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  * version 2 along with this program; If not, see
  18  * http://www.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright  2009 Sun Microsystems, Inc. All rights reserved
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2012, 2014, Intel Corporation.
  27  */
  28 /*
  29  * lustre/lod/lod_lov.c
  30  *
  31  * A set of helpers to maintain Logical Object Volume (LOV)
  32  * Extended Attribute (EA) and known OST targets
  33  *
  34  * Author: Alex Zhuravlev <alexey.zhuravlev@intel.com>
  35  */
  36
  37 #define DEBUG_SUBSYSTEM S_MDS
  38
  39 #include <obd_class.h>
  40 #include <lustre_lfsck.h>
  41 #include <lustre_lmv.h>
  42
  43 #include "lod_internal.h"
  44
  45 /**
  46  * Increase reference count on the target table.
  47  *
  48  * Increase reference count on the target table usage to prevent racing with
  49  * addition/deletion. Any function that expects the table to remain
  50  * stationary must take a ref.
  51  *
  52  * \param[in] ltd       target table (lod_ost_descs or lod_mdt_descs)
  53  */
  54 void lod_getref(struct lod_tgt_descs *ltd)
  55 {
  56         down_read(&ltd->ltd_rw_sem);
  57         mutex_lock(&ltd->ltd_mutex);
  58         ltd->ltd_refcount++;
  59         mutex_unlock(&ltd->ltd_mutex);
  60 }
  61
  62 /**
  63  * Decrease reference count on the target table.
  64  *
  65  * Companion of lod_getref() to release a reference on the target table.
  66  * If this is the last reference and the OST entry was scheduled for deletion,
  67  * the descriptor is removed from the table.
  68  *
  69  * \param[in] lod       LOD device from which we release a reference
  70  * \param[in] ltd       target table (lod_ost_descs or lod_mdt_descs)
  71  */
  72 void lod_putref(struct lod_device *lod, struct lod_tgt_descs *ltd)
  73 {
  74         mutex_lock(&ltd->ltd_mutex);
  75         ltd->ltd_refcount--;
  76         if (ltd->ltd_refcount == 0 && ltd->ltd_death_row) {
  77                 struct lod_tgt_desc *tgt_desc, *tmp;
  78                 struct list_head kill;
  79                 unsigned int idx;
  80
  81                 CDEBUG(D_CONFIG, "destroying %d ltd desc\n",
  82                        ltd->ltd_death_row);
  83
  84                 INIT_LIST_HEAD(&kill);
  85
  86                 cfs_foreach_bit(ltd->ltd_tgt_bitmap, idx) {
  87                         tgt_desc = LTD_TGT(ltd, idx);
  88                         LASSERT(tgt_desc);
  89
  90                         if (!tgt_desc->ltd_reap)
  91                                 continue;
  92
  93                         list_add(&tgt_desc->ltd_kill, &kill);
  94                         LTD_TGT(ltd, idx) = NULL;
  95                         /*FIXME: only support ost pool for now */
  96                         if (ltd == &lod->lod_ost_descs) {
  97                                 lod_ost_pool_remove(&lod->lod_pool_info, idx);
  98                                 if (tgt_desc->ltd_active)
  99                                         lod->lod_desc.ld_active_tgt_count--;
 100                         }
 101                         ltd->ltd_tgtnr--;
 102                         cfs_bitmap_clear(ltd->ltd_tgt_bitmap, idx);
 103                         ltd->ltd_death_row--;
 104                 }
 105                 mutex_unlock(&ltd->ltd_mutex);
 106                 up_read(&ltd->ltd_rw_sem);
 107
 108                 list_for_each_entry_safe(tgt_desc, tmp, &kill, ltd_kill) {
 109                         int rc;
 110                         list_del(&tgt_desc->ltd_kill);
 111                         if (ltd == &lod->lod_ost_descs) {
 112                                 /* remove from QoS structures */
 113                                 rc = qos_del_tgt(lod, tgt_desc);
 114                                 if (rc)
 115                                         CERROR("%s: qos_del_tgt(%s) failed:"
 116                                                "rc = %d\n",
 117                                                lod2obd(lod)->obd_name,
 118                                               obd_uuid2str(&tgt_desc->ltd_uuid),
 119                                                rc);
 120                         }
 121                         rc = obd_disconnect(tgt_desc->ltd_exp);
 122                         if (rc)
 123                                 CERROR("%s: failed to disconnect %s: rc = %d\n",
 124                                        lod2obd(lod)->obd_name,
 125                                        obd_uuid2str(&tgt_desc->ltd_uuid), rc);
 126                         OBD_FREE_PTR(tgt_desc);
 127                 }
 128         } else {
 129                 mutex_unlock(&ltd->ltd_mutex);
 130                 up_read(&ltd->ltd_rw_sem);
 131         }
 132 }
 133
 134 /**
 135  * Expand size of target table.
 136  *
 137  * When the target table is full, we have to extend the table. To do so,
 138  * we allocate new memory with some reserve, move data from the old table
 139  * to the new one and release memory consumed by the old table.
 140  * Notice we take ltd_rw_sem exclusively to ensure atomic switch.
 141  *
 142  * \param[in] ltd               target table
 143  * \param[in] newsize           new size of the table
 144  *
 145  * \retval                      0 on success
 146  * \retval                      -ENOMEM if reallocation failed
 147  */
 148 static int ltd_bitmap_resize(struct lod_tgt_descs *ltd, __u32 newsize)
 149 {
 150         cfs_bitmap_t *new_bitmap, *old_bitmap = NULL;
 151         int           rc = 0;
 152         ENTRY;
 153
 154         /* grab write reference on the lod. Relocating the array requires
 155          * exclusive access */
 156
 157         down_write(&ltd->ltd_rw_sem);
 158         if (newsize <= ltd->ltd_tgts_size)
 159                 /* someone else has already resize the array */
 160                 GOTO(out, rc = 0);
 161
 162         /* allocate new bitmap */
 163         new_bitmap = CFS_ALLOCATE_BITMAP(newsize);
 164         if (!new_bitmap)
 165                 GOTO(out, rc = -ENOMEM);
 166
 167         if (ltd->ltd_tgts_size > 0) {
 168                 /* the bitmap already exists, we need
 169                  * to copy data from old one */
 170                 cfs_bitmap_copy(new_bitmap, ltd->ltd_tgt_bitmap);
 171                 old_bitmap = ltd->ltd_tgt_bitmap;
 172         }
 173
 174         ltd->ltd_tgts_size  = newsize;
 175         ltd->ltd_tgt_bitmap = new_bitmap;
 176
 177         if (old_bitmap)
 178                 CFS_FREE_BITMAP(old_bitmap);
 179
 180         CDEBUG(D_CONFIG, "tgt size: %d\n", ltd->ltd_tgts_size);
 181
 182         EXIT;
 183 out:
 184         up_write(&ltd->ltd_rw_sem);
 185         return rc;
 186 }
 187
 188 /**
 189  * Connect LOD to a new OSP and add it to the target table.
 190  *
 191  * Connect to the OSP device passed, initialize all the internal
 192  * structures related to the device and add it to the target table.
