lustre/lod/lod_lov.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,
  11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  13  * GNU General Public License version 2 for more details.  A copy is
  14  * included in the COPYING file that accompanied this code.
  15
  16  * You should have received a copy of the GNU General Public License
  17  * along with this program; if not, write to the Free Software
  18  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright  2009 Sun Microsystems, Inc. All rights reserved
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2012, 2013, Intel Corporation.
  27  */
  28 /*
  29  * lustre/lod/lod_lov.c
  30  *
  31  * Author: Alex Zhuravlev <alexey.zhuravlev@intel.com>
  32  */
  33
  34 #define DEBUG_SUBSYSTEM S_MDS
  35
  36 #include <obd_class.h>
  37 #include <lustre_lfsck.h>
  38
  39 #include "lod_internal.h"
  40
  41 /*
  42  * Keep a refcount of lod->ltd_tgts usage to prevent racing with
  43  * addition/deletion. Any function that expects lov_tgts to remain stationary
  44  * must take a ref.
  45  *
  46  * \param lod - is the lod device from which we want to grab a reference
  47  */
  48 void lod_getref(struct lod_tgt_descs *ltd)
  49 {
  50         down_read(&ltd->ltd_rw_sem);
  51         mutex_lock(&ltd->ltd_mutex);
  52         ltd->ltd_refcount++;
  53         mutex_unlock(&ltd->ltd_mutex);
  54 }
  55
  56 /*
  57  * Companion of lod_getref() to release a reference on the lod table.
  58  * If this is the last reference and the ost entry was scheduled for deletion,
  59  * the descriptor is removed from the array.
  60  *
  61  * \param lod - is the lod device from which we release a reference
  62  */
  63 void lod_putref(struct lod_device *lod, struct lod_tgt_descs *ltd)
  64 {
  65         mutex_lock(&ltd->ltd_mutex);
  66         ltd->ltd_refcount--;
  67         if (ltd->ltd_refcount == 0 && ltd->ltd_death_row) {
  68                 struct lod_tgt_desc *tgt_desc, *tmp;
  69                 int                  idx;
  70                 CFS_LIST_HEAD(kill);
  71
  72                 CDEBUG(D_CONFIG, "destroying %d ltd desc\n",
  73                        ltd->ltd_death_row);
  74
  75                 cfs_foreach_bit(ltd->ltd_tgt_bitmap, idx) {
  76                         tgt_desc = LTD_TGT(ltd, idx);
  77                         LASSERT(tgt_desc);
  78
  79                         if (!tgt_desc->ltd_reap)
  80                                 continue;
  81
  82                         cfs_list_add(&tgt_desc->ltd_kill, &kill);
  83                         LTD_TGT(ltd, idx) = NULL;
  84                         /*FIXME: only support ost pool for now */
  85                         if (ltd == &lod->lod_ost_descs) {
  86                                 lod_ost_pool_remove(&lod->lod_pool_info, idx);
  87                                 if (tgt_desc->ltd_active)
  88                                         lod->lod_desc.ld_active_tgt_count--;
  89                         }
  90                         ltd->ltd_tgtnr--;
  91                         cfs_bitmap_clear(ltd->ltd_tgt_bitmap, idx);
  92                         ltd->ltd_death_row--;
  93                 }
  94                 mutex_unlock(&ltd->ltd_mutex);
  95                 up_read(&ltd->ltd_rw_sem);
  96
  97                 cfs_list_for_each_entry_safe(tgt_desc, tmp, &kill, ltd_kill) {
  98                         int rc;
  99                         cfs_list_del(&tgt_desc->ltd_kill);
 100                         if (ltd == &lod->lod_ost_descs) {
 101                                 /* remove from QoS structures */
 102                                 rc = qos_del_tgt(lod, tgt_desc);
 103                                 if (rc)
 104                                         CERROR("%s: qos_del_tgt(%s) failed:"
 105                                                "rc = %d\n",
 106                                                lod2obd(lod)->obd_name,
 107                                               obd_uuid2str(&tgt_desc->ltd_uuid),
 108                                                rc);
 109                         }
 110                         rc = obd_disconnect(tgt_desc->ltd_exp);
 111                         if (rc)
 112                                 CERROR("%s: failed to disconnect %s: rc = %d\n",
 113                                        lod2obd(lod)->obd_name,
 114                                        obd_uuid2str(&tgt_desc->ltd_uuid), rc);
 115                         OBD_FREE_PTR(tgt_desc);
 116                 }
 117         } else {
 118                 mutex_unlock(&ltd->ltd_mutex);
 119                 up_read(&ltd->ltd_rw_sem);
 120         }
 121 }
 122
 123 static int ltd_bitmap_resize(struct lod_tgt_descs *ltd, __u32 newsize)
 124 {
 125         cfs_bitmap_t *new_bitmap, *old_bitmap = NULL;
 126         int           rc = 0;
 127         ENTRY;
 128
 129         /* grab write reference on the lod. Relocating the array requires
 130          * exclusive access */
 131
 132         down_write(&ltd->ltd_rw_sem);
 133         if (newsize <= ltd->ltd_tgts_size)
 134                 /* someone else has already resize the array */
 135                 GOTO(out, rc = 0);
 136
 137         /* allocate new bitmap */
 138         new_bitmap = CFS_ALLOCATE_BITMAP(newsize);
 139         if (!new_bitmap)
 140                 GOTO(out, rc = -ENOMEM);
 141
 142         if (ltd->ltd_tgts_size > 0) {
 143                 /* the bitmap already exists, we need
 144                  * to copy data from old one */
 145                 cfs_bitmap_copy(new_bitmap, ltd->ltd_tgt_bitmap);
 146                 old_bitmap = ltd->ltd_tgt_bitmap;
 147         }
 148
 149         ltd->ltd_tgts_size  = newsize;
 150         ltd->ltd_tgt_bitmap = new_bitmap;
 151
 152         if (old_bitmap)
 153                 CFS_FREE_BITMAP(old_bitmap);
 154
 155         CDEBUG(D_CONFIG, "tgt size: %d\n", ltd->ltd_tgts_size);
 156
 157         EXIT;
 158 out:
 159         up_write(&ltd->ltd_rw_sem);
 160         return rc;
 161 }
 162
 163 /*
 164  * Connect LOD to a new OSP and add it to the device table.
