959                 if (!desc)
 960                         GOTO(out, rc = -EINVAL);
 961
 962                 lprocfs_lov_init_vars(&lvars);
 963
 964                 rc = class_process_proc_param(PARAM_LOV, lvars.obd_vars,
 965                                               lcfg, obd);
 966                 if (rc > 0)
 967                         rc = 0;
 968                 GOTO(out, rc);
 969         }
 970         case LCFG_POOL_NEW:
 971         case LCFG_POOL_ADD:
 972         case LCFG_POOL_DEL:
 973         case LCFG_POOL_REM:
 974                 GOTO(out, rc);
 975
 976         default: {
 977                 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
 978                 GOTO(out, rc = -EINVAL);
 979
 980         }
 981         }
 982 out:
 983         RETURN(rc);
 984 }
 985
 986 #ifndef log2
 987 #define log2(n) ffz(~(n))
 988 #endif
 989
 990 static int lov_clear_orphans(struct obd_export *export, struct obdo *src_oa,
 991                              struct lov_stripe_md **ea,
 992                              struct obd_trans_info *oti)
 993 {
 994         struct lov_obd *lov;
 995         struct obdo *tmp_oa;
 996         struct obd_uuid *ost_uuid = NULL;
 997         int rc = 0, i;
 998         ENTRY;
 999
1000         LASSERT(src_oa->o_valid & OBD_MD_FLFLAGS &&
1001                 src_oa->o_flags == OBD_FL_DELORPHAN);
1002
1003         lov = &export->exp_obd->u.lov;
1004
1005         OBDO_ALLOC(tmp_oa);
1006         if (tmp_oa == NULL)
1007                 RETURN(-ENOMEM);
1008
1009         if (oti->oti_ost_uuid) {
1010                 ost_uuid = oti->oti_ost_uuid;
1011                 CDEBUG(D_HA, "clearing orphans only for %s\n",
1012                        ost_uuid->uuid);
1013         }
1014
1015         lov_getref(export->exp_obd);
1016         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
1017                 struct lov_stripe_md obj_md;
1018                 struct lov_stripe_md *obj_mdp = &obj_md;
1019                 struct lov_tgt_desc *tgt;
1020                 int err;
1021
1022                 tgt = lov->lov_tgts[i];
1023                 if (!tgt)
1024                         continue;
1025
1026                 /* if called for a specific target, we don't
1027                    care if it is not active. */
1028                 if (!lov->lov_tgts[i]->ltd_active && ost_uuid == NULL) {
1029                         CDEBUG(D_HA, "lov idx %d inactive\n", i);
1030                         continue;
1031                 }
1032
1033                 if (ost_uuid && !obd_uuid_equals(ost_uuid, &tgt->ltd_uuid))
1034                         continue;
1035
1036                 CDEBUG(D_CONFIG,"Clear orphans for %d:%s\n", i,
1037                        obd_uuid2str(ost_uuid));
1038
1039                 memcpy(tmp_oa, src_oa, sizeof(*tmp_oa));
1040
1041                 LASSERT(lov->lov_tgts[i]->ltd_exp);
1042                 /* XXX: LOV STACKING: use real "obj_mdp" sub-data */
1043                 err = obd_create(lov->lov_tgts[i]->ltd_exp,
1044                                  tmp_oa, &obj_mdp, oti);
1045                 if (err) {
1046                         /* This export will be disabled until it is recovered,
1047                            and then orphan recovery will be completed. */
1048                         CERROR("error in orphan recovery on OST idx %d/%d: "
1049                                "rc = %d\n", i, lov->desc.ld_tgt_count, err);
1050                         rc = err;
1051                 }
1052
1053                 if (ost_uuid)
1054                         break;
1055         }
1056         lov_putref(export->exp_obd);
1057
1058         OBDO_FREE(tmp_oa);
1059         RETURN(rc);
1060 }
1061
1062 static int lov_recreate(struct obd_export *exp, struct obdo *src_oa,
1063                         struct lov_stripe_md **ea, struct obd_trans_info *oti)
1064 {
1065         struct lov_stripe_md *obj_mdp, *lsm;
1066         struct lov_obd *lov = &exp->exp_obd->u.lov;
1067         unsigned ost_idx;
1068         int rc, i;
1069         ENTRY;
1070
1071         LASSERT(src_oa->o_valid & OBD_MD_FLFLAGS &&
1072                 src_oa->o_flags & OBD_FL_RECREATE_OBJS);
1073
1074         OBD_ALLOC(obj_mdp, sizeof(*obj_mdp));
1075         if (obj_mdp == NULL)
1076                 RETURN(-ENOMEM);
1077
1078         ost_idx = src_oa->o_nlink;
1079         lsm = *ea;
1080         if (lsm == NULL)
1081                 GOTO(out, rc = -EINVAL);
1082         if (ost_idx >= lov->desc.ld_tgt_count ||
1083             !lov->lov_tgts[ost_idx])
1084                 GOTO(out, rc = -EINVAL);
1085
1086         for (i = 0; i < lsm->lsm_stripe_count; i++) {
1087                 if (lsm->lsm_oinfo[i]->loi_ost_idx == ost_idx) {
1088                         if (lsm->lsm_oinfo[i]->loi_id != src_oa->o_id)
1089                                 GOTO(out, rc = -EINVAL);
1090                         break;
1091                 }
1092         }
1093         if (i == lsm->lsm_stripe_count)
1094                 GOTO(out, rc = -EINVAL);
1095
1096         rc = obd_create(lov->lov_tgts[ost_idx]->ltd_exp, src_oa, &obj_mdp, oti);
1097 out:
1098         OBD_FREE(obj_mdp, sizeof(*obj_mdp));
1099         RETURN(rc);
1100 }
1101
1102 /* the LOV expects oa->o_id to be set to the LOV object id */
1103 static int lov_create(struct obd_export *exp, struct obdo *src_oa,
1104                       struct lov_stripe_md **ea, struct obd_trans_info *oti)
1105 {
1106         struct lov_obd *lov;
1107         struct obd_info oinfo;
1108         struct lov_request_set *set = NULL;
1109         struct lov_request *req;
1110         int rc = 0;
1111         ENTRY;
1112
1113         LASSERT(ea != NULL);
1114         if (exp == NULL)
1115                 RETURN(-EINVAL);
1116
1117         if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
1118             src_oa->o_flags == OBD_FL_DELORPHAN) {
1119                 rc = lov_clear_orphans(exp, src_oa, ea, oti);
1120                 RETURN(rc);
1121         }
1122
1123         lov = &exp->exp_obd->u.lov;
1124         if (!lov->desc.ld_active_tgt_count)
1125                 RETURN(-EIO);
1126
1127         lov_getref(exp->exp_obd);
1128         /* Recreate a specific object id at the given OST index */
1129         if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
1130             (src_oa->o_flags & OBD_FL_RECREATE_OBJS)) {
1131                  rc = lov_recreate(exp, src_oa, ea, oti);
1132                  GOTO(out, rc);
1133         }
1134
1135         /* issue statfs rpcs if the osfs data is older than qos_maxage - 1s,
1136          * later in alloc_qos(), we will wait for those rpcs to complete if
1137          * the osfs age is older than 2 * qos_maxage */
1138         qos_statfs_update(exp->exp_obd,
1139                           cfs_time_shift_64(-lov->desc.ld_qos_maxage) + HZ, 0);
1140
1141         rc = lov_prep_create_set(exp, &oinfo, ea, src_oa, oti, &set);
1142         if (rc)
1143                 GOTO(out, rc);
1144
1145         list_for_each_entry(req, &set->set_list, rq_link) {
1146                 /* XXX: LOV STACKING: use real "obj_mdp" sub-data */
1147                 rc = obd_create(lov->lov_tgts[req->rq_idx]->ltd_exp,
1148                                 req->rq_oi.oi_oa, &req->rq_oi.oi_md, oti);
1149                 lov_update_create_set(set, req, rc);
1150         }
1151         rc = lov_fini_create_set(set, ea);
1152 out:
1153         lov_putref(exp->exp_obd);
1154         RETURN(rc);
1155 }
1156
1157 #define ASSERT_LSM_MAGIC(lsmp)                                                  \
1158 do {                                                                            \
1159         LASSERT((lsmp) != NULL);                                                \
1160         LASSERTF(((lsmp)->lsm_magic == LOV_MAGIC_V1 ||                          \
1161                  (lsmp)->lsm_magic == LOV_MAGIC_V3 ||                           \
1162                  (lsmp)->lsm_magic == LOV_MAGIC_JOIN), "%p->lsm_magic=%x\n",    \
1163                  (lsmp), (lsmp)->lsm_magic);                                    \
1164 } while (0)
1165
1166 static int lov_destroy(struct obd_export *exp, struct obdo *oa,
1167                        struct lov_stripe_md *lsm, struct obd_trans_info *oti,
1168                        struct obd_export *md_exp, void *capa)
1169 {
1170         struct lov_request_set *set;
1171         struct obd_info oinfo;
1172         struct lov_request *req;
1173         struct list_head *pos;
1174         struct lov_obd *lov;
1175         int rc = 0, err = 0;
1176         ENTRY;
1177
1178         ASSERT_LSM_MAGIC(lsm);
1179
1180         if (!exp || !exp->exp_obd)
1181                 RETURN(-ENODEV);
1182
1183         if (oa->o_valid & OBD_MD_FLCOOKIE) {
1184                 LASSERT(oti);
1185                 LASSERT(oti->oti_logcookies);
1186         }
1187
1188         lov = &exp->exp_obd->u.lov;
1189         lov_getref(exp->exp_obd);