 193  *
 194  * \param[in] env               execution environment for this thread
 195  * \param[in] lod               LOD device to be connected to the new OSP
 196  * \param[in] osp               name of OSP device name to be added
 197  * \param[in] index             index of the new target
 198  * \param[in] gen               target's generation number
 199  * \param[in] tgt_index         OSP's group
 200  * \param[in] type              type of device (mdc or osc)
 201  * \param[in] active            state of OSP: 0 - inactive, 1 - active
 202  *
 203  * \retval                      0 if added successfully
 204  * \retval                      negative error number on failure
 205  */
 206 int lod_add_device(const struct lu_env *env, struct lod_device *lod,
 207                    char *osp, unsigned index, unsigned gen, int tgt_index,
 208                    char *type, int active)
 209 {
 210         struct obd_connect_data *data = NULL;
 211         struct obd_export       *exp = NULL;
 212         struct obd_device       *obd;
 213         struct lu_device        *ldev;
 214         struct dt_device        *d;
 215         int                      rc;
 216         struct lod_tgt_desc     *tgt_desc;
 217         struct lod_tgt_descs    *ltd;
 218         struct obd_uuid         obd_uuid;
 219         bool                    for_ost;
 220         ENTRY;
 221
 222         CDEBUG(D_CONFIG, "osp:%s idx:%d gen:%d\n", osp, index, gen);
 223
 224         if (gen <= 0) {
 225                 CERROR("request to add OBD %s with invalid generation: %d\n",
 226                        osp, gen);
 227                 RETURN(-EINVAL);
 228         }
 229
 230         obd_str2uuid(&obd_uuid, osp);
 231
 232         obd = class_find_client_obd(&obd_uuid, LUSTRE_OSP_NAME,
 233                                 &lod->lod_dt_dev.dd_lu_dev.ld_obd->obd_uuid);
 234         if (obd == NULL) {
 235                 CERROR("can't find %s device\n", osp);
 236                 RETURN(-EINVAL);
 237         }
 238
 239         OBD_ALLOC_PTR(data);
 240         if (data == NULL)
 241                 RETURN(-ENOMEM);
 242
 243         data->ocd_connect_flags = OBD_CONNECT_INDEX | OBD_CONNECT_VERSION;
 244         data->ocd_version = LUSTRE_VERSION_CODE;
 245         data->ocd_index = index;
 246
 247         if (strcmp(LUSTRE_OSC_NAME, type) == 0) {
 248                 for_ost = true;
 249                 data->ocd_connect_flags |= OBD_CONNECT_AT |
 250                                            OBD_CONNECT_FULL20 |
 251                                            OBD_CONNECT_INDEX |
 252 #ifdef HAVE_LRU_RESIZE_SUPPORT
 253                                            OBD_CONNECT_LRU_RESIZE |
 254 #endif
 255                                            OBD_CONNECT_MDS |
 256                                            OBD_CONNECT_REQPORTAL |
 257                                            OBD_CONNECT_SKIP_ORPHAN |
 258                                            OBD_CONNECT_FID |
 259                                            OBD_CONNECT_LVB_TYPE |
 260                                            OBD_CONNECT_VERSION |
 261                                            OBD_CONNECT_PINGLESS |
 262                                            OBD_CONNECT_LFSCK;
 263
 264                 data->ocd_group = tgt_index;
 265                 ltd = &lod->lod_ost_descs;
 266         } else {
 267                 struct obd_import *imp = obd->u.cli.cl_import;
 268
 269                 for_ost = false;
 270                 data->ocd_ibits_known = MDS_INODELOCK_UPDATE;
 271                 data->ocd_connect_flags |= OBD_CONNECT_ACL |
 272                                            OBD_CONNECT_IBITS |
 273                                            OBD_CONNECT_MDS_MDS |
 274                                            OBD_CONNECT_FID |
 275                                            OBD_CONNECT_AT |
 276                                            OBD_CONNECT_FULL20 |
 277                                            OBD_CONNECT_LFSCK;
 278                 spin_lock(&imp->imp_lock);
 279                 imp->imp_server_timeout = 1;
 280                 spin_unlock(&imp->imp_lock);
 281                 imp->imp_client->cli_request_portal = OUT_PORTAL;
 282                 CDEBUG(D_OTHER, "%s: Set 'mds' portal and timeout\n",
 283                       obd->obd_name);
 284                 ltd = &lod->lod_mdt_descs;
 285         }
 286
 287         rc = obd_connect(env, &exp, obd, &obd->obd_uuid, data, NULL);
 288         OBD_FREE_PTR(data);
 289         if (rc) {
 290                 CERROR("%s: cannot connect to next dev %s (%d)\n",
 291                        obd->obd_name, osp, rc);
 292                 GOTO(out_free, rc);
 293         }
 294
 295         LASSERT(obd->obd_lu_dev);
 296         LASSERT(obd->obd_lu_dev->ld_site == lod->lod_dt_dev.dd_lu_dev.ld_site);
 297
 298         ldev = obd->obd_lu_dev;
 299         d = lu2dt_dev(ldev);
 300
 301         /* Allocate ost descriptor and fill it */
 302         OBD_ALLOC_PTR(tgt_desc);
 303         if (!tgt_desc)
 304                 GOTO(out_conn, rc = -ENOMEM);
 305
 306         tgt_desc->ltd_tgt    = d;
 307         tgt_desc->ltd_exp    = exp;
 308         tgt_desc->ltd_uuid   = obd->u.cli.cl_target_uuid;
 309         tgt_desc->ltd_gen    = gen;
 310         tgt_desc->ltd_index  = index;
 311         tgt_desc->ltd_active = active;
 312
 313         lod_getref(ltd);
 314         if (index >= ltd->ltd_tgts_size) {
 315                 /* we have to increase the size of the lod_osts array */
 316                 __u32  newsize;
 317
 318                 newsize = max(ltd->ltd_tgts_size, (__u32)2);
 319                 while (newsize < index + 1)
 320                         newsize = newsize << 1;
 321
 322                 /* lod_bitmap_resize() needs lod_rw_sem
 323                  * which we hold with th reference */
 324                 lod_putref(lod, ltd);
 325
 326                 rc = ltd_bitmap_resize(ltd, newsize);
 327                 if (rc)
 328                         GOTO(out_desc, rc);
 329
 330                 lod_getref(ltd);
 331         }
 332
 333         mutex_lock(&ltd->ltd_mutex);
 334         if (cfs_bitmap_check(ltd->ltd_tgt_bitmap, index)) {
 335                 CERROR("%s: device %d is registered already\n", obd->obd_name,
 336                        index);
 337                 GOTO(out_mutex, rc = -EEXIST);
 338         }
 339
 340         if (ltd->ltd_tgt_idx[index / TGT_PTRS_PER_BLOCK] == NULL) {
 341                 OBD_ALLOC_PTR(ltd->ltd_tgt_idx[index / TGT_PTRS_PER_BLOCK]);
 342                 if (ltd->ltd_tgt_idx[index / TGT_PTRS_PER_BLOCK] == NULL) {
 343                         CERROR("can't allocate index to add %s\n",
 344                                obd->obd_name);
 345                         GOTO(out_mutex, rc = -ENOMEM);
 346                 }
 347         }
 348
 349         if (!strcmp(LUSTRE_OSC_NAME, type)) {
 350                 /* pool and qos are not supported for MDS stack yet */
 351                 rc = lod_ost_pool_add(&lod->lod_pool_info, index,
 352                                       lod->lod_osts_size);
 353                 if (rc) {