 165  *
 166  * \param env - is the environment passed by the caller
 167  * \param lod - is the LOD device to be connected to the new OSP
 168  * \param osp - is the name of OSP device name about to be added
 169  * \param index - is the OSP index
 170  * \param gen - is the generation number
 171  * \param tgt_index - is the group of the OSP.
 172  * \param type - is the type of device (mdc or osc)
 173  */
 174 int lod_add_device(const struct lu_env *env, struct lod_device *lod,
 175                    char *osp, unsigned index, unsigned gen, int tgt_index,
 176                    char *type, int active)
 177 {
 178         struct obd_connect_data *data = NULL;
 179         struct obd_export       *exp = NULL;
 180         struct obd_device       *obd;
 181         struct lu_device        *ldev;
 182         struct dt_device        *d;
 183         int                      rc;
 184         struct lod_tgt_desc     *tgt_desc;
 185         struct lod_tgt_descs    *ltd;
 186         struct obd_uuid         obd_uuid;
 187         bool                    for_ost;
 188         ENTRY;
 189
 190         CDEBUG(D_CONFIG, "osp:%s idx:%d gen:%d\n", osp, index, gen);
 191
 192         if (gen <= 0) {
 193                 CERROR("request to add OBD %s with invalid generation: %d\n",
 194                        osp, gen);
 195                 RETURN(-EINVAL);
 196         }
 197
 198         obd_str2uuid(&obd_uuid, osp);
 199
 200         obd = class_find_client_obd(&obd_uuid, LUSTRE_OSP_NAME,
 201                                 &lod->lod_dt_dev.dd_lu_dev.ld_obd->obd_uuid);
 202         if (obd == NULL) {
 203                 CERROR("can't find %s device\n", osp);
 204                 RETURN(-EINVAL);
 205         }
 206
 207         OBD_ALLOC_PTR(data);
 208         if (data == NULL)
 209                 RETURN(-ENOMEM);
 210
 211         data->ocd_connect_flags = OBD_CONNECT_INDEX | OBD_CONNECT_VERSION;
 212         data->ocd_version = LUSTRE_VERSION_CODE;
 213         data->ocd_index = index;
 214
 215         if (strcmp(LUSTRE_OSC_NAME, type) == 0) {
 216                 for_ost = true;
 217                 data->ocd_connect_flags |= OBD_CONNECT_AT |
 218                                            OBD_CONNECT_FULL20 |
 219                                            OBD_CONNECT_INDEX |
 220 #ifdef HAVE_LRU_RESIZE_SUPPORT
 221                                            OBD_CONNECT_LRU_RESIZE |
 222 #endif
 223                                            OBD_CONNECT_MDS |
 224                                            OBD_CONNECT_OSS_CAPA |
 225                                            OBD_CONNECT_REQPORTAL |
 226                                            OBD_CONNECT_SKIP_ORPHAN |
 227                                            OBD_CONNECT_FID |
 228                                            OBD_CONNECT_LVB_TYPE |
 229                                            OBD_CONNECT_VERSION |
 230                                            OBD_CONNECT_PINGLESS |
 231                                            OBD_CONNECT_LFSCK;
 232
 233                 data->ocd_group = tgt_index;
 234                 ltd = &lod->lod_ost_descs;
 235         } else {
 236                 struct obd_import *imp = obd->u.cli.cl_import;
 237
 238                 for_ost = false;
 239                 data->ocd_ibits_known = MDS_INODELOCK_UPDATE;
 240                 data->ocd_connect_flags |= OBD_CONNECT_ACL |
 241                                            OBD_CONNECT_MDS_CAPA |
 242                                            OBD_CONNECT_OSS_CAPA |
 243                                            OBD_CONNECT_IBITS |
 244                                            OBD_CONNECT_MDS_MDS |
 245                                            OBD_CONNECT_FID |
 246                                            OBD_CONNECT_AT |
 247                                            OBD_CONNECT_FULL20 |
 248                                            OBD_CONNECT_LFSCK;
 249                 /* XXX set MDS-MDS flags, remove this when running this
 250                  * on client*/
 251                 data->ocd_connect_flags |= OBD_CONNECT_MDS_MDS;
 252                 spin_lock(&imp->imp_lock);
 253                 imp->imp_server_timeout = 1;
 254                 spin_unlock(&imp->imp_lock);
 255                 imp->imp_client->cli_request_portal = OUT_PORTAL;
 256                 CDEBUG(D_OTHER, "%s: Set 'mds' portal and timeout\n",
 257                       obd->obd_name);
 258                 ltd = &lod->lod_mdt_descs;
 259         }
 260
 261         rc = obd_connect(env, &exp, obd, &obd->obd_uuid, data, NULL);
 262         OBD_FREE_PTR(data);
 263         if (rc) {
 264                 CERROR("%s: cannot connect to next dev %s (%d)\n",
 265                        obd->obd_name, osp, rc);
 266                 GOTO(out_free, rc);
 267         }
 268
 269         LASSERT(obd->obd_lu_dev);
 270         LASSERT(obd->obd_lu_dev->ld_site == lod->lod_dt_dev.dd_lu_dev.ld_site);
 271
 272         ldev = obd->obd_lu_dev;
 273         d = lu2dt_dev(ldev);
 274
 275         /* Allocate ost descriptor and fill it */
 276         OBD_ALLOC_PTR(tgt_desc);
 277         if (!tgt_desc)
 278                 GOTO(out_conn, rc = -ENOMEM);
 279
 280         tgt_desc->ltd_tgt    = d;
 281         tgt_desc->ltd_exp    = exp;
 282         tgt_desc->ltd_uuid   = obd->u.cli.cl_target_uuid;
