drivers/staging/lustre/lustre/lov/lov_obd.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2011, 2015, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  * Lustre is a trademark of Sun Microsystems, Inc.
  31  *
  32  * lustre/lov/lov_obd.c
  33  *
  34  * Author: Phil Schwan <phil@clusterfs.com>
  35  * Author: Peter Braam <braam@clusterfs.com>
  36  * Author: Mike Shaver <shaver@clusterfs.com>
  37  * Author: Nathan Rutman <nathan@clusterfs.com>
  38  */
  39
  40 #define DEBUG_SUBSYSTEM S_LOV
  41 #include "../../include/linux/libcfs/libcfs.h"
  42
  43 #include "../include/obd_support.h"
  44 #include "../include/lustre/lustre_ioctl.h"
  45 #include "../include/lustre_lib.h"
  46 #include "../include/lustre_net.h"
  47 #include "../include/lustre/lustre_idl.h"
  48 #include "../include/lustre_dlm.h"
  49 #include "../include/lustre_mds.h"
  50 #include "../include/obd_class.h"
  51 #include "../include/lprocfs_status.h"
  52 #include "../include/lustre_param.h"
  53 #include "../include/cl_object.h"
  54 #include "../include/lustre/ll_fiemap.h"
  55 #include "../include/lustre_fid.h"
  56
  57 #include "lov_internal.h"
  58
  59 /* Keep a refcount of lov->tgt usage to prevent racing with addition/deletion.
  60  * Any function that expects lov_tgts to remain stationary must take a ref.
  61  */
  62 static void lov_getref(struct obd_device *obd)
  63 {
  64         struct lov_obd *lov = &obd->u.lov;
  65
  66         /* nobody gets through here until lov_putref is done */
  67         mutex_lock(&lov->lov_lock);
  68         atomic_inc(&lov->lov_refcount);
  69         mutex_unlock(&lov->lov_lock);
  70         return;
  71 }
  72
  73 static void __lov_del_obd(struct obd_device *obd, struct lov_tgt_desc *tgt);
  74
  75 static void lov_putref(struct obd_device *obd)
  76 {
  77         struct lov_obd *lov = &obd->u.lov;
  78
  79         mutex_lock(&lov->lov_lock);
  80         /* ok to dec to 0 more than once -- ltd_exp's will be null */
  81         if (atomic_dec_and_test(&lov->lov_refcount) && lov->lov_death_row) {
  82                 LIST_HEAD(kill);
  83                 int i;
  84                 struct lov_tgt_desc *tgt, *n;
  85
  86                 CDEBUG(D_CONFIG, "destroying %d lov targets\n",
  87                        lov->lov_death_row);
  88                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
  89                         tgt = lov->lov_tgts[i];
  90
  91                         if (!tgt || !tgt->ltd_reap)
  92                                 continue;
  93                         list_add(&tgt->ltd_kill, &kill);
  94                         /* XXX - right now there is a dependency on ld_tgt_count
  95                          * being the maximum tgt index for computing the
  96                          * mds_max_easize. So we can't shrink it.
  97                          */
  98                         lov_ost_pool_remove(&lov->lov_packed, i);
  99                         lov->lov_tgts[i] = NULL;
 100                         lov->lov_death_row--;
 101                 }
 102                 mutex_unlock(&lov->lov_lock);
 103
 104                 list_for_each_entry_safe(tgt, n, &kill, ltd_kill) {
 105                         list_del(&tgt->ltd_kill);
 106                         /* Disconnect */
 107                         __lov_del_obd(obd, tgt);
 108                 }
 109
 110                 if (lov->lov_tgts_kobj)
 111                         kobject_put(lov->lov_tgts_kobj);
 112
 113         } else {
 114                 mutex_unlock(&lov->lov_lock);
 115         }
 116 }
 117
 118 static int lov_set_osc_active(struct obd_device *obd, struct obd_uuid *uuid,
 119                               enum obd_notify_event ev);
 120 static int lov_notify(struct obd_device *obd, struct obd_device *watched,
 121                       enum obd_notify_event ev, void *data);
 122
 123 int lov_connect_obd(struct obd_device *obd, __u32 index, int activate,
 124                     struct obd_connect_data *data)
 125 {
 126         struct lov_obd *lov = &obd->u.lov;
 127         struct obd_uuid *tgt_uuid;
 128         struct obd_device *tgt_obd;
 129         static struct obd_uuid lov_osc_uuid = { "LOV_OSC_UUID" };
 130         struct obd_import *imp;
 131         int rc;
 132
 133         if (!lov->lov_tgts[index])
 134                 return -EINVAL;
 135
 136         tgt_uuid = &lov->lov_tgts[index]->ltd_uuid;
 137         tgt_obd = lov->lov_tgts[index]->ltd_obd;
 138
 139         if (!tgt_obd->obd_set_up) {
 140                 CERROR("Target %s not set up\n", obd_uuid2str(tgt_uuid));
 141                 return -EINVAL;
 142         }
 143
 144         /* override the sp_me from lov */
 145         tgt_obd->u.cli.cl_sp_me = lov->lov_sp_me;
 146
 147         if (data && (data->ocd_connect_flags & OBD_CONNECT_INDEX))
 148                 data->ocd_index = index;
 149
 150         /*
 151          * Divine LOV knows that OBDs under it are OSCs.
 152          */
 153         imp = tgt_obd->u.cli.cl_import;
 154
 155         if (activate) {
 156                 tgt_obd->obd_no_recov = 0;
 157                 /* FIXME this is probably supposed to be
 158                  * ptlrpc_set_import_active.  Horrible naming.
 159                  */
 160                 ptlrpc_activate_import(imp);
 161         }
 162
 163         rc = obd_register_observer(tgt_obd, obd);
 164         if (rc) {
 165                 CERROR("Target %s register_observer error %d\n",
 166                        obd_uuid2str(tgt_uuid), rc);
 167                 return rc;
 168         }
 169
 170         if (imp->imp_invalid) {
 171                 CDEBUG(D_CONFIG, "not connecting OSC %s; administratively disabled\n",
 172                        obd_uuid2str(tgt_uuid));
 173                 return 0;
 174         }
 175
 176         rc = obd_connect(NULL, &lov->lov_tgts[index]->ltd_exp, tgt_obd,
 177                          &lov_osc_uuid, data, NULL);
 178         if (rc || !lov->lov_tgts[index]->ltd_exp) {
 179                 CERROR("Target %s connect error %d\n",
 180                        obd_uuid2str(tgt_uuid), rc);
 181                 return -ENODEV;
 182         }
 183
 184         lov->lov_tgts[index]->ltd_reap = 0;
 185
 186         CDEBUG(D_CONFIG, "Connected tgt idx %d %s (%s) %sactive\n", index,
 187                obd_uuid2str(tgt_uuid), tgt_obd->obd_name, activate ? "":"in");
 188
 189         if (lov->lov_tgts_kobj)
 190                 /* Even if we failed, that's ok */
 191                 rc = sysfs_create_link(lov->lov_tgts_kobj, &tgt_obd->obd_kobj,
 192                                        tgt_obd->obd_name);
 193
 194         return 0;
 195 }
 196
 197 static int lov_connect(const struct lu_env *env,
 198                        struct obd_export **exp, struct obd_device *obd,
 199                        struct obd_uuid *cluuid, struct obd_connect_data *data,
 200                        void *localdata)
 201 {
 202         struct lov_obd *lov = &obd->u.lov;
 203         struct lov_tgt_desc *tgt;
 204         struct lustre_handle conn;
 205         int i, rc;
 206
 207         CDEBUG(D_CONFIG, "connect #%d\n", lov->lov_connects);
 208
 209         rc = class_connect(&conn, obd, cluuid);
 210         if (rc)
 211                 return rc;
 212
 213         *exp = class_conn2export(&conn);
 214
 215         /* Why should there ever be more than 1 connect? */
 216         lov->lov_connects++;
 217         LASSERT(lov->lov_connects == 1);
 218
 219         memset(&lov->lov_ocd, 0, sizeof(lov->lov_ocd));
 220         if (data)
 221                 lov->lov_ocd = *data;
 222
 223         obd_getref(obd);
 224
 225         lov->lov_tgts_kobj = kobject_create_and_add("target_obds",
 226                                                     &obd->obd_kobj);
 227
 228         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
 229                 tgt = lov->lov_tgts[i];
 230                 if (!tgt || obd_uuid_empty(&tgt->ltd_uuid))
 231                         continue;
 232                 /* Flags will be lowest common denominator */
 233                 rc = lov_connect_obd(obd, i, tgt->ltd_activate, &lov->lov_ocd);
 234                 if (rc) {
 235                         CERROR("%s: lov connect tgt %d failed: %d\n",
 236                                obd->obd_name, i, rc);
 237                         continue;
 238                 }
 239                 /* connect to administrative disabled ost */
 240                 if (!lov->lov_tgts[i]->ltd_exp)
 241                         continue;
 242
 243                 rc = lov_notify(obd, lov->lov_tgts[i]->ltd_exp->exp_obd,
 244                                 OBD_NOTIFY_CONNECT, (void *)&i);
 245                 if (rc) {
 246                         CERROR("%s error sending notify %d\n",
 247                                obd->obd_name, rc);
 248                 }
 249         }
 250         obd_putref(obd);
 251
 252         return 0;
 253 }
 254
 255 static int lov_disconnect_obd(struct obd_device *obd, struct lov_tgt_desc *tgt)
 256 {
 257         struct lov_obd *lov = &obd->u.lov;
 258         struct obd_device *osc_obd;
 259         int rc;
 260
 261         osc_obd = class_exp2obd(tgt->ltd_exp);
 262         CDEBUG(D_CONFIG, "%s: disconnecting target %s\n",
 263                obd->obd_name, osc_obd ? osc_obd->obd_name : "NULL");
 264
 265         if (tgt->ltd_active) {
 266                 tgt->ltd_active = 0;
 267                 lov->desc.ld_active_tgt_count--;
 268                 tgt->ltd_exp->exp_obd->obd_inactive = 1;
 269         }
 270
 271         if (osc_obd) {
 272                 if (lov->lov_tgts_kobj)
 273                         sysfs_remove_link(lov->lov_tgts_kobj,
 274                                           osc_obd->obd_name);
 275
 276                 /* Pass it on to our clients.