 354                         CERROR("%s: can't set up pool, failed with %d\n",
 355                                obd->obd_name, rc);
 356                         GOTO(out_mutex, rc);
 357                 }
 358
 359                 rc = qos_add_tgt(lod, tgt_desc);
 360                 if (rc) {
 361                         CERROR("%s: qos_add_tgt failed with %d\n",
 362                                 obd->obd_name, rc);
 363                         GOTO(out_pool, rc);
 364                 }
 365
 366                 /* The new OST is now a full citizen */
 367                 if (index >= lod->lod_desc.ld_tgt_count)
 368                         lod->lod_desc.ld_tgt_count = index + 1;
 369                 if (active)
 370                         lod->lod_desc.ld_active_tgt_count++;
 371         }
 372
 373         LTD_TGT(ltd, index) = tgt_desc;
 374         cfs_bitmap_set(ltd->ltd_tgt_bitmap, index);
 375         ltd->ltd_tgtnr++;
 376         mutex_unlock(&ltd->ltd_mutex);
 377         lod_putref(lod, ltd);
 378         if (lod->lod_recovery_completed)
 379                 ldev->ld_ops->ldo_recovery_complete(env, ldev);
 380
 381         rc = lfsck_add_target(env, lod->lod_child, d, exp, index, for_ost);
 382         if (rc != 0)
 383                 CERROR("Fail to add LFSCK target: name = %s, type = %s, "
 384                        "index = %u, rc = %d\n", osp, type, index, rc);
 385
 386         RETURN(rc);
 387
 388 out_pool:
 389         lod_ost_pool_remove(&lod->lod_pool_info, index);
 390 out_mutex:
 391         mutex_unlock(&ltd->ltd_mutex);
 392         lod_putref(lod, ltd);
 393 out_desc:
 394         OBD_FREE_PTR(tgt_desc);
 395 out_conn:
 396         obd_disconnect(exp);
 397 out_free:
 398         return rc;
 399 }
 400
 401 /**
 402  * Schedule target removal from the target table.
 403  *
 404  * Mark the device as dead. The device is not removed here because it may
 405  * still be in use. The device will be removed in lod_putref() when the
 406  * last reference is released.
 407  *
 408  * \param[in] env               execution environment for this thread
 409  * \param[in] lod               LOD device the target table belongs to
 410  * \param[in] ltd               target table
 411  * \param[in] idx               index of the target
 412  * \param[in] for_ost           type of the target: 0 - MDT, 1 - OST
 413  */
 414 static void __lod_del_device(const struct lu_env *env, struct lod_device *lod,
 415                              struct lod_tgt_descs *ltd, unsigned idx,
 416                              bool for_ost)
 417 {
 418         LASSERT(LTD_TGT(ltd, idx));
 419
 420         lfsck_del_target(env, lod->lod_child, LTD_TGT(ltd, idx)->ltd_tgt,
 421                          idx, for_ost);
 422
 423         if (LTD_TGT(ltd, idx)->ltd_reap == 0) {
 424                 LTD_TGT(ltd, idx)->ltd_reap = 1;
 425                 ltd->ltd_death_row++;
 426         }
 427 }
 428
 429 /**
 430  * Schedule removal of all the targets from the given target table.
 431  *
 432  * See more details in the description for __lod_del_device()
 433  *
 434  * \param[in] env               execution environment for this thread
 435  * \param[in] lod               LOD device the target table belongs to
 436  * \param[in] ltd               target table
 437  * \param[in] for_ost           type of the target: MDT or OST
 438  *
 439  * \retval                      0 always
 440  */
 441 int lod_fini_tgt(const struct lu_env *env, struct lod_device *lod,
 442                  struct lod_tgt_descs *ltd, bool for_ost)
 443 {
 444         unsigned int idx;
 445
 446         if (ltd->ltd_tgts_size <= 0)
 447                 return 0;
 448         lod_getref(ltd);
 449         mutex_lock(&ltd->ltd_mutex);
 450         cfs_foreach_bit(ltd->ltd_tgt_bitmap, idx)
 451                 __lod_del_device(env, lod, ltd, idx, for_ost);
 452         mutex_unlock(&ltd->ltd_mutex);
 453         lod_putref(lod, ltd);
 454         CFS_FREE_BITMAP(ltd->ltd_tgt_bitmap);
 455         for (idx = 0; idx < TGT_PTRS; idx++) {
 456                 if (ltd->ltd_tgt_idx[idx])
 457                         OBD_FREE_PTR(ltd->ltd_tgt_idx[idx]);
 458         }
 459         ltd->ltd_tgts_size = 0;
 460         return 0;
 461 }
 462
 463 /**
 464  * Remove device by name.
 465  *
 466  * Remove a device identified by \a osp from the target table. Given
 467  * the device can be in use, the real deletion happens in lod_putref().
 468  *
 469  * \param[in] env               execution environment for this thread
 470  * \param[in] lod               LOD device to be connected to the new OSP
 471  * \param[in] ltd               target table
 472  * \param[in] osp               name of OSP device to be removed
 473  * \param[in] idx               index of the target
 474  * \param[in] gen               generation number, not used currently
 475  * \param[in] for_ost           type of the target: 0 - MDT, 1 - OST
 476  *
 477  * \retval                      0 if the device was scheduled for removal
 478  * \retval                      -EINVAL if no device was found
 479  */
 480 int lod_del_device(const struct lu_env *env, struct lod_device *lod,
 481                    struct lod_tgt_descs *ltd, char *osp, unsigned idx,
 482                    unsigned gen, bool for_ost)
 483 {
 484         struct obd_device *obd;
 485         int                rc = 0;
 486         struct obd_uuid    uuid;
 487         ENTRY;
 488
 489         CDEBUG(D_CONFIG, "osp:%s idx:%d gen:%d\n", osp, idx, gen);
 490
 491         obd_str2uuid(&uuid, osp);
 492
 493         obd = class_find_client_obd(&uuid, LUSTRE_OSP_NAME,
 494                                    &lod->lod_dt_dev.dd_lu_dev.ld_obd->obd_uuid);
 495         if (obd == NULL) {
 496                 CERROR("can't find %s device\n", osp);
 497                 RETURN(-EINVAL);
 498         }
 499
 500         if (gen <= 0) {
 501                 CERROR("%s: request to remove OBD %s with invalid generation %d"
 502                        "\n", obd->obd_name, osp, gen);
 503                 RETURN(-EINVAL);
 504         }
 505
 506         obd_str2uuid(&uuid,  osp);
 507
 508         lod_getref(ltd);
 509         mutex_lock(&ltd->ltd_mutex);
 510         /* check that the index is allocated in the bitmap */
 511         if (!cfs_bitmap_check(ltd->ltd_tgt_bitmap, idx) ||
 512             !LTD_TGT(ltd, idx)) {
 513                 CERROR("%s: device %d is not set up\n", obd->obd_name, idx);
 514                 GOTO(out, rc = -EINVAL);
 515         }
 516
 517         /* check that the UUID matches */
 518         if (!obd_uuid_equals(&uuid, &LTD_TGT(ltd, idx)->ltd_uuid)) {
 519                 CERROR("%s: LOD target UUID %s at index %d does not match %s\n",
 520                        obd->obd_name, obd_uuid2str(&LTD_TGT(ltd,idx)->ltd_uuid),
 521                        idx, osp);
 522                 GOTO(out, rc = -EINVAL);
 523         }
 524
 525         __lod_del_device(env, lod, ltd, idx, for_ost);
 526         EXIT;
 527 out:
 528         mutex_unlock(&ltd->ltd_mutex);
 529         lod_putref(lod, ltd);
 530         return(rc);
 531 }
 532
 533 /**
 534  * Resize per-thread storage to hold specified size.