 283         tgt_desc->ltd_gen    = gen;
 284         tgt_desc->ltd_index  = index;
 285         tgt_desc->ltd_active = active;
 286
 287         lod_getref(ltd);
 288         if (index >= ltd->ltd_tgts_size) {
 289                 /* we have to increase the size of the lod_osts array */
 290                 __u32  newsize;
 291
 292                 newsize = max(ltd->ltd_tgts_size, (__u32)2);
 293                 while (newsize < index + 1)
 294                         newsize = newsize << 1;
 295
 296                 /* lod_bitmap_resize() needs lod_rw_sem
 297                  * which we hold with th reference */
 298                 lod_putref(lod, ltd);
 299
 300                 rc = ltd_bitmap_resize(ltd, newsize);
 301                 if (rc)
 302                         GOTO(out_desc, rc);
 303
 304                 lod_getref(ltd);
 305         }
 306
 307         mutex_lock(&ltd->ltd_mutex);
 308         if (cfs_bitmap_check(ltd->ltd_tgt_bitmap, index)) {
 309                 CERROR("%s: device %d is registered already\n", obd->obd_name,
 310                        index);
 311                 GOTO(out_mutex, rc = -EEXIST);
 312         }
 313
 314         if (ltd->ltd_tgt_idx[index / TGT_PTRS_PER_BLOCK] == NULL) {
 315                 OBD_ALLOC_PTR(ltd->ltd_tgt_idx[index / TGT_PTRS_PER_BLOCK]);
 316                 if (ltd->ltd_tgt_idx[index / TGT_PTRS_PER_BLOCK] == NULL) {
 317                         CERROR("can't allocate index to add %s\n",
 318                                obd->obd_name);
 319                         GOTO(out_mutex, rc = -ENOMEM);
 320                 }
 321         }
 322
 323         if (!strcmp(LUSTRE_OSC_NAME, type)) {
 324                 /* pool and qos are not supported for MDS stack yet */
 325                 rc = lod_ost_pool_add(&lod->lod_pool_info, index,
 326                                       lod->lod_osts_size);
 327                 if (rc) {
 328                         CERROR("%s: can't set up pool, failed with %d\n",
 329                                obd->obd_name, rc);
 330                         GOTO(out_mutex, rc);
 331                 }
 332
 333                 rc = qos_add_tgt(lod, tgt_desc);
 334                 if (rc) {
 335                         CERROR("%s: qos_add_tgt failed with %d\n",
 336                                 obd->obd_name, rc);
 337                         GOTO(out_pool, rc);
 338                 }
 339
 340                 /* The new OST is now a full citizen */
 341                 if (index >= lod->lod_desc.ld_tgt_count)
 342                         lod->lod_desc.ld_tgt_count = index + 1;
 343                 if (active)
 344                         lod->lod_desc.ld_active_tgt_count++;
 345         }
 346
 347         LTD_TGT(ltd, index) = tgt_desc;
 348         cfs_bitmap_set(ltd->ltd_tgt_bitmap, index);
 349         ltd->ltd_tgtnr++;
 350         mutex_unlock(&ltd->ltd_mutex);
 351         lod_putref(lod, ltd);
 352         if (lod->lod_recovery_completed)
 353                 ldev->ld_ops->ldo_recovery_complete(env, ldev);
 354
 355         rc = lfsck_add_target(env, lod->lod_child, d, exp, index, for_ost);
 356         if (rc != 0)
 357                 CERROR("Fail to add LFSCK target: name = %s, type = %s, "
 358                        "index = %u, rc = %d\n", osp, type, index, rc);
 359
 360         RETURN(rc);
 361
 362 out_pool:
 363         lod_ost_pool_remove(&lod->lod_pool_info, index);
 364 out_mutex:
 365         mutex_unlock(&ltd->ltd_mutex);
 366         lod_putref(lod, ltd);
 367 out_desc:
 368         OBD_FREE_PTR(tgt_desc);
 369 out_conn:
 370         obd_disconnect(exp);
 371 out_free:
 372         return rc;
 373 }
 374
 375 /*
 376  * helper function to schedule OST removal from the device table
 377  */
 378 static void __lod_del_device(const struct lu_env *env, struct lod_device *lod,
 379                              struct lod_tgt_descs *ltd, unsigned idx,
 380                              bool for_ost)
 381 {
 382         LASSERT(LTD_TGT(ltd, idx));
 383
 384         lfsck_del_target(env, lod->lod_child, LTD_TGT(ltd, idx)->ltd_tgt,
 385                          idx, for_ost);
 386
 387         if (LTD_TGT(ltd, idx)->ltd_reap == 0) {
 388                 LTD_TGT(ltd, idx)->ltd_reap = 1;
 389                 ltd->ltd_death_row++;
 390         }
 391 }
 392
 393 int lod_fini_tgt(const struct lu_env *env, struct lod_device *lod,
 394                  struct lod_tgt_descs *ltd, bool for_ost)
 395 {
 396         int idx;
 397
 398         if (ltd->ltd_tgts_size <= 0)
 399                 return 0;
 400         lod_getref(ltd);
 401         mutex_lock(&ltd->ltd_mutex);
 402         cfs_foreach_bit(ltd->ltd_tgt_bitmap, idx)
 403                 __lod_del_device(env, lod, ltd, idx, for_ost);
 404         mutex_unlock(&ltd->ltd_mutex);
 405         lod_putref(lod, ltd);
 406         CFS_FREE_BITMAP(ltd->ltd_tgt_bitmap);
 407         for (idx = 0; idx < TGT_PTRS; idx++) {
 408                 if (ltd->ltd_tgt_idx[idx])
 409                         OBD_FREE_PTR(ltd->ltd_tgt_idx[idx]);
 410         }
 411         ltd->ltd_tgts_size = 0;
 412         return 0;
 413 }
 414
 415 /*
 416  * Add support for administratively disabled OST (through the MGS).
 417  * Schedule a target for deletion.  Disconnection and real removal from the
 418  * table takes place in lod_putref() once the last table user release its
 419  * reference.
