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 #include <lustre_lmv.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                 unsigned 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                                            OBD_CONNECT_LFSCK;
 233
 234                 data->ocd_group = tgt_index;
 235                 ltd = &lod->lod_ost_descs;
 236         } else {
 237                 struct obd_import *imp = obd->u.cli.cl_import;
 238
 239                 for_ost = false;
 240                 data->ocd_ibits_known = MDS_INODELOCK_UPDATE;
 241                 data->ocd_connect_flags |= OBD_CONNECT_ACL |
 242                                            OBD_CONNECT_MDS_CAPA |
 243                                            OBD_CONNECT_OSS_CAPA |
 244                                            OBD_CONNECT_IBITS |
 245                                            OBD_CONNECT_MDS_MDS |
 246                                            OBD_CONNECT_FID |
 247                                            OBD_CONNECT_AT |
 248                                            OBD_CONNECT_FULL20 |
 249                                            OBD_CONNECT_LFSCK;
 250                 /* XXX set MDS-MDS flags, remove this when running this
 251                  * on client*/
 252                 data->ocd_connect_flags |= OBD_CONNECT_MDS_MDS;
 253                 spin_lock(&imp->imp_lock);
 254                 imp->imp_server_timeout = 1;
 255                 spin_unlock(&imp->imp_lock);
 256                 imp->imp_client->cli_request_portal = OUT_PORTAL;
 257                 CDEBUG(D_OTHER, "%s: Set 'mds' portal and timeout\n",
 258                       obd->obd_name);
 259                 ltd = &lod->lod_mdt_descs;
 260         }
 261
 262         rc = obd_connect(env, &exp, obd, &obd->obd_uuid, data, NULL);
 263         OBD_FREE_PTR(data);
 264         if (rc) {
 265                 CERROR("%s: cannot connect to next dev %s (%d)\n",
 266                        obd->obd_name, osp, rc);
 267                 GOTO(out_free, rc);
 268         }
 269
 270         LASSERT(obd->obd_lu_dev);
 271         LASSERT(obd->obd_lu_dev->ld_site == lod->lod_dt_dev.dd_lu_dev.ld_site);
 272
 273         ldev = obd->obd_lu_dev;
 274         d = lu2dt_dev(ldev);
 275
 276         /* Allocate ost descriptor and fill it */
 277         OBD_ALLOC_PTR(tgt_desc);
 278         if (!tgt_desc)
 279                 GOTO(out_conn, rc = -ENOMEM);
 280
 281         tgt_desc->ltd_tgt    = d;
 282         tgt_desc->ltd_exp    = exp;
 283         tgt_desc->ltd_uuid   = obd->u.cli.cl_target_uuid;
 284         tgt_desc->ltd_gen    = gen;
 285         tgt_desc->ltd_index  = index;
 286         tgt_desc->ltd_active = active;
 287
 288         lod_getref(ltd);
 289         if (index >= ltd->ltd_tgts_size) {
 290                 /* we have to increase the size of the lod_osts array */
 291                 __u32  newsize;
 292
 293                 newsize = max(ltd->ltd_tgts_size, (__u32)2);
 294                 while (newsize < index + 1)
 295                         newsize = newsize << 1;
 296
 297                 /* lod_bitmap_resize() needs lod_rw_sem
 298                  * which we hold with th reference */
 299                 lod_putref(lod, ltd);
 300
 301                 rc = ltd_bitmap_resize(ltd, newsize);
 302                 if (rc)
 303                         GOTO(out_desc, rc);
 304
 305                 lod_getref(ltd);
 306         }
 307
 308         mutex_lock(&ltd->ltd_mutex);
 309         if (cfs_bitmap_check(ltd->ltd_tgt_bitmap, index)) {
 310                 CERROR("%s: device %d is registered already\n", obd->obd_name,
 311                        index);
 312                 GOTO(out_mutex, rc = -EEXIST);
 313         }
 314
 315         if (ltd->ltd_tgt_idx[index / TGT_PTRS_PER_BLOCK] == NULL) {
 316                 OBD_ALLOC_PTR(ltd->ltd_tgt_idx[index / TGT_PTRS_PER_BLOCK]);
 317                 if (ltd->ltd_tgt_idx[index / TGT_PTRS_PER_BLOCK] == NULL) {
 318                         CERROR("can't allocate index to add %s\n",
 319                                obd->obd_name);
 320                         GOTO(out_mutex, rc = -ENOMEM);
 321                 }
 322         }
 323
 324         if (!strcmp(LUSTRE_OSC_NAME, type)) {
 325                 /* pool and qos are not supported for MDS stack yet */
 326                 rc = lod_ost_pool_add(&lod->lod_pool_info, index,
 327                                       lod->lod_osts_size);
 328                 if (rc) {
 329                         CERROR("%s: can't set up pool, failed with %d\n",
 330                                obd->obd_name, rc);
 331                         GOTO(out_mutex, rc);
 332                 }
 333
 334                 rc = qos_add_tgt(lod, tgt_desc);
 335                 if (rc) {
 336                         CERROR("%s: qos_add_tgt failed with %d\n",
 337                                 obd->obd_name, rc);
 338                         GOTO(out_pool, rc);
 339                 }