 535  *
 536  * A helper function to resize per-thread temporary storage. This storage
 537  * is used to process LOV/LVM EAs and may be quite large. We do not want to
 538  * allocate/release it every time, so instead we put it into the env and
 539  * reallocate on demand. The memory is released when the correspondent thread
 540  * is finished.
 541  *
 542  * \param[in] info              LOD-specific storage in the environment
 543  * \param[in] size              new size to grow the buffer to
 544
 545  * \retval                      0 on success, -ENOMEM if reallocation failed
 546  */
 547 int lod_ea_store_resize(struct lod_thread_info *info, size_t size)
 548 {
 549         __u32 round = size_roundup_power2(size);
 550
 551         LASSERT(round <=
 552                 lov_mds_md_size(LOV_MAX_STRIPE_COUNT, LOV_MAGIC_V3));
 553         if (info->lti_ea_store) {
 554                 LASSERT(info->lti_ea_store_size);
 555                 LASSERT(info->lti_ea_store_size < round);
 556                 CDEBUG(D_INFO, "EA store size %d is not enough, need %d\n",
 557                        info->lti_ea_store_size, round);
 558                 OBD_FREE_LARGE(info->lti_ea_store, info->lti_ea_store_size);
 559                 info->lti_ea_store = NULL;
 560                 info->lti_ea_store_size = 0;
 561         }
 562
 563         OBD_ALLOC_LARGE(info->lti_ea_store, round);
 564         if (info->lti_ea_store == NULL)
 565                 RETURN(-ENOMEM);
 566         info->lti_ea_store_size = round;
 567         RETURN(0);
 568 }
 569
 570 /**
 571  * Make LOV EA for striped object.
 572  *
 573  * Generate striping information and store it in the LOV EA of the given
 574  * object. The caller must ensure nobody else is calling the function
 575  * against the object concurrently. The transaction must be started.
 576  * FLDB service must be running as well; it's used to map FID to the target,
 577  * which is stored in LOV EA.
 578  *
 579  * \param[in] env               execution environment for this thread
 580  * \param[in] lo                LOD object
 581  * \param[in] th                transaction handle
 582  *
 583  * \retval                      0 if LOV EA is stored successfully
 584  * \retval                      negative error number on failure
 585  */
 586 int lod_generate_and_set_lovea(const struct lu_env *env,
 587                                struct lod_object *lo, struct thandle *th)
 588 {
 589         struct lod_thread_info  *info = lod_env_info(env);
 590         struct dt_object        *next = dt_object_child(&lo->ldo_obj);
 591         const struct lu_fid     *fid  = lu_object_fid(&lo->ldo_obj.do_lu);
 592         struct lov_mds_md_v1    *lmm;
 593         struct lov_ost_data_v1  *objs;
 594         __u32                    magic;
 595         int                      i, rc;
 596         size_t                   lmm_size;
 597         ENTRY;
 598
 599         LASSERT(lo);
 600
 601         magic = lo->ldo_pool != NULL ? LOV_MAGIC_V3 : LOV_MAGIC_V1;
 602         lmm_size = lov_mds_md_size(lo->ldo_stripenr, magic);
 603         if (info->lti_ea_store_size < lmm_size) {
 604                 rc = lod_ea_store_resize(info, lmm_size);
 605                 if (rc)
 606                         RETURN(rc);
 607         }
 608
 609         if (lo->ldo_pattern == 0) /* default striping */
 610                 lo->ldo_pattern = LOV_PATTERN_RAID0;
 611
 612         lmm = info->lti_ea_store;
 613
 614         lmm->lmm_magic = cpu_to_le32(magic);
 615         lmm->lmm_pattern = cpu_to_le32(lo->ldo_pattern);
 616         fid_to_lmm_oi(fid, &lmm->lmm_oi);
 617         if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_LMMOI))
 618                 lmm->lmm_oi.oi.oi_id++;
 619         lmm_oi_cpu_to_le(&lmm->lmm_oi, &lmm->lmm_oi);
 620         lmm->lmm_stripe_size = cpu_to_le32(lo->ldo_stripe_size);
 621         lmm->lmm_stripe_count = cpu_to_le16(lo->ldo_stripenr);
 622         if (lo->ldo_pattern & LOV_PATTERN_F_RELEASED)
 623                 lmm->lmm_stripe_count = cpu_to_le16(lo->ldo_released_stripenr);
 624         lmm->lmm_layout_gen = 0;
 625         if (magic == LOV_MAGIC_V1) {
 626                 objs = &lmm->lmm_objects[0];
 627         } else {
 628                 struct lov_mds_md_v3 *v3 = (struct lov_mds_md_v3 *) lmm;
 629                 size_t cplen = strlcpy(v3->lmm_pool_name, lo->ldo_pool,
 630                                 sizeof(v3->lmm_pool_name));
 631                 if (cplen >= sizeof(v3->lmm_pool_name))
 632                         RETURN(-E2BIG);
 633                 objs = &v3->lmm_objects[0];
 634         }
 635
 636         for (i = 0; i < lo->ldo_stripenr; i++) {
 637                 struct lu_fid           *fid    = &info->lti_fid;
 638                 struct lod_device       *lod;
 639                 __u32                   index;
 640                 int                     type    = LU_SEQ_RANGE_OST;
 641
 642                 lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
 643                 LASSERT(lo->ldo_stripe[i]);
 644
 645                 *fid = *lu_object_fid(&lo->ldo_stripe[i]->do_lu);
 646                 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_MULTIPLE_REF)) {
 647                         if (cfs_fail_val == 0)
 648                                 cfs_fail_val = fid->f_oid;
 649                         else
 650                                 fid->f_oid = cfs_fail_val;
 651                 }
 652
 653                 rc = fid_to_ostid(fid, &info->lti_ostid);
 654                 LASSERT(rc == 0);
 655
 656                 ostid_cpu_to_le(&info->lti_ostid, &objs[i].l_ost_oi);
 657                 objs[i].l_ost_gen    = cpu_to_le32(0);
 658                 rc = lod_fld_lookup(env, lod, fid, &index, &type);
 659                 if (rc < 0) {
 660                         CERROR("%s: Can not locate "DFID": rc = %d\n",
 661                                lod2obd(lod)->obd_name, PFID(fid), rc);
 662                         lod_object_free_striping(env, lo);
 663                         RETURN(rc);
 664                 }
 665                 objs[i].l_ost_idx = cpu_to_le32(index);
 666         }
 667
 668         info->lti_buf.lb_buf = lmm;
 669         info->lti_buf.lb_len = lmm_size;
 670         rc = lod_sub_object_xattr_set(env, next, &info->lti_buf, XATTR_NAME_LOV,
 671                                       0, th);
 672         if (rc < 0) {
 673                 lod_object_free_striping(env, lo);
 674                 RETURN(rc);
 675         }
 676
 677         RETURN(rc);
 678 }
 679
 680 /**
 681  * Get LOV EA.