 420  *
 421  * \param env - is the environment passed by the caller
 422  * \param lod - is the lod device currently connected to the OSP about to be
 423  *              removed
 424  * \param osp - is the name of OSP device about to be removed
 425  * \param idx - is the OSP index
 426  * \param gen - is the generation number, not used currently
 427  */
 428 int lod_del_device(const struct lu_env *env, struct lod_device *lod,
 429                    struct lod_tgt_descs *ltd, char *osp, unsigned idx,
 430                    unsigned gen, bool for_ost)
 431 {
 432         struct obd_device *obd;
 433         int                rc = 0;
 434         struct obd_uuid    uuid;
 435         ENTRY;
 436
 437         CDEBUG(D_CONFIG, "osp:%s idx:%d gen:%d\n", osp, idx, gen);
 438
 439         obd_str2uuid(&uuid, osp);
 440
 441         obd = class_find_client_obd(&uuid, LUSTRE_OSP_NAME,
 442                                    &lod->lod_dt_dev.dd_lu_dev.ld_obd->obd_uuid);
 443         if (obd == NULL) {
 444                 CERROR("can't find %s device\n", osp);
 445                 RETURN(-EINVAL);
 446         }
 447
 448         if (gen <= 0) {
 449                 CERROR("%s: request to remove OBD %s with invalid generation %d"
 450                        "\n", obd->obd_name, osp, gen);
 451                 RETURN(-EINVAL);
 452         }
 453
 454         obd_str2uuid(&uuid,  osp);
 455
 456         lod_getref(ltd);
 457         mutex_lock(&ltd->ltd_mutex);
 458         /* check that the index is allocated in the bitmap */
 459         if (!cfs_bitmap_check(ltd->ltd_tgt_bitmap, idx) ||
 460             !LTD_TGT(ltd, idx)) {
 461                 CERROR("%s: device %d is not set up\n", obd->obd_name, idx);
 462                 GOTO(out, rc = -EINVAL);
 463         }
 464
 465         /* check that the UUID matches */
 466         if (!obd_uuid_equals(&uuid, &LTD_TGT(ltd, idx)->ltd_uuid)) {
 467                 CERROR("%s: LOD target UUID %s at index %d does not match %s\n",
 468                        obd->obd_name, obd_uuid2str(&LTD_TGT(ltd,idx)->ltd_uuid),
 469                        idx, osp);
 470                 GOTO(out, rc = -EINVAL);
 471         }
 472
 473         __lod_del_device(env, lod, ltd, idx, for_ost);
 474         EXIT;
 475 out:
 476         mutex_unlock(&ltd->ltd_mutex);
 477         lod_putref(lod, ltd);
 478         return(rc);
 479 }
 480
 481 int lod_ea_store_resize(struct lod_thread_info *info, int size)
 482 {
 483         int round = size_roundup_power2(size);
 484
 485         LASSERT(round <=
 486                 lov_mds_md_size(LOV_MAX_STRIPE_COUNT, LOV_MAGIC_V3));
 487         if (info->lti_ea_store) {
 488                 LASSERT(info->lti_ea_store_size);
 489                 LASSERT(info->lti_ea_store_size < round);
 490                 CDEBUG(D_INFO, "EA store size %d is not enough, need %d\n",
 491                        info->lti_ea_store_size, round);
 492                 OBD_FREE_LARGE(info->lti_ea_store, info->lti_ea_store_size);
 493                 info->lti_ea_store = NULL;
 494                 info->lti_ea_store_size = 0;
 495         }
 496
 497         OBD_ALLOC_LARGE(info->lti_ea_store, round);
 498         if (info->lti_ea_store == NULL)
 499                 RETURN(-ENOMEM);
 500         info->lti_ea_store_size = round;
 501         RETURN(0);
 502 }
 503
 504 /*
 505  * generate and write LOV EA for given striped object
 506  */
 507 int lod_generate_and_set_lovea(const struct lu_env *env,
 508                                struct lod_object *lo, struct thandle *th)
 509 {
 510         struct lod_thread_info  *info = lod_env_info(env);
 511         struct dt_object        *next = dt_object_child(&lo->ldo_obj);
 512         const struct lu_fid     *fid  = lu_object_fid(&lo->ldo_obj.do_lu);
 513         struct lov_mds_md_v1    *lmm;
 514         struct lov_ost_data_v1  *objs;
 515         __u32                    magic;
 516         int                      i, rc, lmm_size;
 517         int                      cplen = 0;
 518         ENTRY;
 519
 520         LASSERT(lo);
 521
 522         magic = lo->ldo_pool != NULL ? LOV_MAGIC_V3 : LOV_MAGIC_V1;
 523         lmm_size = lov_mds_md_size(lo->ldo_stripenr, magic);
 524         if (info->lti_ea_store_size < lmm_size) {
 525                 rc = lod_ea_store_resize(info, lmm_size);
 526                 if (rc)
 527                         RETURN(rc);
 528         }
 529
 530         if (lo->ldo_pattern == 0) /* default striping */
 531                 lo->ldo_pattern = LOV_PATTERN_RAID0;
 532
 533         lmm = info->lti_ea_store;
 534
 535         lmm->lmm_magic = cpu_to_le32(magic);
 536         lmm->lmm_pattern = cpu_to_le32(lo->ldo_pattern);
 537         fid_to_lmm_oi(fid, &lmm->lmm_oi);
 538         if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_LMMOI))
 539                 lmm->lmm_oi.oi.oi_id++;
 540         lmm_oi_cpu_to_le(&lmm->lmm_oi, &lmm->lmm_oi);
 541         lmm->lmm_stripe_size = cpu_to_le32(lo->ldo_stripe_size);
 542         lmm->lmm_stripe_count = cpu_to_le16(lo->ldo_stripenr);
 543         if (lo->ldo_pattern & LOV_PATTERN_F_RELEASED)
 544                 lmm->lmm_stripe_count = cpu_to_le16(lo->ldo_released_stripenr);
 545         lmm->lmm_layout_gen = 0;
 546         if (magic == LOV_MAGIC_V1) {
 547                 objs = &lmm->lmm_objects[0];
 548         } else {
 549                 struct lov_mds_md_v3 *v3 = (struct lov_mds_md_v3 *) lmm;
 550                 cplen = strlcpy(v3->lmm_pool_name, lo->ldo_pool,
 551                                 sizeof(v3->lmm_pool_name));
 552                 if (cplen >= sizeof(v3->lmm_pool_name))
 553                         RETURN(-E2BIG);
 554                 objs = &v3->lmm_objects[0];
 555         }
 556
 557         for (i = 0; i < lo->ldo_stripenr; i++) {
 558                 struct lu_fid           *fid    = &info->lti_fid;
 559                 struct lod_device       *lod;
 560                 __u32                   index;
 561                 int                     type    = LU_SEQ_RANGE_OST;
 562
 563                 lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