 277                  * XXX This should be an argument to disconnect,
 278                  * XXX not a back-door flag on the OBD.  Ah well.
 279                  */
 280                 osc_obd->obd_force = obd->obd_force;
 281                 osc_obd->obd_fail = obd->obd_fail;
 282                 osc_obd->obd_no_recov = obd->obd_no_recov;
 283         }
 284
 285         obd_register_observer(osc_obd, NULL);
 286
 287         rc = obd_disconnect(tgt->ltd_exp);
 288         if (rc) {
 289                 CERROR("Target %s disconnect error %d\n",
 290                        tgt->ltd_uuid.uuid, rc);
 291                 rc = 0;
 292         }
 293
 294         tgt->ltd_exp = NULL;
 295         return 0;
 296 }
 297
 298 static int lov_disconnect(struct obd_export *exp)
 299 {
 300         struct obd_device *obd = class_exp2obd(exp);
 301         struct lov_obd *lov = &obd->u.lov;
 302         int i, rc;
 303
 304         if (!lov->lov_tgts)
 305                 goto out;
 306
 307         /* Only disconnect the underlying layers on the final disconnect. */
 308         lov->lov_connects--;
 309         if (lov->lov_connects != 0) {
 310                 /* why should there be more than 1 connect? */
 311                 CERROR("disconnect #%d\n", lov->lov_connects);
 312                 goto out;
 313         }
 314
 315         /* Let's hold another reference so lov_del_obd doesn't spin through
 316          * putref every time
 317          */
 318         obd_getref(obd);
 319
 320         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
 321                 if (lov->lov_tgts[i] && lov->lov_tgts[i]->ltd_exp) {
 322                         /* Disconnection is the last we know about an obd */
 323                         lov_del_target(obd, i, NULL, lov->lov_tgts[i]->ltd_gen);
 324                 }
 325         }
 326
 327         obd_putref(obd);
 328
 329 out:
 330         rc = class_disconnect(exp); /* bz 9811 */
 331         return rc;
 332 }
 333
 334 /* Error codes:
 335  *
 336  *  -EINVAL  : UUID can't be found in the LOV's target list
 337  *  -ENOTCONN: The UUID is found, but the target connection is bad (!)
 338  *  -EBADF   : The UUID is found, but the OBD is the wrong type (!)
 339  *  any >= 0 : is log target index
 340  */
 341 static int lov_set_osc_active(struct obd_device *obd, struct obd_uuid *uuid,
 342                               enum obd_notify_event ev)
 343 {
 344         struct lov_obd *lov = &obd->u.lov;
 345         struct lov_tgt_desc *tgt;
 346         int index, activate, active;
 347
 348         CDEBUG(D_INFO, "Searching in lov %p for uuid %s event(%d)\n",
 349                lov, uuid->uuid, ev);
 350
 351         obd_getref(obd);
 352         for (index = 0; index < lov->desc.ld_tgt_count; index++) {
 353                 tgt = lov->lov_tgts[index];
 354                 if (!tgt)
 355                         continue;
 356                 /*
 357                  * LU-642, initially inactive OSC could miss the obd_connect,
 358                  * we make up for it here.
 359                  */
 360                 if (ev == OBD_NOTIFY_ACTIVATE && !tgt->ltd_exp &&
 361                     obd_uuid_equals(uuid, &tgt->ltd_uuid)) {
 362                         struct obd_uuid lov_osc_uuid = {"LOV_OSC_UUID"};
 363
 364                         obd_connect(NULL, &tgt->ltd_exp, tgt->ltd_obd,
 365                                     &lov_osc_uuid, &lov->lov_ocd, NULL);
 366                 }
 367                 if (!tgt->ltd_exp)
 368                         continue;
 369
 370                 CDEBUG(D_INFO, "lov idx %d is %s conn %#llx\n",
 371                        index, obd_uuid2str(&tgt->ltd_uuid),
 372                        tgt->ltd_exp->exp_handle.h_cookie);
 373                 if (obd_uuid_equals(uuid, &tgt->ltd_uuid))
 374                         break;
 375         }
 376
 377         if (index == lov->desc.ld_tgt_count) {
 378                 index = -EINVAL;
 379                 goto out;
 380         }
 381
 382         if (ev == OBD_NOTIFY_DEACTIVATE || ev == OBD_NOTIFY_ACTIVATE) {
 383                 activate = (ev == OBD_NOTIFY_ACTIVATE) ? 1 : 0;
 384
 385                 if (lov->lov_tgts[index]->ltd_activate == activate) {
 386                         CDEBUG(D_INFO, "OSC %s already %sactivate!\n",
 387                                uuid->uuid, activate ? "" : "de");
 388                 } else {
 389                         lov->lov_tgts[index]->ltd_activate = activate;
 390                         CDEBUG(D_CONFIG, "%sactivate OSC %s\n",
 391                                activate ? "" : "de", obd_uuid2str(uuid));
 392                 }
 393
 394         } else if (ev == OBD_NOTIFY_INACTIVE || ev == OBD_NOTIFY_ACTIVE) {
 395                 active = (ev == OBD_NOTIFY_ACTIVE) ? 1 : 0;
 396
 397                 if (lov->lov_tgts[index]->ltd_active == active) {
 398                         CDEBUG(D_INFO, "OSC %s already %sactive!\n",
 399                                uuid->uuid, active ? "" : "in");
 400                         goto out;
 401                 }
 402                 CDEBUG(D_CONFIG, "Marking OSC %s %sactive\n",
 403                        obd_uuid2str(uuid), active ? "" : "in");
 404
 405                 lov->lov_tgts[index]->ltd_active = active;
 406                 if (active) {
 407                         lov->desc.ld_active_tgt_count++;
 408                         lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 0;
 409                 } else {
 410                         lov->desc.ld_active_tgt_count--;
 411                         lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 1;
 412                 }
 413         } else {
 414                 CERROR("Unknown event(%d) for uuid %s", ev, uuid->uuid);
 415         }
 416
 417  out:
 418         obd_putref(obd);
 419         return index;
 420 }
 421
 422 static int lov_notify(struct obd_device *obd, struct obd_device *watched,
 423                       enum obd_notify_event ev, void *data)
 424 {
 425         int rc = 0;
 426         struct lov_obd *lov = &obd->u.lov;
 427
 428         down_read(&lov->lov_notify_lock);
 429         if (!lov->lov_connects) {
 430                 up_read(&lov->lov_notify_lock);
 431                 return rc;
 432         }
 433
 434         if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE ||
 435             ev == OBD_NOTIFY_ACTIVATE || ev == OBD_NOTIFY_DEACTIVATE) {
 436                 struct obd_uuid *uuid;
 437
 438                 LASSERT(watched);
 439
 440                 if (strcmp(watched->obd_type->typ_name, LUSTRE_OSC_NAME)) {
 441                         up_read(&lov->lov_notify_lock);
 442                         CERROR("unexpected notification of %s %s!\n",
 443                                watched->obd_type->typ_name,
 444                                watched->obd_name);
 445                         return -EINVAL;
 446                 }
 447                 uuid = &watched->u.cli.cl_target_uuid;
 448
 449                 /* Set OSC as active before notifying the observer, so the
 450                  * observer can use the OSC normally.
 451                  */
 452                 rc = lov_set_osc_active(obd, uuid, ev);
 453                 if (rc < 0) {
 454                         up_read(&lov->lov_notify_lock);
 455                         CERROR("event(%d) of %s failed: %d\n", ev,
 456                                obd_uuid2str(uuid), rc);
 457                         return rc;
 458                 }
 459                 /* active event should be pass lov target index as data */
 460                 data = &rc;
 461         }
 462
 463         /* Pass the notification up the chain. */
 464         if (watched) {
 465                 rc = obd_notify_observer(obd, watched, ev, data);
 466         } else {
 467                 /* NULL watched means all osc's in the lov (only for syncs) */
 468                 /* sync event should be send lov idx as data */
 469                 struct lov_obd *lov = &obd->u.lov;
 470                 int i, is_sync;
 471
 472                 data = &i;
 473                 is_sync = (ev == OBD_NOTIFY_SYNC) ||
 474                           (ev == OBD_NOTIFY_SYNC_NONBLOCK);
 475
 476                 obd_getref(obd);
 477                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
 478                         if (!lov->lov_tgts[i])
 479                                 continue;
 480
 481                         /* don't send sync event if target not
 482                          * connected/activated
 483                          */
 484                         if (is_sync &&  !lov->lov_tgts[i]->ltd_active)
 485                                 continue;
 486
 487                         rc = obd_notify_observer(obd, lov->lov_tgts[i]->ltd_obd,
 488                                                  ev, data);
 489                         if (rc) {
 490                                 CERROR("%s: notify %s of %s failed %d\n",
 491                                        obd->obd_name,
 492                                        obd->obd_observer->obd_name,
 493                                        lov->lov_tgts[i]->ltd_obd->obd_name,
 494                                        rc);
 495                         }
 496                 }
 497                 obd_putref(obd);
 498         }
 499
 500         up_read(&lov->lov_notify_lock);
 501         return rc;
 502 }
 503
 504 static int lov_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
 505                           __u32 index, int gen, int active)
 506 {
 507         struct lov_obd *lov = &obd->u.lov;
 508         struct lov_tgt_desc *tgt;
 509         struct obd_device *tgt_obd;
 510         int rc;
 511
 512         CDEBUG(D_CONFIG, "uuid:%s idx:%d gen:%d active:%d\n",
 513                uuidp->uuid, index, gen, active);
 514
 515         if (gen <= 0) {
 516                 CERROR("request to add OBD %s with invalid generation: %d\n",
 517                        uuidp->uuid, gen);
 518                 return -EINVAL;
 519         }
 520
 521         tgt_obd = class_find_client_obd(uuidp, LUSTRE_OSC_NAME,
 522                                         &obd->obd_uuid);
 523         if (!tgt_obd)
 524                 return -EINVAL;
 525
 526         mutex_lock(&lov->lov_lock);
 527
 528         if ((index < lov->lov_tgt_size) && lov->lov_tgts[index]) {
 529                 tgt = lov->lov_tgts[index];
 530                 CERROR("UUID %s already assigned at LOV target index %d\n",
 531                        obd_uuid2str(&tgt->ltd_uuid), index);
 532                 mutex_unlock(&lov->lov_lock);
 533                 return -EEXIST;
 534         }
 535
 536         if (index >= lov->lov_tgt_size) {
 537                 /* We need to reallocate the lov target array. */
 538                 struct lov_tgt_desc **newtgts, **old = NULL;
 539                 __u32 newsize, oldsize = 0;
 540
 541                 newsize = max_t(__u32, lov->lov_tgt_size, 2);
 542                 while (newsize < index + 1)
 543                         newsize <<= 1;
 544                 newtgts = kcalloc(newsize, sizeof(*newtgts), GFP_NOFS);
 545                 if (!newtgts) {
 546                         mutex_unlock(&lov->lov_lock);
 547                         return -ENOMEM;
 548                 }
 549
 550                 if (lov->lov_tgt_size) {
 551                         memcpy(newtgts, lov->lov_tgts, sizeof(*newtgts) *
 552                                lov->lov_tgt_size);
 553                         old = lov->lov_tgts;
 554                         oldsize = lov->lov_tgt_size;
 555                 }
 556
 557                 lov->lov_tgts = newtgts;
 558                 lov->lov_tgt_size = newsize;
 559                 smp_rmb();
 560                 kfree(old);
 561
 562                 CDEBUG(D_CONFIG, "tgts: %p size: %d\n",
 563                        lov->lov_tgts, lov->lov_tgt_size);
 564         }
 565
 566         tgt = kzalloc(sizeof(*tgt), GFP_NOFS);
 567         if (!tgt) {
 568                 mutex_unlock(&lov->lov_lock);
 569                 return -ENOMEM;
 570         }
 571
 572         rc = lov_ost_pool_add(&lov->lov_packed, index, lov->lov_tgt_size);
 573         if (rc) {
 574                 mutex_unlock(&lov->lov_lock);
 575                 kfree(tgt);
 576                 return rc;
 577         }
 578
 579         tgt->ltd_uuid = *uuidp;
 580         tgt->ltd_obd = tgt_obd;
 581         /* XXX - add a sanity check on the generation number. */
 582         tgt->ltd_gen = gen;
 583         tgt->ltd_index = index;
 584         tgt->ltd_activate = active;
 585         lov->lov_tgts[index] = tgt;
 586         if (index >= lov->desc.ld_tgt_count)
 587                 lov->desc.ld_tgt_count = index + 1;
 588
 589         mutex_unlock(&lov->lov_lock);
 590
 591         CDEBUG(D_CONFIG, "idx=%d ltd_gen=%d ld_tgt_count=%d\n",
 592                index, tgt->ltd_gen, lov->desc.ld_tgt_count);
 593
 594         rc = obd_notify(obd, tgt_obd, OBD_NOTIFY_CREATE, &index);
 595
 596         if (lov->lov_connects == 0) {
 597                 /* lov_connect hasn't been called yet. We'll do the
 598                  * lov_connect_obd on this target when that fn first runs,
 599                  * because we don't know the connect flags yet.