1190         rc = lov_prep_destroy_set(exp, &oinfo, oa, lsm, oti, &set);
1191         if (rc)
1192                 GOTO(out, rc);
1193
1194         list_for_each (pos, &set->set_list) {
1195                 req = list_entry(pos, struct lov_request, rq_link);
1196
1197                 if (oa->o_valid & OBD_MD_FLCOOKIE)
1198                         oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1199
1200                 err = obd_destroy(lov->lov_tgts[req->rq_idx]->ltd_exp,
1201                                   req->rq_oi.oi_oa, NULL, oti, NULL, capa);
1202                 err = lov_update_common_set(set, req, err);
1203                 if (err) {
1204                         CERROR("error: destroying objid "LPX64" subobj "
1205                                LPX64" on OST idx %d: rc = %d\n",
1206                                oa->o_id, req->rq_oi.oi_oa->o_id,
1207                                req->rq_idx, err);
1208                         if (!rc)
1209                                 rc = err;
1210                 }
1211         }
1212
1213         if (rc == 0) {
1214                 LASSERT(lsm_op_find(lsm->lsm_magic) != NULL);
1215                 rc = lsm_op_find(lsm->lsm_magic)->lsm_destroy(lsm, oa, md_exp);
1216         }
1217         err = lov_fini_destroy_set(set);
1218 out:
1219         lov_putref(exp->exp_obd);
1220         RETURN(rc ? rc : err);
1221 }
1222
1223 static int lov_getattr(struct obd_export *exp, struct obd_info *oinfo)
1224 {
1225         struct lov_request_set *set;
1226         struct lov_request *req;
1227         struct list_head *pos;
1228         struct lov_obd *lov;
1229         int err = 0, rc = 0;
1230         ENTRY;
1231
1232         LASSERT(oinfo);
1233         ASSERT_LSM_MAGIC(oinfo->oi_md);
1234
1235         if (!exp || !exp->exp_obd)
1236                 RETURN(-ENODEV);
1237
1238         lov = &exp->exp_obd->u.lov;
1239
1240         rc = lov_prep_getattr_set(exp, oinfo, &set);
1241         if (rc)
1242                 RETURN(rc);
1243
1244         list_for_each (pos, &set->set_list) {
1245                 req = list_entry(pos, struct lov_request, rq_link);
1246
1247                 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
1248                        "%u\n", oinfo->oi_oa->o_id, req->rq_stripe,
1249                        req->rq_oi.oi_oa->o_id, req->rq_idx);
1250
1251                 rc = obd_getattr(lov->lov_tgts[req->rq_idx]->ltd_exp,
1252                                  &req->rq_oi);
1253                 err = lov_update_common_set(set, req, rc);
1254                 if (err) {
1255                         CERROR("error: getattr objid "LPX64" subobj "
1256                                LPX64" on OST idx %d: rc = %d\n",
1257                                oinfo->oi_oa->o_id, req->rq_oi.oi_oa->o_id,
1258                                req->rq_idx, err);
1259                         break;
1260                 }
1261         }
1262
1263         rc = lov_fini_getattr_set(set);
1264         if (err)
1265                 rc = err;
1266         RETURN(rc);
1267 }
1268
1269 static int lov_getattr_interpret(struct ptlrpc_request_set *rqset,
1270                                  void *data, int rc)
1271 {
1272         struct lov_request_set *lovset = (struct lov_request_set *)data;
1273         int err;
1274         ENTRY;
1275
1276         /* don't do attribute merge if this aysnc op failed */
1277         if (rc)
1278                 lovset->set_completes = 0;
1279         err = lov_fini_getattr_set(lovset);
1280         RETURN(rc ? rc : err);
1281 }
1282
1283 static int lov_getattr_async(struct obd_export *exp, struct obd_info *oinfo,
1284                               struct ptlrpc_request_set *rqset)
1285 {
1286         struct lov_request_set *lovset;
1287         struct lov_obd *lov;
1288         struct list_head *pos;
1289         struct lov_request *req;
1290         int rc = 0, err;
1291         ENTRY;
1292
1293         LASSERT(oinfo);
1294         ASSERT_LSM_MAGIC(oinfo->oi_md);
1295
1296         if (!exp || !exp->exp_obd)
1297                 RETURN(-ENODEV);
1298
1299         lov = &exp->exp_obd->u.lov;
1300
1301         rc = lov_prep_getattr_set(exp, oinfo, &lovset);
1302         if (rc)
1303                 RETURN(rc);
1304
1305         CDEBUG(D_INFO, "objid "LPX64": %ux%u byte stripes\n",
1306                oinfo->oi_md->lsm_object_id, oinfo->oi_md->lsm_stripe_count,
1307                oinfo->oi_md->lsm_stripe_size);
1308
1309         list_for_each (pos, &lovset->set_list) {
1310                 req = list_entry(pos, struct lov_request, rq_link);
1311
1312                 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
1313                        "%u\n", oinfo->oi_oa->o_id, req->rq_stripe,
1314                        req->rq_oi.oi_oa->o_id, req->rq_idx);
1315                 rc = obd_getattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1316                                        &req->rq_oi, rqset);
1317                 if (rc) {
1318                         CERROR("error: getattr objid "LPX64" subobj "
1319                                LPX64" on OST idx %d: rc = %d\n",
1320                                oinfo->oi_oa->o_id, req->rq_oi.oi_oa->o_id,
1321                                req->rq_idx, rc);
1322                         GOTO(out, rc);
1323                 }
1324         }
1325
1326         if (!list_empty(&rqset->set_requests)) {
1327                 LASSERT(rc == 0);
1328                 LASSERT (rqset->set_interpret == NULL);
1329                 rqset->set_interpret = lov_getattr_interpret;
1330                 rqset->set_arg = (void *)lovset;
1331                 RETURN(rc);
1332         }
1333 out:
1334         if (rc)
1335                 lovset->set_completes = 0;
1336         err = lov_fini_getattr_set(lovset);
1337         RETURN(rc ? rc : err);
1338 }
1339
1340 static int lov_setattr(struct obd_export *exp, struct obd_info *oinfo,
1341                        struct obd_trans_info *oti)
1342 {
1343         struct lov_request_set *set;
1344         struct lov_obd *lov;
1345         struct list_head *pos;
1346         struct lov_request *req;
1347         int err = 0, rc = 0;
1348         ENTRY;
1349
1350         LASSERT(oinfo);
1351         ASSERT_LSM_MAGIC(oinfo->oi_md);
1352
1353         if (!exp || !exp->exp_obd)
1354                 RETURN(-ENODEV);
1355
1356         /* for now, we only expect the following updates here */
1357         LASSERT(!(oinfo->oi_oa->o_valid & ~(OBD_MD_FLID | OBD_MD_FLTYPE |
1358                                             OBD_MD_FLMODE | OBD_MD_FLATIME |
1359                                             OBD_MD_FLMTIME | OBD_MD_FLCTIME |
1360                                             OBD_MD_FLFLAGS | OBD_MD_FLSIZE |
1361                                             OBD_MD_FLGROUP | OBD_MD_FLUID |
1362                                             OBD_MD_FLGID | OBD_MD_FLFID |
1363                                             OBD_MD_FLGENER)));
1364         lov = &exp->exp_obd->u.lov;
1365         rc = lov_prep_setattr_set(exp, oinfo, oti, &set);
1366         if (rc)
1367                 RETURN(rc);
1368
1369         list_for_each (pos, &set->set_list) {
1370                 req = list_entry(pos, struct lov_request, rq_link);
1371
1372                 rc = obd_setattr(lov->lov_tgts[req->rq_idx]->ltd_exp,
1373                                  &req->rq_oi, NULL);
1374                 err = lov_update_setattr_set(set, req, rc);
1375                 if (err) {
1376                         CERROR("error: setattr objid "LPX64" subobj "
1377                                LPX64" on OST idx %d: rc = %d\n",
1378                                set->set_oi->oi_oa->o_id,
1379                                req->rq_oi.oi_oa->o_id, req->rq_idx, err);
1380                         if (!rc)
1381                                 rc = err;
1382                 }
1383         }
1384         err = lov_fini_setattr_set(set);
1385         if (!rc)
1386                 rc = err;
1387         RETURN(rc);
1388 }
1389
1390 static int lov_setattr_interpret(struct ptlrpc_request_set *rqset,
1391                                  void *data, int rc)
1392 {
1393         struct lov_request_set *lovset = (struct lov_request_set *)data;
1394         int err;
1395         ENTRY;
1396
1397         if (rc)
1398                 lovset->set_completes = 0;
1399         err = lov_fini_setattr_set(lovset);
1400         RETURN(rc ? rc : err);
1401 }
1402
1403 /* If @oti is given, the request goes from MDS and responses from OSTs are not
1404    needed. Otherwise, a client is waiting for responses. */
1405 static int lov_setattr_async(struct obd_export *exp, struct obd_info *oinfo,
1406                              struct obd_trans_info *oti,
1407                              struct ptlrpc_request_set *rqset)
1408 {
1409         struct lov_request_set *set;
1410         struct lov_request *req;
