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_OSS_CAPA |
 257                                            OBD_CONNECT_REQPORTAL |
 258                                            OBD_CONNECT_SKIP_ORPHAN |
 259                                            OBD_CONNECT_FID |
 260                                            OBD_CONNECT_LVB_TYPE |
 261                                            OBD_CONNECT_VERSION |
 262                                            OBD_CONNECT_PINGLESS |
 263                                            OBD_CONNECT_LFSCK;
 264
 265                 data->ocd_group = tgt_index;
 266                 ltd = &lod->lod_ost_descs;
 267         } else {
 268                 struct obd_import *imp = obd->u.cli.cl_import;
 269
 270                 for_ost = false;
 271                 data->ocd_ibits_known = MDS_INODELOCK_UPDATE;
 272                 data->ocd_connect_flags |= OBD_CONNECT_ACL |
 273                                            OBD_CONNECT_MDS_CAPA |
 274                                            OBD_CONNECT_OSS_CAPA |
 275                                            OBD_CONNECT_IBITS |
 276                                            OBD_CONNECT_MDS_MDS |
 277                                            OBD_CONNECT_FID |
 278                                            OBD_CONNECT_AT |
 279                                            OBD_CONNECT_FULL20 |
 280                                            OBD_CONNECT_LFSCK;
 281                 /* XXX set MDS-MDS flags, remove this when running this
 282                  * on client*/
 283                 data->ocd_connect_flags |= OBD_CONNECT_MDS_MDS;
 284                 spin_lock(&imp->imp_lock);
 285                 imp->imp_server_timeout = 1;
 286                 spin_unlock(&imp->imp_lock);
 287                 imp->imp_client->cli_request_portal = OUT_PORTAL;
 288                 CDEBUG(D_OTHER, "%s: Set 'mds' portal and timeout\n",
 289                       obd->obd_name);
 290                 ltd = &lod->lod_mdt_descs;
 291         }
 292
 293         rc = obd_connect(env, &exp, obd, &obd->obd_uuid, data, NULL);
 294         OBD_FREE_PTR(data);
 295         if (rc) {
 296                 CERROR("%s: cannot connect to next dev %s (%d)\n",
 297                        obd->obd_name, osp, rc);
 298                 GOTO(out_free, rc);
 299         }
 300
 301         LASSERT(obd->obd_lu_dev);
 302         LASSERT(obd->obd_lu_dev->ld_site == lod->lod_dt_dev.dd_lu_dev.ld_site);
 303
 304         ldev = obd->obd_lu_dev;
 305         d = lu2dt_dev(ldev);
 306
 307         /* Allocate ost descriptor and fill it */
 308         OBD_ALLOC_PTR(tgt_desc);
 309         if (!tgt_desc)
 310                 GOTO(out_conn, rc = -ENOMEM);
 311
 312         tgt_desc->ltd_tgt    = d;
 313         tgt_desc->ltd_exp    = exp;
 314         tgt_desc->ltd_uuid   = obd->u.cli.cl_target_uuid;
 315         tgt_desc->ltd_gen    = gen;
 316         tgt_desc->ltd_index  = index;
 317         tgt_desc->ltd_active = active;
 318
 319         lod_getref(ltd);
 320         if (index >= ltd->ltd_tgts_size) {
 321                 /* we have to increase the size of the lod_osts array */
 322                 __u32  newsize;
 323
 324                 newsize = max(ltd->ltd_tgts_size, (__u32)2);
 325                 while (newsize < index + 1)
 326                         newsize = newsize << 1;
 327
 328                 /* lod_bitmap_resize() needs lod_rw_sem
 329                  * which we hold with th reference */
 330                 lod_putref(lod, ltd);
 331
 332                 rc = ltd_bitmap_resize(ltd, newsize);
 333                 if (rc)
 334                         GOTO(out_desc, rc);
 335
 336                 lod_getref(ltd);
 337         }
 338
 339         mutex_lock(&ltd->ltd_mutex);
 340         if (cfs_bitmap_check(ltd->ltd_tgt_bitmap, index)) {
 341                 CERROR("%s: device %d is registered already\n", obd->obd_name,