 340
 341                 /* The new OST is now a full citizen */
 342                 if (index >= lod->lod_desc.ld_tgt_count)
 343                         lod->lod_desc.ld_tgt_count = index + 1;
 344                 if (active)
 345                         lod->lod_desc.ld_active_tgt_count++;
 346         }
 347
 348         LTD_TGT(ltd, index) = tgt_desc;
 349         cfs_bitmap_set(ltd->ltd_tgt_bitmap, index);
 350         ltd->ltd_tgtnr++;
 351         mutex_unlock(&ltd->ltd_mutex);
 352         lod_putref(lod, ltd);
 353         if (lod->lod_recovery_completed)
 354                 ldev->ld_ops->ldo_recovery_complete(env, ldev);
 355
 356         rc = lfsck_add_target(env, lod->lod_child, d, exp, index, for_ost);
 357         if (rc != 0)
 358                 CERROR("Fail to add LFSCK target: name = %s, type = %s, "
 359                        "index = %u, rc = %d\n", osp, type, index, rc);
 360
 361         RETURN(rc);
 362
 363 out_pool:
 364         lod_ost_pool_remove(&lod->lod_pool_info, index);
 365 out_mutex:
 366         mutex_unlock(&ltd->ltd_mutex);
 367         lod_putref(lod, ltd);
 368 out_desc:
 369         OBD_FREE_PTR(tgt_desc);
 370 out_conn:
 371         obd_disconnect(exp);
 372 out_free:
 373         return rc;
 374 }
 375
 376 /*
 377  * helper function to schedule OST removal from the device table
 378  */
 379 static void __lod_del_device(const struct lu_env *env, struct lod_device *lod,
 380                              struct lod_tgt_descs *ltd, unsigned idx,
 381                              bool for_ost)
 382 {
 383         LASSERT(LTD_TGT(ltd, idx));
 384
 385         lfsck_del_target(env, lod->lod_child, LTD_TGT(ltd, idx)->ltd_tgt,
 386                          idx, for_ost);
 387
 388         if (LTD_TGT(ltd, idx)->ltd_reap == 0) {
 389                 LTD_TGT(ltd, idx)->ltd_reap = 1;
 390                 ltd->ltd_death_row++;
 391         }
 392 }
 393
 394 int lod_fini_tgt(const struct lu_env *env, struct lod_device *lod,
 395                  struct lod_tgt_descs *ltd, bool for_ost)
 396 {
 397         unsigned int idx;
 398
 399         if (ltd->ltd_tgts_size <= 0)
 400                 return 0;
 401         lod_getref(ltd);
 402         mutex_lock(&ltd->ltd_mutex);
 403         cfs_foreach_bit(ltd->ltd_tgt_bitmap, idx)
 404                 __lod_del_device(env, lod, ltd, idx, for_ost);
 405         mutex_unlock(&ltd->ltd_mutex);
 406         lod_putref(lod, ltd);
 407         CFS_FREE_BITMAP(ltd->ltd_tgt_bitmap);
 408         for (idx = 0; idx < TGT_PTRS; idx++) {
 409                 if (ltd->ltd_tgt_idx[idx])
 410                         OBD_FREE_PTR(ltd->ltd_tgt_idx[idx]);
 411         }
 412         ltd->ltd_tgts_size = 0;
 413         return 0;
 414 }
 415
 416 /*
 417  * Add support for administratively disabled OST (through the MGS).
 418  * Schedule a target for deletion.  Disconnection and real removal from the
 419  * table takes place in lod_putref() once the last table user release its
 420  * reference.
 421  *
 422  * \param env - is the environment passed by the caller
 423  * \param lod - is the lod device currently connected to the OSP about to be
 424  *              removed
 425  * \param osp - is the name of OSP device about to be removed
 426  * \param idx - is the OSP index
 427  * \param gen - is the generation number, not used currently
 428  */
 429 int lod_del_device(const struct lu_env *env, struct lod_device *lod,
 430                    struct lod_tgt_descs *ltd, char *osp, unsigned idx,
 431                    unsigned gen, bool for_ost)
 432 {
 433         struct obd_device *obd;
 434         int                rc = 0;
 435         struct obd_uuid    uuid;
 436         ENTRY;
 437
 438         CDEBUG(D_CONFIG, "osp:%s idx:%d gen:%d\n", osp, idx, gen);
 439
 440         obd_str2uuid(&uuid, osp);
 441
 442         obd = class_find_client_obd(&uuid, LUSTRE_OSP_NAME,
 443                                    &lod->lod_dt_dev.dd_lu_dev.ld_obd->obd_uuid);
 444         if (obd == NULL) {
 445                 CERROR("can't find %s device\n", osp);
 446                 RETURN(-EINVAL);
 447         }
 448
 449         if (gen <= 0) {
 450                 CERROR("%s: request to remove OBD %s with invalid generation %d"
 451                        "\n", obd->obd_name, osp, gen);
 452                 RETURN(-EINVAL);
 453         }
 454
 455         obd_str2uuid(&uuid,  osp);
 456
 457         lod_getref(ltd);
 458         mutex_lock(&ltd->ltd_mutex);
 459         /* check that the index is allocated in the bitmap */
 460         if (!cfs_bitmap_check(ltd->ltd_tgt_bitmap, idx) ||
 461             !LTD_TGT(ltd, idx)) {
 462                 CERROR("%s: device %d is not set up\n", obd->obd_name, idx);
 463                 GOTO(out, rc = -EINVAL);
 464         }
 465
 466         /* check that the UUID matches */
 467         if (!obd_uuid_equals(&uuid, &LTD_TGT(ltd, idx)->ltd_uuid)) {
 468                 CERROR("%s: LOD target UUID %s at index %d does not match %s\n",