1411         struct list_head *pos;
1412         struct lov_obd *lov;
1413         int rc = 0;
1414         ENTRY;
1415
1416         LASSERT(oinfo);
1417         ASSERT_LSM_MAGIC(oinfo->oi_md);
1418         if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE) {
1419                 LASSERT(oti);
1420                 LASSERT(oti->oti_logcookies);
1421         }
1422
1423         if (!exp || !exp->exp_obd)
1424                 RETURN(-ENODEV);
1425
1426         lov = &exp->exp_obd->u.lov;
1427         rc = lov_prep_setattr_set(exp, oinfo, oti, &set);
1428         if (rc)
1429                 RETURN(rc);
1430
1431         CDEBUG(D_INFO, "objid "LPX64": %ux%u byte stripes\n",
1432                oinfo->oi_md->lsm_object_id, oinfo->oi_md->lsm_stripe_count,
1433                oinfo->oi_md->lsm_stripe_size);
1434
1435         list_for_each (pos, &set->set_list) {
1436                 req = list_entry(pos, struct lov_request, rq_link);
1437
1438                 if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE)
1439                         oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1440
1441                 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
1442                        "%u\n", oinfo->oi_oa->o_id, req->rq_stripe,
1443                        req->rq_oi.oi_oa->o_id, req->rq_idx);
1444
1445                 rc = obd_setattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1446                                        &req->rq_oi, oti, rqset);
1447                 if (rc) {
1448                         CERROR("error: setattr objid "LPX64" subobj "
1449                                LPX64" on OST idx %d: rc = %d\n",
1450                                set->set_oi->oi_oa->o_id,
1451                                req->rq_oi.oi_oa->o_id,
1452                                req->rq_idx, rc);
1453                         break;
1454                 }
1455         }
1456
1457         /* If we are not waiting for responses on async requests, return. */
1458         if (rc || !rqset || list_empty(&rqset->set_requests)) {
1459                 int err;
1460                 if (rc)
1461                         set->set_completes = 0;
1462                 err = lov_fini_setattr_set(set);
1463                 RETURN(rc ? rc : err);
1464         }
1465
1466         LASSERT(rqset->set_interpret == NULL);
1467         rqset->set_interpret = lov_setattr_interpret;
1468         rqset->set_arg = (void *)set;
1469
1470         RETURN(0);
1471 }
1472
1473 static int lov_punch_interpret(struct ptlrpc_request_set *rqset,
1474                                void *data, int rc)
1475 {
1476         struct lov_request_set *lovset = (struct lov_request_set *)data;
1477         int err;
1478         ENTRY;
1479
1480         if (rc)
1481                 lovset->set_completes = 0;
1482         err = lov_fini_punch_set(lovset);
1483         RETURN(rc ? rc : err);
1484 }
1485
1486 /* FIXME: maybe we'll just make one node the authoritative attribute node, then
1487  * we can send this 'punch' to just the authoritative node and the nodes
1488  * that the punch will affect. */
1489 static int lov_punch(struct obd_export *exp, struct obd_info *oinfo,
1490                      struct obd_trans_info *oti,
1491                      struct ptlrpc_request_set *rqset)
1492 {
1493         struct lov_request_set *set;
1494         struct lov_obd *lov;
1495         struct list_head *pos;
1496         struct lov_request *req;
1497         int rc = 0;
1498         ENTRY;
1499
1500         LASSERT(oinfo);
1501         ASSERT_LSM_MAGIC(oinfo->oi_md);
1502
1503         if (!exp || !exp->exp_obd)
1504                 RETURN(-ENODEV);
1505
1506         lov = &exp->exp_obd->u.lov;
1507         rc = lov_prep_punch_set(exp, oinfo, oti, &set);
1508         if (rc)
1509                 RETURN(rc);
1510
1511         list_for_each (pos, &set->set_list) {
1512                 req = list_entry(pos, struct lov_request, rq_link);
1513
1514                 rc = obd_punch(lov->lov_tgts[req->rq_idx]->ltd_exp,
1515                                &req->rq_oi, NULL, rqset);
1516                 if (rc) {
1517                         CERROR("error: punch objid "LPX64" subobj "LPX64
1518                                " on OST idx %d: rc = %d\n",
1519                                set->set_oi->oi_oa->o_id,
1520                                req->rq_oi.oi_oa->o_id, req->rq_idx, rc);
1521                         break;
1522                 }
1523         }
1524
1525         if (rc || list_empty(&rqset->set_requests)) {
1526                 int err;
1527                 err = lov_fini_punch_set(set);
1528                 RETURN(rc ? rc : err);
1529         }
1530
1531         LASSERT(rqset->set_interpret == NULL);
1532         rqset->set_interpret = lov_punch_interpret;
1533         rqset->set_arg = (void *)set;
1534
1535         RETURN(0);
1536 }
1537
1538 static int lov_sync(struct obd_export *exp, struct obdo *oa,
1539                     struct lov_stripe_md *lsm, obd_off start, obd_off end,
1540                     void *capa)
1541 {
1542         struct lov_request_set *set;
1543         struct obd_info oinfo;
1544         struct lov_obd *lov;
1545         struct list_head *pos;
1546         struct lov_request *req;
1547         int err = 0, rc = 0;
1548         ENTRY;
1549
1550         ASSERT_LSM_MAGIC(lsm);
1551
1552         if (!exp->exp_obd)
1553                 RETURN(-ENODEV);
1554
1555         lov = &exp->exp_obd->u.lov;
1556         rc = lov_prep_sync_set(exp, &oinfo, oa, lsm, start, end, &set);
1557         if (rc)
1558                 RETURN(rc);
1559
1560         list_for_each (pos, &set->set_list) {
1561                 req = list_entry(pos, struct lov_request, rq_link);
1562
1563                 rc = obd_sync(lov->lov_tgts[req->rq_idx]->ltd_exp,
1564                               req->rq_oi.oi_oa, NULL,
1565                               req->rq_oi.oi_policy.l_extent.start,
1566                               req->rq_oi.oi_policy.l_extent.end, capa);
1567                 err = lov_update_common_set(set, req, rc);
1568                 if (err) {
1569                         CERROR("error: fsync objid "LPX64" subobj "LPX64
1570                                " on OST idx %d: rc = %d\n",
1571                                set->set_oi->oi_oa->o_id,
1572                                req->rq_oi.oi_oa->o_id, req->rq_idx, rc);
1573                         if (!rc)
1574                                 rc = err;
1575                 }
1576         }
1577         err = lov_fini_sync_set(set);
1578         if (!rc)
1579                 rc = err;
1580         RETURN(rc);
1581 }
1582
1583 static int lov_brw_check(struct lov_obd *lov, struct obd_info *lov_oinfo,
1584                          obd_count oa_bufs, struct brw_page *pga)
1585 {
1586         struct obd_info oinfo = { { { 0 } } };
1587         int i, rc = 0;
1588
1589         oinfo.oi_oa = lov_oinfo->oi_oa;
1590
1591         /* The caller just wants to know if there's a chance that this
1592          * I/O can succeed */
1593         for (i = 0; i < oa_bufs; i++) {
1594                 int stripe = lov_stripe_number(lov_oinfo->oi_md, pga[i].off);
1595                 int ost = lov_oinfo->oi_md->lsm_oinfo[stripe]->loi_ost_idx;
1596                 obd_off start, end;
1597
1598                 if (!lov_stripe_intersects(lov_oinfo->oi_md, i, pga[i].off,
1599                                            pga[i].off + pga[i].count - 1,
1600                                            &start, &end))
1601                         continue;
1602
1603                 if (!lov->lov_tgts[ost] || !lov->lov_tgts[ost]->ltd_active) {
1604                         CDEBUG(D_HA, "lov idx %d inactive\n", ost);
1605                         return -EIO;
1606                 }
1607
1608                 rc = obd_brw(OBD_BRW_CHECK, lov->lov_tgts[ost]->ltd_exp, &oinfo,
1609                              1, &pga[i], NULL);
1610                 if (rc)
1611                         break;
1612         }
1613         return rc;
1614 }
1615
1616 static int lov_brw(int cmd, struct obd_export *exp, struct obd_info *oinfo,
1617                    obd_count oa_bufs, struct brw_page *pga,
1618                    struct obd_trans_info *oti)
1619 {
1620         struct lov_request_set *set;
1621         struct lov_request *req;
1622         struct list_head *pos;
1623         struct lov_obd *lov = &exp->exp_obd->u.lov;
1624         int err, rc = 0;
1625         ENTRY;
1626
1627         ASSERT_LSM_MAGIC(oinfo->oi_md);
1628
1629         if (cmd == OBD_BRW_CHECK) {
1630                 rc = lov_brw_check(lov, oinfo, oa_bufs, pga);
1631                 RETURN(rc);
1632         }
1633
1634         rc = lov_prep_brw_set(exp, oinfo, oa_bufs, pga, oti, &set);
1635         if (rc)
1636                 RETURN(rc);
1637
1638         list_for_each (pos, &set->set_list) {
1639                 struct obd_export *sub_exp;