 342                        index);
 343                 GOTO(out_mutex, rc = -EEXIST);
 344         }
 345
 346         if (ltd->ltd_tgt_idx[index / TGT_PTRS_PER_BLOCK] == NULL) {
 347                 OBD_ALLOC_PTR(ltd->ltd_tgt_idx[index / TGT_PTRS_PER_BLOCK]);
 348                 if (ltd->ltd_tgt_idx[index / TGT_PTRS_PER_BLOCK] == NULL) {
 349                         CERROR("can't allocate index to add %s\n",
 350                                obd->obd_name);
 351                         GOTO(out_mutex, rc = -ENOMEM);
 352                 }
 353         }
 354
 355         if (!strcmp(LUSTRE_OSC_NAME, type)) {
 356                 /* pool and qos are not supported for MDS stack yet */
 357                 rc = lod_ost_pool_add(&lod->lod_pool_info, index,
 358                                       lod->lod_osts_size);
 359                 if (rc) {
 360                         CERROR("%s: can't set up pool, failed with %d\n",
 361                                obd->obd_name, rc);
 362                         GOTO(out_mutex, rc);
 363                 }
 364
 365                 rc = qos_add_tgt(lod, tgt_desc);
 366                 if (rc) {
 367                         CERROR("%s: qos_add_tgt failed with %d\n",
 368                                 obd->obd_name, rc);
 369                         GOTO(out_pool, rc);
 370                 }
 371
 372                 /* The new OST is now a full citizen */
 373                 if (index >= lod->lod_desc.ld_tgt_count)
 374                         lod->lod_desc.ld_tgt_count = index + 1;
 375                 if (active)
 376                         lod->lod_desc.ld_active_tgt_count++;
 377         }
 378
 379         LTD_TGT(ltd, index) = tgt_desc;
 380         cfs_bitmap_set(ltd->ltd_tgt_bitmap, index);
 381         ltd->ltd_tgtnr++;
 382         mutex_unlock(&ltd->ltd_mutex);
 383         lod_putref(lod, ltd);
 384         if (lod->lod_recovery_completed)
 385                 ldev->ld_ops->ldo_recovery_complete(env, ldev);
 386
 387         rc = lfsck_add_target(env, lod->lod_child, d, exp, index, for_ost);
 388         if (rc != 0)
 389                 CERROR("Fail to add LFSCK target: name = %s, type = %s, "
 390                        "index = %u, rc = %d\n", osp, type, index, rc);
 391
 392         RETURN(rc);
 393
 394 out_pool:
 395         lod_ost_pool_remove(&lod->lod_pool_info, index);
 396 out_mutex:
 397         mutex_unlock(&ltd->ltd_mutex);
 398         lod_putref(lod, ltd);
 399 out_desc:
 400         OBD_FREE_PTR(tgt_desc);
 401 out_conn:
 402         obd_disconnect(exp);
 403 out_free:
 404         return rc;
 405 }
 406
 407 /**
 408  * Schedule target removal from the target table.
 409  *
 410  * Mark the device as dead. The device is not removed here because it may
 411  * still be in use. The device will be removed in lod_putref() when the
 412  * last reference is released.
 413  *
 414  * \param[in] env               execution environment for this thread
 415  * \param[in] lod               LOD device the target table belongs to
 416  * \param[in] ltd               target table
 417  * \param[in] idx               index of the target
 418  * \param[in] for_ost           type of the target: 0 - MDT, 1 - OST
 419  */
 420 static void __lod_del_device(const struct lu_env *env, struct lod_device *lod,
 421                              struct lod_tgt_descs *ltd, unsigned idx,
 422                              bool for_ost)
 423 {
 424         LASSERT(LTD_TGT(ltd, idx));
 425
 426         lfsck_del_target(env, lod->lod_child, LTD_TGT(ltd, idx)->ltd_tgt,
 427                          idx, for_ost);
 428
 429         if (LTD_TGT(ltd, idx)->ltd_reap == 0) {
 430                 LTD_TGT(ltd, idx)->ltd_reap = 1;
 431                 ltd->ltd_death_row++;
 432         }
 433 }
 434
 435 /**
 436  * Schedule removal of all the targets from the given target table.