 469                        obd->obd_name, obd_uuid2str(&LTD_TGT(ltd,idx)->ltd_uuid),
 470                        idx, osp);
 471                 GOTO(out, rc = -EINVAL);
 472         }
 473
 474         __lod_del_device(env, lod, ltd, idx, for_ost);
 475         EXIT;
 476 out:
 477         mutex_unlock(&ltd->ltd_mutex);
 478         lod_putref(lod, ltd);
 479         return(rc);
 480 }
 481
 482 int lod_ea_store_resize(struct lod_thread_info *info, size_t size)
 483 {
 484         __u32 round = size_roundup_power2(size);
 485
 486         LASSERT(round <=
 487                 lov_mds_md_size(LOV_MAX_STRIPE_COUNT, LOV_MAGIC_V3));
 488         if (info->lti_ea_store) {
 489                 LASSERT(info->lti_ea_store_size);
 490                 LASSERT(info->lti_ea_store_size < round);
 491                 CDEBUG(D_INFO, "EA store size %d is not enough, need %d\n",
 492                        info->lti_ea_store_size, round);
 493                 OBD_FREE_LARGE(info->lti_ea_store, info->lti_ea_store_size);
 494                 info->lti_ea_store = NULL;
 495                 info->lti_ea_store_size = 0;
 496         }
 497
 498         OBD_ALLOC_LARGE(info->lti_ea_store, round);
 499         if (info->lti_ea_store == NULL)
 500                 RETURN(-ENOMEM);
 501         info->lti_ea_store_size = round;
 502         RETURN(0);
 503 }
 504
 505 /*
 506  * generate and write LOV EA for given striped object
 507  */
 508 int lod_generate_and_set_lovea(const struct lu_env *env,
 509                                struct lod_object *lo, struct thandle *th)
 510 {
 511         struct lod_thread_info  *info = lod_env_info(env);
 512         struct dt_object        *next = dt_object_child(&lo->ldo_obj);
 513         const struct lu_fid     *fid  = lu_object_fid(&lo->ldo_obj.do_lu);
 514         struct lov_mds_md_v1    *lmm;
 515         struct lov_ost_data_v1  *objs;
 516         __u32                    magic;
 517         int                      i, rc;
 518         size_t                   lmm_size;
 519         ENTRY;
 520
 521         LASSERT(lo);
 522
 523         magic = lo->ldo_pool != NULL ? LOV_MAGIC_V3 : LOV_MAGIC_V1;
 524         lmm_size = lov_mds_md_size(lo->ldo_stripenr, magic);
 525         if (info->lti_ea_store_size < lmm_size) {
 526                 rc = lod_ea_store_resize(info, lmm_size);
 527                 if (rc)
 528                         RETURN(rc);
 529         }
 530
 531         if (lo->ldo_pattern == 0) /* default striping */
 532                 lo->ldo_pattern = LOV_PATTERN_RAID0;
 533
 534         lmm = info->lti_ea_store;
 535
 536         lmm->lmm_magic = cpu_to_le32(magic);
 537         lmm->lmm_pattern = cpu_to_le32(lo->ldo_pattern);
 538         fid_to_lmm_oi(fid, &lmm->lmm_oi);
 539         if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_LMMOI))
 540                 lmm->lmm_oi.oi.oi_id++;
 541         lmm_oi_cpu_to_le(&lmm->lmm_oi, &lmm->lmm_oi);
 542         lmm->lmm_stripe_size = cpu_to_le32(lo->ldo_stripe_size);
 543         lmm->lmm_stripe_count = cpu_to_le16(lo->ldo_stripenr);
 544         if (lo->ldo_pattern & LOV_PATTERN_F_RELEASED)
 545                 lmm->lmm_stripe_count = cpu_to_le16(lo->ldo_released_stripenr);
 546         lmm->lmm_layout_gen = 0;
 547         if (magic == LOV_MAGIC_V1) {
 548                 objs = &lmm->lmm_objects[0];
 549         } else {
 550                 struct lov_mds_md_v3 *v3 = (struct lov_mds_md_v3 *) lmm;
 551                 size_t cplen = strlcpy(v3->lmm_pool_name, lo->ldo_pool,
 552                                 sizeof(v3->lmm_pool_name));
 553                 if (cplen >= sizeof(v3->lmm_pool_name))
 554                         RETURN(-E2BIG);
 555                 objs = &v3->lmm_objects[0];
 556         }
 557
 558         for (i = 0; i < lo->ldo_stripenr; i++) {
 559                 struct lu_fid           *fid    = &info->lti_fid;
 560                 struct lod_device       *lod;
 561                 __u32                   index;
 562                 int                     type    = LU_SEQ_RANGE_OST;
 563
 564                 lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
 565                 LASSERT(lo->ldo_stripe[i]);
 566
 567                 *fid = *lu_object_fid(&lo->ldo_stripe[i]->do_lu);
 568                 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_MULTIPLE_REF)) {
 569                         if (cfs_fail_val == 0)
 570                                 cfs_fail_val = fid->f_oid;
 571                         else
 572                                 fid->f_oid = cfs_fail_val;
 573                 }
 574
 575                 rc = fid_to_ostid(fid, &info->lti_ostid);
 576                 LASSERT(rc == 0);
 577
 578                 ostid_cpu_to_le(&info->lti_ostid, &objs[i].l_ost_oi);
 579                 objs[i].l_ost_gen    = cpu_to_le32(0);
 580                 rc = lod_fld_lookup(env, lod, fid, &index, &type);
 581                 if (rc < 0) {
 582                         CERROR("%s: Can not locate "DFID": rc = %d\n",