1640                 struct brw_page *sub_pga;
1641                 req = list_entry(pos, struct lov_request, rq_link);
1642
1643                 sub_exp = lov->lov_tgts[req->rq_idx]->ltd_exp;
1644                 sub_pga = set->set_pga + req->rq_pgaidx;
1645                 rc = obd_brw(cmd, sub_exp, &req->rq_oi, req->rq_oabufs,
1646                              sub_pga, oti);
1647                 if (rc)
1648                         break;
1649                 lov_update_common_set(set, req, rc);
1650         }
1651
1652         err = lov_fini_brw_set(set);
1653         if (!rc)
1654                 rc = err;
1655         RETURN(rc);
1656 }
1657
1658 static int lov_enqueue_interpret(struct ptlrpc_request_set *rqset,
1659                                  void *data, int rc)
1660 {
1661         struct lov_request_set *lovset = (struct lov_request_set *)data;
1662         ENTRY;
1663         rc = lov_fini_enqueue_set(lovset, lovset->set_ei->ei_mode, rc, rqset);
1664         RETURN(rc);
1665 }
1666
1667 static int lov_enqueue(struct obd_export *exp, struct obd_info *oinfo,
1668                        struct ldlm_enqueue_info *einfo,
1669                        struct ptlrpc_request_set *rqset)
1670 {
1671         ldlm_mode_t mode = einfo->ei_mode;
1672         struct lov_request_set *set;
1673         struct lov_request *req;
1674         struct list_head *pos;
1675         struct lov_obd *lov;
1676         ldlm_error_t rc;
1677         ENTRY;
1678
1679         LASSERT(oinfo);
1680         ASSERT_LSM_MAGIC(oinfo->oi_md);
1681         LASSERT(mode == (mode & -mode));
1682
1683         /* we should never be asked to replay a lock this way. */
1684         LASSERT((oinfo->oi_flags & LDLM_FL_REPLAY) == 0);
1685
1686         if (!exp || !exp->exp_obd)
1687                 RETURN(-ENODEV);
1688
1689         lov = &exp->exp_obd->u.lov;
1690         rc = lov_prep_enqueue_set(exp, oinfo, einfo, &set);
1691         if (rc)
1692                 RETURN(rc);
1693
1694         list_for_each (pos, &set->set_list) {
1695                 req = list_entry(pos, struct lov_request, rq_link);
1696
1697                 rc = obd_enqueue(lov->lov_tgts[req->rq_idx]->ltd_exp,
1698                                  &req->rq_oi, einfo, rqset);
1699                 if (rc != ELDLM_OK)
1700                         GOTO(out, rc);
1701         }
1702
1703         if (rqset && !list_empty(&rqset->set_requests)) {
1704                 LASSERT(rc == 0);
1705                 LASSERT(rqset->set_interpret == NULL);
1706                 rqset->set_interpret = lov_enqueue_interpret;
1707                 rqset->set_arg = (void *)set;
1708                 RETURN(rc);
1709         }
1710 out:
1711         rc = lov_fini_enqueue_set(set, mode, rc, rqset);
1712         RETURN(rc);
1713 }
1714
1715 static int lov_change_cbdata(struct obd_export *exp,
1716                              struct lov_stripe_md *lsm, ldlm_iterator_t it,
1717                              void *data)
1718 {
1719         struct lov_obd *lov;
1720         int rc = 0, i;
1721         ENTRY;
1722
1723         ASSERT_LSM_MAGIC(lsm);
1724
1725         if (!exp || !exp->exp_obd)
1726                 RETURN(-ENODEV);
1727
1728         LASSERT_MDS_GROUP(lsm->lsm_object_gr);
1729
1730         lov = &exp->exp_obd->u.lov;
1731         for (i = 0; i < lsm->lsm_stripe_count; i++) {
1732                 struct lov_stripe_md submd;
1733                 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
1734
1735                 if (!lov->lov_tgts[loi->loi_ost_idx]) {
1736                         CDEBUG(D_HA, "lov idx %d NULL \n", loi->loi_ost_idx);
1737                         continue;
1738                 }
1739
1740                 submd.lsm_object_id = loi->loi_id;
1741                 submd.lsm_object_gr = lsm->lsm_object_gr;
1742                 submd.lsm_stripe_count = 0;
1743                 rc = obd_change_cbdata(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1744                                        &submd, it, data);
1745         }
1746         RETURN(rc);
1747 }
1748
1749 static int lov_cancel(struct obd_export *exp, struct lov_stripe_md *lsm,
1750                       __u32 mode, struct lustre_handle *lockh)
1751 {
1752         struct lov_request_set *set;
1753         struct obd_info oinfo;
1754         struct lov_request *req;
1755         struct list_head *pos;
1756         struct lov_obd *lov;
1757         struct lustre_handle *lov_lockhp;
1758         int err = 0, rc = 0;
1759         ENTRY;
1760
1761         ASSERT_LSM_MAGIC(lsm);
1762
1763         if (!exp || !exp->exp_obd)
1764                 RETURN(-ENODEV);
1765
1766         LASSERT_MDS_GROUP(lsm->lsm_object_gr);
1767         LASSERT(lockh);
1768         lov = &exp->exp_obd->u.lov;
1769         rc = lov_prep_cancel_set(exp, &oinfo, lsm, mode, lockh, &set);
1770         if (rc)
1771                 RETURN(rc);
1772
1773         list_for_each (pos, &set->set_list) {
1774                 req = list_entry(pos, struct lov_request, rq_link);
1775                 lov_lockhp = set->set_lockh->llh_handles + req->rq_stripe;
1776
1777                 rc = obd_cancel(lov->lov_tgts[req->rq_idx]->ltd_exp,
1778                                 req->rq_oi.oi_md, mode, lov_lockhp);
1779                 rc = lov_update_common_set(set, req, rc);
1780                 if (rc) {
1781                         CERROR("error: cancel objid "LPX64" subobj "
1782                                LPX64" on OST idx %d: rc = %d\n",
1783                                lsm->lsm_object_id,
1784                                req->rq_oi.oi_md->lsm_object_id,
1785                                req->rq_idx, rc);
1786                         err = rc;
1787                 }
1788
1789         }
1790         lov_fini_cancel_set(set);
1791         RETURN(err);
1792 }
1793
1794 static int lov_cancel_unused(struct obd_export *exp,
1795                              struct lov_stripe_md *lsm,
1796                              int flags, void *opaque)
1797 {
1798         struct lov_obd *lov;
1799         int rc = 0, i;
1800         ENTRY;
1801
1802         if (!exp || !exp->exp_obd)
1803                 RETURN(-ENODEV);
1804
1805         lov = &exp->exp_obd->u.lov;
1806         if (lsm == NULL) {
1807                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
1808                         int err;
1809                         if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
1810                                 continue;
1811
1812                         err = obd_cancel_unused(lov->lov_tgts[i]->ltd_exp, NULL,
1813                                                 flags, opaque);
1814                         if (!rc)
1815                                 rc = err;
1816                 }
1817                 RETURN(rc);
1818         }
1819
1820         ASSERT_LSM_MAGIC(lsm);
1821
1822         LASSERT_MDS_GROUP(lsm->lsm_object_gr);
1823         for (i = 0; i < lsm->lsm_stripe_count; i++) {
1824                 struct lov_stripe_md submd;
1825                 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
1826                 int err;
1827
1828                 if (!lov->lov_tgts[loi->loi_ost_idx]) {
1829                         CDEBUG(D_HA, "lov idx %d NULL\n", loi->loi_ost_idx);
1830                         continue;
1831                 }
1832
1833                 if (!lov->lov_tgts[loi->loi_ost_idx]->ltd_active)
1834                         CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
1835
1836                 submd.lsm_object_id = loi->loi_id;
1837                 submd.lsm_object_gr = lsm->lsm_object_gr;
1838                 submd.lsm_stripe_count = 0;
1839                 err = obd_cancel_unused(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1840                                         &submd, flags, opaque);
1841                 if (err && lov->lov_tgts[loi->loi_ost_idx]->ltd_active) {
1842                         CERROR("error: cancel unused objid "LPX64" subobj "LPX64
1843                                " on OST idx %d: rc = %d\n", lsm->lsm_object_id,
1844                                loi->loi_id, loi->loi_ost_idx, err);
1845                         if (!rc)
1846                                 rc = err;
1847                 }
1848         }
1849         RETURN(rc);
1850 }
1851
1852 int lov_statfs_interpret(struct ptlrpc_request_set *rqset, void *data, int rc)
1853 {
1854         struct lov_request_set *lovset = (struct lov_request_set *)data;
1855         int err;
1856         ENTRY;
1857
1858         if (rc)
1859                 lovset->set_completes = 0;
1860
1861         err = lov_fini_statfs_set(lovset);
1862         RETURN(rc ? rc : err);
1863 }
1864
1865 static int lov_statfs_async(struct obd_device *obd, struct obd_info *oinfo,
1866                             __u64 max_age, struct ptlrpc_request_set *rqset)
1867 {
1868         struct lov_request_set *set;
1869         struct lov_request *req;