 600                  */
 601                 return 0;
 602         }
 603
 604         obd_getref(obd);
 605
 606         rc = lov_connect_obd(obd, index, active, &lov->lov_ocd);
 607         if (rc)
 608                 goto out;
 609
 610         /* connect to administrative disabled ost */
 611         if (!tgt->ltd_exp) {
 612                 rc = 0;
 613                 goto out;
 614         }
 615
 616         if (lov->lov_cache) {
 617                 rc = obd_set_info_async(NULL, tgt->ltd_exp,
 618                                         sizeof(KEY_CACHE_SET), KEY_CACHE_SET,
 619                                         sizeof(struct cl_client_cache),
 620                                         lov->lov_cache, NULL);
 621                 if (rc < 0)
 622                         goto out;
 623         }
 624
 625         rc = lov_notify(obd, tgt->ltd_exp->exp_obd,
 626                         active ? OBD_NOTIFY_CONNECT : OBD_NOTIFY_INACTIVE,
 627                         (void *)&index);
 628
 629 out:
 630         if (rc) {
 631                 CERROR("add failed (%d), deleting %s\n", rc,
 632                        obd_uuid2str(&tgt->ltd_uuid));
 633                 lov_del_target(obd, index, NULL, 0);
 634         }
 635         obd_putref(obd);
 636         return rc;
 637 }
 638
 639 /* Schedule a target for deletion */
 640 int lov_del_target(struct obd_device *obd, __u32 index,
 641                    struct obd_uuid *uuidp, int gen)
 642 {
 643         struct lov_obd *lov = &obd->u.lov;
 644         int count = lov->desc.ld_tgt_count;
 645         int rc = 0;
 646
 647         if (index >= count) {
 648                 CERROR("LOV target index %d >= number of LOV OBDs %d.\n",
 649                        index, count);
 650                 return -EINVAL;
 651         }
 652
 653         /* to make sure there's no ongoing lov_notify() now */
 654         down_write(&lov->lov_notify_lock);
 655         obd_getref(obd);
 656
 657         if (!lov->lov_tgts[index]) {
 658                 CERROR("LOV target at index %d is not setup.\n", index);
 659                 rc = -EINVAL;
 660                 goto out;
 661         }
 662
 663         if (uuidp && !obd_uuid_equals(uuidp, &lov->lov_tgts[index]->ltd_uuid)) {
 664                 CERROR("LOV target UUID %s at index %d doesn't match %s.\n",
 665                        lov_uuid2str(lov, index), index,
 666                        obd_uuid2str(uuidp));
 667                 rc = -EINVAL;
 668                 goto out;
 669         }
 670
 671         CDEBUG(D_CONFIG, "uuid: %s idx: %d gen: %d exp: %p active: %d\n",
 672                lov_uuid2str(lov, index), index,
 673                lov->lov_tgts[index]->ltd_gen, lov->lov_tgts[index]->ltd_exp,
 674                lov->lov_tgts[index]->ltd_active);
 675
 676         lov->lov_tgts[index]->ltd_reap = 1;
 677         lov->lov_death_row++;
 678         /* we really delete it from obd_putref */
 679 out:
 680         obd_putref(obd);
 681         up_write(&lov->lov_notify_lock);
 682
 683         return rc;
 684 }
 685
 686 static void __lov_del_obd(struct obd_device *obd, struct lov_tgt_desc *tgt)
 687 {
 688         struct obd_device *osc_obd;
 689
 690         LASSERT(tgt);
 691         LASSERT(tgt->ltd_reap);
 692
 693         osc_obd = class_exp2obd(tgt->ltd_exp);
 694
 695         CDEBUG(D_CONFIG, "Removing tgt %s : %s\n",
 696                tgt->ltd_uuid.uuid,
 697                osc_obd ? osc_obd->obd_name : "<no obd>");
 698
 699         if (tgt->ltd_exp)
 700                 lov_disconnect_obd(obd, tgt);
 701
 702         kfree(tgt);
 703
 704         /* Manual cleanup - no cleanup logs to clean up the osc's.  We must
 705          * do it ourselves. And we can't do it from lov_cleanup,
 706          * because we just lost our only reference to it.
 707          */
 708         if (osc_obd)
 709                 class_manual_cleanup(osc_obd);
 710 }
 711
 712 void lov_fix_desc_stripe_size(__u64 *val)
 713 {
 714         if (*val < LOV_MIN_STRIPE_SIZE) {
 715                 if (*val != 0)
 716                         LCONSOLE_INFO("Increasing default stripe size to minimum %u\n",
 717                                       LOV_DESC_STRIPE_SIZE_DEFAULT);
 718                 *val = LOV_DESC_STRIPE_SIZE_DEFAULT;
 719         } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) {
 720                 *val &= ~(LOV_MIN_STRIPE_SIZE - 1);
 721                 LCONSOLE_WARN("Changing default stripe size to %llu (a multiple of %u)\n",
 722                               *val, LOV_MIN_STRIPE_SIZE);
 723         }
 724 }
 725
 726 void lov_fix_desc_stripe_count(__u32 *val)
 727 {
 728         if (*val == 0)
 729                 *val = 1;
 730 }
 731
 732 void lov_fix_desc_pattern(__u32 *val)
 733 {
 734         /* from lov_setstripe */
 735         if ((*val != 0) && (*val != LOV_PATTERN_RAID0)) {
 736                 LCONSOLE_WARN("Unknown stripe pattern: %#x\n", *val);
 737                 *val = 0;
 738         }
 739 }
 740
 741 void lov_fix_desc_qos_maxage(__u32 *val)
 742 {
 743         if (*val == 0)
 744                 *val = LOV_DESC_QOS_MAXAGE_DEFAULT;
 745 }
 746
 747 void lov_fix_desc(struct lov_desc *desc)
 748 {
 749         lov_fix_desc_stripe_size(&desc->ld_default_stripe_size);
 750         lov_fix_desc_stripe_count(&desc->ld_default_stripe_count);
 751         lov_fix_desc_pattern(&desc->ld_pattern);
 752         lov_fix_desc_qos_maxage(&desc->ld_qos_maxage);
 753 }
 754
 755 int lov_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
 756 {
 757         struct lprocfs_static_vars lvars = { NULL };
 758         struct lov_desc *desc;
 759         struct lov_obd *lov = &obd->u.lov;
 760         int rc;
 761
 762         if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
 763                 CERROR("LOV setup requires a descriptor\n");
 764                 return -EINVAL;
 765         }
 766
 767         desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
 768
 769         if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
 770                 CERROR("descriptor size wrong: %d > %d\n",
 771                        (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
 772                 return -EINVAL;
 773         }
 774
 775         if (desc->ld_magic != LOV_DESC_MAGIC) {
 776                 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
 777                         CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
 778                                obd->obd_name, desc);
 779                         lustre_swab_lov_desc(desc);
 780                 } else {
 781                         CERROR("%s: Bad lov desc magic: %#x\n",
 782                                obd->obd_name, desc->ld_magic);
 783                         return -EINVAL;
 784                 }
 785         }
 786
 787         lov_fix_desc(desc);
 788
 789         desc->ld_active_tgt_count = 0;
 790         lov->desc = *desc;
 791         lov->lov_tgt_size = 0;
 792
 793         mutex_init(&lov->lov_lock);
 794         atomic_set(&lov->lov_refcount, 0);
 795         lov->lov_sp_me = LUSTRE_SP_CLI;
 796
 797         init_rwsem(&lov->lov_notify_lock);
 798
 799         lov->lov_pools_hash_body = cfs_hash_create("POOLS", HASH_POOLS_CUR_BITS,
 800                                                    HASH_POOLS_MAX_BITS,
 801                                                    HASH_POOLS_BKT_BITS, 0,
 802                                                    CFS_HASH_MIN_THETA,
 803                                                    CFS_HASH_MAX_THETA,
 804                                                    &pool_hash_operations,
 805                                                    CFS_HASH_DEFAULT);
 806         INIT_LIST_HEAD(&lov->lov_pool_list);
 807         lov->lov_pool_count = 0;
 808         rc = lov_ost_pool_init(&lov->lov_packed, 0);
 809         if (rc)
 810                 goto out;
 811
 812         lprocfs_lov_init_vars(&lvars);
 813         lprocfs_obd_setup(obd, lvars.obd_vars, lvars.sysfs_vars);
 814
 815         rc = ldebugfs_seq_create(obd->obd_debugfs_entry, "target_obd",
 816                                  0444, &lov_proc_target_fops, obd);
 817         if (rc)
 818                 CWARN("Error adding the target_obd file\n");
 819
 820         lov->lov_pool_debugfs_entry = ldebugfs_register("pools",
 821                                                      obd->obd_debugfs_entry,
 822                                                      NULL, NULL);
 823         return 0;
 824
 825 out:
 826         return rc;
 827 }
 828
 829 static int lov_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
 830 {
 831         struct lov_obd *lov = &obd->u.lov;
 832
 833         switch (stage) {
 834         case OBD_CLEANUP_EARLY: {
 835                 int i;
 836
 837                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
 838                         if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_active)
 839                                 continue;