 583                                lod2obd(lod)->obd_name, PFID(fid), rc);
 584                         lod_object_free_striping(env, lo);
 585                         RETURN(rc);
 586                 }
 587                 objs[i].l_ost_idx = cpu_to_le32(index);
 588         }
 589
 590         info->lti_buf.lb_buf = lmm;
 591         info->lti_buf.lb_len = lmm_size;
 592         rc = dt_xattr_set(env, next, &info->lti_buf, XATTR_NAME_LOV, 0,
 593                           th, BYPASS_CAPA);
 594         if (rc < 0)
 595                 lod_object_free_striping(env, lo);
 596
 597         RETURN(rc);
 598 }
 599
 600 int lod_get_ea(const struct lu_env *env, struct lod_object *lo,
 601                const char *name)
 602 {
 603         struct lod_thread_info  *info = lod_env_info(env);
 604         struct dt_object        *next = dt_object_child(&lo->ldo_obj);
 605         int                     rc;
 606         ENTRY;
 607
 608         LASSERT(info);
 609
 610         if (unlikely(info->lti_ea_store == NULL)) {
 611                 /* just to enter in allocation block below */
 612                 rc = -ERANGE;
 613         } else {
 614 repeat:
 615                 info->lti_buf.lb_buf = info->lti_ea_store;
 616                 info->lti_buf.lb_len = info->lti_ea_store_size;
 617                 rc = dt_xattr_get(env, next, &info->lti_buf, name, BYPASS_CAPA);
 618         }
 619
 620         /* if object is not striped or inaccessible */
 621         if (rc == -ENODATA || rc == -ENOENT)
 622                 RETURN(0);
 623
 624         if (rc == -ERANGE) {
 625                 /* EA doesn't fit, reallocate new buffer */
 626                 rc = dt_xattr_get(env, next, &LU_BUF_NULL, name,
 627                                   BYPASS_CAPA);
 628                 if (rc == -ENODATA || rc == -ENOENT)
 629                         RETURN(0);
 630                 else if (rc < 0)
 631                         RETURN(rc);
 632
 633                 LASSERT(rc > 0);
 634                 rc = lod_ea_store_resize(info, rc);
 635                 if (rc)
 636                         RETURN(rc);
 637                 goto repeat;
 638         }
 639
 640         RETURN(rc);
 641 }
 642
 643 int lod_store_def_striping(const struct lu_env *env, struct dt_object *dt,
 644                            struct thandle *th)
 645 {
 646         struct lod_thread_info  *info = lod_env_info(env);
 647         struct lod_object       *lo = lod_dt_obj(dt);
 648         struct dt_object        *next = dt_object_child(dt);
 649         struct lov_user_md_v3   *v3;
 650         int                      rc;
 651         ENTRY;
 652
 653         if (S_ISDIR(dt->do_lu.lo_header->loh_attr))
 654                 RETURN(-ENOTDIR);
 655         /*
 656          * store striping defaults into new directory
 657          * used to implement defaults inheritance
 658          */
 659
 660         /* probably nothing to inherite */
 661         if (lo->ldo_striping_cached == 0)
 662                 RETURN(0);
 663
 664         if (LOVEA_DELETE_VALUES(lo->ldo_def_stripe_size, lo->ldo_def_stripenr,
 665                                 lo->ldo_def_stripe_offset))
 666                 RETURN(0);
 667
 668         v3 = info->lti_ea_store;
 669         if (info->lti_ea_store_size < sizeof(*v3)) {
 670                 rc = lod_ea_store_resize(info, sizeof(*v3));
 671                 if (rc != 0)
 672                         RETURN(rc);
 673                 v3 = info->lti_ea_store;
 674         }
 675         memset(v3, 0, sizeof(*v3));
 676         v3->lmm_magic = cpu_to_le32(LOV_USER_MAGIC_V3);
 677         v3->lmm_stripe_count = cpu_to_le16(lo->ldo_def_stripenr);
 678         v3->lmm_stripe_offset = cpu_to_le16(lo->ldo_def_stripe_offset);
 679         v3->lmm_stripe_size = cpu_to_le32(lo->ldo_def_stripe_size);
 680         if (lo->ldo_pool != NULL)
 681                 strlcpy(v3->lmm_pool_name, lo->ldo_pool,
 682                         sizeof(v3->lmm_pool_name));
 683         info->lti_buf.lb_buf = v3;
 684         info->lti_buf.lb_len = sizeof(*v3);
 685         rc = dt_xattr_set(env, next, &info->lti_buf, XATTR_NAME_LOV, 0, th,
 686                         BYPASS_CAPA);
 687
 688         RETURN(rc);
 689 }
 690
 691 static int validate_lod_and_idx(struct lod_device *md, __u32 idx)
 692 {
 693         if (unlikely(idx >= md->lod_ost_descs.ltd_tgts_size ||
 694                      !cfs_bitmap_check(md->lod_ost_bitmap, idx))) {
 695                 CERROR("%s: bad idx: %d of %d\n", lod2obd(md)->obd_name, idx,
 696                        md->lod_ost_descs.ltd_tgts_size);
 697                 return -EINVAL;
 698         }
 699
 700         if (unlikely(OST_TGT(md, idx) == NULL)) {
 701                 CERROR("%s: bad lod_tgt_desc for idx: %d\n",
 702                        lod2obd(md)->obd_name, idx);
 703                 return -EINVAL;
 704         }
 705
 706         if (unlikely(OST_TGT(md, idx)->ltd_ost == NULL)) {
 707                 CERROR("%s: invalid lod device, for idx: %d\n",