 437  *
 438  * See more details in the description for __lod_del_device()
 439  *
 440  * \param[in] env               execution environment for this thread
 441  * \param[in] lod               LOD device the target table belongs to
 442  * \param[in] ltd               target table
 443  * \param[in] for_ost           type of the target: MDT or OST
 444  *
 445  * \retval                      0 always
 446  */
 447 int lod_fini_tgt(const struct lu_env *env, struct lod_device *lod,
 448                  struct lod_tgt_descs *ltd, bool for_ost)
 449 {
 450         unsigned int idx;
 451
 452         if (ltd->ltd_tgts_size <= 0)
 453                 return 0;
 454         lod_getref(ltd);
 455         mutex_lock(&ltd->ltd_mutex);
 456         cfs_foreach_bit(ltd->ltd_tgt_bitmap, idx)
 457                 __lod_del_device(env, lod, ltd, idx, for_ost);
 458         mutex_unlock(&ltd->ltd_mutex);
 459         lod_putref(lod, ltd);
 460         CFS_FREE_BITMAP(ltd->ltd_tgt_bitmap);
 461         for (idx = 0; idx < TGT_PTRS; idx++) {
 462                 if (ltd->ltd_tgt_idx[idx])
 463                         OBD_FREE_PTR(ltd->ltd_tgt_idx[idx]);
 464         }
 465         ltd->ltd_tgts_size = 0;
 466         return 0;
 467 }
 468
 469 /**
 470  * Remove device by name.
 471  *
 472  * Remove a device identified by \a osp from the target table. Given
 473  * the device can be in use, the real deletion happens in lod_putref().
 474  *
 475  * \param[in] env               execution environment for this thread
 476  * \param[in] lod               LOD device to be connected to the new OSP
 477  * \param[in] ltd               target table
 478  * \param[in] osp               name of OSP device to be removed
 479  * \param[in] idx               index of the target
 480  * \param[in] gen               generation number, not used currently
 481  * \param[in] for_ost           type of the target: 0 - MDT, 1 - OST
 482  *
 483  * \retval                      0 if the device was scheduled for removal
 484  * \retval                      -EINVAL if no device was found
 485  */
 486 int lod_del_device(const struct lu_env *env, struct lod_device *lod,
 487                    struct lod_tgt_descs *ltd, char *osp, unsigned idx,
 488                    unsigned gen, bool for_ost)
 489 {
 490         struct obd_device *obd;
 491         int                rc = 0;
 492         struct obd_uuid    uuid;
 493         ENTRY;
 494
 495         CDEBUG(D_CONFIG, "osp:%s idx:%d gen:%d\n", osp, idx, gen);
 496
 497         obd_str2uuid(&uuid, osp);
 498
 499         obd = class_find_client_obd(&uuid, LUSTRE_OSP_NAME,
 500                                    &lod->lod_dt_dev.dd_lu_dev.ld_obd->obd_uuid);
 501         if (obd == NULL) {
 502                 CERROR("can't find %s device\n", osp);
 503                 RETURN(-EINVAL);
 504         }
 505
 506         if (gen <= 0) {
 507                 CERROR("%s: request to remove OBD %s with invalid generation %d"
 508                        "\n", obd->obd_name, osp, gen);
 509                 RETURN(-EINVAL);
 510         }
 511
 512         obd_str2uuid(&uuid,  osp);
 513
 514         lod_getref(ltd);
 515         mutex_lock(&ltd->ltd_mutex);
 516         /* check that the index is allocated in the bitmap */
 517         if (!cfs_bitmap_check(ltd->ltd_tgt_bitmap, idx) ||
 518             !LTD_TGT(ltd, idx)) {
 519                 CERROR("%s: device %d is not set up\n", obd->obd_name, idx);
 520                 GOTO(out, rc = -EINVAL);
 521         }
 522
 523         /* check that the UUID matches */
 524         if (!obd_uuid_equals(&uuid, &LTD_TGT(ltd, idx)->ltd_uuid)) {
 525                 CERROR("%s: LOD target UUID %s at index %d does not match %s\n",
 526                        obd->obd_name, obd_uuid2str(&LTD_TGT(ltd,idx)->ltd_uuid),
 527                        idx, osp);
 528                 GOTO(out, rc = -EINVAL);
 529         }
 530
 531         __lod_del_device(env, lod, ltd, idx, for_ost);
 532         EXIT;
 533 out:
 534         mutex_unlock(&ltd->ltd_mutex);
 535         lod_putref(lod, ltd);
 536         return(rc);
 537 }
 538
 539 /**
 540  * Resize per-thread storage to hold specified size.
 541  *
 542  * A helper function to resize per-thread temporary storage. This storage
 543  * is used to process LOV/LVM EAs and may be quite large. We do not want to
 544  * allocate/release it every time, so instead we put it into the env and
 545  * reallocate on demand. The memory is released when the correspondent thread
 546  * is finished.