 840                         obd_precleanup(class_exp2obd(lov->lov_tgts[i]->ltd_exp),
 841                                        OBD_CLEANUP_EARLY);
 842                 }
 843                 break;
 844         }
 845         default:
 846                 break;
 847         }
 848
 849         return 0;
 850 }
 851
 852 static int lov_cleanup(struct obd_device *obd)
 853 {
 854         struct lov_obd *lov = &obd->u.lov;
 855         struct list_head *pos, *tmp;
 856         struct pool_desc *pool;
 857
 858         list_for_each_safe(pos, tmp, &lov->lov_pool_list) {
 859                 pool = list_entry(pos, struct pool_desc, pool_list);
 860                 /* free pool structs */
 861                 CDEBUG(D_INFO, "delete pool %p\n", pool);
 862                 /* In the function below, .hs_keycmp resolves to
 863                  * pool_hashkey_keycmp()
 864                  */
 865                 /* coverity[overrun-buffer-val] */
 866                 lov_pool_del(obd, pool->pool_name);
 867         }
 868         cfs_hash_putref(lov->lov_pools_hash_body);
 869         lov_ost_pool_free(&lov->lov_packed);
 870
 871         lprocfs_obd_cleanup(obd);
 872         if (lov->lov_tgts) {
 873                 int i;
 874
 875                 obd_getref(obd);
 876                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
 877                         if (!lov->lov_tgts[i])
 878                                 continue;
 879
 880                         /* Inactive targets may never have connected */
 881                         if (lov->lov_tgts[i]->ltd_active ||
 882                             atomic_read(&lov->lov_refcount))
 883                             /* We should never get here - these
 884                              * should have been removed in the
 885                              * disconnect.
 886                              */
 887                                 CERROR("lov tgt %d not cleaned! deathrow=%d, lovrc=%d\n",
 888                                        i, lov->lov_death_row,
 889                                        atomic_read(&lov->lov_refcount));
 890                         lov_del_target(obd, i, NULL, 0);
 891                 }
 892                 obd_putref(obd);
 893                 kfree(lov->lov_tgts);
 894                 lov->lov_tgt_size = 0;
 895         }
 896
 897         if (lov->lov_cache) {
 898                 cl_cache_decref(lov->lov_cache);
 899                 lov->lov_cache = NULL;
 900         }
 901
 902         return 0;
 903 }
 904
 905 int lov_process_config_base(struct obd_device *obd, struct lustre_cfg *lcfg,
 906                             __u32 *indexp, int *genp)
 907 {
 908         struct obd_uuid obd_uuid;
 909         int cmd;
 910         int rc = 0;
 911
 912         switch (cmd = lcfg->lcfg_command) {
 913         case LCFG_LOV_ADD_OBD:
 914         case LCFG_LOV_ADD_INA:
 915         case LCFG_LOV_DEL_OBD: {
 916                 __u32 index;
 917                 int gen;
 918                 /* lov_modify_tgts add  0:lov_mdsA  1:ost1_UUID  2:0  3:1 */
 919                 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid)) {
 920                         rc = -EINVAL;
 921                         goto out;
 922                 }
 923
 924                 obd_str2uuid(&obd_uuid,  lustre_cfg_buf(lcfg, 1));
 925
 926                 rc = kstrtoint(lustre_cfg_buf(lcfg, 2), 10, indexp);
 927                 if (rc < 0)
 928                         goto out;
 929                 rc = kstrtoint(lustre_cfg_buf(lcfg, 3), 10, genp);
 930                 if (rc < 0)
 931                         goto out;
 932                 index = *indexp;
 933                 gen = *genp;
 934                 if (cmd == LCFG_LOV_ADD_OBD)
 935                         rc = lov_add_target(obd, &obd_uuid, index, gen, 1);
 936                 else if (cmd == LCFG_LOV_ADD_INA)
 937                         rc = lov_add_target(obd, &obd_uuid, index, gen, 0);
 938                 else
 939                         rc = lov_del_target(obd, index, &obd_uuid, gen);
 940                 goto out;
 941         }
 942         case LCFG_PARAM: {
 943                 struct lprocfs_static_vars lvars = { NULL };
 944                 struct lov_desc *desc = &obd->u.lov.desc;
 945
 946                 if (!desc) {
 947                         rc = -EINVAL;
 948                         goto out;
 949                 }
 950
 951                 lprocfs_lov_init_vars(&lvars);
 952
 953                 rc = class_process_proc_param(PARAM_LOV, lvars.obd_vars,
 954                                               lcfg, obd);
 955                 if (rc > 0)
 956                         rc = 0;
 957                 goto out;
 958         }
 959         case LCFG_POOL_NEW:
 960         case LCFG_POOL_ADD:
 961         case LCFG_POOL_DEL:
 962         case LCFG_POOL_REM:
 963                 goto out;
 964
 965         default: {
 966                 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
 967                 rc = -EINVAL;
 968                 goto out;
 969         }
 970         }
 971 out:
 972         return rc;
 973 }
 974
 975 #define ASSERT_LSM_MAGIC(lsmp)                                            \
 976 do {                                                                        \
 977         LASSERT((lsmp));                                                \
 978         LASSERTF(((lsmp)->lsm_magic == LOV_MAGIC_V1 ||                    \
 979                  (lsmp)->lsm_magic == LOV_MAGIC_V3),                        \
 980                  "%p->lsm_magic=%x\n", (lsmp), (lsmp)->lsm_magic);            \
 981 } while (0)
 982
 983 static int lov_getattr_interpret(struct ptlrpc_request_set *rqset,
 984                                  void *data, int rc)
 985 {
 986         struct lov_request_set *lovset = (struct lov_request_set *)data;
 987         int err;
 988
 989         /* don't do attribute merge if this async op failed */
 990         if (rc)
 991                 atomic_set(&lovset->set_completes, 0);
 992         err = lov_fini_getattr_set(lovset);
 993         return rc ? rc : err;
 994 }
 995
 996 static int lov_getattr_async(struct obd_export *exp, struct obd_info *oinfo,
 997                              struct ptlrpc_request_set *rqset)
 998 {
 999         struct lov_request_set *lovset;
1000         struct lov_obd *lov;
1001         struct lov_request *req;
1002         int rc = 0, err;
1003
1004         LASSERT(oinfo);
1005         ASSERT_LSM_MAGIC(oinfo->oi_md);
1006
1007         if (!exp || !exp->exp_obd)
1008                 return -ENODEV;
1009
1010         lov = &exp->exp_obd->u.lov;
1011
1012         rc = lov_prep_getattr_set(exp, oinfo, &lovset);
1013         if (rc)
1014                 return rc;
1015
1016         CDEBUG(D_INFO, "objid "DOSTID": %ux%u byte stripes\n",
1017                POSTID(&oinfo->oi_md->lsm_oi), oinfo->oi_md->lsm_stripe_count,
1018                oinfo->oi_md->lsm_stripe_size);
1019
1020         list_for_each_entry(req, &lovset->set_list, rq_link) {
1021                 CDEBUG(D_INFO, "objid " DOSTID "[%d] has subobj " DOSTID " at idx%u\n",
1022                        POSTID(&oinfo->oi_oa->o_oi), req->rq_stripe,
1023                        POSTID(&req->rq_oi.oi_oa->o_oi), req->rq_idx);
1024                 rc = obd_getattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1025                                        &req->rq_oi, rqset);
1026                 if (rc) {
1027                         CERROR("%s: getattr objid "DOSTID" subobj"
1028                                DOSTID" on OST idx %d: rc = %d\n",
1029                                exp->exp_obd->obd_name,
1030                                POSTID(&oinfo->oi_oa->o_oi),
1031                                POSTID(&req->rq_oi.oi_oa->o_oi),
1032                                req->rq_idx, rc);
1033                         goto out;
1034                 }
1035         }
1036
1037         if (!list_empty(&rqset->set_requests)) {
1038                 LASSERT(rc == 0);
1039                 LASSERT(!rqset->set_interpret);
1040                 rqset->set_interpret = lov_getattr_interpret;
1041                 rqset->set_arg = (void *)lovset;
1042                 return rc;
1043         }
1044 out:
1045         if (rc)
1046                 atomic_set(&lovset->set_completes, 0);
1047         err = lov_fini_getattr_set(lovset);
1048         return rc ? rc : err;
1049 }
1050
1051 static int lov_setattr_interpret(struct ptlrpc_request_set *rqset,
1052                                  void *data, int rc)
1053 {
1054         struct lov_request_set *lovset = (struct lov_request_set *)data;
1055         int err;
1056
1057         if (rc)
1058                 atomic_set(&lovset->set_completes, 0);
1059         err = lov_fini_setattr_set(lovset);
1060         return rc ? rc : err;
1061 }
1062
1063 /* If @oti is given, the request goes from MDS and responses from OSTs are not
1064  * needed. Otherwise, a client is waiting for responses.
