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 (for_ost) {
 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         if (!for_ost && lod->lod_initialized) {
 382                 rc = lod_sub_init_llog(env, lod, tgt_desc->ltd_tgt);
 383                 if (rc != 0) {
 384                         CERROR("%s: cannot start llog on %s:rc = %d\n",
 385                                lod2obd(lod)->obd_name, osp, rc);
 386                         GOTO(out_pool, rc);
 387                 }
 388         }
 389
 390         rc = lfsck_add_target(env, lod->lod_child, d, exp, index, for_ost);
 391         if (rc != 0) {
 392                 CERROR("Fail to add LFSCK target: name = %s, type = %s, "
 393                        "index = %u, rc = %d\n", osp, type, index, rc);
 394                 GOTO(out_fini_llog, rc);
 395         }
 396         RETURN(rc);
 397 out_fini_llog:
 398         lod_sub_fini_llog(env, tgt_desc->ltd_tgt,
 399                           tgt_desc->ltd_recovery_thread);
 400 out_pool:
 401         lod_ost_pool_remove(&lod->lod_pool_info, index);
 402 out_mutex:
 403         mutex_unlock(&ltd->ltd_mutex);
 404         lod_putref(lod, ltd);
 405 out_desc:
 406         OBD_FREE_PTR(tgt_desc);
 407 out_conn:
 408         obd_disconnect(exp);
 409 out_free:
 410         return rc;
 411 }
 412
 413 /**
 414  * Schedule target removal from the target table.
 415  *
 416  * Mark the device as dead. The device is not removed here because it may
 417  * still be in use. The device will be removed in lod_putref() when the
 418  * last reference is released.
 419  *
 420  * \param[in] env               execution environment for this thread
 421  * \param[in] lod               LOD device the target table belongs to
 422  * \param[in] ltd               target table
 423  * \param[in] idx               index of the target
 424  * \param[in] for_ost           type of the target: 0 - MDT, 1 - OST
 425  */
 426 static void __lod_del_device(const struct lu_env *env, struct lod_device *lod,
 427                              struct lod_tgt_descs *ltd, unsigned idx,
 428                              bool for_ost)
 429 {
 430         LASSERT(LTD_TGT(ltd, idx));
 431
 432         lfsck_del_target(env, lod->lod_child, LTD_TGT(ltd, idx)->ltd_tgt,
 433                          idx, for_ost);
 434
 435         if (LTD_TGT(ltd, idx)->ltd_reap == 0) {
 436                 LTD_TGT(ltd, idx)->ltd_reap = 1;
 437                 ltd->ltd_death_row++;
 438         }
 439 }
 440
 441 /**
 442  * Schedule removal of all the targets from the given target table.
 443  *
 444  * See more details in the description for __lod_del_device()
 445  *
 446  * \param[in] env               execution environment for this thread
 447  * \param[in] lod               LOD device the target table belongs to
 448  * \param[in] ltd               target table
 449  * \param[in] for_ost           type of the target: MDT or OST
 450  *
 451  * \retval                      0 always
 452  */
 453 int lod_fini_tgt(const struct lu_env *env, struct lod_device *lod,
 454                  struct lod_tgt_descs *ltd, bool for_ost)
 455 {
 456         unsigned int idx;
 457
 458         if (ltd->ltd_tgts_size <= 0)
 459                 return 0;
 460         lod_getref(ltd);
 461         mutex_lock(&ltd->ltd_mutex);
 462         cfs_foreach_bit(ltd->ltd_tgt_bitmap, idx)
 463                 __lod_del_device(env, lod, ltd, idx, for_ost);
 464         mutex_unlock(&ltd->ltd_mutex);
 465         lod_putref(lod, ltd);
 466         CFS_FREE_BITMAP(ltd->ltd_tgt_bitmap);
 467         for (idx = 0; idx < TGT_PTRS; idx++) {
 468                 if (ltd->ltd_tgt_idx[idx])
 469                         OBD_FREE_PTR(ltd->ltd_tgt_idx[idx]);
 470         }
 471         ltd->ltd_tgts_size = 0;
 472         return 0;
 473 }
 474
 475 /**
 476  * Remove device by name.
 477  *
 478  * Remove a device identified by \a osp from the target table. Given
 479  * the device can be in use, the real deletion happens in lod_putref().