 682  *
 683  * Fill lti_ea_store buffer in the environment with a value for the given
 684  * EA. The buffer is reallocated if the value doesn't fit.
 685  *
 686  * \param[in,out] env           execution environment for this thread
 687  *                              .lti_ea_store buffer is filled with EA's value
 688  * \param[in] lo                LOD object
 689  * \param[in] name              name of the EA
 690  *
 691  * \retval                      0 if EA is fetched successfully
 692  * \retval                      negative error number on failure
 693  */
 694 int lod_get_ea(const struct lu_env *env, struct lod_object *lo,
 695                const char *name)
 696 {
 697         struct lod_thread_info  *info = lod_env_info(env);
 698         struct dt_object        *next = dt_object_child(&lo->ldo_obj);
 699         int                     rc;
 700         ENTRY;
 701
 702         LASSERT(info);
 703
 704         if (unlikely(info->lti_ea_store == NULL)) {
 705                 /* just to enter in allocation block below */
 706                 rc = -ERANGE;
 707         } else {
 708 repeat:
 709                 info->lti_buf.lb_buf = info->lti_ea_store;
 710                 info->lti_buf.lb_len = info->lti_ea_store_size;
 711                 rc = dt_xattr_get(env, next, &info->lti_buf, name);
 712         }
 713
 714         /* if object is not striped or inaccessible */
 715         if (rc == -ENODATA || rc == -ENOENT)
 716                 RETURN(0);
 717
 718         if (rc == -ERANGE) {
 719                 /* EA doesn't fit, reallocate new buffer */
 720                 rc = dt_xattr_get(env, next, &LU_BUF_NULL, name);
 721                 if (rc == -ENODATA || rc == -ENOENT)
 722                         RETURN(0);
 723                 else if (rc < 0)
 724                         RETURN(rc);
 725
 726                 LASSERT(rc > 0);
 727                 rc = lod_ea_store_resize(info, rc);
 728                 if (rc)
 729                         RETURN(rc);
 730                 goto repeat;
 731         }
 732
 733         RETURN(rc);
 734 }
 735
 736 /**
 737  * Verify the target index is present in the current configuration.
 738  *
 739  * \param[in] md                LOD device where the target table is stored
 740  * \param[in] idx               target's index
 741  *
 742  * \retval                      0 if the index is present
 743  * \retval                      -EINVAL if not
 744  */
 745 static int validate_lod_and_idx(struct lod_device *md, __u32 idx)
 746 {
 747         if (unlikely(idx >= md->lod_ost_descs.ltd_tgts_size ||
 748                      !cfs_bitmap_check(md->lod_ost_bitmap, idx))) {
 749                 CERROR("%s: bad idx: %d of %d\n", lod2obd(md)->obd_name, idx,
 750                        md->lod_ost_descs.ltd_tgts_size);
 751                 return -EINVAL;
 752         }
 753
 754         if (unlikely(OST_TGT(md, idx) == NULL)) {
 755                 CERROR("%s: bad lod_tgt_desc for idx: %d\n",
 756                        lod2obd(md)->obd_name, idx);
 757                 return -EINVAL;
 758         }
 759
 760         if (unlikely(OST_TGT(md, idx)->ltd_ost == NULL)) {
 761                 CERROR("%s: invalid lod device, for idx: %d\n",
 762                        lod2obd(md)->obd_name , idx);
 763                 return -EINVAL;
 764         }
 765
 766         return 0;
 767 }
 768
 769 /**
 770  * Instantiate objects for stripes.
 771  *
 772  * Allocate and initialize LU-objects representing the stripes. The number
 773  * of the stripes (ldo_stripenr) must be initialized already. The caller
 774  * must ensure nobody else is calling the function on the object at the same
 775  * time. FLDB service must be running to be able to map a FID to the targets
 776  * and find appropriate device representing that target.
 777  *
 778  * \param[in] env               execution environment for this thread
 779  * \param[in,out] lo            LOD object
 780  * \param[in] objs              an array of IDs to creates the objects from
 781  *
 782  * \retval                      0 if the objects are instantiated successfully
 783  * \retval                      negative error number on failure
 784  */
 785 int lod_initialize_objects(const struct lu_env *env, struct lod_object *lo,
 786                            struct lov_ost_data_v1 *objs)
 787 {
 788         struct lod_thread_info  *info = lod_env_info(env);
 789         struct lod_device       *md;
 790         struct lu_object        *o, *n;
 791         struct lu_device        *nd;
 792         struct dt_object       **stripe;
 793         int                      stripe_len;
 794         int                      i, rc = 0;
 795         __u32                   idx;
 796         ENTRY;
 797
 798         LASSERT(lo != NULL);
 799         md = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
 800         LASSERT(lo->ldo_stripe == NULL);
 801         LASSERT(lo->ldo_stripenr > 0);
 802         LASSERT(lo->ldo_stripe_size > 0);
 803
 804         stripe_len = lo->ldo_stripenr;
 805         OBD_ALLOC(stripe, sizeof(stripe[0]) * stripe_len);
 806         if (stripe == NULL)
 807                 RETURN(-ENOMEM);
 808
 809         for (i = 0; i < lo->ldo_stripenr; i++) {
 810                 if (unlikely(lovea_slot_is_dummy(&objs[i])))
 811                         continue;
 812
 813                 ostid_le_to_cpu(&objs[i].l_ost_oi, &info->lti_ostid);
 814                 idx = le32_to_cpu(objs[i].l_ost_idx);
 815                 rc = ostid_to_fid(&info->lti_fid, &info->lti_ostid, idx);
 816                 if (rc != 0)
 817                         GOTO(out, rc);
 818                 LASSERTF(fid_is_sane(&info->lti_fid), ""DFID" insane!\n",
 819                          PFID(&info->lti_fid));
 820                 lod_getref(&md->lod_ost_descs);
 821
 822                 rc = validate_lod_and_idx(md, idx);
 823                 if (unlikely(rc != 0)) {
 824                         lod_putref(md, &md->lod_ost_descs);
 825                         GOTO(out, rc);
 826                 }
 827
 828                 nd = &OST_TGT(md,idx)->ltd_ost->dd_lu_dev;
 829                 lod_putref(md, &md->lod_ost_descs);
 830
 831                 /* In the function below, .hs_keycmp resolves to
 832                  * u_obj_hop_keycmp() */
 833                 /* coverity[overrun-buffer-val] */
 834                 o = lu_object_find_at(env, nd, &info->lti_fid, NULL);
 835                 if (IS_ERR(o))
 836                         GOTO(out, rc = PTR_ERR(o));
 837
 838                 n = lu_object_locate(o->lo_header, nd->ld_type);
 839                 LASSERT(n);
 840
 841                 stripe[i] = container_of(n, struct dt_object, do_lu);
 842         }
 843
 844 out:
 845         if (rc != 0) {
 846                 for (i = 0; i < stripe_len; i++)
 847                         if (stripe[i] != NULL)
 848                                 lu_object_put(env, &stripe[i]->do_lu);
 849
 850                 OBD_FREE(stripe, sizeof(stripe[0]) * stripe_len);
 851                 lo->ldo_stripenr = 0;
 852         } else {
 853                 lo->ldo_stripe = stripe;
 854                 lo->ldo_stripes_allocated = stripe_len;
 855         }
 856
 857         RETURN(rc);
 858 }
 859
 860 /**
 861  * Instantiate objects for striping.