1870         struct list_head *pos;
1871         struct lov_obd *lov;
1872         int rc = 0;
1873         ENTRY;
1874
1875         LASSERT(oinfo != NULL);
1876         LASSERT(oinfo->oi_osfs != NULL);
1877
1878         lov = &obd->u.lov;
1879         rc = lov_prep_statfs_set(obd, oinfo, &set);
1880         if (rc)
1881                 RETURN(rc);
1882
1883         list_for_each (pos, &set->set_list) {
1884                 struct obd_device *osc_obd;
1885
1886                 req = list_entry(pos, struct lov_request, rq_link);
1887
1888                 osc_obd = class_exp2obd(lov->lov_tgts[req->rq_idx]->ltd_exp);
1889                 rc = obd_statfs_async(osc_obd, &req->rq_oi, max_age, rqset);
1890                 if (rc)
1891                         break;
1892         }
1893
1894         if (rc || list_empty(&rqset->set_requests)) {
1895                 int err;
1896                 if (rc)
1897                         set->set_completes = 0;
1898                 err = lov_fini_statfs_set(set);
1899                 RETURN(rc ? rc : err);
1900         }
1901
1902         LASSERT(rqset->set_interpret == NULL);
1903         rqset->set_interpret = lov_statfs_interpret;
1904         rqset->set_arg = (void *)set;
1905         RETURN(0);
1906 }
1907
1908 static int lov_statfs(struct obd_device *obd, struct obd_statfs *osfs,
1909                       __u64 max_age, __u32 flags)
1910 {
1911         struct ptlrpc_request_set *set = NULL;
1912         struct obd_info oinfo = { { { 0 } } };
1913         int rc = 0;
1914         ENTRY;
1915
1916
1917         /* for obdclass we forbid using obd_statfs_rqset, but prefer using async
1918          * statfs requests */
1919         set = ptlrpc_prep_set();
1920         if (set == NULL)
1921                 RETURN(-ENOMEM);
1922
1923         oinfo.oi_osfs = osfs;
1924         oinfo.oi_flags = flags;
1925         rc = lov_statfs_async(obd, &oinfo, max_age, set);
1926         if (rc == 0)
1927                 rc = ptlrpc_set_wait(set);
1928         ptlrpc_set_destroy(set);
1929
1930         RETURN(rc);
1931 }
1932
1933 static int lov_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1934                          void *karg, void *uarg)
1935 {
1936         struct obd_device *obddev = class_exp2obd(exp);
1937         struct lov_obd *lov = &obddev->u.lov;
1938         int i = 0, rc = 0, count = lov->desc.ld_tgt_count;
1939         struct obd_uuid *uuidp;
1940         ENTRY;
1941
1942         switch (cmd) {
1943         case IOC_OBD_STATFS: {
1944                 struct obd_ioctl_data *data = karg;
1945                 struct obd_device *osc_obd;
1946                 struct obd_statfs stat_buf = {0};
1947                 __u32 index;
1948
1949                 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
1950                 LASSERT(data->ioc_plen1 == sizeof(struct obd_statfs));
1951
1952                 if ((index >= count))
1953                         RETURN(-ENODEV);
1954
1955                 if (!lov->lov_tgts[index])
1956                         /* Try again with the next index */
1957                         RETURN(-EAGAIN);
1958                 if (!lov->lov_tgts[index]->ltd_active)
1959                         RETURN(-ENODATA);
1960
1961                 osc_obd = class_exp2obd(lov->lov_tgts[index]->ltd_exp);
1962                 if (!osc_obd)
1963                         RETURN(-EINVAL);
1964
1965                 /* got statfs data */
1966                 rc = obd_statfs(osc_obd, &stat_buf,
1967                                 cfs_time_current_64() - HZ, 0);
1968                 if (rc)
1969                         RETURN(rc);
1970                 if (copy_to_user(data->ioc_pbuf1, &stat_buf, data->ioc_plen1))
1971                         RETURN(-EFAULT);
1972                 /* copy UUID */
1973                 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(osc_obd),
1974                                  data->ioc_plen2))
1975                         RETURN(-EFAULT);
1976                 break;
1977         }
1978         case OBD_IOC_LOV_GET_CONFIG: {
1979                 struct obd_ioctl_data *data;
1980                 struct lov_desc *desc;
1981                 char *buf = NULL;
1982                 __u32 *genp;
1983
1984                 len = 0;
1985                 if (obd_ioctl_getdata(&buf, &len, (void *)uarg))
1986                         RETURN(-EINVAL);
1987
1988                 data = (struct obd_ioctl_data *)buf;
1989
1990                 if (sizeof(*desc) > data->ioc_inllen1) {
1991                         obd_ioctl_freedata(buf, len);
1992                         RETURN(-EINVAL);
1993                 }
1994
1995                 if (sizeof(uuidp->uuid) * count > data->ioc_inllen2) {
1996                         obd_ioctl_freedata(buf, len);
1997                         RETURN(-EINVAL);
1998                 }
1999
2000                 if (sizeof(__u32) * count > data->ioc_inllen3) {
2001                         obd_ioctl_freedata(buf, len);
2002                         RETURN(-EINVAL);
2003                 }
2004
2005                 desc = (struct lov_desc *)data->ioc_inlbuf1;
2006                 memcpy(desc, &(lov->desc), sizeof(*desc));
2007
2008                 uuidp = (struct obd_uuid *)data->ioc_inlbuf2;
2009                 genp = (__u32 *)data->ioc_inlbuf3;
2010                 /* the uuid will be empty for deleted OSTs */
2011                 for (i = 0; i < count; i++, uuidp++, genp++) {
2012                         if (!lov->lov_tgts[i])
2013                                 continue;
2014                         *uuidp = lov->lov_tgts[i]->ltd_uuid;
2015                         *genp = lov->lov_tgts[i]->ltd_gen;
2016                 }
2017
2018                 if (copy_to_user((void *)uarg, buf, len))
2019                         rc = -EFAULT;
2020                 obd_ioctl_freedata(buf, len);
2021                 break;
2022         }
2023         case LL_IOC_LOV_SETSTRIPE:
2024                 rc = lov_setstripe(exp, karg, uarg);
2025                 break;
2026         case LL_IOC_LOV_GETSTRIPE:
2027                 rc = lov_getstripe(exp, karg, uarg);
2028                 break;
2029         case LL_IOC_LOV_SETEA:
2030                 rc = lov_setea(exp, karg, uarg);
2031                 break;
2032         case OBD_IOC_QUOTACTL: {
2033                 struct if_quotactl *qctl = karg;
2034                 struct lov_tgt_desc *tgt = NULL;
2035                 struct obd_quotactl *oqctl;
2036
2037                 if (qctl->qc_valid == QC_OSTIDX) {
2038                         if (qctl->qc_idx < 0 || count <= qctl->qc_idx)
2039                                 RETURN(-EINVAL);
2040
2041                         tgt = lov->lov_tgts[qctl->qc_idx];
2042                         if (!tgt || !tgt->ltd_exp)
2043                                 RETURN(-EINVAL);
2044                 } else if (qctl->qc_valid == QC_UUID) {
2045                         for (i = 0; i < count; i++) {
2046                                 tgt = lov->lov_tgts[i];
2047                                 if (!tgt ||
2048                                     !obd_uuid_equals(&tgt->ltd_uuid,
2049                                                      &qctl->obd_uuid))
2050                                         continue;
2051
2052                                 if (tgt->ltd_exp == NULL)
2053                                         RETURN(-EINVAL);
2054
2055                                 break;
2056                         }
2057                 } else {
2058                         RETURN(-EINVAL);
2059                 }
2060
2061                 if (i >= count)
2062                         RETURN(-EAGAIN);
2063
2064                 LASSERT(tgt && tgt->ltd_exp);
2065                 OBD_ALLOC_PTR(oqctl);
2066                 if (!oqctl)
2067                         RETURN(-ENOMEM);
2068
2069                 QCTL_COPY(oqctl, qctl);
2070                 rc = obd_quotactl(tgt->ltd_exp, oqctl);
2071                 if (rc == 0) {
2072                         QCTL_COPY(qctl, oqctl);
2073                         qctl->qc_valid = QC_OSTIDX;
2074                         qctl->obd_uuid = tgt->ltd_uuid;
2075                 }
2076                 OBD_FREE_PTR(oqctl);
2077                 break;
2078         }
2079         default: {
2080                 int set = 0;
2081
2082                 if (count == 0)
2083                         RETURN(-ENOTTY);
2084
2085                 for (i = 0; i < count; i++) {
2086                         int err;
2087
2088                         /* OST was disconnected */
2089                         if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
2090                                 continue;
2091
2092                         err = obd_iocontrol(cmd, lov->lov_tgts[i]->ltd_exp,
2093                                             len, karg, uarg);
2094                         if (err == -ENODATA && cmd == OBD_IOC_POLL_QUOTACHECK) {