1065  */
1066 static int lov_setattr_async(struct obd_export *exp, struct obd_info *oinfo,
1067                              struct obd_trans_info *oti,
1068                              struct ptlrpc_request_set *rqset)
1069 {
1070         struct lov_request_set *set;
1071         struct lov_request *req;
1072         struct lov_obd *lov;
1073         int rc = 0;
1074
1075         LASSERT(oinfo);
1076         ASSERT_LSM_MAGIC(oinfo->oi_md);
1077         if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE) {
1078                 LASSERT(oti);
1079                 LASSERT(oti->oti_logcookies);
1080         }
1081
1082         if (!exp || !exp->exp_obd)
1083                 return -ENODEV;
1084
1085         lov = &exp->exp_obd->u.lov;
1086         rc = lov_prep_setattr_set(exp, oinfo, oti, &set);
1087         if (rc)
1088                 return rc;
1089
1090         CDEBUG(D_INFO, "objid "DOSTID": %ux%u byte stripes\n",
1091                POSTID(&oinfo->oi_md->lsm_oi),
1092                oinfo->oi_md->lsm_stripe_count,
1093                oinfo->oi_md->lsm_stripe_size);
1094
1095         list_for_each_entry(req, &set->set_list, rq_link) {
1096                 if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE)
1097                         oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1098
1099                 CDEBUG(D_INFO, "objid " DOSTID "[%d] has subobj " DOSTID " at idx%u\n",
1100                        POSTID(&oinfo->oi_oa->o_oi), req->rq_stripe,
1101                        POSTID(&req->rq_oi.oi_oa->o_oi), req->rq_idx);
1102
1103                 rc = obd_setattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1104                                        &req->rq_oi, oti, rqset);
1105                 if (rc) {
1106                         CERROR("error: setattr objid "DOSTID" subobj"
1107                                DOSTID" on OST idx %d: rc = %d\n",
1108                                POSTID(&set->set_oi->oi_oa->o_oi),
1109                                POSTID(&req->rq_oi.oi_oa->o_oi),
1110                                req->rq_idx, rc);
1111                         break;
1112                 }
1113         }
1114
1115         /* If we are not waiting for responses on async requests, return. */
1116         if (rc || !rqset || list_empty(&rqset->set_requests)) {
1117                 int err;
1118
1119                 if (rc)
1120                         atomic_set(&set->set_completes, 0);
1121                 err = lov_fini_setattr_set(set);
1122                 return rc ? rc : err;
1123         }
1124
1125         LASSERT(!rqset->set_interpret);
1126         rqset->set_interpret = lov_setattr_interpret;
1127         rqset->set_arg = (void *)set;
1128
1129         return 0;
1130 }
1131
1132 int lov_statfs_interpret(struct ptlrpc_request_set *rqset, void *data, int rc)
1133 {
1134         struct lov_request_set *lovset = (struct lov_request_set *)data;
1135         int err;
1136
1137         if (rc)
1138                 atomic_set(&lovset->set_completes, 0);
1139
1140         err = lov_fini_statfs_set(lovset);
1141         return rc ? rc : err;
1142 }
1143
1144 static int lov_statfs_async(struct obd_export *exp, struct obd_info *oinfo,
1145                             __u64 max_age, struct ptlrpc_request_set *rqset)
1146 {
1147         struct obd_device      *obd = class_exp2obd(exp);
1148         struct lov_request_set *set;
1149         struct lov_request *req;
1150         struct lov_obd *lov;
1151         int rc = 0;
1152
1153         LASSERT(oinfo->oi_osfs);
1154
1155         lov = &obd->u.lov;
1156         rc = lov_prep_statfs_set(obd, oinfo, &set);
1157         if (rc)
1158                 return rc;
1159
1160         list_for_each_entry(req, &set->set_list, rq_link) {
1161                 rc = obd_statfs_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1162                                       &req->rq_oi, max_age, rqset);
1163                 if (rc)
1164                         break;
1165         }
1166
1167         if (rc || list_empty(&rqset->set_requests)) {
1168                 int err;
1169
1170                 if (rc)
1171                         atomic_set(&set->set_completes, 0);
1172                 err = lov_fini_statfs_set(set);
1173                 return rc ? rc : err;
1174         }
1175
1176         LASSERT(!rqset->set_interpret);
1177         rqset->set_interpret = lov_statfs_interpret;
1178         rqset->set_arg = (void *)set;
1179         return 0;
1180 }
1181
1182 static int lov_statfs(const struct lu_env *env, struct obd_export *exp,
1183                       struct obd_statfs *osfs, __u64 max_age, __u32 flags)
1184 {
1185         struct ptlrpc_request_set *set = NULL;
1186         struct obd_info oinfo = { };
1187         int rc = 0;
1188
1189         /* for obdclass we forbid using obd_statfs_rqset, but prefer using async
1190          * statfs requests
1191          */
1192         set = ptlrpc_prep_set();
1193         if (!set)
1194                 return -ENOMEM;
1195
1196         oinfo.oi_osfs = osfs;
1197         oinfo.oi_flags = flags;
1198         rc = lov_statfs_async(exp, &oinfo, max_age, set);
1199         if (rc == 0)
1200                 rc = ptlrpc_set_wait(set);
1201         ptlrpc_set_destroy(set);
1202
1203         return rc;
1204 }
1205
1206 static int lov_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1207                          void *karg, void __user *uarg)
1208 {
1209         struct obd_device *obddev = class_exp2obd(exp);
1210         struct lov_obd *lov = &obddev->u.lov;
1211         int i = 0, rc = 0, count = lov->desc.ld_tgt_count;
1212         struct obd_uuid *uuidp;
1213
1214         switch (cmd) {
1215         case IOC_OBD_STATFS: {
1216                 struct obd_ioctl_data *data = karg;
1217                 struct obd_device *osc_obd;
1218                 struct obd_statfs stat_buf = {0};
1219                 __u32 index;
1220                 __u32 flags;
1221
1222                 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
1223                 if (index >= count)
1224                         return -ENODEV;
1225
1226                 if (!lov->lov_tgts[index])
1227                         /* Try again with the next index */
1228                         return -EAGAIN;
1229                 if (!lov->lov_tgts[index]->ltd_active)
1230                         return -ENODATA;
1231
1232                 osc_obd = class_exp2obd(lov->lov_tgts[index]->ltd_exp);
1233                 if (!osc_obd)
1234                         return -EINVAL;
1235
1236                 /* copy UUID */
1237                 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(osc_obd),
1238                                  min((int)data->ioc_plen2,
1239                                      (int)sizeof(struct obd_uuid))))
1240                         return -EFAULT;
1241
1242                 memcpy(&flags, data->ioc_inlbuf1, sizeof(__u32));
1243                 flags = flags & LL_STATFS_NODELAY ? OBD_STATFS_NODELAY : 0;
1244
1245                 /* got statfs data */
1246                 rc = obd_statfs(NULL, lov->lov_tgts[index]->ltd_exp, &stat_buf,
1247                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1248                                 flags);
1249                 if (rc)
1250                         return rc;
1251                 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
1252                                  min((int)data->ioc_plen1,
1253                                      (int)sizeof(stat_buf))))
1254                         return -EFAULT;
1255                 break;
1256         }
1257         case OBD_IOC_LOV_GET_CONFIG: {
1258                 struct obd_ioctl_data *data;
1259                 struct lov_desc *desc;
1260                 char *buf = NULL;
1261                 __u32 *genp;
1262
1263                 len = 0;
1264                 if (obd_ioctl_getdata(&buf, &len, uarg))
1265                         return -EINVAL;
1266
1267                 data = (struct obd_ioctl_data *)buf;
1268
1269                 if (sizeof(*desc) > data->ioc_inllen1) {
1270                         obd_ioctl_freedata(buf, len);
1271                         return -EINVAL;
1272                 }
1273
1274                 if (sizeof(uuidp->uuid) * count > data->ioc_inllen2) {
1275                         obd_ioctl_freedata(buf, len);
1276                         return -EINVAL;
1277                 }
1278
1279                 if (sizeof(__u32) * count > data->ioc_inllen3) {
1280                         obd_ioctl_freedata(buf, len);
1281                         return -EINVAL;
1282                 }
1283
1284                 desc = (struct lov_desc *)data->ioc_inlbuf1;
1285                 memcpy(desc, &lov->desc, sizeof(*desc));
1286
1287                 uuidp = (struct obd_uuid *)data->ioc_inlbuf2;
1288                 genp = (__u32 *)data->ioc_inlbuf3;
1289                 /* the uuid will be empty for deleted OSTs */
1290                 for (i = 0; i < count; i++, uuidp++, genp++) {
1291                         if (!lov->lov_tgts[i])
1292                                 continue;
1293                         *uuidp = lov->lov_tgts[i]->ltd_uuid;
1294                         *genp = lov->lov_tgts[i]->ltd_gen;
1295                 }
1296
1297                 if (copy_to_user(uarg, buf, len))
1298                         rc = -EFAULT;
1299                 obd_ioctl_freedata(buf, len);
1300                 break;
1301         }
1302         case OBD_IOC_QUOTACTL: {
1303                 struct if_quotactl *qctl = karg;
1304                 struct lov_tgt_desc *tgt = NULL;
1305                 struct obd_quotactl *oqctl;
1306
1307                 if (qctl->qc_valid == QC_OSTIDX) {
1308                         if (count <= qctl->qc_idx)
1309                                 return -EINVAL;
1310
1311                         tgt = lov->lov_tgts[qctl->qc_idx];
1312                         if (!tgt || !tgt->ltd_exp)
1313                                 return -EINVAL;
1314                 } else if (qctl->qc_valid == QC_UUID) {
1315                         for (i = 0; i < count; i++) {
1316                                 tgt = lov->lov_tgts[i];
1317                                 if (!tgt ||
1318                                     !obd_uuid_equals(&tgt->ltd_uuid,
1319                                                      &qctl->obd_uuid))
1320                                         continue;
1321
1322                                 if (!tgt->ltd_exp)
1323                                         return -EINVAL;
1324
1325                                 break;
1326                         }
1327                 } else {
1328                         return -EINVAL;
1329                 }
1330
1331                 if (i >= count)
1332                         return -EAGAIN;
1333
1334                 LASSERT(tgt && tgt->ltd_exp);
1335                 oqctl = kzalloc(sizeof(*oqctl), GFP_NOFS);
1336                 if (!oqctl)
1337                         return -ENOMEM;
1338
1339                 QCTL_COPY(oqctl, qctl);
1340                 rc = obd_quotactl(tgt->ltd_exp, oqctl);
1341                 if (rc == 0) {
1342                         QCTL_COPY(qctl, oqctl);
1343                         qctl->qc_valid = QC_OSTIDX;
1344                         qctl->obd_uuid = tgt->ltd_uuid;
1345                 }
1346                 kfree(oqctl);
1347                 break;
1348         }
1349         default: {
1350                 int set = 0;
1351
1352                 if (count == 0)
1353                         return -ENOTTY;
1354
1355                 for (i = 0; i < count; i++) {
1356                         int err;
1357                         struct obd_device *osc_obd;
1358
1359                         /* OST was disconnected */
1360                         if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
1361                                 continue;
1362
1363                         /* ll_umount_begin() sets force flag but for lov, not
1364                          * osc. Let's pass it through
1365                          */
1366                         osc_obd = class_exp2obd(lov->lov_tgts[i]->ltd_exp);
1367                         osc_obd->obd_force = obddev->obd_force;
1368                         err = obd_iocontrol(cmd, lov->lov_tgts[i]->ltd_exp,
1369                                             len, karg, uarg);
1370                         if (err == -ENODATA && cmd == OBD_IOC_POLL_QUOTACHECK)
1371                                 return err;