 708                        lod2obd(md)->obd_name , idx);
 709                 return -EINVAL;
 710         }
 711
 712         return 0;
 713 }
 714
 715 /*
 716  * allocate array of objects pointers, find/create objects
 717  * stripenr and other fields should be initialized by this moment
 718  */
 719 int lod_initialize_objects(const struct lu_env *env, struct lod_object *lo,
 720                            struct lov_ost_data_v1 *objs)
 721 {
 722         struct lod_thread_info  *info = lod_env_info(env);
 723         struct lod_device       *md;
 724         struct lu_object        *o, *n;
 725         struct lu_device        *nd;
 726         struct dt_object       **stripe;
 727         int                      stripe_len;
 728         int                      i, rc = 0;
 729         __u32                   idx;
 730         ENTRY;
 731
 732         LASSERT(lo != NULL);
 733         md = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
 734         LASSERT(lo->ldo_stripe == NULL);
 735         LASSERT(lo->ldo_stripenr > 0);
 736         LASSERT(lo->ldo_stripe_size > 0);
 737
 738         stripe_len = lo->ldo_stripenr;
 739         OBD_ALLOC(stripe, sizeof(stripe[0]) * stripe_len);
 740         if (stripe == NULL)
 741                 RETURN(-ENOMEM);
 742
 743         for (i = 0; i < lo->ldo_stripenr; i++) {
 744                 if (unlikely(lovea_slot_is_dummy(&objs[i])))
 745                         continue;
 746
 747                 ostid_le_to_cpu(&objs[i].l_ost_oi, &info->lti_ostid);
 748                 idx = le32_to_cpu(objs[i].l_ost_idx);
 749                 rc = ostid_to_fid(&info->lti_fid, &info->lti_ostid, idx);
 750                 if (rc != 0)
 751                         GOTO(out, rc);
 752                 LASSERTF(fid_is_sane(&info->lti_fid), ""DFID" insane!\n",
 753                          PFID(&info->lti_fid));
 754                 lod_getref(&md->lod_ost_descs);
 755
 756                 rc = validate_lod_and_idx(md, idx);
 757                 if (unlikely(rc != 0)) {
 758                         lod_putref(md, &md->lod_ost_descs);
 759                         GOTO(out, rc);
 760                 }
 761
 762                 nd = &OST_TGT(md,idx)->ltd_ost->dd_lu_dev;
 763                 lod_putref(md, &md->lod_ost_descs);
 764
 765                 /* In the function below, .hs_keycmp resolves to
 766                  * u_obj_hop_keycmp() */
 767                 /* coverity[overrun-buffer-val] */
 768                 o = lu_object_find_at(env, nd, &info->lti_fid, NULL);
 769                 if (IS_ERR(o))
 770                         GOTO(out, rc = PTR_ERR(o));
 771
 772                 n = lu_object_locate(o->lo_header, nd->ld_type);
 773                 LASSERT(n);
 774
 775                 stripe[i] = container_of(n, struct dt_object, do_lu);
 776         }
 777
 778 out:
 779         if (rc != 0) {
 780                 for (i = 0; i < stripe_len; i++)
 781                         if (stripe[i] != NULL)
 782                                 lu_object_put(env, &stripe[i]->do_lu);
 783
 784                 OBD_FREE(stripe, sizeof(stripe[0]) * stripe_len);
 785                 lo->ldo_stripenr = 0;
 786         } else {
 787                 lo->ldo_stripe = stripe;
 788                 lo->ldo_stripes_allocated = stripe_len;
 789         }
 790
 791         RETURN(rc);
 792 }
 793
 794 /*
 795  * Parse striping information stored in lti_ea_store
 796  */
 797 int lod_parse_striping(const struct lu_env *env, struct lod_object *lo,
 798                        const struct lu_buf *buf)
 799 {
 800         struct lov_mds_md_v1    *lmm;
 801         struct lov_ost_data_v1  *objs;
 802         __u32                    magic;
 803         __u32                    pattern;
 804         int                      rc = 0;
 805         ENTRY;
 806
 807         LASSERT(buf);
 808         LASSERT(buf->lb_buf);
 809         LASSERT(buf->lb_len);
 810
 811         lmm = (struct lov_mds_md_v1 *) buf->lb_buf;
 812         magic = le32_to_cpu(lmm->lmm_magic);
 813         pattern = le32_to_cpu(lmm->lmm_pattern);
 814
 815         if (magic != LOV_MAGIC_V1 && magic != LOV_MAGIC_V3)
 816                 GOTO(out, rc = -EINVAL);
 817         if (lov_pattern(pattern) != LOV_PATTERN_RAID0)
 818                 GOTO(out, rc = -EINVAL);
 819
 820         lo->ldo_pattern = pattern;
 821         lo->ldo_stripe_size = le32_to_cpu(lmm->lmm_stripe_size);
 822         lo->ldo_layout_gen = le16_to_cpu(lmm->lmm_layout_gen);
 823         lo->ldo_stripenr = le16_to_cpu(lmm->lmm_stripe_count);
 824         /* released file stripenr fixup. */
 825         if (pattern & LOV_PATTERN_F_RELEASED)
 826                 lo->ldo_stripenr = 0;
 827
 828         LASSERT(buf->lb_len >= lov_mds_md_size(lo->ldo_stripenr, magic));
 829
 830         if (magic == LOV_MAGIC_V3) {
 831                 struct lov_mds_md_v3 *v3 = (struct lov_mds_md_v3 *) lmm;
 832                 objs = &v3->lmm_objects[0];
 833                 lod_object_set_pool(lo, v3->lmm_pool_name);