 564                 LASSERT(lo->ldo_stripe[i]);
 565
 566                 *fid = *lu_object_fid(&lo->ldo_stripe[i]->do_lu);
 567                 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_MULTIPLE_REF)) {
 568                         if (cfs_fail_val == 0)
 569                                 cfs_fail_val = fid->f_oid;
 570                         else
 571                                 fid->f_oid = cfs_fail_val;
 572                 }
 573
 574                 rc = fid_to_ostid(fid, &info->lti_ostid);
 575                 LASSERT(rc == 0);
 576
 577                 ostid_cpu_to_le(&info->lti_ostid, &objs[i].l_ost_oi);
 578                 objs[i].l_ost_gen    = cpu_to_le32(0);
 579                 rc = lod_fld_lookup(env, lod, fid, &index, &type);
 580                 if (rc < 0) {
 581                         CERROR("%s: Can not locate "DFID": rc = %d\n",
 582                                lod2obd(lod)->obd_name, PFID(fid), rc);
 583                         lod_object_free_striping(env, lo);
 584                         RETURN(rc);
 585                 }
 586                 objs[i].l_ost_idx = cpu_to_le32(index);
 587         }
 588
 589         info->lti_buf.lb_buf = lmm;
 590         info->lti_buf.lb_len = lmm_size;
 591         rc = dt_xattr_set(env, next, &info->lti_buf, XATTR_NAME_LOV, 0,
 592                           th, BYPASS_CAPA);
 593
 594         RETURN(rc);
 595 }
 596
 597 int lod_get_ea(const struct lu_env *env, struct lod_object *lo,
 598                const char *name)
 599 {
 600         struct lod_thread_info  *info = lod_env_info(env);
 601         struct dt_object        *next = dt_object_child(&lo->ldo_obj);
 602         int                     rc;
 603         ENTRY;
 604
 605         LASSERT(info);
 606
 607         if (unlikely(info->lti_ea_store == NULL)) {
 608                 /* just to enter in allocation block below */
 609                 rc = -ERANGE;
 610         } else {
 611 repeat:
 612                 info->lti_buf.lb_buf = info->lti_ea_store;
 613                 info->lti_buf.lb_len = info->lti_ea_store_size;
 614                 rc = dt_xattr_get(env, next, &info->lti_buf, name, BYPASS_CAPA);
 615         }
 616
 617         /* if object is not striped or inaccessible */
 618         if (rc == -ENODATA || rc == -ENOENT)
 619                 RETURN(0);
 620
 621         if (rc == -ERANGE) {
 622                 /* EA doesn't fit, reallocate new buffer */
 623                 rc = dt_xattr_get(env, next, &LU_BUF_NULL, name,
 624                                   BYPASS_CAPA);
 625                 if (rc == -ENODATA || rc == -ENOENT)
 626                         RETURN(0);
 627                 else if (rc < 0)
 628                         RETURN(rc);
 629
 630                 LASSERT(rc > 0);
 631                 rc = lod_ea_store_resize(info, rc);
 632                 if (rc)
 633                         RETURN(rc);
 634                 goto repeat;
 635         }
 636
 637         RETURN(rc);
 638 }
 639
 640 int lod_store_def_striping(const struct lu_env *env, struct dt_object *dt,
 641                            struct thandle *th)
 642 {
 643         struct lod_thread_info  *info = lod_env_info(env);
 644         struct lod_object       *lo = lod_dt_obj(dt);
 645         struct dt_object        *next = dt_object_child(dt);
 646         struct lov_user_md_v3   *v3;
 647         int                      rc;
 648         ENTRY;
 649
 650         if (S_ISDIR(dt->do_lu.lo_header->loh_attr))
 651                 RETURN(-ENOTDIR);
 652         /*
 653          * store striping defaults into new directory
 654          * used to implement defaults inheritance
 655          */
 656
 657         /* probably nothing to inherite */
 658         if (lo->ldo_striping_cached == 0)
 659                 RETURN(0);
 660
 661         if (LOVEA_DELETE_VALUES(lo->ldo_def_stripe_size, lo->ldo_def_stripenr,
 662                                 lo->ldo_def_stripe_offset))
 663                 RETURN(0);
 664
 665         v3 = info->lti_ea_store;
 666         if (info->lti_ea_store_size < sizeof(*v3)) {
 667                 rc = lod_ea_store_resize(info, sizeof(*v3));
 668                 if (rc != 0)
 669                         RETURN(rc);
 670                 v3 = info->lti_ea_store;
 671         }
 672         memset(v3, 0, sizeof(*v3));
 673         v3->lmm_magic = cpu_to_le32(LOV_USER_MAGIC_V3);
 674         v3->lmm_stripe_count = cpu_to_le16(lo->ldo_def_stripenr);
 675         v3->lmm_stripe_offset = cpu_to_le16(lo->ldo_def_stripe_offset);
 676         v3->lmm_stripe_size = cpu_to_le32(lo->ldo_def_stripe_size);
 677         if (lo->ldo_pool)
 678                 strncpy(v3->lmm_pool_name, lo->ldo_pool,
 679                         sizeof(v3->lmm_pool_name));
 680         info->lti_buf.lb_buf = v3;
 681         info->lti_buf.lb_len = sizeof(*v3);
 682         rc = dt_xattr_set(env, next, &info->lti_buf, XATTR_NAME_LOV, 0, th,
 683                         BYPASS_CAPA);
 684
 685         RETURN(rc);
 686 }
 687
 688 static int validate_lod_and_idx(struct lod_device *md, int idx)
 689 {
 690         if (unlikely(idx >= md->lod_ost_descs.ltd_tgts_size ||
 691                      !cfs_bitmap_check(md->lod_ost_bitmap, idx))) {
 692                 CERROR("%s: bad idx: %d of %d\n", lod2obd(md)->obd_name, idx,
 693                        md->lod_ost_descs.ltd_tgts_size);
 694                 return -EINVAL;
 695         }
 696
 697         if (unlikely(OST_TGT(md, idx) == NULL)) {
 698                 CERROR("%s: bad lod_tgt_desc for idx: %d\n",
 699                        lod2obd(md)->obd_name, idx);
 700                 return -EINVAL;
 701         }
 702
 703         if (unlikely(OST_TGT(md, idx)->ltd_ost == NULL)) {
 704                 CERROR("%s: invalid lod device, for idx: %d\n",
 705                        lod2obd(md)->obd_name , idx);
 706                 return -EINVAL;
 707         }
 708
 709         return 0;
 710 }
 711
 712 /*
 713  * allocate array of objects pointers, find/create objects
 714  * stripenr and other fields should be initialized by this moment
 715  */