1372                         if (err) {
1373                                 if (lov->lov_tgts[i]->ltd_active) {
1374                                         CDEBUG(err == -ENOTTY ?
1375                                                D_IOCTL : D_WARNING,
1376                                                "iocontrol OSC %s on OST idx %d cmd %x: err = %d\n",
1377                                                lov_uuid2str(lov, i),
1378                                                i, cmd, err);
1379                                         if (!rc)
1380                                                 rc = err;
1381                                 }
1382                         } else {
1383                                 set = 1;
1384                         }
1385                 }
1386                 if (!set && !rc)
1387                         rc = -EIO;
1388         }
1389         }
1390
1391         return rc;
1392 }
1393
1394 #define FIEMAP_BUFFER_SIZE 4096
1395
1396 /**
1397  * Non-zero fe_logical indicates that this is a continuation FIEMAP
1398  * call. The local end offset and the device are sent in the first
1399  * fm_extent. This function calculates the stripe number from the index.
1400  * This function returns a stripe_no on which mapping is to be restarted.
1401  *
1402  * This function returns fm_end_offset which is the in-OST offset at which
1403  * mapping should be restarted. If fm_end_offset=0 is returned then caller
1404  * will re-calculate proper offset in next stripe.
1405  * Note that the first extent is passed to lov_get_info via the value field.
1406  *
1407  * \param fiemap fiemap request header
1408  * \param lsm striping information for the file
1409  * \param fm_start logical start of mapping
1410  * \param fm_end logical end of mapping
1411  * \param start_stripe starting stripe will be returned in this
1412  */
1413 static u64 fiemap_calc_fm_end_offset(struct ll_user_fiemap *fiemap,
1414                                      struct lov_stripe_md *lsm, u64 fm_start,
1415                                      u64 fm_end, int *start_stripe)
1416 {
1417         u64 local_end = fiemap->fm_extents[0].fe_logical;
1418         u64 lun_start, lun_end;
1419         u64 fm_end_offset;
1420         int stripe_no = -1, i;
1421
1422         if (fiemap->fm_extent_count == 0 ||
1423             fiemap->fm_extents[0].fe_logical == 0)
1424                 return 0;
1425
1426         /* Find out stripe_no from ost_index saved in the fe_device */
1427         for (i = 0; i < lsm->lsm_stripe_count; i++) {
1428                 struct lov_oinfo *oinfo = lsm->lsm_oinfo[i];
1429
1430                 if (lov_oinfo_is_dummy(oinfo))
1431                         continue;
1432
1433                 if (oinfo->loi_ost_idx == fiemap->fm_extents[0].fe_device) {
1434                         stripe_no = i;
1435                         break;
1436                 }
1437         }
1438         if (stripe_no == -1)
1439                 return -EINVAL;
1440
1441         /* If we have finished mapping on previous device, shift logical
1442          * offset to start of next device
1443          */
1444         if ((lov_stripe_intersects(lsm, stripe_no, fm_start, fm_end,
1445                                    &lun_start, &lun_end)) != 0 &&
1446                                    local_end < lun_end) {
1447                 fm_end_offset = local_end;
1448                 *start_stripe = stripe_no;
1449         } else {
1450                 /* This is a special value to indicate that caller should
1451                  * calculate offset in next stripe.
1452                  */
1453                 fm_end_offset = 0;
1454                 *start_stripe = (stripe_no + 1) % lsm->lsm_stripe_count;
1455         }
1456
1457         return fm_end_offset;
1458 }
1459
1460 /**
1461  * We calculate on which OST the mapping will end. If the length of mapping
1462  * is greater than (stripe_size * stripe_count) then the last_stripe will
1463  * will be one just before start_stripe. Else we check if the mapping
1464  * intersects each OST and find last_stripe.
1465  * This function returns the last_stripe and also sets the stripe_count
1466  * over which the mapping is spread
1467  *
1468  * \param lsm striping information for the file
1469  * \param fm_start logical start of mapping
1470  * \param fm_end logical end of mapping
1471  * \param start_stripe starting stripe of the mapping
1472  * \param stripe_count the number of stripes across which to map is returned
1473  *
1474  * \retval last_stripe return the last stripe of the mapping
1475  */
1476 static int fiemap_calc_last_stripe(struct lov_stripe_md *lsm, u64 fm_start,
1477                                    u64 fm_end, int start_stripe,
1478                                    int *stripe_count)
1479 {
1480         int last_stripe;
1481         u64 obd_start, obd_end;
1482         int i, j;
1483
1484         if (fm_end - fm_start > lsm->lsm_stripe_size * lsm->lsm_stripe_count) {
1485                 last_stripe = start_stripe < 1 ? lsm->lsm_stripe_count - 1 :
1486                                                               start_stripe - 1;
1487                 *stripe_count = lsm->lsm_stripe_count;
1488         } else {
1489                 for (j = 0, i = start_stripe; j < lsm->lsm_stripe_count;
1490                      i = (i + 1) % lsm->lsm_stripe_count, j++) {
1491                         if ((lov_stripe_intersects(lsm, i, fm_start, fm_end,
1492                                                    &obd_start, &obd_end)) == 0)
1493                                 break;
1494                 }
1495                 *stripe_count = j;
1496                 last_stripe = (start_stripe + j - 1) % lsm->lsm_stripe_count;
1497         }
1498
1499         return last_stripe;
1500 }
1501
1502 /**
1503  * Set fe_device and copy extents from local buffer into main return buffer.
1504  *
1505  * \param fiemap fiemap request header
1506  * \param lcl_fm_ext array of local fiemap extents to be copied
1507  * \param ost_index OST index to be written into the fm_device field for each
1508                     extent
1509  * \param ext_count number of extents to be copied
1510  * \param current_extent where to start copying in main extent array
1511  */
1512 static void fiemap_prepare_and_copy_exts(struct ll_user_fiemap *fiemap,
1513                                          struct ll_fiemap_extent *lcl_fm_ext,
1514                                          int ost_index, unsigned int ext_count,
1515                                          int current_extent)
1516 {
1517         char *to;
1518         int ext;
1519
1520         for (ext = 0; ext < ext_count; ext++) {
1521                 lcl_fm_ext[ext].fe_device = ost_index;
1522                 lcl_fm_ext[ext].fe_flags |= FIEMAP_EXTENT_NET;
1523         }
1524
1525         /* Copy fm_extent's from fm_local to return buffer */
1526         to = (char *)fiemap + fiemap_count_to_size(current_extent);
1527         memcpy(to, lcl_fm_ext, ext_count * sizeof(struct ll_fiemap_extent));
1528 }
1529
1530 /**
1531  * Break down the FIEMAP request and send appropriate calls to individual OSTs.
1532  * This also handles the restarting of FIEMAP calls in case mapping overflows
1533  * the available number of extents in single call.
1534  */
1535 static int lov_fiemap(struct lov_obd *lov, __u32 keylen, void *key,
1536                       __u32 *vallen, void *val, struct lov_stripe_md *lsm)
1537 {
1538         struct ll_fiemap_info_key *fm_key = key;
1539         struct ll_user_fiemap *fiemap = val;
1540         struct ll_user_fiemap *fm_local = NULL;
1541         struct ll_fiemap_extent *lcl_fm_ext;
1542         int count_local;
1543         unsigned int get_num_extents = 0;
1544         int ost_index = 0, actual_start_stripe, start_stripe;
1545         u64 fm_start, fm_end, fm_length, fm_end_offset;
1546         u64 curr_loc;
1547         int current_extent = 0, rc = 0, i;
1548         /* Whether have we collected enough extents */
1549         bool enough = false;
1550         int ost_eof = 0; /* EOF for object */
1551         int ost_done = 0; /* done with required mapping for this OST? */
1552         int last_stripe;
1553         int cur_stripe = 0, cur_stripe_wrap = 0, stripe_count;
1554         unsigned int buffer_size = FIEMAP_BUFFER_SIZE;
1555
1556         if (!lsm_has_objects(lsm)) {
1557                 if (lsm && lsm_is_released(lsm) && (fm_key->fiemap.fm_start <
1558                     fm_key->oa.o_size)) {
1559                         /*
1560                          * released file, return a minimal FIEMAP if
1561                          * request fits in file-size.