 480  *
 481  * \param[in] env               execution environment for this thread
 482  * \param[in] lod               LOD device to be connected to the new OSP
 483  * \param[in] ltd               target table
 484  * \param[in] osp               name of OSP device to be removed
 485  * \param[in] idx               index of the target
 486  * \param[in] gen               generation number, not used currently
 487  * \param[in] for_ost           type of the target: 0 - MDT, 1 - OST
 488  *
 489  * \retval                      0 if the device was scheduled for removal
 490  * \retval                      -EINVAL if no device was found
 491  */
 492 int lod_del_device(const struct lu_env *env, struct lod_device *lod,
 493                    struct lod_tgt_descs *ltd, char *osp, unsigned idx,
 494                    unsigned gen, bool for_ost)
 495 {
 496         struct obd_device *obd;
 497         int                rc = 0;
 498         struct obd_uuid    uuid;
 499         ENTRY;
 500
 501         CDEBUG(D_CONFIG, "osp:%s idx:%d gen:%d\n", osp, idx, gen);
 502
 503         obd_str2uuid(&uuid, osp);
 504
 505         obd = class_find_client_obd(&uuid, LUSTRE_OSP_NAME,
 506                                    &lod->lod_dt_dev.dd_lu_dev.ld_obd->obd_uuid);
 507         if (obd == NULL) {
 508                 CERROR("can't find %s device\n", osp);
 509                 RETURN(-EINVAL);
 510         }
 511
 512         if (gen <= 0) {
 513                 CERROR("%s: request to remove OBD %s with invalid generation %d"
 514                        "\n", obd->obd_name, osp, gen);
 515                 RETURN(-EINVAL);
 516         }
 517
 518         obd_str2uuid(&uuid,  osp);
 519
 520         lod_getref(ltd);
 521         mutex_lock(&ltd->ltd_mutex);
 522         /* check that the index is allocated in the bitmap */
 523         if (!cfs_bitmap_check(ltd->ltd_tgt_bitmap, idx) ||
 524             !LTD_TGT(ltd, idx)) {
 525                 CERROR("%s: device %d is not set up\n", obd->obd_name, idx);
 526                 GOTO(out, rc = -EINVAL);
 527         }
 528
 529         /* check that the UUID matches */
 530         if (!obd_uuid_equals(&uuid, &LTD_TGT(ltd, idx)->ltd_uuid)) {
 531                 CERROR("%s: LOD target UUID %s at index %d does not match %s\n",
 532                        obd->obd_name, obd_uuid2str(&LTD_TGT(ltd,idx)->ltd_uuid),
 533                        idx, osp);
 534                 GOTO(out, rc = -EINVAL);
 535         }
 536
 537         __lod_del_device(env, lod, ltd, idx, for_ost);
 538         EXIT;
 539 out:
 540         mutex_unlock(&ltd->ltd_mutex);
 541         lod_putref(lod, ltd);
 542         return(rc);
 543 }
 544
 545 /**
 546  * Resize per-thread storage to hold specified size.
 547  *
 548  * A helper function to resize per-thread temporary storage. This storage
 549  * is used to process LOV/LVM EAs and may be quite large. We do not want to
 550  * allocate/release it every time, so instead we put it into the env and
 551  * reallocate on demand. The memory is released when the correspondent thread
 552  * is finished.
 553  *
 554  * \param[in] info              LOD-specific storage in the environment
 555  * \param[in] size              new size to grow the buffer to
 556
 557  * \retval                      0 on success, -ENOMEM if reallocation failed
 558  */
 559 int lod_ea_store_resize(struct lod_thread_info *info, size_t size)
 560 {
 561         __u32 round = size_roundup_power2(size);
 562
 563         LASSERT(round <=
 564                 lov_mds_md_size(LOV_MAX_STRIPE_COUNT, LOV_MAGIC_V3));
 565         if (info->lti_ea_store) {
 566                 LASSERT(info->lti_ea_store_size);
 567                 LASSERT(info->lti_ea_store_size < round);
 568                 CDEBUG(D_INFO, "EA store size %d is not enough, need %d\n",
 569                        info->lti_ea_store_size, round);
 570                 OBD_FREE_LARGE(info->lti_ea_store, info->lti_ea_store_size);
 571                 info->lti_ea_store = NULL;
 572                 info->lti_ea_store_size = 0;
 573         }
 574
 575         OBD_ALLOC_LARGE(info->lti_ea_store, round);
 576         if (info->lti_ea_store == NULL)
 577                 RETURN(-ENOMEM);
 578         info->lti_ea_store_size = round;
 579         RETURN(0);
 580 }
 581
 582 /**
 583  * Make LOV EA for striped object.
 584  *
 585  * Generate striping information and store it in the LOV EA of the given
 586  * object. The caller must ensure nobody else is calling the function
 587  * against the object concurrently. The transaction must be started.
 588  * FLDB service must be running as well; it's used to map FID to the target,
 589  * which is stored in LOV EA.