2095                                 RETURN(err);
2096                         } else if (err) {
2097                                 if (lov->lov_tgts[i]->ltd_active) {
2098                                         CDEBUG(err == -ENOTTY ?
2099                                                D_IOCTL : D_WARNING,
2100                                                "iocontrol OSC %s on OST "
2101                                                "idx %d cmd %x: err = %d\n",
2102                                                lov_uuid2str(lov, i),
2103                                                i, cmd, err);
2104                                         if (!rc)
2105                                                 rc = err;
2106                                 }
2107                         } else {
2108                                 set = 1;
2109                         }
2110                 }
2111                 if (!set && !rc)
2112                         rc = -EIO;
2113         }
2114         }
2115
2116         RETURN(rc);
2117 }
2118
2119 #define FIEMAP_BUFFER_SIZE 4096
2120
2121 /**
2122  * Non-zero fe_logical indicates that this is a continuation FIEMAP
2123  * call. The local end offset and the device are sent in the first
2124  * fm_extent. This function calculates the stripe number from the index.
2125  * This function returns a stripe_no on which mapping is to be restarted.
2126  *
2127  * This function returns fm_end_offset which is the in-OST offset at which
2128  * mapping should be restarted. If fm_end_offset=0 is returned then caller
2129  * will re-calculate proper offset in next stripe.
2130  * Note that the first extent is passed to lov_get_info via the value field.
2131  *
2132  * \param fiemap fiemap request header
2133  * \param lsm striping information for the file
2134  * \param fm_start logical start of mapping
2135  * \param fm_end logical end of mapping
2136  * \param start_stripe starting stripe will be returned in this
2137  */
2138 obd_size fiemap_calc_fm_end_offset(struct ll_user_fiemap *fiemap,
2139                                    struct lov_stripe_md *lsm, obd_size fm_start,
2140                                    obd_size fm_end, int *start_stripe)
2141 {
2142         obd_size local_end = fiemap->fm_extents[0].fe_logical;
2143         obd_off lun_start, lun_end;
2144         obd_size fm_end_offset;
2145         int stripe_no = -1, i;
2146
2147         if (fiemap->fm_extent_count == 0 ||
2148             fiemap->fm_extents[0].fe_logical == 0)
2149                 return 0;
2150
2151         /* Find out stripe_no from ost_index saved in the fe_device */
2152         for (i = 0; i < lsm->lsm_stripe_count; i++) {
2153                 if (lsm->lsm_oinfo[i]->loi_ost_idx ==
2154                                         fiemap->fm_extents[0].fe_device) {
2155                         stripe_no = i;
2156                         break;
2157                 }
2158         }
2159
2160         /* If we have finished mapping on previous device, shift logical
2161          * offset to start of next device */
2162         if ((lov_stripe_intersects(lsm, stripe_no, fm_start, fm_end,
2163                                    &lun_start, &lun_end)) != 0 &&
2164                                    local_end < lun_end) {
2165                 fm_end_offset = local_end;
2166                 *start_stripe = stripe_no;
2167         } else {
2168                 /* This is a special value to indicate that caller should
2169                  * calculate offset in next stripe. */
2170                 fm_end_offset = 0;
2171                 *start_stripe = (stripe_no + 1) % lsm->lsm_stripe_count;
2172         }
2173
2174         return fm_end_offset;
2175 }
2176
2177 /**
2178  * We calculate on which OST the mapping will end. If the length of mapping
2179  * is greater than (stripe_size * stripe_count) then the last_stripe will
2180  * will be one just before start_stripe. Else we check if the mapping
2181  * intersects each OST and find last_stripe.
2182  * This function returns the last_stripe and also sets the stripe_count
2183  * over which the mapping is spread
2184  *
2185  * \param lsm striping information for the file
2186  * \param fm_start logical start of mapping
2187  * \param fm_end logical end of mapping
2188  * \param start_stripe starting stripe of the mapping
2189  * \param stripe_count the number of stripes across which to map is returned
2190  *
2191  * \retval last_stripe return the last stripe of the mapping
2192  */
2193 int fiemap_calc_last_stripe(struct lov_stripe_md *lsm, obd_size fm_start,
2194                             obd_size fm_end, int start_stripe,
2195                             int *stripe_count)
2196 {
2197         int last_stripe;
2198         obd_off obd_start, obd_end;
2199         int i, j;
2200
2201         if (fm_end - fm_start > lsm->lsm_stripe_size * lsm->lsm_stripe_count) {
2202                 last_stripe = (start_stripe < 1 ? lsm->lsm_stripe_count - 1 :
2203                                                               start_stripe - 1);
2204                 *stripe_count = lsm->lsm_stripe_count;
2205         } else {
2206                 for (j = 0, i = start_stripe; j < lsm->lsm_stripe_count;
2207                      i = (i + 1) % lsm->lsm_stripe_count, j++) {
2208                         if ((lov_stripe_intersects(lsm, i, fm_start, fm_end,
2209                                                    &obd_start, &obd_end)) == 0)
2210                                 break;
2211                 }
2212                 *stripe_count = j;
2213                 last_stripe = (start_stripe + j - 1) %lsm->lsm_stripe_count;
2214         }
2215
2216         return last_stripe;
2217 }
2218
2219 /**
2220  * Set fe_device and copy extents from local buffer into main return buffer.
2221  *
2222  * \param fiemap fiemap request header
2223  * \param lcl_fm_ext array of local fiemap extents to be copied
2224  * \param ost_index OST index to be written into the fm_device field for each
2225                     extent
2226  * \param ext_count number of extents to be copied
2227  * \param current_extent where to start copying in main extent array
2228  */
2229 void fiemap_prepare_and_copy_exts(struct ll_user_fiemap *fiemap,
2230                                   struct ll_fiemap_extent *lcl_fm_ext,
2231                                   int ost_index, unsigned int ext_count,
2232                                   int current_extent)
2233 {
2234         char *to;
2235         int ext;
2236
2237         for (ext = 0; ext < ext_count; ext++) {
2238                 lcl_fm_ext[ext].fe_device = ost_index;
2239                 lcl_fm_ext[ext].fe_flags |= FIEMAP_EXTENT_NET;
2240         }
2241
2242         /* Copy fm_extent's from fm_local to return buffer */
2243         to = (char *)fiemap + fiemap_count_to_size(current_extent);
2244         memcpy(to, lcl_fm_ext, ext_count * sizeof(struct ll_fiemap_extent));
2245 }
2246
2247 /**
2248  * Break down the FIEMAP request and send appropriate calls to individual OSTs.
2249  * This also handles the restarting of FIEMAP calls in case mapping overflows
2250  * the available number of extents in single call.