1562                          */
1563                         fiemap->fm_mapped_extents = 1;
1564                         fiemap->fm_extents[0].fe_logical =
1565                                         fm_key->fiemap.fm_start;
1566                         if (fm_key->fiemap.fm_start + fm_key->fiemap.fm_length <
1567                             fm_key->oa.o_size) {
1568                                 fiemap->fm_extents[0].fe_length =
1569                                         fm_key->fiemap.fm_length;
1570                         } else {
1571                                 fiemap->fm_extents[0].fe_length =
1572                                         fm_key->oa.o_size - fm_key->fiemap.fm_start;
1573                                 fiemap->fm_extents[0].fe_flags |=
1574                                                 (FIEMAP_EXTENT_UNKNOWN |
1575                                                  FIEMAP_EXTENT_LAST);
1576                         }
1577                 }
1578                 rc = 0;
1579                 goto out;
1580         }
1581
1582         if (fiemap_count_to_size(fm_key->fiemap.fm_extent_count) < buffer_size)
1583                 buffer_size = fiemap_count_to_size(fm_key->fiemap.fm_extent_count);
1584
1585         fm_local = libcfs_kvzalloc(buffer_size, GFP_NOFS);
1586         if (!fm_local) {
1587                 rc = -ENOMEM;
1588                 goto out;
1589         }
1590         lcl_fm_ext = &fm_local->fm_extents[0];
1591
1592         count_local = fiemap_size_to_count(buffer_size);
1593
1594         memcpy(fiemap, &fm_key->fiemap, sizeof(*fiemap));
1595         fm_start = fiemap->fm_start;
1596         fm_length = fiemap->fm_length;
1597         /* Calculate start stripe, last stripe and length of mapping */
1598         start_stripe = lov_stripe_number(lsm, fm_start);
1599         actual_start_stripe = start_stripe;
1600         fm_end = (fm_length == ~0ULL ? fm_key->oa.o_size :
1601                                                 fm_start + fm_length - 1);
1602         /* If fm_length != ~0ULL but fm_start+fm_length-1 exceeds file size */
1603         if (fm_end > fm_key->oa.o_size)
1604                 fm_end = fm_key->oa.o_size;
1605
1606         last_stripe = fiemap_calc_last_stripe(lsm, fm_start, fm_end,
1607                                               actual_start_stripe,
1608                                               &stripe_count);
1609
1610         fm_end_offset = fiemap_calc_fm_end_offset(fiemap, lsm, fm_start,
1611                                                   fm_end, &start_stripe);
1612         if (fm_end_offset == -EINVAL) {
1613                 rc = -EINVAL;
1614                 goto out;
1615         }
1616
1617         if (fiemap_count_to_size(fiemap->fm_extent_count) > *vallen)
1618                 fiemap->fm_extent_count = fiemap_size_to_count(*vallen);
1619         if (fiemap->fm_extent_count == 0) {
1620                 get_num_extents = 1;
1621                 count_local = 0;
1622         }
1623         /* Check each stripe */
1624         for (cur_stripe = start_stripe, i = 0; i < stripe_count;
1625              i++, cur_stripe = (cur_stripe + 1) % lsm->lsm_stripe_count) {
1626                 u64 req_fm_len; /* Stores length of required mapping */
1627                 u64 len_mapped_single_call;
1628                 u64 lun_start, lun_end, obd_object_end;
1629                 unsigned int ext_count;
1630
1631                 cur_stripe_wrap = cur_stripe;
1632
1633                 /* Find out range of mapping on this stripe */
1634                 if ((lov_stripe_intersects(lsm, cur_stripe, fm_start, fm_end,
1635                                            &lun_start, &obd_object_end)) == 0)
1636                         continue;
1637
1638                 if (lov_oinfo_is_dummy(lsm->lsm_oinfo[cur_stripe])) {
1639                         rc = -EIO;
1640                         goto out;
1641                 }
1642
1643                 /* If this is a continuation FIEMAP call and we are on
1644                  * starting stripe then lun_start needs to be set to
1645                  * fm_end_offset
1646                  */
1647                 if (fm_end_offset != 0 && cur_stripe == start_stripe)
1648                         lun_start = fm_end_offset;
1649
1650                 if (fm_length != ~0ULL) {
1651                         /* Handle fm_start + fm_length overflow */
1652                         if (fm_start + fm_length < fm_start)
1653                                 fm_length = ~0ULL - fm_start;
1654                         lun_end = lov_size_to_stripe(lsm, fm_start + fm_length,
1655                                                      cur_stripe);
1656                 } else {
1657                         lun_end = ~0ULL;
1658                 }
1659
1660                 if (lun_start == lun_end)
1661                         continue;
1662
1663                 req_fm_len = obd_object_end - lun_start;
1664                 fm_local->fm_length = 0;
1665                 len_mapped_single_call = 0;
1666
1667                 /* If the output buffer is very large and the objects have many
1668                  * extents we may need to loop on a single OST repeatedly
1669                  */
1670                 ost_eof = 0;
1671                 ost_done = 0;
1672                 do {
1673                         if (get_num_extents == 0) {
1674                                 /* Don't get too many extents. */
1675                                 if (current_extent + count_local >
1676                                     fiemap->fm_extent_count)
1677                                         count_local = fiemap->fm_extent_count -
1678                                                                  current_extent;
1679                         }
1680
1681                         lun_start += len_mapped_single_call;
1682                         fm_local->fm_length = req_fm_len - len_mapped_single_call;
1683                         req_fm_len = fm_local->fm_length;
1684                         fm_local->fm_extent_count = enough ? 1 : count_local;
1685                         fm_local->fm_mapped_extents = 0;
1686                         fm_local->fm_flags = fiemap->fm_flags;
1687
1688                         fm_key->oa.o_oi = lsm->lsm_oinfo[cur_stripe]->loi_oi;
1689                         ost_index = lsm->lsm_oinfo[cur_stripe]->loi_ost_idx;
1690
1691                         if (ost_index < 0 ||
1692                             ost_index >= lov->desc.ld_tgt_count) {
1693                                 rc = -EINVAL;
1694                                 goto out;
1695                         }
1696
1697                         /* If OST is inactive, return extent with UNKNOWN flag */
1698                         if (!lov->lov_tgts[ost_index]->ltd_active) {
1699                                 fm_local->fm_flags |= FIEMAP_EXTENT_LAST;
1700                                 fm_local->fm_mapped_extents = 1;
1701
1702                                 lcl_fm_ext[0].fe_logical = lun_start;
1703                                 lcl_fm_ext[0].fe_length = obd_object_end -
1704                                                                       lun_start;
1705                                 lcl_fm_ext[0].fe_flags |= FIEMAP_EXTENT_UNKNOWN;
1706
1707                                 goto inactive_tgt;
1708                         }
1709
1710                         fm_local->fm_start = lun_start;
1711                         fm_local->fm_flags &= ~FIEMAP_FLAG_DEVICE_ORDER;
1712                         memcpy(&fm_key->fiemap, fm_local, sizeof(*fm_local));
1713                         *vallen = fiemap_count_to_size(fm_local->fm_extent_count);
1714                         rc = obd_get_info(NULL,
1715                                           lov->lov_tgts[ost_index]->ltd_exp,
1716                                           keylen, key, vallen, fm_local, lsm);
1717                         if (rc != 0)
1718                                 goto out;
1719
1720 inactive_tgt:
1721                         ext_count = fm_local->fm_mapped_extents;
1722                         if (ext_count == 0) {
1723                                 ost_done = 1;
1724                                 /* If last stripe has hole at the end,
1725                                  * then we need to return
1726                                  */
1727                                 if (cur_stripe_wrap == last_stripe) {
1728                                         fiemap->fm_mapped_extents = 0;
1729                                         goto finish;
1730                                 }
1731                                 break;
1732                         } else if (enough) {
1733                                 /*
1734                                  * We've collected enough extents and there are
1735                                  * more extents after it.
1736                                  */
1737                                 goto finish;
1738                         }
1739
1740                         /* If we just need num of extents then go to next device */
1741                         if (get_num_extents) {
1742                                 current_extent += ext_count;
1743                                 break;
1744                         }
1745
1746                         len_mapped_single_call =
1747                                 lcl_fm_ext[ext_count - 1].fe_logical -
1748                                 lun_start + lcl_fm_ext[ext_count - 1].fe_length;
1749
1750                         /* Have we finished mapping on this device? */
1751                         if (req_fm_len <= len_mapped_single_call)
1752                                 ost_done = 1;
1753
1754                         /* Clear the EXTENT_LAST flag which can be present on
1755                          * last extent
1756                          */
1757                         if (lcl_fm_ext[ext_count - 1].fe_flags &
1758                             FIEMAP_EXTENT_LAST)
1759                                 lcl_fm_ext[ext_count - 1].fe_flags &=
1760                                                             ~FIEMAP_EXTENT_LAST;
1761
1762                         curr_loc = lov_stripe_size(lsm,
1763                                         lcl_fm_ext[ext_count - 1].fe_logical +
1764                                         lcl_fm_ext[ext_count - 1].fe_length,
1765                                         cur_stripe);
1766                         if (curr_loc >= fm_key->oa.o_size)
1767                                 ost_eof = 1;
1768
1769                         fiemap_prepare_and_copy_exts(fiemap, lcl_fm_ext,
1770                                                      ost_index, ext_count,
1771                                                      current_extent);
1772
1773                         current_extent += ext_count;
1774
1775                         /* Ran out of available extents? */
1776                         if (current_extent >= fiemap->fm_extent_count)
1777                                 enough = true;
1778                 } while (ost_done == 0 && ost_eof == 0);
1779
1780                 if (cur_stripe_wrap == last_stripe)
1781                         goto finish;
1782         }
1783
1784 finish:
1785         /* Indicate that we are returning device offsets unless file just has
1786          * single stripe
1787          */
1788         if (lsm->lsm_stripe_count > 1)
1789                 fiemap->fm_flags |= FIEMAP_FLAG_DEVICE_ORDER;
1790
1791         if (get_num_extents)
1792                 goto skip_last_device_calc;
1793
1794         /* Check if we have reached the last stripe and whether mapping for that
1795          * stripe is done.