 590  *
 591  * \param[in] env               execution environment for this thread
 592  * \param[in] lo                LOD object
 593  * \param[in] th                transaction handle
 594  *
 595  * \retval                      0 if LOV EA is stored successfully
 596  * \retval                      negative error number on failure
 597  */
 598 int lod_generate_and_set_lovea(const struct lu_env *env,
 599                                struct lod_object *lo, struct thandle *th)
 600 {
 601         struct lod_thread_info  *info = lod_env_info(env);
 602         struct dt_object        *next = dt_object_child(&lo->ldo_obj);
 603         const struct lu_fid     *fid  = lu_object_fid(&lo->ldo_obj.do_lu);
 604         struct lov_mds_md_v1    *lmm;
 605         struct lov_ost_data_v1  *objs;
 606         __u32                    magic;
 607         int                      i, rc;
 608         size_t                   lmm_size;
 609         ENTRY;
 610
 611         LASSERT(lo);
 612
 613         magic = lo->ldo_pool != NULL ? LOV_MAGIC_V3 : LOV_MAGIC_V1;
 614         lmm_size = lov_mds_md_size(lo->ldo_stripenr, magic);
 615         if (info->lti_ea_store_size < lmm_size) {
 616                 rc = lod_ea_store_resize(info, lmm_size);
 617                 if (rc)
 618                         RETURN(rc);
 619         }
 620
 621         if (lo->ldo_pattern == 0) /* default striping */
 622                 lo->ldo_pattern = LOV_PATTERN_RAID0;
 623
 624         lmm = info->lti_ea_store;
 625
 626         lmm->lmm_magic = cpu_to_le32(magic);
 627         lmm->lmm_pattern = cpu_to_le32(lo->ldo_pattern);
 628         fid_to_lmm_oi(fid, &lmm->lmm_oi);
 629         if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_LMMOI))
 630                 lmm->lmm_oi.oi.oi_id++;
 631         lmm_oi_cpu_to_le(&lmm->lmm_oi, &lmm->lmm_oi);
 632         lmm->lmm_stripe_size = cpu_to_le32(lo->ldo_stripe_size);
 633         lmm->lmm_stripe_count = cpu_to_le16(lo->ldo_stripenr);
 634         if (lo->ldo_pattern & LOV_PATTERN_F_RELEASED)
 635                 lmm->lmm_stripe_count = cpu_to_le16(lo->ldo_released_stripenr);
 636         lmm->lmm_layout_gen = 0;
 637         if (magic == LOV_MAGIC_V1) {
 638                 objs = &lmm->lmm_objects[0];
 639         } else {
 640                 struct lov_mds_md_v3 *v3 = (struct lov_mds_md_v3 *) lmm;
 641                 size_t cplen = strlcpy(v3->lmm_pool_name, lo->ldo_pool,
 642                                 sizeof(v3->lmm_pool_name));
 643                 if (cplen >= sizeof(v3->lmm_pool_name))
 644                         RETURN(-E2BIG);
 645                 objs = &v3->lmm_objects[0];
 646         }
 647
 648         for (i = 0; i < lo->ldo_stripenr; i++) {
 649                 struct lu_fid           *fid    = &info->lti_fid;
 650                 struct lod_device       *lod;
 651                 __u32                   index;
 652                 int                     type    = LU_SEQ_RANGE_OST;
 653
 654                 lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
 655                 LASSERT(lo->ldo_stripe[i]);
 656
 657                 *fid = *lu_object_fid(&lo->ldo_stripe[i]->do_lu);
 658                 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_MULTIPLE_REF)) {
 659                         if (cfs_fail_val == 0)
 660                                 cfs_fail_val = fid->f_oid;
 661                         else
 662                                 fid->f_oid = cfs_fail_val;
 663                 }
 664
 665                 rc = fid_to_ostid(fid, &info->lti_ostid);
 666                 LASSERT(rc == 0);
 667
 668                 ostid_cpu_to_le(&info->lti_ostid, &objs[i].l_ost_oi);
 669                 objs[i].l_ost_gen    = cpu_to_le32(0);
 670                 rc = lod_fld_lookup(env, lod, fid, &index, &type);
 671                 if (rc < 0) {
 672                         CERROR("%s: Can not locate "DFID": rc = %d\n",
 673                                lod2obd(lod)->obd_name, PFID(fid), rc);
 674                         lod_object_free_striping(env, lo);
 675                         RETURN(rc);
 676                 }
 677                 objs[i].l_ost_idx = cpu_to_le32(index);
 678         }
 679
 680         info->lti_buf.lb_buf = lmm;
 681         info->lti_buf.lb_len = lmm_size;
 682         rc = lod_sub_object_xattr_set(env, next, &info->lti_buf, XATTR_NAME_LOV,
 683                                       0, th);
 684         if (rc < 0) {
 685                 lod_object_free_striping(env, lo);
 686                 RETURN(rc);
 687         }
 688
 689         RETURN(rc);
 690 }
 691
 692 /**
 693  * Get LOV EA.
 694  *
 695  * Fill lti_ea_store buffer in the environment with a value for the given
 696  * EA. The buffer is reallocated if the value doesn't fit.
 697  *
 698  * \param[in,out] env           execution environment for this thread
 699  *                              .lti_ea_store buffer is filled with EA's value
 700  * \param[in] lo                LOD object
 701  * \param[in] name              name of the EA
 702  *
 703  * \retval                      0 if EA is fetched successfully
 704  * \retval                      negative error number on failure
 705  */
 706 int lod_get_ea(const struct lu_env *env, struct lod_object *lo,
 707                const char *name)
 708 {
 709         struct lod_thread_info  *info = lod_env_info(env);
 710         struct dt_object        *next = dt_object_child(&lo->ldo_obj);
 711         int                     rc;
 712         ENTRY;
 713
 714         LASSERT(info);
 715
 716         if (unlikely(info->lti_ea_store == NULL)) {
 717                 /* just to enter in allocation block below */
 718                 rc = -ERANGE;
 719         } else {
 720 repeat:
 721                 info->lti_buf.lb_buf = info->lti_ea_store;
 722                 info->lti_buf.lb_len = info->lti_ea_store_size;
 723                 rc = dt_xattr_get(env, next, &info->lti_buf, name);
 724         }
 725
 726         /* if object is not striped or inaccessible */
 727         if (rc == -ENODATA || rc == -ENOENT)
 728                 RETURN(0);
 729
 730         if (rc == -ERANGE) {
 731                 /* EA doesn't fit, reallocate new buffer */
 732                 rc = dt_xattr_get(env, next, &LU_BUF_NULL, name);
 733                 if (rc == -ENODATA || rc == -ENOENT)
 734                         RETURN(0);
 735                 else if (rc < 0)
 736                         RETURN(rc);
 737
 738                 LASSERT(rc > 0);
 739                 rc = lod_ea_store_resize(info, rc);
 740                 if (rc)
 741                         RETURN(rc);
 742                 goto repeat;
 743         }
 744
 745         RETURN(rc);
 746 }
 747
 748 /**
 749  * Verify the target index is present in the current configuration.