 834         } else {
 835                 objs = &lmm->lmm_objects[0];
 836         }
 837
 838         if (lo->ldo_stripenr > 0)
 839                 rc = lod_initialize_objects(env, lo, objs);
 840
 841 out:
 842         RETURN(rc);
 843 }
 844
 845 int lod_load_striping_locked(const struct lu_env *env, struct lod_object *lo)
 846 {
 847         struct lod_thread_info  *info = lod_env_info(env);
 848         struct lu_buf           *buf  = &info->lti_buf;
 849         struct dt_object        *next = dt_object_child(&lo->ldo_obj);
 850         int                      rc = 0;
 851         ENTRY;
 852
 853         /* already initialized? */
 854         if (lo->ldo_stripe != NULL)
 855                 GOTO(out, rc = 0);
 856
 857         if (!dt_object_exists(next))
 858                 GOTO(out, rc = 0);
 859
 860         /* Do not load stripe for slaves of striped dir */
 861         if (lo->ldo_dir_slave_stripe)
 862                 GOTO(out, rc = 0);
 863
 864         if (S_ISREG(lu_object_attr(lod2lu_obj(lo)))) {
 865                 rc = lod_get_lov_ea(env, lo);
 866                 if (rc <= 0)
 867                         GOTO(out, rc);
 868                 /*
 869                  * there is LOV EA (striping information) in this object
 870                  * let's parse it and create in-core objects for the stripes
 871                  */
 872                 buf->lb_buf = info->lti_ea_store;
 873                 buf->lb_len = info->lti_ea_store_size;
 874                 rc = lod_parse_striping(env, lo, buf);
 875         } else if (S_ISDIR(lu_object_attr(lod2lu_obj(lo)))) {
 876                 rc = lod_get_lmv_ea(env, lo);
 877                 if (rc < (typeof(rc))sizeof(struct lmv_mds_md_v1))
 878                         GOTO(out, rc = rc > 0 ? -EINVAL : rc);
 879
 880                 buf->lb_buf = info->lti_ea_store;
 881                 buf->lb_len = info->lti_ea_store_size;
 882                 if (rc == sizeof(struct lmv_mds_md_v1)) {
 883                         rc = lod_load_lmv_shards(env, lo, buf, true);
 884                         if (buf->lb_buf != info->lti_ea_store) {
 885                                 OBD_FREE_LARGE(info->lti_ea_store,
 886                                                info->lti_ea_store_size);
 887                                 info->lti_ea_store = buf->lb_buf;
 888                                 info->lti_ea_store_size = buf->lb_len;
 889                         }
 890
 891                         if (rc < 0)
 892                                 GOTO(out, rc);
 893                 }
 894
 895                 /*
 896                  * there is LOV EA (striping information) in this object
 897                  * let's parse it and create in-core objects for the stripes
 898                  */
 899                 rc = lod_parse_dir_striping(env, lo, buf);
 900         }
 901 out:
 902         RETURN(rc);
 903 }
 904
 905 /**
 906  * Load and parse striping information, create in-core representation for the
 907  * stripes
 908  **/
 909 int lod_load_striping(const struct lu_env *env, struct lod_object *lo)
 910 {
 911         struct dt_object        *next = dt_object_child(&lo->ldo_obj);
 912         int                     rc = 0;
 913
 914         /* currently this code is supposed to be called from declaration
 915          * phase only, thus the object is not expected to be locked by caller */
 916         dt_write_lock(env, next, 0);
 917         rc = lod_load_striping_locked(env, lo);
 918         dt_write_unlock(env, next);
 919         return rc;
 920 }
 921
 922 /* verify the striping information for directory */
 923 int lod_verify_striping(struct lod_device *d, const struct lu_buf *buf,
 924                         bool is_from_disk)
 925 {
 926         struct lov_user_md_v1   *lum;
 927         struct lov_user_md_v3   *lum3;
 928         struct pool_desc        *pool = NULL;
 929         __u32                    magic;
 930         __u32                    stripe_size;
 931         __u16                    stripe_count;
 932         __u16                    stripe_offset;
 933         size_t                   lum_size;
 934         int                      rc = 0;
 935         ENTRY;
 936
 937         lum = buf->lb_buf;
 938
 939         LASSERT(sizeof(*lum) < sizeof(*lum3));
 940
 941         if (buf->lb_len < sizeof(*lum)) {
 942                 CDEBUG(D_IOCTL, "buf len %zd too small for lov_user_md\n",
 943                        buf->lb_len);
 944                 GOTO(out, rc = -EINVAL);
 945         }
 946
 947         magic = le32_to_cpu(lum->lmm_magic);
 948         if (magic != LOV_USER_MAGIC_V1 &&
 949             magic != LOV_USER_MAGIC_V3 &&
 950             magic != LOV_MAGIC_V1_DEF &&
 951             magic != LOV_MAGIC_V3_DEF) {
 952                 CDEBUG(D_IOCTL, "bad userland LOV MAGIC: %#x\n", magic);
 953                 GOTO(out, rc = -EINVAL);
 954         }
 955
 956         /* the user uses "0" for default stripe pattern normally. */
 957         if (!is_from_disk && lum->lmm_pattern == 0)