 716 int lod_initialize_objects(const struct lu_env *env, struct lod_object *lo,
 717                            struct lov_ost_data_v1 *objs)
 718 {
 719         struct lod_thread_info  *info = lod_env_info(env);
 720         struct lod_device       *md;
 721         struct lu_object        *o, *n;
 722         struct lu_device        *nd;
 723         struct dt_object       **stripe;
 724         int                      stripe_len;
 725         int                      i, idx, rc = 0;
 726         ENTRY;
 727
 728         LASSERT(lo != NULL);
 729         md = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
 730         LASSERT(lo->ldo_stripe == NULL);
 731         LASSERT(lo->ldo_stripenr > 0);
 732         LASSERT(lo->ldo_stripe_size > 0);
 733
 734         stripe_len = lo->ldo_stripenr;
 735         OBD_ALLOC(stripe, sizeof(stripe[0]) * stripe_len);
 736         if (stripe == NULL)
 737                 RETURN(-ENOMEM);
 738
 739         for (i = 0; i < lo->ldo_stripenr; i++) {
 740                 ostid_le_to_cpu(&objs[i].l_ost_oi, &info->lti_ostid);
 741                 idx = le32_to_cpu(objs[i].l_ost_idx);
 742                 rc = ostid_to_fid(&info->lti_fid, &info->lti_ostid, idx);
 743                 if (rc != 0)
 744                         GOTO(out, rc);
 745                 LASSERTF(fid_is_sane(&info->lti_fid), ""DFID" insane!\n",
 746                          PFID(&info->lti_fid));
 747                 lod_getref(&md->lod_ost_descs);
 748
 749                 rc = validate_lod_and_idx(md, idx);
 750                 if (unlikely(rc != 0)) {
 751                         lod_putref(md, &md->lod_ost_descs);
 752                         GOTO(out, rc);
 753                 }
 754
 755                 nd = &OST_TGT(md,idx)->ltd_ost->dd_lu_dev;
 756                 lod_putref(md, &md->lod_ost_descs);
 757
 758                 /* In the function below, .hs_keycmp resolves to
 759                  * u_obj_hop_keycmp() */
 760                 /* coverity[overrun-buffer-val] */
 761                 o = lu_object_find_at(env, nd, &info->lti_fid, NULL);
 762                 if (IS_ERR(o))
 763                         GOTO(out, rc = PTR_ERR(o));
 764
 765                 n = lu_object_locate(o->lo_header, nd->ld_type);
 766                 LASSERT(n);
 767
 768                 stripe[i] = container_of(n, struct dt_object, do_lu);
 769         }
 770
 771 out:
 772         if (rc != 0) {
 773                 for (i = 0; i < stripe_len; i++)
 774                         if (stripe[i] != NULL)
 775                                 lu_object_put(env, &stripe[i]->do_lu);
 776
 777                 OBD_FREE(stripe, sizeof(stripe[0]) * stripe_len);
 778         } else {
 779                 lo->ldo_stripe = stripe;
 780                 lo->ldo_stripes_allocated = stripe_len;
 781         }
 782
 783         RETURN(rc);
 784 }
 785
 786 /*
 787  * Parse striping information stored in lti_ea_store
 788  */
 789 int lod_parse_striping(const struct lu_env *env, struct lod_object *lo,
 790                        const struct lu_buf *buf)
 791 {
 792         struct lov_mds_md_v1    *lmm;
 793         struct lov_ost_data_v1  *objs;
 794         __u32                    magic;
 795         __u32                    pattern;
 796         int                      rc = 0;
 797         ENTRY;
 798
 799         LASSERT(buf);
 800         LASSERT(buf->lb_buf);
 801         LASSERT(buf->lb_len);
 802
 803         lmm = (struct lov_mds_md_v1 *) buf->lb_buf;
 804         magic = le32_to_cpu(lmm->lmm_magic);
 805         pattern = le32_to_cpu(lmm->lmm_pattern);
 806
 807         if (magic != LOV_MAGIC_V1 && magic != LOV_MAGIC_V3)
 808                 GOTO(out, rc = -EINVAL);
 809         if (lov_pattern(pattern) != LOV_PATTERN_RAID0)
 810                 GOTO(out, rc = -EINVAL);
 811
 812         lo->ldo_pattern = pattern;
 813         lo->ldo_stripe_size = le32_to_cpu(lmm->lmm_stripe_size);
 814         lo->ldo_layout_gen = le16_to_cpu(lmm->lmm_layout_gen);
 815         lo->ldo_stripenr = le16_to_cpu(lmm->lmm_stripe_count);
 816         /* released file stripenr fixup. */
 817         if (pattern & LOV_PATTERN_F_RELEASED)
 818                 lo->ldo_stripenr = 0;
 819
 820         LASSERT(buf->lb_len >= lov_mds_md_size(lo->ldo_stripenr, magic));
 821
 822         if (magic == LOV_MAGIC_V3) {
 823                 struct lov_mds_md_v3 *v3 = (struct lov_mds_md_v3 *) lmm;
 824                 objs = &v3->lmm_objects[0];
 825                 lod_object_set_pool(lo, v3->lmm_pool_name);
 826         } else {
 827                 objs = &lmm->lmm_objects[0];
 828         }
 829
 830         if (lo->ldo_stripenr > 0)
 831                 rc = lod_initialize_objects(env, lo, objs);
 832
 833 out:
 834         RETURN(rc);
 835 }
 836
 837 /*
 838  * Load and parse striping information, create in-core representation for the
 839  * stripes
 840  */
 841 int lod_load_striping(const struct lu_env *env, struct lod_object *lo)
 842 {
 843         struct lod_thread_info  *info = lod_env_info(env);
 844         struct dt_object        *next = dt_object_child(&lo->ldo_obj);
 845         int                      rc = 0;
 846         ENTRY;
 847
 848         /*
 849          * currently this code is supposed to be called from declaration
 850          * phase only, thus the object is not expected to be locked by caller
 851          */
 852         dt_write_lock(env, next, 0);
 853         /* already initialized? */
 854         if (lo->ldo_stripe != NULL)
 855                 GOTO(out, rc = 0);
 856
 857         /* Do not load stripe for slaves of striped dir */
 858         if (!dt_object_exists(next) || lo->ldo_dir_slave_stripe)
 859                 GOTO(out, rc = 0);
 860
 861         /* only regular files can be striped */
 862         if (lu_object_attr(lod2lu_obj(lo)) & S_IFREG) {
 863                 rc = lod_get_lov_ea(env, lo);
 864                 if (rc <= 0)
 865                         GOTO(out, rc);
 866                 /*
 867                  * there is LOV EA (striping information) in this object
 868                  * let's parse it and create in-core objects for the stripes