 750  *
 751  * \param[in] md                LOD device where the target table is stored
 752  * \param[in] idx               target's index
 753  *
 754  * \retval                      0 if the index is present
 755  * \retval                      -EINVAL if not
 756  */
 757 static int validate_lod_and_idx(struct lod_device *md, __u32 idx)
 758 {
 759         if (unlikely(idx >= md->lod_ost_descs.ltd_tgts_size ||
 760                      !cfs_bitmap_check(md->lod_ost_bitmap, idx))) {
 761                 CERROR("%s: bad idx: %d of %d\n", lod2obd(md)->obd_name, idx,
 762                        md->lod_ost_descs.ltd_tgts_size);
 763                 return -EINVAL;
 764         }
 765
 766         if (unlikely(OST_TGT(md, idx) == NULL)) {
 767                 CERROR("%s: bad lod_tgt_desc for idx: %d\n",
 768                        lod2obd(md)->obd_name, idx);
 769                 return -EINVAL;
 770         }
 771
 772         if (unlikely(OST_TGT(md, idx)->ltd_ost == NULL)) {
 773                 CERROR("%s: invalid lod device, for idx: %d\n",
 774                        lod2obd(md)->obd_name , idx);
 775                 return -EINVAL;
 776         }
 777
 778         return 0;
 779 }
 780
 781 /**
 782  * Instantiate objects for stripes.
 783  *
 784  * Allocate and initialize LU-objects representing the stripes. The number
 785  * of the stripes (ldo_stripenr) must be initialized already. The caller
 786  * must ensure nobody else is calling the function on the object at the same
 787  * time. FLDB service must be running to be able to map a FID to the targets
 788  * and find appropriate device representing that target.
 789  *
 790  * \param[in] env               execution environment for this thread
 791  * \param[in,out] lo            LOD object
 792  * \param[in] objs              an array of IDs to creates the objects from
 793  *
 794  * \retval                      0 if the objects are instantiated successfully
 795  * \retval                      negative error number on failure
 796  */
 797 int lod_initialize_objects(const struct lu_env *env, struct lod_object *lo,
 798                            struct lov_ost_data_v1 *objs)
 799 {
 800         struct lod_thread_info  *info = lod_env_info(env);
 801         struct lod_device       *md;
 802         struct lu_object        *o, *n;
 803         struct lu_device        *nd;
 804         struct dt_object       **stripe;
 805         int                      stripe_len;
 806         int                      i, rc = 0;
 807         __u32                   idx;
 808         ENTRY;
 809
 810         LASSERT(lo != NULL);
 811         md = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
 812         LASSERT(lo->ldo_stripe == NULL);
 813         LASSERT(lo->ldo_stripenr > 0);
 814         LASSERT(lo->ldo_stripe_size > 0);
 815
 816         stripe_len = lo->ldo_stripenr;
 817         OBD_ALLOC(stripe, sizeof(stripe[0]) * stripe_len);
 818         if (stripe == NULL)
 819                 RETURN(-ENOMEM);
 820
 821         for (i = 0; i < lo->ldo_stripenr; i++) {
 822                 if (unlikely(lovea_slot_is_dummy(&objs[i])))
 823                         continue;
 824
 825                 ostid_le_to_cpu(&objs[i].l_ost_oi, &info->lti_ostid);
 826                 idx = le32_to_cpu(objs[i].l_ost_idx);
 827                 rc = ostid_to_fid(&info->lti_fid, &info->lti_ostid, idx);
 828                 if (rc != 0)
 829                         GOTO(out, rc);
 830                 LASSERTF(fid_is_sane(&info->lti_fid), ""DFID" insane!\n",
 831                          PFID(&info->lti_fid));
 832                 lod_getref(&md->lod_ost_descs);
 833
 834                 rc = validate_lod_and_idx(md, idx);
 835                 if (unlikely(rc != 0)) {
 836                         lod_putref(md, &md->lod_ost_descs);
 837                         GOTO(out, rc);
 838                 }
 839
 840                 nd = &OST_TGT(md,idx)->ltd_ost->dd_lu_dev;
 841                 lod_putref(md, &md->lod_ost_descs);
 842
 843                 /* In the function below, .hs_keycmp resolves to
 844                  * u_obj_hop_keycmp() */
 845                 /* coverity[overrun-buffer-val] */
 846                 o = lu_object_find_at(env, nd, &info->lti_fid, NULL);
 847                 if (IS_ERR(o))
 848                         GOTO(out, rc = PTR_ERR(o));
 849
 850                 n = lu_object_locate(o->lo_header, nd->ld_type);
 851                 LASSERT(n);
 852
 853                 stripe[i] = container_of(n, struct dt_object, do_lu);
 854         }
 855
 856 out:
 857         if (rc != 0) {
 858                 for (i = 0; i < stripe_len; i++)
 859                         if (stripe[i] != NULL)
 860                                 lu_object_put(env, &stripe[i]->do_lu);
 861
 862                 OBD_FREE(stripe, sizeof(stripe[0]) * stripe_len);
 863                 lo->ldo_stripenr = 0;
 864         } else {
 865                 lo->ldo_stripe = stripe;
 866                 lo->ldo_stripes_allocated = stripe_len;
 867         }
 868
 869         RETURN(rc);
 870 }
 871
 872 /**
 873  * Instantiate objects for striping.