 958                 lum->lmm_pattern = cpu_to_le32(LOV_PATTERN_RAID0);
 959
 960         if (le32_to_cpu(lum->lmm_pattern) != LOV_PATTERN_RAID0) {
 961                 CDEBUG(D_IOCTL, "bad userland stripe pattern: %#x\n",
 962                        le32_to_cpu(lum->lmm_pattern));
 963                 GOTO(out, rc = -EINVAL);
 964         }
 965
 966         /* 64kB is the largest common page size we see (ia64), and matches the
 967          * check in lfs */
 968         stripe_size = le32_to_cpu(lum->lmm_stripe_size);
 969         if (stripe_size & (LOV_MIN_STRIPE_SIZE - 1)) {
 970                 CDEBUG(D_IOCTL, "stripe size %u not a multiple of %u\n",
 971                        stripe_size, LOV_MIN_STRIPE_SIZE);
 972                 GOTO(out, rc = -EINVAL);
 973         }
 974
 975         /* an offset of -1 is treated as a "special" valid offset */
 976         stripe_offset = le16_to_cpu(lum->lmm_stripe_offset);
 977         if (stripe_offset != (typeof(stripe_offset))-1) {
 978                 /* if offset is not within valid range [0, osts_size) */
 979                 if (stripe_offset >= d->lod_osts_size) {
 980                         CDEBUG(D_IOCTL, "stripe offset %u >= bitmap size %u\n",
 981                                stripe_offset, d->lod_osts_size);
 982                         GOTO(out, rc = -EINVAL);
 983                 }
 984
 985                 /* if lmm_stripe_offset is *not* in bitmap */
 986                 if (!cfs_bitmap_check(d->lod_ost_bitmap, stripe_offset)) {
 987                         CDEBUG(D_IOCTL, "stripe offset %u not in bitmap\n",
 988                                stripe_offset);
 989                         GOTO(out, rc = -EINVAL);
 990                 }
 991         }
 992
 993         if (magic == LOV_USER_MAGIC_V1 || magic == LOV_MAGIC_V1_DEF)
 994                 lum_size = offsetof(struct lov_user_md_v1,
 995                                     lmm_objects[0]);
 996         else if (magic == LOV_USER_MAGIC_V3 || magic == LOV_MAGIC_V3_DEF)
 997                 lum_size = offsetof(struct lov_user_md_v3,
 998                                     lmm_objects[0]);
 999         else
1000                 GOTO(out, rc = -EINVAL);
1001
1002         stripe_count = le16_to_cpu(lum->lmm_stripe_count);
1003         if (buf->lb_len != lum_size) {
1004                 CDEBUG(D_IOCTL, "invalid buf len %zd for lov_user_md with "
1005                        "magic %#x and stripe_count %u\n",
1006                        buf->lb_len, magic, stripe_count);
1007                 GOTO(out, rc = -EINVAL);
1008         }
1009
1010         if (!(magic == LOV_USER_MAGIC_V3 || magic == LOV_MAGIC_V3_DEF))
1011                 goto out;
1012
1013         lum3 = buf->lb_buf;
1014         if (buf->lb_len < sizeof(*lum3)) {
1015                 CDEBUG(D_IOCTL, "buf len %zd too small for lov_user_md_v3\n",
1016                        buf->lb_len);
1017                 GOTO(out, rc = -EINVAL);
1018         }
1019
1020         /* In the function below, .hs_keycmp resolves to
1021          * pool_hashkey_keycmp() */
1022         /* coverity[overrun-buffer-val] */
1023         pool = lod_find_pool(d, lum3->lmm_pool_name);
1024         if (pool == NULL)
1025                 goto out;
1026
1027         if (stripe_offset != (typeof(stripe_offset))-1) {
1028                 rc = lod_check_index_in_pool(stripe_offset, pool);
1029                 if (rc < 0)
1030                         GOTO(out, rc = -EINVAL);
1031         }
1032
1033         if (is_from_disk && stripe_count > pool_tgt_count(pool)) {
1034                 CDEBUG(D_IOCTL,
1035                        "stripe count %u > # OSTs %u in the pool\n",
1036                        stripe_count, pool_tgt_count(pool));
1037                 GOTO(out, rc = -EINVAL);
1038         }
1039
1040 out:
1041         if (pool != NULL)
1042                 lod_pool_putref(pool);
1043
1044         RETURN(rc);
1045 }
1046
1047 void lod_fix_desc_stripe_size(__u64 *val)
1048 {
1049         if (*val < LOV_MIN_STRIPE_SIZE) {
1050                 if (*val != 0)
1051                         LCONSOLE_INFO("Increasing default stripe size to "
1052                                       "minimum value %u\n",
1053                                       LOV_DESC_STRIPE_SIZE_DEFAULT);
1054                 *val = LOV_DESC_STRIPE_SIZE_DEFAULT;
1055         } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) {
1056                 *val &= ~(LOV_MIN_STRIPE_SIZE - 1);
1057                 LCONSOLE_WARN("Changing default stripe size to "LPU64" (a "
1058                               "multiple of %u)\n",
1059                               *val, LOV_MIN_STRIPE_SIZE);
1060         }
1061 }
1062
1063 void lod_fix_desc_stripe_count(__u32 *val)
1064 {
1065         if (*val == 0)
1066                 *val = 1;
1067 }
1068
1069 void lod_fix_desc_pattern(__u32 *val)
1070 {
1071         /* from lov_setstripe */
1072         if ((*val != 0) && (*val != LOV_PATTERN_RAID0)) {