 869                  */
 870                 info->lti_buf.lb_buf = info->lti_ea_store;
 871                 info->lti_buf.lb_len = info->lti_ea_store_size;
 872                 rc = lod_parse_striping(env, lo, &info->lti_buf);
 873         } else if (lu_object_attr(lod2lu_obj(lo)) & S_IFDIR) {
 874                 rc = lod_get_lmv_ea(env, lo);
 875                 if (rc <= 0)
 876                         GOTO(out, rc);
 877                 /*
 878                  * there is LOV EA (striping information) in this object
 879                  * let's parse it and create in-core objects for the stripes
 880                  */
 881                 info->lti_buf.lb_buf = info->lti_ea_store;
 882                 info->lti_buf.lb_len = info->lti_ea_store_size;
 883                 rc = lod_parse_dir_striping(env, lo, &info->lti_buf);
 884         }
 885 out:
 886         dt_write_unlock(env, next);
 887         RETURN(rc);
 888 }
 889
 890 int lod_verify_striping(struct lod_device *d, const struct lu_buf *buf,
 891                         int specific)
 892 {
 893         struct lov_user_md_v1   *lum;
 894         struct lov_user_md_v3   *lum3;
 895         struct pool_desc        *pool = NULL;
 896         __u32                    magic;
 897         __u32                    stripe_size;
 898         __u16                    stripe_count;
 899         __u16                    stripe_offset;
 900         size_t                   lum_size;
 901         int                      rc = 0;
 902         ENTRY;
 903
 904         lum = buf->lb_buf;
 905
 906         LASSERT(sizeof(*lum) < sizeof(*lum3));
 907
 908         if (buf->lb_len < sizeof(*lum)) {
 909                 CDEBUG(D_IOCTL, "buf len %zd too small for lov_user_md\n",
 910                        buf->lb_len);
 911                 GOTO(out, rc = -EINVAL);
 912         }
 913
 914         magic = le32_to_cpu(lum->lmm_magic);
 915         if (magic != LOV_USER_MAGIC_V1 &&
 916             magic != LOV_USER_MAGIC_V3 &&
 917             magic != LOV_MAGIC_V1_DEF &&
 918             magic != LOV_MAGIC_V3_DEF) {
 919                 CDEBUG(D_IOCTL, "bad userland LOV MAGIC: %#x\n", magic);
 920                 GOTO(out, rc = -EINVAL);
 921         }
 922
 923         if ((specific && le32_to_cpu(lum->lmm_pattern) != LOV_PATTERN_RAID0) ||
 924             (!specific && lum->lmm_pattern != 0)) {
 925                 CDEBUG(D_IOCTL, "bad userland stripe pattern: %#x\n",
 926                        le32_to_cpu(lum->lmm_pattern));
 927                 GOTO(out, rc = -EINVAL);
 928         }
 929
 930         /* 64kB is the largest common page size we see (ia64), and matches the
 931          * check in lfs */
 932         stripe_size = le32_to_cpu(lum->lmm_stripe_size);
 933         if (stripe_size & (LOV_MIN_STRIPE_SIZE - 1)) {
 934                 CDEBUG(D_IOCTL, "stripe size %u not a multiple of %u\n",
 935                        stripe_size, LOV_MIN_STRIPE_SIZE);
 936                 GOTO(out, rc = -EINVAL);
 937         }
 938
 939         /* an offset of -1 is treated as a "special" valid offset */
 940         stripe_offset = le16_to_cpu(lum->lmm_stripe_offset);
 941         if (stripe_offset != (typeof(stripe_offset))-1) {
 942                 /* if offset is not within valid range [0, osts_size) */
 943                 if (stripe_offset >= d->lod_osts_size) {
 944                         CDEBUG(D_IOCTL, "stripe offset %u >= bitmap size %u\n",
 945                                stripe_offset, d->lod_osts_size);
 946                         GOTO(out, rc = -EINVAL);
 947                 }
 948
 949                 /* if lmm_stripe_offset is *not* in bitmap */
 950                 if (!cfs_bitmap_check(d->lod_ost_bitmap, stripe_offset)) {
 951                         CDEBUG(D_IOCTL, "stripe offset %u not in bitmap\n",
 952                                stripe_offset);
 953                         GOTO(out, rc = -EINVAL);
 954                 }
 955         }
 956
 957         stripe_count = le16_to_cpu(lum->lmm_stripe_count);
 958         if (magic == LOV_USER_MAGIC_V1 || magic == LOV_MAGIC_V1_DEF)
 959                 lum_size = offsetof(struct lov_user_md_v1,
 960                                     lmm_objects[stripe_count]);
 961         else if (magic == LOV_USER_MAGIC_V3 || magic == LOV_MAGIC_V3_DEF)
 962                 lum_size = offsetof(struct lov_user_md_v3,
 963                                     lmm_objects[stripe_count]);
 964         else
 965                 LBUG();
 966
 967         if (specific && buf->lb_len != lum_size) {
 968                 CDEBUG(D_IOCTL, "invalid buf len %zd for lov_user_md with "
 969                        "magic %#x and stripe_count %u\n",
 970                        buf->lb_len, magic, stripe_count);
 971                 GOTO(out, rc = -EINVAL);
 972         }
 973
 974         if (!(magic == LOV_USER_MAGIC_V3 || magic == LOV_MAGIC_V3_DEF))
 975                 goto out;
 976
 977         lum3 = buf->lb_buf;
 978         if (buf->lb_len < sizeof(*lum3)) {
 979                 CDEBUG(D_IOCTL, "buf len %zd too small for lov_user_md_v3\n",
 980                        buf->lb_len);
 981                 GOTO(out, rc = -EINVAL);
 982         }
 983
 984         /* In the function below, .hs_keycmp resolves to
 985          * pool_hashkey_keycmp() */
 986         /* coverity[overrun-buffer-val] */
 987         pool = lod_find_pool(d, lum3->lmm_pool_name);
 988         if (pool == NULL)
 989                 goto out;
 990
 991         if (stripe_offset != (typeof(stripe_offset))-1) {
 992                 rc = lod_check_index_in_pool(stripe_offset, pool);
 993                 if (rc < 0)
 994                         GOTO(out, rc = -EINVAL);
 995         }
 996
 997         if (specific && stripe_count > pool_tgt_count(pool)) {
 998                 CDEBUG(D_IOCTL,
 999                        "stripe count %u > # OSTs %u in the pool\n",
1000                        stripe_count, pool_tgt_count(pool));
1001                 GOTO(out, rc = -EINVAL);
1002         }
1003
1004 out:
1005         if (pool != NULL)
1006                 lod_pool_putref(pool);