 874  *
 875  * Parse striping information in \a buf and instantiate the objects
 876  * representing the stripes.
 877  *
 878  * \param[in] env               execution environment for this thread
 879  * \param[in] lo                LOD object
 880  * \param[in] buf               buffer storing LOV EA to parse
 881  *
 882  * \retval                      0 if parsing and objects creation succeed
 883  * \retval                      negative error number on failure
 884  */
 885 int lod_parse_striping(const struct lu_env *env, struct lod_object *lo,
 886                        const struct lu_buf *buf)
 887 {
 888         struct lov_mds_md_v1    *lmm;
 889         struct lov_ost_data_v1  *objs;
 890         __u32                    magic;
 891         __u32                    pattern;
 892         int                      rc = 0;
 893         ENTRY;
 894
 895         LASSERT(buf);
 896         LASSERT(buf->lb_buf);
 897         LASSERT(buf->lb_len);
 898
 899         lmm = (struct lov_mds_md_v1 *) buf->lb_buf;
 900         magic = le32_to_cpu(lmm->lmm_magic);
 901         pattern = le32_to_cpu(lmm->lmm_pattern);
 902
 903         if (magic != LOV_MAGIC_V1 && magic != LOV_MAGIC_V3)
 904                 GOTO(out, rc = -EINVAL);
 905         if (lov_pattern(pattern) != LOV_PATTERN_RAID0)
 906                 GOTO(out, rc = -EINVAL);
 907
 908         lo->ldo_pattern = pattern;
 909         lo->ldo_stripe_size = le32_to_cpu(lmm->lmm_stripe_size);
 910         lo->ldo_layout_gen = le16_to_cpu(lmm->lmm_layout_gen);
 911         lo->ldo_stripenr = le16_to_cpu(lmm->lmm_stripe_count);
 912         /* released file stripenr fixup. */
 913         if (pattern & LOV_PATTERN_F_RELEASED)
 914                 lo->ldo_stripenr = 0;
 915
 916         LASSERT(buf->lb_len >= lov_mds_md_size(lo->ldo_stripenr, magic));
 917
 918         if (magic == LOV_MAGIC_V3) {
 919                 struct lov_mds_md_v3 *v3 = (struct lov_mds_md_v3 *) lmm;
 920                 objs = &v3->lmm_objects[0];
 921                 lod_object_set_pool(lo, v3->lmm_pool_name);
 922         } else {
 923                 objs = &lmm->lmm_objects[0];
 924         }
 925
 926         if (lo->ldo_stripenr > 0)
 927                 rc = lod_initialize_objects(env, lo, objs);
 928
 929 out:
 930         RETURN(rc);
 931 }
 932
 933 /**
 934  * Initialize the object representing the stripes.
 935  *
 936  * Unless the stripes are initialized already, fetch LOV (for regular
 937  * objects) or LMV (for directory objects) EA and call lod_parse_striping()
 938  * to instantiate the objects representing the stripes.
 939  *
 940  * \param[in] env               execution environment for this thread
 941  * \param[in,out] lo            LOD object
 942  *
 943  * \retval                      0 if parsing and object creation succeed
 944  * \retval                      negative error number on failure
 945  */
 946 int lod_load_striping_locked(const struct lu_env *env, struct lod_object *lo)
 947 {
 948         struct lod_thread_info  *info = lod_env_info(env);
 949         struct lu_buf           *buf  = &info->lti_buf;
 950         struct dt_object        *next = dt_object_child(&lo->ldo_obj);
 951         int                      rc = 0;
 952         ENTRY;
 953
 954         /* already initialized? */
 955         if (lo->ldo_stripe != NULL)
 956                 GOTO(out, rc = 0);
 957
 958         if (!dt_object_exists(next))
 959                 GOTO(out, rc = 0);
 960
 961         /* Do not load stripe for slaves of striped dir */
 962         if (lo->ldo_dir_slave_stripe)
 963                 GOTO(out, rc = 0);
 964
 965         if (S_ISREG(lu_object_attr(lod2lu_obj(lo)))) {
 966                 rc = lod_get_lov_ea(env, lo);
 967                 if (rc <= 0)
 968                         GOTO(out, rc);
 969                 /*
 970                  * there is LOV EA (striping information) in this object
 971                  * let's parse it and create in-core objects for the stripes
 972                  */
 973                 buf->lb_buf = info->lti_ea_store;
 974                 buf->lb_len = info->lti_ea_store_size;
 975                 rc = lod_parse_striping(env, lo, buf);
 976         } else if (S_ISDIR(lu_object_attr(lod2lu_obj(lo)))) {
 977                 rc = lod_get_lmv_ea(env, lo);
 978                 if (rc < (typeof(rc))sizeof(struct lmv_mds_md_v1))
 979                         GOTO(out, rc = rc > 0 ? -EINVAL : rc);
 980
 981                 buf->lb_buf = info->lti_ea_store;
 982                 buf->lb_len = info->lti_ea_store_size;
 983                 if (rc == sizeof(struct lmv_mds_md_v1)) {
 984                         rc = lod_load_lmv_shards(env, lo, buf, true);
 985                         if (buf->lb_buf != info->lti_ea_store) {
 986                                 OBD_FREE_LARGE(info->lti_ea_store,
 987                                                info->lti_ea_store_size);
 988                                 info->lti_ea_store = buf->lb_buf;
 989                                 info->lti_ea_store_size = buf->lb_len;
 990                         }
 991
 992                         if (rc < 0)
 993                                 GOTO(out, rc);
 994                 }
 995
 996                 /*
 997                  * there is LOV EA (striping information) in this object
 998                  * let's parse it and create in-core objects for the stripes
 999                  */
1000                 rc = lod_parse_dir_striping(env, lo, buf);
1001         }
1002
1003         if (rc == 0)
1004                 lo->ldo_striping_cached = 1;
1005 out:
1006         RETURN(rc);
1007 }
1008
1009 /**
1010  * A generic function to initialize the stripe objects.