 862  *
 863  * Parse striping information in \a buf and instantiate the objects
 864  * representing the stripes.
 865  *
 866  * \param[in] env               execution environment for this thread
 867  * \param[in] lo                LOD object
 868  * \param[in] buf               buffer storing LOV EA to parse
 869  *
 870  * \retval                      0 if parsing and objects creation succeed
 871  * \retval                      negative error number on failure
 872  */
 873 int lod_parse_striping(const struct lu_env *env, struct lod_object *lo,
 874                        const struct lu_buf *buf)
 875 {
 876         struct lov_mds_md_v1    *lmm;
 877         struct lov_ost_data_v1  *objs;
 878         __u32                    magic;
 879         __u32                    pattern;
 880         int                      rc = 0;
 881         ENTRY;
 882
 883         LASSERT(buf);
 884         LASSERT(buf->lb_buf);
 885         LASSERT(buf->lb_len);
 886
 887         lmm = (struct lov_mds_md_v1 *) buf->lb_buf;
 888         magic = le32_to_cpu(lmm->lmm_magic);
 889         pattern = le32_to_cpu(lmm->lmm_pattern);
 890
 891         if (magic != LOV_MAGIC_V1 && magic != LOV_MAGIC_V3)
 892                 GOTO(out, rc = -EINVAL);
 893         if (lov_pattern(pattern) != LOV_PATTERN_RAID0)
 894                 GOTO(out, rc = -EINVAL);
 895
 896         lo->ldo_pattern = pattern;
 897         lo->ldo_stripe_size = le32_to_cpu(lmm->lmm_stripe_size);
 898         lo->ldo_layout_gen = le16_to_cpu(lmm->lmm_layout_gen);
 899         lo->ldo_stripenr = le16_to_cpu(lmm->lmm_stripe_count);
 900         /* released file stripenr fixup. */
 901         if (pattern & LOV_PATTERN_F_RELEASED)
 902                 lo->ldo_stripenr = 0;
 903
 904         LASSERT(buf->lb_len >= lov_mds_md_size(lo->ldo_stripenr, magic));
 905
 906         if (magic == LOV_MAGIC_V3) {
 907                 struct lov_mds_md_v3 *v3 = (struct lov_mds_md_v3 *) lmm;
 908                 objs = &v3->lmm_objects[0];
 909                 lod_object_set_pool(lo, v3->lmm_pool_name);
 910         } else {
 911                 objs = &lmm->lmm_objects[0];
 912         }
 913
 914         if (lo->ldo_stripenr > 0)
 915                 rc = lod_initialize_objects(env, lo, objs);
 916
 917 out:
 918         RETURN(rc);
 919 }
 920
 921 /**
 922  * Initialize the object representing the stripes.
 923  *
 924  * Unless the stripes are initialized already, fetch LOV (for regular
 925  * objects) or LMV (for directory objects) EA and call lod_parse_striping()
 926  * to instantiate the objects representing the stripes.
 927  *
 928  * \param[in] env               execution environment for this thread
 929  * \param[in,out] lo            LOD object
 930  *
 931  * \retval                      0 if parsing and object creation succeed
 932  * \retval                      negative error number on failure
 933  */
 934 int lod_load_striping_locked(const struct lu_env *env, struct lod_object *lo)
 935 {
 936         struct lod_thread_info  *info = lod_env_info(env);
 937         struct lu_buf           *buf  = &info->lti_buf;
 938         struct dt_object        *next = dt_object_child(&lo->ldo_obj);
 939         int                      rc = 0;
 940         ENTRY;
 941
 942         /* already initialized? */
 943         if (lo->ldo_stripe != NULL)
 944                 GOTO(out, rc = 0);
 945
 946         if (!dt_object_exists(next))
 947                 GOTO(out, rc = 0);
 948
 949         /* Do not load stripe for slaves of striped dir */
 950         if (lo->ldo_dir_slave_stripe)
 951                 GOTO(out, rc = 0);
 952
 953         if (S_ISREG(lu_object_attr(lod2lu_obj(lo)))) {
 954                 rc = lod_get_lov_ea(env, lo);
 955                 if (rc <= 0)
 956                         GOTO(out, rc);
 957                 /*
 958                  * there is LOV EA (striping information) in this object
 959                  * let's parse it and create in-core objects for the stripes
 960                  */
 961                 buf->lb_buf = info->lti_ea_store;
 962                 buf->lb_len = info->lti_ea_store_size;
 963                 rc = lod_parse_striping(env, lo, buf);
 964         } else if (S_ISDIR(lu_object_attr(lod2lu_obj(lo)))) {
 965                 rc = lod_get_lmv_ea(env, lo);
 966                 if (rc < (typeof(rc))sizeof(struct lmv_mds_md_v1))
 967                         GOTO(out, rc = rc > 0 ? -EINVAL : rc);
 968
 969                 buf->lb_buf = info->lti_ea_store;
 970                 buf->lb_len = info->lti_ea_store_size;
 971                 if (rc == sizeof(struct lmv_mds_md_v1)) {
 972                         rc = lod_load_lmv_shards(env, lo, buf, true);
 973                         if (buf->lb_buf != info->lti_ea_store) {
 974                                 OBD_FREE_LARGE(info->lti_ea_store,
 975                                                info->lti_ea_store_size);
 976                                 info->lti_ea_store = buf->lb_buf;
 977                                 info->lti_ea_store_size = buf->lb_len;
 978                         }
 979
 980                         if (rc < 0)
 981                                 GOTO(out, rc);
 982                 }
 983
 984                 /*
 985                  * there is LOV EA (striping information) in this object
 986                  * let's parse it and create in-core objects for the stripes
 987                  */
 988                 rc = lod_parse_dir_striping(env, lo, buf);
 989         }
 990
 991         if (rc == 0)
 992                 lo->ldo_striping_cached = 1;
 993 out:
 994         RETURN(rc);
 995 }
 996
 997 /**
 998  * A generic function to initialize the stripe objects.
 999  *
1000  * A protected version of lod_load_striping_locked() - load the striping
1001  * information from storage, parse that and instantiate LU objects to
1002  * represent the stripes.  The LOD object \a lo supplies a pointer to the
1003  * next sub-object in the LU stack so we can lock it. Also use \a lo to
1004  * return an array of references to the newly instantiated objects.