1073                 LCONSOLE_WARN("Unknown stripe pattern: %#x\n", *val);
1074                 *val = 0;
1075         }
1076 }
1077
1078 void lod_fix_desc_qos_maxage(__u32 *val)
1079 {
1080         /* fix qos_maxage */
1081         if (*val == 0)
1082                 *val = LOV_DESC_QOS_MAXAGE_DEFAULT;
1083 }
1084
1085 void lod_fix_desc(struct lov_desc *desc)
1086 {
1087         lod_fix_desc_stripe_size(&desc->ld_default_stripe_size);
1088         lod_fix_desc_stripe_count(&desc->ld_default_stripe_count);
1089         lod_fix_desc_pattern(&desc->ld_pattern);
1090         lod_fix_desc_qos_maxage(&desc->ld_qos_maxage);
1091 }
1092
1093 int lod_pools_init(struct lod_device *lod, struct lustre_cfg *lcfg)
1094 {
1095         struct obd_device          *obd;
1096         struct lov_desc            *desc;
1097         int                         rc;
1098         ENTRY;
1099
1100         obd = class_name2obd(lustre_cfg_string(lcfg, 0));
1101         LASSERT(obd != NULL);
1102         obd->obd_lu_dev = &lod->lod_dt_dev.dd_lu_dev;
1103
1104         if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
1105                 CERROR("LOD setup requires a descriptor\n");
1106                 RETURN(-EINVAL);
1107         }
1108
1109         desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
1110
1111         if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
1112                 CERROR("descriptor size wrong: %d > %d\n",
1113                        (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
1114                 RETURN(-EINVAL);
1115         }
1116
1117         if (desc->ld_magic != LOV_DESC_MAGIC) {
1118                 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
1119                         CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
1120                                obd->obd_name, desc);
1121                         lustre_swab_lov_desc(desc);
1122                 } else {
1123                         CERROR("%s: Bad lov desc magic: %#x\n",
1124                                obd->obd_name, desc->ld_magic);
1125                         RETURN(-EINVAL);
1126                 }
1127         }
1128
1129         lod_fix_desc(desc);
1130
1131         desc->ld_active_tgt_count = 0;
1132         lod->lod_desc = *desc;
1133
1134         lod->lod_sp_me = LUSTRE_SP_CLI;
1135
1136         /* Set up allocation policy (QoS and RR) */
1137         CFS_INIT_LIST_HEAD(&lod->lod_qos.lq_oss_list);
1138         init_rwsem(&lod->lod_qos.lq_rw_sem);
1139         lod->lod_qos.lq_dirty = 1;
1140         lod->lod_qos.lq_rr.lqr_dirty = 1;
1141         lod->lod_qos.lq_reset = 1;
1142         /* Default priority is toward free space balance */
1143         lod->lod_qos.lq_prio_free = 232;
1144         /* Default threshold for rr (roughly 17%) */
1145         lod->lod_qos.lq_threshold_rr = 43;
1146
1147         /* Set up OST pool environment */
1148         lod->lod_pools_hash_body = cfs_hash_create("POOLS", HASH_POOLS_CUR_BITS,
1149                                                    HASH_POOLS_MAX_BITS,
1150                                                    HASH_POOLS_BKT_BITS, 0,
1151                                                    CFS_HASH_MIN_THETA,
1152                                                    CFS_HASH_MAX_THETA,
1153                                                    &pool_hash_operations,
1154                                                    CFS_HASH_DEFAULT);
1155         if (lod->lod_pools_hash_body == NULL)
1156                 RETURN(-ENOMEM);
1157
1158         CFS_INIT_LIST_HEAD(&lod->lod_pool_list);
1159         lod->lod_pool_count = 0;
1160         rc = lod_ost_pool_init(&lod->lod_pool_info, 0);
1161         if (rc)
1162                 GOTO(out_hash, rc);
1163         rc = lod_ost_pool_init(&lod->lod_qos.lq_rr.lqr_pool, 0);
1164         if (rc)
1165                 GOTO(out_pool_info, rc);
1166
1167         RETURN(0);
1168
1169 out_pool_info:
1170         lod_ost_pool_free(&lod->lod_pool_info);
1171 out_hash:
1172         cfs_hash_putref(lod->lod_pools_hash_body);
1173
1174         return rc;
1175 }
1176
1177 int lod_pools_fini(struct lod_device *lod)
1178 {
1179         struct obd_device   *obd = lod2obd(lod);
1180         cfs_list_t          *pos, *tmp;
1181         struct pool_desc    *pool;
1182         ENTRY;
1183
1184         cfs_list_for_each_safe(pos, tmp, &lod->lod_pool_list) {
1185                 pool = cfs_list_entry(pos, struct pool_desc, pool_list);
1186                 /* free pool structs */
1187                 CDEBUG(D_INFO, "delete pool %p\n", pool);
1188                 /* In the function below, .hs_keycmp resolves to
1189                  * pool_hashkey_keycmp() */
1190                 /* coverity[overrun-buffer-val] */
1191                 lod_pool_del(obd, pool->pool_name);
1192         }
1193
1194         cfs_hash_putref(lod->lod_pools_hash_body);
1195         lod_ost_pool_free(&(lod->lod_qos.lq_rr.lqr_pool));
1196         lod_ost_pool_free(&lod->lod_pool_info);
1197
1198         RETURN(0);
1199 }
1200