1011  *
1012  * A protected version of lod_load_striping_locked() - load the striping
1013  * information from storage, parse that and instantiate LU objects to
1014  * represent the stripes.  The LOD object \a lo supplies a pointer to the
1015  * next sub-object in the LU stack so we can lock it. Also use \a lo to
1016  * return an array of references to the newly instantiated objects.
1017  *
1018  * \param[in] env               execution environment for this thread
1019  * \param[in,out] lo            LOD object, where striping is stored and
1020  *                              which gets an array of references
1021  *
1022  * \retval                      0 if parsing and object creation succeed
1023  * \retval                      negative error number on failure
1024  **/
1025 int lod_load_striping(const struct lu_env *env, struct lod_object *lo)
1026 {
1027         struct dt_object        *next = dt_object_child(&lo->ldo_obj);
1028         int                     rc = 0;
1029
1030         /* currently this code is supposed to be called from declaration
1031          * phase only, thus the object is not expected to be locked by caller */
1032         dt_write_lock(env, next, 0);
1033         rc = lod_load_striping_locked(env, lo);
1034         dt_write_unlock(env, next);
1035         return rc;
1036 }
1037
1038 /**
1039  * Verify striping.
1040  *
1041  * Check the validity of all fields including the magic, stripe size,
1042  * stripe count, stripe offset and that the pool is present.  Also check
1043  * that each target index points to an existing target. The additional
1044  * \a is_from_disk turns additional checks. In some cases zero fields
1045  * are allowed (like pattern=0).
1046  *
1047  * \param[in] d                 LOD device
1048  * \param[in] buf               buffer with LOV EA to verify
1049  * \param[in] is_from_disk      0 - from user, allow some fields to be 0
1050  *                              1 - from disk, do not allow
1051  *
1052  * \retval                      0 if the striping is valid
1053  * \retval                      -EINVAL if striping is invalid
1054  */
1055 int lod_verify_striping(struct lod_device *d, const struct lu_buf *buf,
1056                         bool is_from_disk)
1057 {
1058         struct lov_user_md_v1   *lum;
1059         struct lov_user_md_v3   *lum3;
1060         struct pool_desc        *pool = NULL;
1061         __u32                    magic;
1062         __u32                    stripe_size;
1063         __u16                    stripe_count;
1064         __u16                    stripe_offset;
1065         size_t                   lum_size;
1066         int                      rc = 0;
1067         ENTRY;
1068
1069         lum = buf->lb_buf;
1070
1071         LASSERT(sizeof(*lum) < sizeof(*lum3));
1072
1073         if (buf->lb_len < sizeof(*lum)) {
1074                 CDEBUG(D_IOCTL, "buf len %zu too small for lov_user_md\n",
1075                        buf->lb_len);
1076                 GOTO(out, rc = -EINVAL);
1077         }
1078
1079         magic = le32_to_cpu(lum->lmm_magic);
1080         if (magic != LOV_USER_MAGIC_V1 &&
1081             magic != LOV_USER_MAGIC_V3 &&
1082             magic != LOV_MAGIC_V1_DEF &&
1083             magic != LOV_MAGIC_V3_DEF) {
1084                 CDEBUG(D_IOCTL, "bad userland LOV MAGIC: %#x\n", magic);
1085                 GOTO(out, rc = -EINVAL);
1086         }
1087
1088         /* the user uses "0" for default stripe pattern normally. */
1089         if (!is_from_disk && lum->lmm_pattern == 0)
1090                 lum->lmm_pattern = cpu_to_le32(LOV_PATTERN_RAID0);
1091
1092         if (le32_to_cpu(lum->lmm_pattern) != LOV_PATTERN_RAID0) {
1093                 CDEBUG(D_IOCTL, "bad userland stripe pattern: %#x\n",
1094                        le32_to_cpu(lum->lmm_pattern));
1095                 GOTO(out, rc = -EINVAL);
1096         }
1097
1098         /* 64kB is the largest common page size we see (ia64), and matches the
1099          * check in lfs */
1100         stripe_size = le32_to_cpu(lum->lmm_stripe_size);
1101         if (stripe_size & (LOV_MIN_STRIPE_SIZE - 1)) {
1102                 CDEBUG(D_IOCTL, "stripe size %u not a multiple of %u\n",
1103                        stripe_size, LOV_MIN_STRIPE_SIZE);
1104                 GOTO(out, rc = -EINVAL);
1105         }
1106
1107         stripe_offset = le16_to_cpu(lum->lmm_stripe_offset);