1005  *
1006  * \param[in] env               execution environment for this thread
1007  * \param[in,out] lo            LOD object, where striping is stored and
1008  *                              which gets an array of references
1009  *
1010  * \retval                      0 if parsing and object creation succeed
1011  * \retval                      negative error number on failure
1012  **/
1013 int lod_load_striping(const struct lu_env *env, struct lod_object *lo)
1014 {
1015         struct dt_object        *next = dt_object_child(&lo->ldo_obj);
1016         int                     rc = 0;
1017
1018         /* currently this code is supposed to be called from declaration
1019          * phase only, thus the object is not expected to be locked by caller */
1020         dt_write_lock(env, next, 0);
1021         rc = lod_load_striping_locked(env, lo);
1022         dt_write_unlock(env, next);
1023         return rc;
1024 }
1025
1026 /**
1027  * Verify striping.
1028  *
1029  * Check the validity of all fields including the magic, stripe size,
1030  * stripe count, stripe offset and that the pool is present.  Also check
1031  * that each target index points to an existing target. The additional
1032  * \a is_from_disk turns additional checks. In some cases zero fields
1033  * are allowed (like pattern=0).
1034  *
1035  * \param[in] d                 LOD device
1036  * \param[in] buf               buffer with LOV EA to verify
1037  * \param[in] is_from_disk      0 - from user, allow some fields to be 0
1038  *                              1 - from disk, do not allow
1039  *
1040  * \retval                      0 if the striping is valid
1041  * \retval                      -EINVAL if striping is invalid
1042  */
1043 int lod_verify_striping(struct lod_device *d, const struct lu_buf *buf,
1044                         bool is_from_disk)
1045 {
1046         struct lov_user_md_v1   *lum;
1047         struct lov_user_md_v3   *lum3;
1048         struct pool_desc        *pool = NULL;
1049         __u32                    magic;
1050         __u32                    stripe_size;
1051         __u16                    stripe_count;
1052         __u16                    stripe_offset;
1053         size_t                   lum_size;
1054         int                      rc = 0;
1055         ENTRY;
1056
1057         lum = buf->lb_buf;
1058
1059         LASSERT(sizeof(*lum) < sizeof(*lum3));
1060
1061         if (buf->lb_len < sizeof(*lum)) {
1062                 CDEBUG(D_IOCTL, "buf len %zu too small for lov_user_md\n",
1063                        buf->lb_len);
1064                 GOTO(out, rc = -EINVAL);
1065         }
1066
1067         magic = le32_to_cpu(lum->lmm_magic);
1068         if (magic != LOV_USER_MAGIC_V1 &&
1069             magic != LOV_USER_MAGIC_V3 &&
1070             magic != LOV_MAGIC_V1_DEF &&
1071             magic != LOV_MAGIC_V3_DEF) {
1072                 CDEBUG(D_IOCTL, "bad userland LOV MAGIC: %#x\n", magic);
1073                 GOTO(out, rc = -EINVAL);
1074         }
1075
1076         /* the user uses "0" for default stripe pattern normally. */
1077         if (!is_from_disk && lum->lmm_pattern == 0)
1078                 lum->lmm_pattern = cpu_to_le32(LOV_PATTERN_RAID0);
1079
1080         if (le32_to_cpu(lum->lmm_pattern) != LOV_PATTERN_RAID0) {
1081                 CDEBUG(D_IOCTL, "bad userland stripe pattern: %#x\n",
1082                        le32_to_cpu(lum->lmm_pattern));
1083                 GOTO(out, rc = -EINVAL);
1084         }
1085
1086         /* 64kB is the largest common page size we see (ia64), and matches the
1087          * check in lfs */
1088         stripe_size = le32_to_cpu(lum->lmm_stripe_size);
1089         if (stripe_size & (LOV_MIN_STRIPE_SIZE - 1)) {
1090                 CDEBUG(D_IOCTL, "stripe size %u not a multiple of %u\n",
1091                        stripe_size, LOV_MIN_STRIPE_SIZE);
1092                 GOTO(out, rc = -EINVAL);
1093         }
1094
1095         stripe_offset = le16_to_cpu(lum->lmm_stripe_offset);
1096         if (stripe_offset != LOV_OFFSET_DEFAULT) {
1097                 /* if offset is not within valid range [0, osts_size) */
1098                 if (stripe_offset >= d->lod_osts_size) {
1099                         CDEBUG(D_IOCTL, "stripe offset %u >= bitmap size %u\n",
1100                                stripe_offset, d->lod_osts_size);
1101                         GOTO(out, rc = -EINVAL);
1102                 }
1103
1104                 /* if lmm_stripe_offset is *not* in bitmap */
1105                 if (!cfs_bitmap_check(d->lod_ost_bitmap, stripe_offset)) {
1106                         CDEBUG(D_IOCTL, "stripe offset %u not in bitmap\n",
1107                                stripe_offset);
1108                         GOTO(out, rc = -EINVAL);
1109                 }
1110         }
1111
1112         if (magic == LOV_USER_MAGIC_V1 || magic == LOV_MAGIC_V1_DEF)
1113                 lum_size = offsetof(struct lov_user_md_v1,
1114                                     lmm_objects[0]);
1115         else if (magic == LOV_USER_MAGIC_V3 || magic == LOV_MAGIC_V3_DEF)
1116                 lum_size = offsetof(struct lov_user_md_v3,
1117                                     lmm_objects[0]);
1118         else
1119                 GOTO(out, rc = -EINVAL);
1120
1121         stripe_count = le16_to_cpu(lum->lmm_stripe_count);
1122         if (buf->lb_len != lum_size) {
1123                 CDEBUG(D_IOCTL, "invalid buf len %zu for lov_user_md with "
1124                        "magic %#x and stripe_count %u\n",
1125                        buf->lb_len, magic, stripe_count);
1126                 GOTO(out, rc = -EINVAL);
1127         }
1128
1129         if (!(magic == LOV_USER_MAGIC_V3 || magic == LOV_MAGIC_V3_DEF))
1130                 goto out;
1131
1132         lum3 = buf->lb_buf;
1133         if (buf->lb_len < sizeof(*lum3)) {
1134                 CDEBUG(D_IOCTL, "buf len %zu too small for lov_user_md_v3\n",
1135                        buf->lb_len);
1136                 GOTO(out, rc = -EINVAL);
1137         }
1138
1139         /* In the function below, .hs_keycmp resolves to
1140          * pool_hashkey_keycmp() */
1141         /* coverity[overrun-buffer-val] */
1142         pool = lod_find_pool(d, lum3->lmm_pool_name);
1143         if (pool == NULL)
1144                 goto out;
1145
1146         if (stripe_offset != LOV_OFFSET_DEFAULT) {
1147                 rc = lod_check_index_in_pool(stripe_offset, pool);
1148                 if (rc < 0)
1149                         GOTO(out, rc = -EINVAL);
1150         }
1151
1152         if (is_from_disk && stripe_count > pool_tgt_count(pool)) {
1153                 CDEBUG(D_IOCTL,
1154                        "stripe count %u > # OSTs %u in the pool\n",
1155                        stripe_count, pool_tgt_count(pool));
1156                 GOTO(out, rc = -EINVAL);
1157         }
1158
1159 out:
1160         if (pool != NULL)
1161                 lod_pool_putref(pool);
1162
1163         RETURN(rc);
1164 }
1165
1166 void lod_fix_desc_stripe_size(__u64 *val)
1167 {
1168         if (*val < LOV_MIN_STRIPE_SIZE) {
1169                 if (*val != 0)
1170                         LCONSOLE_INFO("Increasing default stripe size to "
1171                                       "minimum value %u\n",
1172                                       LOV_DESC_STRIPE_SIZE_DEFAULT);
1173                 *val = LOV_DESC_STRIPE_SIZE_DEFAULT;
1174         } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) {
1175                 *val &= ~(LOV_MIN_STRIPE_SIZE - 1);
1176                 LCONSOLE_WARN("Changing default stripe size to "LPU64" (a "
1177                               "multiple of %u)\n",
1178                               *val, LOV_MIN_STRIPE_SIZE);
1179         }
1180 }
1181
1182 void lod_fix_desc_stripe_count(__u32 *val)
1183 {
1184         if (*val == 0)
1185                 *val = 1;
1186 }
1187
1188 void lod_fix_desc_pattern(__u32 *val)
1189 {
1190         /* from lov_setstripe */
1191         if ((*val != 0) && (*val != LOV_PATTERN_RAID0)) {
1192                 LCONSOLE_WARN("Unknown stripe pattern: %#x\n", *val);
1193                 *val = 0;
1194         }
1195 }
1196
1197 void lod_fix_desc_qos_maxage(__u32 *val)
1198 {
1199         /* fix qos_maxage */
1200         if (*val == 0)
1201                 *val = LOV_DESC_QOS_MAXAGE_DEFAULT;
1202 }
1203
1204 /**
1205  * Used to fix insane default striping.
