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