1108         if (stripe_offset != LOV_OFFSET_DEFAULT) {
1109                 /* if offset is not within valid range [0, osts_size) */
1110                 if (stripe_offset >= d->lod_osts_size) {
1111                         CDEBUG(D_IOCTL, "stripe offset %u >= bitmap size %u\n",
1112                                stripe_offset, d->lod_osts_size);
1113                         GOTO(out, rc = -EINVAL);
1114                 }
1115
1116                 /* if lmm_stripe_offset is *not* in bitmap */
1117                 if (!cfs_bitmap_check(d->lod_ost_bitmap, stripe_offset)) {
1118                         CDEBUG(D_IOCTL, "stripe offset %u not in bitmap\n",
1119                                stripe_offset);
1120                         GOTO(out, rc = -EINVAL);
1121                 }
1122         }
1123
1124         if (magic == LOV_USER_MAGIC_V1 || magic == LOV_MAGIC_V1_DEF)
1125                 lum_size = offsetof(struct lov_user_md_v1,
1126                                     lmm_objects[0]);
1127         else if (magic == LOV_USER_MAGIC_V3 || magic == LOV_MAGIC_V3_DEF)
1128                 lum_size = offsetof(struct lov_user_md_v3,
1129                                     lmm_objects[0]);
1130         else
1131                 GOTO(out, rc = -EINVAL);
1132
1133         stripe_count = le16_to_cpu(lum->lmm_stripe_count);
1134         if (buf->lb_len != lum_size) {
1135                 CDEBUG(D_IOCTL, "invalid buf len %zu for lov_user_md with "
1136                        "magic %#x and stripe_count %u\n",
1137                        buf->lb_len, magic, stripe_count);
1138                 GOTO(out, rc = -EINVAL);
1139         }
1140
1141         if (!(magic == LOV_USER_MAGIC_V3 || magic == LOV_MAGIC_V3_DEF))
1142                 goto out;
1143
1144         lum3 = buf->lb_buf;
1145         if (buf->lb_len < sizeof(*lum3)) {
1146                 CDEBUG(D_IOCTL, "buf len %zu too small for lov_user_md_v3\n",
1147                        buf->lb_len);
1148                 GOTO(out, rc = -EINVAL);
1149         }
1150
1151         /* In the function below, .hs_keycmp resolves to
1152          * pool_hashkey_keycmp() */
1153         /* coverity[overrun-buffer-val] */
1154         pool = lod_find_pool(d, lum3->lmm_pool_name);
1155         if (pool == NULL)
1156                 goto out;
1157
1158         if (stripe_offset != LOV_OFFSET_DEFAULT) {
1159                 rc = lod_check_index_in_pool(stripe_offset, pool);
1160                 if (rc < 0)
1161                         GOTO(out, rc = -EINVAL);
1162         }
1163
1164         if (is_from_disk && stripe_count > pool_tgt_count(pool)) {
1165                 CDEBUG(D_IOCTL,
1166                        "stripe count %u > # OSTs %u in the pool\n",
1167                        stripe_count, pool_tgt_count(pool));
1168                 GOTO(out, rc = -EINVAL);
1169         }
1170
1171 out:
1172         if (pool != NULL)
1173                 lod_pool_putref(pool);
1174
1175         RETURN(rc);
1176 }
1177
1178 void lod_fix_desc_stripe_size(__u64 *val)
1179 {
1180         if (*val < LOV_MIN_STRIPE_SIZE) {
1181                 if (*val != 0)
1182                         LCONSOLE_INFO("Increasing default stripe size to "
1183                                       "minimum value %u\n",
1184                                       LOV_DESC_STRIPE_SIZE_DEFAULT);
1185                 *val = LOV_DESC_STRIPE_SIZE_DEFAULT;
1186         } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) {
1187                 *val &= ~(LOV_MIN_STRIPE_SIZE - 1);
1188                 LCONSOLE_WARN("Changing default stripe size to "LPU64" (a "
1189                               "multiple of %u)\n",
1190                               *val, LOV_MIN_STRIPE_SIZE);
1191         }
1192 }
1193
1194 void lod_fix_desc_stripe_count(__u32 *val)
1195 {
1196         if (*val == 0)
1197                 *val = 1;
1198 }
1199
1200 void lod_fix_desc_pattern(__u32 *val)
1201 {
1202         /* from lov_setstripe */
1203         if ((*val != 0) && (*val != LOV_PATTERN_RAID0)) {
1204                 LCONSOLE_WARN("Unknown stripe pattern: %#x\n", *val);
1205                 *val = 0;
1206         }
1207 }
1208
1209 void lod_fix_desc_qos_maxage(__u32 *val)
1210 {
1211         /* fix qos_maxage */
1212         if (*val == 0)
1213                 *val = LOV_DESC_QOS_MAXAGE_DEFAULT;
1214 }
1215
1216 /**
1217  * Used to fix insane default striping.