2251  */
2252 static int lov_fiemap(struct lov_obd *lov, __u32 keylen, void *key,
2253                       __u32 *vallen, void *val, struct lov_stripe_md *lsm)
2254 {
2255         struct ll_fiemap_info_key *fm_key = key;
2256         struct ll_user_fiemap *fiemap = val;
2257         struct ll_user_fiemap *fm_local = NULL;
2258         struct ll_fiemap_extent *lcl_fm_ext;
2259         int count_local;
2260         unsigned int get_num_extents = 0;
2261         int ost_index = 0, actual_start_stripe, start_stripe;
2262         obd_size fm_start, fm_end, fm_length, fm_end_offset = 0;
2263         obd_size curr_loc;
2264         int current_extent = 0, rc = 0, i;
2265         int ost_eof = 0; /* EOF for object */
2266         int ost_done = 0; /* done with required mapping for this OST? */
2267         int last_stripe;
2268         int cur_stripe = 0, cur_stripe_wrap = 0, stripe_count;
2269         unsigned int buffer_size = FIEMAP_BUFFER_SIZE;
2270
2271         if (lsm == NULL)
2272                 GOTO(out, rc = 0);
2273
2274         if (fiemap_count_to_size(fm_key->fiemap.fm_extent_count) < buffer_size)
2275                 buffer_size = fiemap_count_to_size(fm_key->fiemap.fm_extent_count);
2276
2277         OBD_ALLOC(fm_local, buffer_size);
2278         if (fm_local == NULL)
2279                 GOTO(out, rc = -ENOMEM);
2280         lcl_fm_ext = &fm_local->fm_extents[0];
2281
2282         count_local = fiemap_size_to_count(buffer_size);
2283
2284         memcpy(fiemap, &fm_key->fiemap, sizeof(*fiemap));
2285         fm_start = fiemap->fm_start;
2286         fm_length = fiemap->fm_length;
2287         /* Calculate start stripe, last stripe and length of mapping */
2288         actual_start_stripe = start_stripe = lov_stripe_number(lsm, fm_start);
2289         fm_end = (fm_length == ~0ULL ? fm_key->oa.o_size :
2290                                                 fm_start + fm_length - 1);
2291         /* If fm_length != ~0ULL but fm_start+fm_length-1 exceeds file size */
2292         if (fm_end > fm_key->oa.o_size)
2293                 fm_end = fm_key->oa.o_size;
2294
2295         last_stripe = fiemap_calc_last_stripe(lsm, fm_start, fm_end,
2296                                             actual_start_stripe, &stripe_count);
2297
2298         fm_end_offset = fiemap_calc_fm_end_offset(fiemap, lsm, fm_start, fm_end,
2299                                                   &start_stripe);
2300
2301         if (fiemap->fm_extent_count == 0) {
2302                 get_num_extents = 1;
2303                 count_local = 0;
2304         }
2305
2306         /* Check each stripe */
2307         for (cur_stripe = start_stripe, i = 0; i < stripe_count;
2308              i++, cur_stripe = (cur_stripe + 1) % lsm->lsm_stripe_count) {
2309                 obd_size req_fm_len; /* Stores length of required mapping */
2310                 obd_size len_mapped_single_call;
2311                 obd_off lun_start, lun_end, obd_object_end;
2312                 unsigned int ext_count;
2313
2314                 cur_stripe_wrap = cur_stripe;
2315
2316                 /* Find out range of mapping on this stripe */
2317                 if ((lov_stripe_intersects(lsm, cur_stripe, fm_start, fm_end,
2318                                            &lun_start, &obd_object_end)) == 0)
2319                         continue;
2320
2321                 /* If this is a continuation FIEMAP call and we are on
2322                  * starting stripe then lun_start needs to be set to
2323                  * fm_end_offset */
2324                 if (fm_end_offset != 0 && cur_stripe == start_stripe)
2325                         lun_start = fm_end_offset;
2326
2327                 if (fm_length != ~0ULL) {
2328                         /* Handle fm_start + fm_length overflow */
2329                         if (fm_start + fm_length < fm_start)
2330                                 fm_length = ~0ULL - fm_start;
2331                         lun_end = lov_size_to_stripe(lsm, fm_start + fm_length,
2332                                                      cur_stripe);
2333                 } else {
2334                         lun_end = ~0ULL;
2335                 }
2336
2337                 if (lun_start == lun_end)
2338                         continue;
2339
2340                 req_fm_len = obd_object_end - lun_start;
2341                 fm_local->fm_length = 0;
2342                 len_mapped_single_call = 0;
2343
2344                 /* If the output buffer is very large and the objects have many
2345                  * extents we may need to loop on a single OST repeatedly */
2346                 ost_eof = 0;
2347                 ost_done = 0;
2348                 do {
2349                         if (get_num_extents == 0) {
2350                                 /* Don't get too many extents. */
2351                                 if (current_extent + count_local >
2352                                     fiemap->fm_extent_count)
2353                                         count_local = fiemap->fm_extent_count -
2354                                                                  current_extent;
2355                         }
2356
2357                         lun_start += len_mapped_single_call;
2358                         fm_local->fm_length = req_fm_len - len_mapped_single_call;
2359                         req_fm_len = fm_local->fm_length;
2360                         fm_local->fm_extent_count = count_local;
2361                         fm_local->fm_mapped_extents = 0;
2362                         fm_local->fm_flags = fiemap->fm_flags;
2363
2364                         fm_key->oa.o_id = lsm->lsm_oinfo[cur_stripe]->loi_id;
2365                         ost_index = lsm->lsm_oinfo[cur_stripe]->loi_ost_idx;
2366
2367                         if (ost_index < 0 || ost_index >=lov->desc.ld_tgt_count)
2368                                 GOTO(out, rc = -EINVAL);
2369
2370                         /* If OST is inactive, return extent with UNKNOWN flag */
2371                         if (!lov->lov_tgts[ost_index]->ltd_active) {
2372                                 fm_local->fm_flags |= FIEMAP_EXTENT_LAST;
2373                                 fm_local->fm_mapped_extents = 1;
2374
2375                                 lcl_fm_ext[0].fe_logical = lun_start;
2376                                 lcl_fm_ext[0].fe_length = obd_object_end -
2377                                                                       lun_start;
2378                                 lcl_fm_ext[0].fe_flags |= FIEMAP_EXTENT_UNKNOWN;
2379
2380                                 goto inactive_tgt;
2381                         }
2382
2383                         fm_local->fm_start = lun_start;
2384                         fm_local->fm_flags &= ~FIEMAP_FLAG_DEVICE_ORDER;
2385                         memcpy(&fm_key->fiemap, fm_local, sizeof(*fm_local));
2386                         *vallen=fiemap_count_to_size(fm_local->fm_extent_count);
2387                         rc = obd_get_info(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(fm_local, buffer_size);
2468         return rc;
2469 }
2470
2471 static int lov_get_info(struct obd_export *exp, __u32 keylen,
2472                         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         lov_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                             osc_res_name_eq(loi->loi_id, loi->loi_gr, 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(tgt->ltd_exp, keylen, key, &size, info->data, NULL);
2530                 GOTO(out, rc = 0);
2531         } else if (KEY_IS(KEY_LOVDESC)) {
2532                 struct lov_desc *desc_ret = val;
2533                 *desc_ret = lov->desc;
2534
2535                 GOTO(out, rc = 0);
2536         } else if (KEY_IS(KEY_FIEMAP)) {
2537                 rc = lov_fiemap(lov, keylen, key, vallen, val, lsm);
2538                 GOTO(out, rc);
2539         }
2540
2541         rc = -EINVAL;
2542
2543 out:
2544         lov_putref(obddev);
2545         RETURN(rc);
2546 }
2547
2548 static int lov_set_info_async(struct obd_export *exp, obd_count keylen,
2549                               void *key, obd_count vallen, void *val,
2550                               struct ptlrpc_request_set *set)
2551 {
2552         struct obd_device *obddev = class_exp2obd(exp);
2553         struct lov_obd *lov = &obddev->u.lov;
2554         obd_count count;
2555         int i, rc = 0, err;
2556         struct lov_tgt_desc *tgt;
2557         unsigned incr, check_uuid,
2558                  do_inactive, no_set;
2559         unsigned next_id = 0,  mds_con = 0;
2560         ENTRY;
2561
2562         incr = check_uuid = do_inactive = no_set = 0;
2563         if (set == NULL) {
2564                 no_set = 1;
2565                 set = ptlrpc_prep_set();
2566                 if (!set)
2567                         RETURN(-ENOMEM);
2568         }
2569
2570         lov_getref(obddev);
2571         count = lov->desc.ld_tgt_count;
2572
2573         if (KEY_IS(KEY_NEXT_ID)) {
2574                 count = vallen / sizeof(struct obd_id_info);
2575                 vallen = sizeof(obd_id);
2576                 incr = sizeof(struct obd_id_info);
2577                 do_inactive = 1;
2578                 next_id = 1;
2579         } else if (KEY_IS(KEY_CHECKSUM)) {
2580                 do_inactive = 1;
2581         } else if (KEY_IS(KEY_UNLINKED)) {
2582                 check_uuid = val ? 1 : 0;