1206  *
1207  * \param[in] desc      striping description
1208  */
1209 void lod_fix_desc(struct lov_desc *desc)
1210 {
1211         lod_fix_desc_stripe_size(&desc->ld_default_stripe_size);
1212         lod_fix_desc_stripe_count(&desc->ld_default_stripe_count);
1213         lod_fix_desc_pattern(&desc->ld_pattern);
1214         lod_fix_desc_qos_maxage(&desc->ld_qos_maxage);
1215 }
1216
1217 /**
1218  * Initialize the structures used to store pools and default striping.
1219  *
1220  * \param[in] lod       LOD device
1221  * \param[in] lcfg      configuration structure storing default striping.
1222  *
1223  * \retval              0 if initialization succeeds
1224  * \retval              negative error number on failure
1225  */
1226 int lod_pools_init(struct lod_device *lod, struct lustre_cfg *lcfg)
1227 {
1228         struct obd_device          *obd;
1229         struct lov_desc            *desc;
1230         int                         rc;
1231         ENTRY;
1232
1233         obd = class_name2obd(lustre_cfg_string(lcfg, 0));
1234         LASSERT(obd != NULL);
1235         obd->obd_lu_dev = &lod->lod_dt_dev.dd_lu_dev;
1236
1237         if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
1238                 CERROR("LOD setup requires a descriptor\n");
1239                 RETURN(-EINVAL);
1240         }
1241
1242         desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
1243
1244         if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
1245                 CERROR("descriptor size wrong: %d > %d\n",
1246                        (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
1247                 RETURN(-EINVAL);
1248         }
1249
1250         if (desc->ld_magic != LOV_DESC_MAGIC) {
1251                 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
1252                         CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
1253                                obd->obd_name, desc);
1254                         lustre_swab_lov_desc(desc);
1255                 } else {
1256                         CERROR("%s: Bad lov desc magic: %#x\n",
1257                                obd->obd_name, desc->ld_magic);
1258                         RETURN(-EINVAL);
1259                 }
1260         }
1261
1262         lod_fix_desc(desc);
1263
1264         desc->ld_active_tgt_count = 0;
1265         lod->lod_desc = *desc;
1266
1267         lod->lod_sp_me = LUSTRE_SP_CLI;
1268
1269         /* Set up allocation policy (QoS and RR) */
1270         INIT_LIST_HEAD(&lod->lod_qos.lq_oss_list);
1271         init_rwsem(&lod->lod_qos.lq_rw_sem);
1272         lod->lod_qos.lq_dirty = 1;
1273         lod->lod_qos.lq_rr.lqr_dirty = 1;
1274         lod->lod_qos.lq_reset = 1;
1275         /* Default priority is toward free space balance */
1276         lod->lod_qos.lq_prio_free = 232;
1277         /* Default threshold for rr (roughly 17%) */
1278         lod->lod_qos.lq_threshold_rr = 43;
1279
1280         /* Set up OST pool environment */
1281         lod->lod_pools_hash_body = cfs_hash_create("POOLS", HASH_POOLS_CUR_BITS,
1282                                                    HASH_POOLS_MAX_BITS,
1283                                                    HASH_POOLS_BKT_BITS, 0,
1284                                                    CFS_HASH_MIN_THETA,
1285                                                    CFS_HASH_MAX_THETA,
1286                                                    &pool_hash_operations,
1287                                                    CFS_HASH_DEFAULT);
1288         if (lod->lod_pools_hash_body == NULL)
1289                 RETURN(-ENOMEM);
1290
1291         INIT_LIST_HEAD(&lod->lod_pool_list);
1292         lod->lod_pool_count = 0;
1293         rc = lod_ost_pool_init(&lod->lod_pool_info, 0);
1294         if (rc)
1295                 GOTO(out_hash, rc);
1296         rc = lod_ost_pool_init(&lod->lod_qos.lq_rr.lqr_pool, 0);
1297         if (rc)
1298                 GOTO(out_pool_info, rc);
1299
1300         RETURN(0);
1301
1302 out_pool_info:
1303         lod_ost_pool_free(&lod->lod_pool_info);
1304 out_hash:
1305         cfs_hash_putref(lod->lod_pools_hash_body);
1306
1307         return rc;
1308 }
1309
1310 /**
1311  * Release the structures describing the pools.
1312  *
1313  * \param[in] lod       LOD device from which we release the structures
1314  *
1315  * \retval              0 always
1316  */
1317 int lod_pools_fini(struct lod_device *lod)
1318 {
1319         struct obd_device   *obd = lod2obd(lod);
1320         struct pool_desc    *pool, *tmp;
1321         ENTRY;
1322
1323         list_for_each_entry_safe(pool, tmp, &lod->lod_pool_list, pool_list) {
1324                 /* free pool structs */
1325                 CDEBUG(D_INFO, "delete pool %p\n", pool);
1326                 /* In the function below, .hs_keycmp resolves to
1327                  * pool_hashkey_keycmp() */
1328                 /* coverity[overrun-buffer-val] */
1329                 lod_pool_del(obd, pool->pool_name);
1330         }
1331
1332         cfs_hash_putref(lod->lod_pools_hash_body);
1333         lod_ost_pool_free(&(lod->lod_qos.lq_rr.lqr_pool));
1334         lod_ost_pool_free(&lod->lod_pool_info);
1335
1336         RETURN(0);
1337 }