1218  *
1219  * \param[in] desc      striping description
1220  */
1221 void lod_fix_desc(struct lov_desc *desc)
1222 {
1223         lod_fix_desc_stripe_size(&desc->ld_default_stripe_size);
1224         lod_fix_desc_stripe_count(&desc->ld_default_stripe_count);
1225         lod_fix_desc_pattern(&desc->ld_pattern);
1226         lod_fix_desc_qos_maxage(&desc->ld_qos_maxage);
1227 }
1228
1229 /**
1230  * Initialize the structures used to store pools and default striping.
1231  *
1232  * \param[in] lod       LOD device
1233  * \param[in] lcfg      configuration structure storing default striping.
1234  *
1235  * \retval              0 if initialization succeeds
1236  * \retval              negative error number on failure
1237  */
1238 int lod_pools_init(struct lod_device *lod, struct lustre_cfg *lcfg)
1239 {
1240         struct obd_device          *obd;
1241         struct lov_desc            *desc;
1242         int                         rc;
1243         ENTRY;
1244
1245         obd = class_name2obd(lustre_cfg_string(lcfg, 0));
1246         LASSERT(obd != NULL);
1247         obd->obd_lu_dev = &lod->lod_dt_dev.dd_lu_dev;
1248
1249         if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
1250                 CERROR("LOD setup requires a descriptor\n");
1251                 RETURN(-EINVAL);
1252         }
1253
1254         desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
1255
1256         if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
1257                 CERROR("descriptor size wrong: %d > %d\n",
1258                        (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
1259                 RETURN(-EINVAL);
1260         }
1261
1262         if (desc->ld_magic != LOV_DESC_MAGIC) {
1263                 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
1264                         CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
1265                                obd->obd_name, desc);
1266                         lustre_swab_lov_desc(desc);
1267                 } else {
1268                         CERROR("%s: Bad lov desc magic: %#x\n",
1269                                obd->obd_name, desc->ld_magic);
1270                         RETURN(-EINVAL);
1271                 }
1272         }
1273
1274         lod_fix_desc(desc);
1275
1276         desc->ld_active_tgt_count = 0;
1277         lod->lod_desc = *desc;
1278
1279         lod->lod_sp_me = LUSTRE_SP_CLI;
1280
1281         /* Set up allocation policy (QoS and RR) */
1282         INIT_LIST_HEAD(&lod->lod_qos.lq_oss_list);
1283         init_rwsem(&lod->lod_qos.lq_rw_sem);
1284         lod->lod_qos.lq_dirty = 1;
1285         lod->lod_qos.lq_rr.lqr_dirty = 1;
1286         lod->lod_qos.lq_reset = 1;
1287         /* Default priority is toward free space balance */
1288         lod->lod_qos.lq_prio_free = 232;
1289         /* Default threshold for rr (roughly 17%) */
1290         lod->lod_qos.lq_threshold_rr = 43;
1291
1292         /* Set up OST pool environment */
1293         lod->lod_pools_hash_body = cfs_hash_create("POOLS", HASH_POOLS_CUR_BITS,
1294                                                    HASH_POOLS_MAX_BITS,
1295                                                    HASH_POOLS_BKT_BITS, 0,
1296                                                    CFS_HASH_MIN_THETA,
1297                                                    CFS_HASH_MAX_THETA,
1298                                                    &pool_hash_operations,
1299                                                    CFS_HASH_DEFAULT);
1300         if (lod->lod_pools_hash_body == NULL)
1301                 RETURN(-ENOMEM);
1302
1303         INIT_LIST_HEAD(&lod->lod_pool_list);
1304         lod->lod_pool_count = 0;
1305         rc = lod_ost_pool_init(&lod->lod_pool_info, 0);
1306         if (rc)
1307                 GOTO(out_hash, rc);
1308         rc = lod_ost_pool_init(&lod->lod_qos.lq_rr.lqr_pool, 0);
1309         if (rc)
1310                 GOTO(out_pool_info, rc);
1311
1312         RETURN(0);
1313
1314 out_pool_info:
1315         lod_ost_pool_free(&lod->lod_pool_info);
1316 out_hash:
1317         cfs_hash_putref(lod->lod_pools_hash_body);
1318
1319         return rc;
1320 }
1321
1322 /**
1323  * Release the structures describing the pools.
1324  *
1325  * \param[in] lod       LOD device from which we release the structures
1326  *
1327  * \retval              0 always
1328  */
1329 int lod_pools_fini(struct lod_device *lod)
1330 {
1331         struct obd_device   *obd = lod2obd(lod);
1332         struct pool_desc    *pool, *tmp;
1333         ENTRY;
1334
1335         list_for_each_entry_safe(pool, tmp, &lod->lod_pool_list, pool_list) {
1336                 /* free pool structs */
1337                 CDEBUG(D_INFO, "delete pool %p\n", pool);
1338                 /* In the function below, .hs_keycmp resolves to
1339                  * pool_hashkey_keycmp() */
1340                 /* coverity[overrun-buffer-val] */
1341                 lod_pool_del(obd, pool->pool_name);
1342         }
1343
1344         cfs_hash_putref(lod->lod_pools_hash_body);
1345         lod_ost_pool_free(&(lod->lod_qos.lq_rr.lqr_pool));
1346         lod_ost_pool_free(&lod->lod_pool_info);
1347
1348         RETURN(0);
1349 }