2583         } else if (KEY_IS(KEY_EVICT_BY_NID)) {
2584                 /* use defaults:  do_inactive = incr = 0; */
2585         } else if (KEY_IS(KEY_MDS_CONN)) {
2586                 mds_con = 1;
2587         }
2588
2589         for (i = 0; i < count; i++, val = (char *)val + incr) {
2590                 if (next_id) {
2591                         tgt = lov->lov_tgts[((struct obd_id_info*)val)->idx];
2592                 } else {
2593                         tgt = lov->lov_tgts[i];
2594                 }
2595                 /* OST was disconnected */
2596                 if (!tgt || !tgt->ltd_exp)
2597                         continue;
2598
2599                 /* OST is inactive and we don't want inactive OSCs */
2600                 if (!tgt->ltd_active && !do_inactive)
2601                         continue;
2602
2603                 if (mds_con) {
2604                         struct mds_group_info *mgi;
2605
2606                         LASSERT(vallen == sizeof(*mgi));
2607                         mgi = (struct mds_group_info *)val;
2608
2609                         /* Only want a specific OSC */
2610                         if (mgi->uuid && !obd_uuid_equals(mgi->uuid,
2611                                                 &tgt->ltd_uuid))
2612                                 continue;
2613
2614                         err = obd_set_info_async(tgt->ltd_exp,
2615                                          keylen, key, sizeof(int),
2616                                          &mgi->group, set);
2617                 } else if (next_id) {
2618                         err = obd_set_info_async(tgt->ltd_exp,
2619                                          keylen, key, vallen,
2620                                          ((struct obd_id_info*)val)->data, set);
2621                 } else  {
2622                         /* Only want a specific OSC */
2623                         if (check_uuid &&
2624                             !obd_uuid_equals(val, &tgt->ltd_uuid))
2625                                 continue;
2626
2627                         err = obd_set_info_async(tgt->ltd_exp,
2628                                          keylen, key, vallen, val, set);
2629                 }
2630
2631                 if (!rc)
2632                         rc = err;
2633         }
2634
2635         lov_putref(obddev);
2636         if (no_set) {
2637                 err = ptlrpc_set_wait(set);
2638                 if (!rc)
2639                         rc = err;
2640                 ptlrpc_set_destroy(set);
2641         }
2642         RETURN(rc);
2643 }
2644
2645 static int lov_checkmd(struct obd_export *exp, struct obd_export *md_exp,
2646                        struct lov_stripe_md *lsm)
2647 {
2648         int rc;
2649         ENTRY;
2650
2651         if (!lsm)
2652                 RETURN(0);
2653         LASSERT(md_exp);
2654         LASSERT(lsm_op_find(lsm->lsm_magic) != NULL);
2655         rc = lsm_op_find(lsm->lsm_magic)->lsm_revalidate(lsm, md_exp->exp_obd);
2656
2657         RETURN(rc);
2658 }
2659
2660 int lov_test_and_clear_async_rc(struct lov_stripe_md *lsm)
2661 {
2662         int i, rc = 0;
2663         ENTRY;
2664
2665         for (i = 0; i < lsm->lsm_stripe_count; i++) {
2666                 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
2667                 if (loi->loi_ar.ar_rc && !rc)
2668                         rc = loi->loi_ar.ar_rc;
2669                 loi->loi_ar.ar_rc = 0;
2670         }
2671         RETURN(rc);
2672 }
2673 EXPORT_SYMBOL(lov_test_and_clear_async_rc);
2674
2675
2676 static int lov_extent_calc(struct obd_export *exp, struct lov_stripe_md *lsm,
2677                            int cmd, __u64 *offset)
2678 {
2679         __u32 ssize = lsm->lsm_stripe_size;
2680         __u64 start;
2681
2682         start = *offset;
2683         do_div(start, ssize);
2684         start = start * ssize;
2685
2686         CDEBUG(D_DLMTRACE, "offset "LPU64", stripe %u, start "LPU64
2687                            ", end "LPU64"\n", *offset, ssize, start,
2688                            start + ssize - 1);
2689         if (cmd == OBD_CALC_STRIPE_END) {
2690                 *offset = start + ssize - 1;
2691         } else if (cmd == OBD_CALC_STRIPE_START) {
2692                 *offset = start;
2693         } else {
2694                 LBUG();
2695         }
2696
2697         RETURN(0);
2698 }
2699
2700 void lov_stripe_lock(struct lov_stripe_md *md)
2701 {
2702         LASSERT(md->lsm_lock_owner != cfs_curproc_pid());
2703         spin_lock(&md->lsm_lock);
2704         LASSERT(md->lsm_lock_owner == 0);
2705         md->lsm_lock_owner = cfs_curproc_pid();
2706 }
2707 EXPORT_SYMBOL(lov_stripe_lock);
2708
2709 void lov_stripe_unlock(struct lov_stripe_md *md)
2710 {
2711         LASSERT(md->lsm_lock_owner == cfs_curproc_pid());
2712         md->lsm_lock_owner = 0;
2713         spin_unlock(&md->lsm_lock);
2714 }
2715 EXPORT_SYMBOL(lov_stripe_unlock);
2716
2717
2718 struct obd_ops lov_obd_ops = {
2719         .o_owner               = THIS_MODULE,
2720         .o_setup               = lov_setup,
2721         .o_precleanup          = lov_precleanup,
2722         .o_cleanup             = lov_cleanup,
2723         //.o_process_config      = lov_process_config,
2724         .o_connect             = lov_connect,
2725         .o_disconnect          = lov_disconnect,
2726         .o_statfs              = lov_statfs,
2727         .o_statfs_async        = lov_statfs_async,
2728         .o_packmd              = lov_packmd,
2729         .o_unpackmd            = lov_unpackmd,
2730         .o_checkmd             = lov_checkmd,
2731         .o_create              = lov_create,
2732         .o_destroy             = lov_destroy,
2733         .o_getattr             = lov_getattr,
2734         .o_getattr_async       = lov_getattr_async,
2735         .o_setattr             = lov_setattr,
2736         .o_setattr_async       = lov_setattr_async,
2737         .o_brw                 = lov_brw,
2738         .o_merge_lvb           = lov_merge_lvb,
2739         .o_adjust_kms          = lov_adjust_kms,
2740         .o_punch               = lov_punch,
2741         .o_sync                = lov_sync,
2742         .o_enqueue             = lov_enqueue,
2743         .o_change_cbdata       = lov_change_cbdata,
2744         .o_cancel              = lov_cancel,
2745         .o_cancel_unused       = lov_cancel_unused,
2746         .o_iocontrol           = lov_iocontrol,
2747         .o_get_info            = lov_get_info,
2748         .o_set_info_async      = lov_set_info_async,
2749         .o_extent_calc         = lov_extent_calc,
2750         .o_llog_init           = lov_llog_init,
2751         .o_llog_finish         = lov_llog_finish,
2752         .o_notify              = lov_notify,
2753         .o_pool_new            = lov_pool_new,
2754         .o_pool_rem            = lov_pool_remove,
2755         .o_pool_add            = lov_pool_add,
2756         .o_pool_del            = lov_pool_del,
2757 };
2758
2759 static quota_interface_t *quota_interface;
2760 extern quota_interface_t lov_quota_interface;
2761
2762 cfs_mem_cache_t *lov_oinfo_slab;
2763
2764 extern struct lu_kmem_descr lov_caches[];
2765
2766 int __init lov_init(void)
2767 {
2768         struct lprocfs_static_vars lvars = { 0 };
2769         int rc, rc2;
2770         ENTRY;
2771
2772         /* print an address of _any_ initialized kernel symbol from this
2773          * module, to allow debugging with gdb that doesn't support data
2774          * symbols from modules.*/
2775         CDEBUG(D_CONSOLE, "Lustre LOV module (%p).\n", &lov_caches);
2776
2777         rc = lu_kmem_init(lov_caches);
2778         if (rc)
2779                 return rc;
2780
2781         lov_oinfo_slab = cfs_mem_cache_create("lov_oinfo",
2782                                               sizeof(struct lov_oinfo),
2783                                               0, SLAB_HWCACHE_ALIGN);
2784         if (lov_oinfo_slab == NULL) {
2785                 lu_kmem_fini(lov_caches);
2786                 return -ENOMEM;
2787         }
2788         lprocfs_lov_init_vars(&lvars);
2789
2790         request_module("lquota");
2791         quota_interface = PORTAL_SYMBOL_GET(lov_quota_interface);
2792         init_obd_quota_ops(quota_interface, &lov_obd_ops);
2793
2794         rc = class_register_type(&lov_obd_ops, NULL, lvars.module_vars,
2795                                  LUSTRE_LOV_NAME, &lov_device_type);
2796
2797         if (rc) {
2798                 if (quota_interface)
2799                         PORTAL_SYMBOL_PUT(lov_quota_interface);
2800                 rc2 = cfs_mem_cache_destroy(lov_oinfo_slab);
2801                 LASSERT(rc2 == 0);
2802                 lu_kmem_fini(lov_caches);
2803         }
2804
2805         RETURN(rc);
2806 }
2807
2808 #ifdef __KERNEL__
2809 static void /*__exit*/ lov_exit(void)
2810 {
2811         int rc;
2812
2813         lu_device_type_fini(&lov_device_type);
2814         lu_kmem_fini(lov_caches);
2815
2816         if (quota_interface)
2817                 PORTAL_SYMBOL_PUT(lov_quota_interface);
2818
2819         class_unregister_type(LUSTRE_LOV_NAME);
2820         rc = cfs_mem_cache_destroy(lov_oinfo_slab);
2821         LASSERT(rc == 0);
2822 }
2823
2824 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2825 MODULE_DESCRIPTION("Lustre Logical Object Volume OBD driver");
2826 MODULE_LICENSE("GPL");
2827
2828 cfs_module(lov, LUSTRE_VERSION_STRING, lov_init, lov_exit);
2829 #endif