1796          */
1797         if (cur_stripe_wrap == last_stripe) {
1798                 if (ost_done || ost_eof)
1799                         fiemap->fm_extents[current_extent - 1].fe_flags |=
1800                                                              FIEMAP_EXTENT_LAST;
1801         }
1802
1803 skip_last_device_calc:
1804         fiemap->fm_mapped_extents = current_extent;
1805
1806 out:
1807         kvfree(fm_local);
1808         return rc;
1809 }
1810
1811 static int lov_get_info(const struct lu_env *env, struct obd_export *exp,
1812                         __u32 keylen, void *key, __u32 *vallen, void *val,
1813                         struct lov_stripe_md *lsm)
1814 {
1815         struct obd_device *obddev = class_exp2obd(exp);
1816         struct lov_obd *lov = &obddev->u.lov;
1817         int rc;
1818
1819         if (!vallen || !val)
1820                 return -EFAULT;
1821
1822         obd_getref(obddev);
1823
1824         if (KEY_IS(KEY_LOVDESC)) {
1825                 struct lov_desc *desc_ret = val;
1826                 *desc_ret = lov->desc;
1827
1828                 rc = 0;
1829                 goto out;
1830         } else if (KEY_IS(KEY_FIEMAP)) {
1831                 rc = lov_fiemap(lov, keylen, key, vallen, val, lsm);
1832                 goto out;
1833         } else if (KEY_IS(KEY_TGT_COUNT)) {
1834                 *((int *)val) = lov->desc.ld_tgt_count;
1835                 rc = 0;
1836                 goto out;
1837         }
1838
1839         rc = -EINVAL;
1840
1841 out:
1842         obd_putref(obddev);
1843         return rc;
1844 }
1845
1846 static int lov_set_info_async(const struct lu_env *env, struct obd_export *exp,
1847                               u32 keylen, void *key, u32 vallen,
1848                               void *val, struct ptlrpc_request_set *set)
1849 {
1850         struct obd_device *obddev = class_exp2obd(exp);
1851         struct lov_obd *lov = &obddev->u.lov;
1852         u32 count;
1853         int i, rc = 0, err;
1854         struct lov_tgt_desc *tgt;
1855         int do_inactive = 0, no_set = 0;
1856
1857         if (!set) {
1858                 no_set = 1;
1859                 set = ptlrpc_prep_set();
1860                 if (!set)
1861                         return -ENOMEM;
1862         }
1863
1864         obd_getref(obddev);
1865         count = lov->desc.ld_tgt_count;
1866
1867         if (KEY_IS(KEY_CHECKSUM)) {
1868                 do_inactive = 1;
1869         } else if (KEY_IS(KEY_CACHE_SET)) {
1870                 LASSERT(!lov->lov_cache);
1871                 lov->lov_cache = val;
1872                 do_inactive = 1;
1873                 cl_cache_incref(lov->lov_cache);
1874         }
1875
1876         for (i = 0; i < count; i++) {
1877                 tgt = lov->lov_tgts[i];
1878
1879                 /* OST was disconnected */
1880                 if (!tgt || !tgt->ltd_exp)
1881                         continue;
1882
1883                 /* OST is inactive and we don't want inactive OSCs */
1884                 if (!tgt->ltd_active && !do_inactive)
1885                         continue;
1886
1887                 err = obd_set_info_async(env, tgt->ltd_exp, keylen, key,
1888                                          vallen, val, set);
1889                 if (!rc)
1890                         rc = err;
1891         }
1892
1893         obd_putref(obddev);
1894         if (no_set) {
1895                 err = ptlrpc_set_wait(set);
1896                 if (!rc)
1897                         rc = err;
1898                 ptlrpc_set_destroy(set);
1899         }
1900         return rc;
1901 }
1902
1903 void lov_stripe_lock(struct lov_stripe_md *md)
1904                 __acquires(&md->lsm_lock)
1905 {
1906         LASSERT(md->lsm_lock_owner != current_pid());
1907         spin_lock(&md->lsm_lock);
1908         LASSERT(md->lsm_lock_owner == 0);
1909         md->lsm_lock_owner = current_pid();
1910 }
1911
1912 void lov_stripe_unlock(struct lov_stripe_md *md)
1913                 __releases(&md->lsm_lock)
1914 {
1915         LASSERT(md->lsm_lock_owner == current_pid());
1916         md->lsm_lock_owner = 0;
1917         spin_unlock(&md->lsm_lock);
1918 }
1919
1920 static int lov_quotactl(struct obd_device *obd, struct obd_export *exp,
1921                         struct obd_quotactl *oqctl)
1922 {
1923         struct lov_obd      *lov = &obd->u.lov;
1924         struct lov_tgt_desc *tgt;
1925         __u64           curspace = 0;
1926         __u64           bhardlimit = 0;
1927         int               i, rc = 0;
1928
1929         if (oqctl->qc_cmd != LUSTRE_Q_QUOTAON &&
1930             oqctl->qc_cmd != LUSTRE_Q_QUOTAOFF &&
1931             oqctl->qc_cmd != Q_GETOQUOTA &&
1932             oqctl->qc_cmd != Q_INITQUOTA &&
1933             oqctl->qc_cmd != LUSTRE_Q_SETQUOTA &&
1934             oqctl->qc_cmd != Q_FINVALIDATE) {
1935                 CERROR("bad quota opc %x for lov obd\n", oqctl->qc_cmd);
1936                 return -EFAULT;
1937         }
1938
1939         /* for lov tgt */
1940         obd_getref(obd);
1941         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
1942                 int err;
1943
1944                 tgt = lov->lov_tgts[i];
1945
1946                 if (!tgt)
1947                         continue;
1948
1949                 if (!tgt->ltd_active || tgt->ltd_reap) {
1950                         if (oqctl->qc_cmd == Q_GETOQUOTA &&
1951                             lov->lov_tgts[i]->ltd_activate) {
1952                                 rc = -EREMOTEIO;
1953                                 CERROR("ost %d is inactive\n", i);
1954                         } else {
1955                                 CDEBUG(D_HA, "ost %d is inactive\n", i);
1956                         }
1957                         continue;
1958                 }
1959
1960                 err = obd_quotactl(tgt->ltd_exp, oqctl);
1961                 if (err) {
1962                         if (tgt->ltd_active && !rc)
1963                                 rc = err;
1964                         continue;
1965                 }
1966
1967                 if (oqctl->qc_cmd == Q_GETOQUOTA) {
1968                         curspace += oqctl->qc_dqblk.dqb_curspace;
1969                         bhardlimit += oqctl->qc_dqblk.dqb_bhardlimit;
1970                 }
1971         }
1972         obd_putref(obd);
1973
1974         if (oqctl->qc_cmd == Q_GETOQUOTA) {
1975                 oqctl->qc_dqblk.dqb_curspace = curspace;
1976                 oqctl->qc_dqblk.dqb_bhardlimit = bhardlimit;
1977         }
1978         return rc;
1979 }
1980
1981 static int lov_quotacheck(struct obd_device *obd, struct obd_export *exp,
1982                           struct obd_quotactl *oqctl)
1983 {
1984         struct lov_obd *lov = &obd->u.lov;
1985         int          i, rc = 0;
1986
1987         obd_getref(obd);
1988
1989         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
1990                 if (!lov->lov_tgts[i])
1991                         continue;
1992
1993                 /* Skip quota check on the administratively disabled OSTs. */
1994                 if (!lov->lov_tgts[i]->ltd_activate) {
1995                         CWARN("lov idx %d was administratively disabled, skip quotacheck on it.\n",
1996                               i);
1997                         continue;
1998                 }
1999
2000                 if (!lov->lov_tgts[i]->ltd_active) {
2001                         CERROR("lov idx %d inactive\n", i);
2002                         rc = -EIO;
2003                         goto out;
2004                 }
2005         }
2006
2007         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2008                 int err;
2009
2010                 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_activate)
2011                         continue;
2012
2013                 err = obd_quotacheck(lov->lov_tgts[i]->ltd_exp, oqctl);
2014                 if (err && !rc)
2015                         rc = err;
2016         }
2017
2018 out:
2019         obd_putref(obd);
2020
2021         return rc;
2022 }
2023
2024 static struct obd_ops lov_obd_ops = {
2025         .owner          = THIS_MODULE,
2026         .setup          = lov_setup,
2027         .precleanup     = lov_precleanup,
2028         .cleanup        = lov_cleanup,
2029         /*.process_config       = lov_process_config,*/
2030         .connect        = lov_connect,
2031         .disconnect     = lov_disconnect,
2032         .statfs         = lov_statfs,
2033         .statfs_async   = lov_statfs_async,
2034         .packmd         = lov_packmd,
2035         .unpackmd       = lov_unpackmd,
2036         .getattr_async  = lov_getattr_async,
2037         .setattr_async  = lov_setattr_async,
2038         .iocontrol      = lov_iocontrol,
2039         .get_info       = lov_get_info,
2040         .set_info_async = lov_set_info_async,
2041         .notify         = lov_notify,
2042         .pool_new       = lov_pool_new,
2043         .pool_rem       = lov_pool_remove,
2044         .pool_add       = lov_pool_add,
2045         .pool_del       = lov_pool_del,
2046         .getref         = lov_getref,
2047         .putref         = lov_putref,
2048         .quotactl       = lov_quotactl,
2049         .quotacheck     = lov_quotacheck,
2050 };
2051
2052 struct kmem_cache *lov_oinfo_slab;
2053
2054 static int __init lov_init(void)
2055 {
2056         struct lprocfs_static_vars lvars = { NULL };
2057         int rc;
2058
2059         /* print an address of _any_ initialized kernel symbol from this
2060          * module, to allow debugging with gdb that doesn't support data
2061          * symbols from modules.
2062          */
2063         CDEBUG(D_INFO, "Lustre LOV module (%p).\n", &lov_caches);
2064
2065         rc = lu_kmem_init(lov_caches);
2066         if (rc)
2067                 return rc;
2068
2069         lov_oinfo_slab = kmem_cache_create("lov_oinfo",
2070                                            sizeof(struct lov_oinfo),
2071                                            0, SLAB_HWCACHE_ALIGN, NULL);
2072         if (!lov_oinfo_slab) {
2073                 lu_kmem_fini(lov_caches);
2074                 return -ENOMEM;
2075         }
2076         lprocfs_lov_init_vars(&lvars);
2077
2078         rc = class_register_type(&lov_obd_ops, NULL,
2079                                  LUSTRE_LOV_NAME, &lov_device_type);
2080
2081         if (rc) {
2082                 kmem_cache_destroy(lov_oinfo_slab);
2083                 lu_kmem_fini(lov_caches);
2084         }
2085
2086         return rc;
2087 }
2088
2089 static void /*__exit*/ lov_exit(void)
2090 {
2091         class_unregister_type(LUSTRE_LOV_NAME);
2092         kmem_cache_destroy(lov_oinfo_slab);
2093
2094         lu_kmem_fini(lov_caches);
2095 }
2096
2097 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
2098 MODULE_DESCRIPTION("Lustre Logical Object Volume");
2099 MODULE_LICENSE("GPL");
2100 MODULE_VERSION(LUSTRE_VERSION_STRING);
2101
2102 module_init(lov_init);
2103 module_exit(lov_exit);