1007
1008         RETURN(rc);
1009 }
1010
1011 void lod_fix_desc_stripe_size(__u64 *val)
1012 {
1013         if (*val < LOV_MIN_STRIPE_SIZE) {
1014                 if (*val != 0)
1015                         LCONSOLE_INFO("Increasing default stripe size to "
1016                                       "minimum value %u\n",
1017                                       LOV_DESC_STRIPE_SIZE_DEFAULT);
1018                 *val = LOV_DESC_STRIPE_SIZE_DEFAULT;
1019         } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) {
1020                 *val &= ~(LOV_MIN_STRIPE_SIZE - 1);
1021                 LCONSOLE_WARN("Changing default stripe size to "LPU64" (a "
1022                               "multiple of %u)\n",
1023                               *val, LOV_MIN_STRIPE_SIZE);
1024         }
1025 }
1026
1027 void lod_fix_desc_stripe_count(__u32 *val)
1028 {
1029         if (*val == 0)
1030                 *val = 1;
1031 }
1032
1033 void lod_fix_desc_pattern(__u32 *val)
1034 {
1035         /* from lov_setstripe */
1036         if ((*val != 0) && (*val != LOV_PATTERN_RAID0)) {
1037                 LCONSOLE_WARN("Unknown stripe pattern: %#x\n", *val);
1038                 *val = 0;
1039         }
1040 }
1041
1042 void lod_fix_desc_qos_maxage(__u32 *val)
1043 {
1044         /* fix qos_maxage */
1045         if (*val == 0)
1046                 *val = LOV_DESC_QOS_MAXAGE_DEFAULT;
1047 }
1048
1049 void lod_fix_desc(struct lov_desc *desc)
1050 {
1051         lod_fix_desc_stripe_size(&desc->ld_default_stripe_size);
1052         lod_fix_desc_stripe_count(&desc->ld_default_stripe_count);
1053         lod_fix_desc_pattern(&desc->ld_pattern);
1054         lod_fix_desc_qos_maxage(&desc->ld_qos_maxage);
1055 }
1056
1057 int lod_pools_init(struct lod_device *lod, struct lustre_cfg *lcfg)
1058 {
1059         struct obd_device          *obd;
1060         struct lov_desc            *desc;
1061         int                         rc;
1062         ENTRY;
1063
1064         obd = class_name2obd(lustre_cfg_string(lcfg, 0));
1065         LASSERT(obd != NULL);
1066         obd->obd_lu_dev = &lod->lod_dt_dev.dd_lu_dev;
1067
1068         if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
1069                 CERROR("LOD setup requires a descriptor\n");
1070                 RETURN(-EINVAL);
1071         }
1072
1073         desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
1074
1075         if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
1076                 CERROR("descriptor size wrong: %d > %d\n",
1077                        (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
1078                 RETURN(-EINVAL);
1079         }
1080
1081         if (desc->ld_magic != LOV_DESC_MAGIC) {
1082                 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
1083                         CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
1084                                obd->obd_name, desc);
1085                         lustre_swab_lov_desc(desc);
1086                 } else {
1087                         CERROR("%s: Bad lov desc magic: %#x\n",
1088                                obd->obd_name, desc->ld_magic);
1089                         RETURN(-EINVAL);
1090                 }
1091         }
1092
1093         lod_fix_desc(desc);
1094
1095         desc->ld_active_tgt_count = 0;
1096         lod->lod_desc = *desc;
1097
1098         lod->lod_sp_me = LUSTRE_SP_CLI;
1099
1100         /* Set up allocation policy (QoS and RR) */
1101         CFS_INIT_LIST_HEAD(&lod->lod_qos.lq_oss_list);
1102         init_rwsem(&lod->lod_qos.lq_rw_sem);
1103         lod->lod_qos.lq_dirty = 1;
1104         lod->lod_qos.lq_rr.lqr_dirty = 1;
1105         lod->lod_qos.lq_reset = 1;
1106         /* Default priority is toward free space balance */
1107         lod->lod_qos.lq_prio_free = 232;
1108         /* Default threshold for rr (roughly 17%) */
1109         lod->lod_qos.lq_threshold_rr = 43;
1110
1111         /* Set up OST pool environment */
1112         lod->lod_pools_hash_body = cfs_hash_create("POOLS", HASH_POOLS_CUR_BITS,
1113                                                    HASH_POOLS_MAX_BITS,
1114                                                    HASH_POOLS_BKT_BITS, 0,
1115                                                    CFS_HASH_MIN_THETA,
1116                                                    CFS_HASH_MAX_THETA,
1117                                                    &pool_hash_operations,
1118                                                    CFS_HASH_DEFAULT);
1119         if (lod->lod_pools_hash_body == NULL)
1120                 RETURN(-ENOMEM);
1121
1122         CFS_INIT_LIST_HEAD(&lod->lod_pool_list);
1123         lod->lod_pool_count = 0;
1124         rc = lod_ost_pool_init(&lod->lod_pool_info, 0);
1125         if (rc)
1126                 GOTO(out_hash, rc);
1127         rc = lod_ost_pool_init(&lod->lod_qos.lq_rr.lqr_pool, 0);
1128         if (rc)
1129                 GOTO(out_pool_info, rc);
1130
1131         RETURN(0);
1132
1133 out_pool_info:
1134         lod_ost_pool_free(&lod->lod_pool_info);
1135 out_hash:
1136         cfs_hash_putref(lod->lod_pools_hash_body);
1137
1138         return rc;
1139 }
1140
1141 int lod_pools_fini(struct lod_device *lod)
1142 {
1143         struct obd_device   *obd = lod2obd(lod);
1144         cfs_list_t          *pos, *tmp;
1145         struct pool_desc    *pool;
1146         ENTRY;
1147
1148         cfs_list_for_each_safe(pos, tmp, &lod->lod_pool_list) {
1149                 pool = cfs_list_entry(pos, struct pool_desc, pool_list);
1150                 /* free pool structs */
1151                 CDEBUG(D_INFO, "delete pool %p\n", pool);
1152                 /* In the function below, .hs_keycmp resolves to
1153                  * pool_hashkey_keycmp() */
1154                 /* coverity[overrun-buffer-val] */
1155                 lod_pool_del(obd, pool->pool_name);
1156         }
1157
1158         cfs_hash_putref(lod->lod_pools_hash_body);
1159         lod_ost_pool_free(&(lod->lod_qos.lq_rr.lqr_pool));
1160         lod_ost_pool_free(&lod->lod_pool_info);
1161
1162         RETURN(0);
1163 }
1164