 547  *
 548  * \param[in] info              LOD-specific storage in the environment
 549  * \param[in] size              new size to grow the buffer to
 550
 551  * \retval                      0 on success, -ENOMEM if reallocation failed
 552  */
 553 int lod_ea_store_resize(struct lod_thread_info *info, size_t size)
 554 {
 555         __u32 round = size_roundup_power2(size);
 556
 557         LASSERT(round <=
 558                 lov_mds_md_size(LOV_MAX_STRIPE_COUNT, LOV_MAGIC_V3));
 559         if (info->lti_ea_store) {
 560                 LASSERT(info->lti_ea_store_size);
 561                 LASSERT(info->lti_ea_store_size < round);
 562                 CDEBUG(D_INFO, "EA store size %d is not enough, need %d\n",
 563                        info->lti_ea_store_size, round);
 564                 OBD_FREE_LARGE(info->lti_ea_store, info->lti_ea_store_size);
 565                 info->lti_ea_store = NULL;
 566                 info->lti_ea_store_size = 0;
 567         }
 568
 569         OBD_ALLOC_LARGE(info->lti_ea_store, round);
 570         if (info->lti_ea_store == NULL)
 571                 RETURN(-ENOMEM);
 572         info->lti_ea_store_size = round;
 573         RETURN(0);
 574 }
 575
 576 /**
 577  * Make LOV EA for striped object.
 578  *
 579  * Generate striping information and store it in the LOV EA of the given
 580  * object. The caller must ensure nobody else is calling the function
 581  * against the object concurrently. The transaction must be started.
 582  * FLDB service must be running as well; it's used to map FID to the target,
 583  * which is stored in LOV EA.
 584  *
 585  * \param[in] env               execution environment for this thread
 586  * \param[in] lo                LOD object
 587  * \param[in] th                transaction handle
 588  *
 589  * \retval                      0 if LOV EA is stored successfully
 590  * \retval                      negative error number on failure
 591  */
 592 int lod_generate_and_set_lovea(const struct lu_env *env,
 593                                struct lod_object *lo, struct thandle *th)
 594 {
 595         struct lod_thread_info  *info = lod_env_info(env);
 596         struct dt_object        *next = dt_object_child(&lo->ldo_obj);
 597         const struct lu_fid     *fid  = lu_object_fid(&lo->ldo_obj.do_lu);
 598         struct lov_mds_md_v1    *lmm;
 599         struct lov_ost_data_v1  *objs;
 600         __u32                    magic;
 601         int                      i, rc;
 602         size_t                   lmm_size;
 603         ENTRY;
 604
 605         LASSERT(lo);
 606
 607         magic = lo->ldo_pool != NULL ? LOV_MAGIC_V3 : LOV_MAGIC_V1;
 608         lmm_size = lov_mds_md_size(lo->ldo_stripenr, magic);
 609         if (info->lti_ea_store_size < lmm_size) {
 610                 rc = lod_ea_store_resize(info, lmm_size);
 611                 if (rc)
 612                         RETURN(rc);
 613         }
 614
 615         if (lo->ldo_pattern == 0) /* default striping */
 616                 lo->ldo_pattern = LOV_PATTERN_RAID0;
 617
 618         lmm = info->lti_ea_store;
 619
 620         lmm->lmm_magic = cpu_to_le32(magic);
 621         lmm->lmm_pattern = cpu_to_le32(lo->ldo_pattern);
 622         fid_to_lmm_oi(fid, &lmm->lmm_oi);
 623         if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_LMMOI))
 624                 lmm->lmm_oi.oi.oi_id++;
 625         lmm_oi_cpu_to_le(&lmm->lmm_oi, &lmm->lmm_oi);
 626         lmm->lmm_stripe_size = cpu_to_le32(lo->ldo_stripe_size);
 627         lmm->lmm_stripe_count = cpu_to_le16(lo->ldo_stripenr);
 628         if (lo->ldo_pattern & LOV_PATTERN_F_RELEASED)
 629                 lmm->lmm_stripe_count = cpu_to_le16(lo->ldo_released_stripenr);
 630         lmm->lmm_layout_gen = 0;
 631         if (magic == LOV_MAGIC_V1) {
 632                 objs = &lmm->lmm_objects[0];
 633         } else {
 634                 struct lov_mds_md_v3 *v3 = (struct lov_mds_md_v3 *) lmm;
 635                 size_t cplen = strlcpy(v3->lmm_pool_name, lo->ldo_pool,
 636                                 sizeof(v3->lmm_pool_name));
 637                 if (cplen >= sizeof(v3->lmm_pool_name))
 638                         RETURN(-E2BIG);
 639                 objs = &v3->lmm_objects[0];
 640         }
 641
 642         for (i = 0; i < lo->ldo_stripenr; i++) {
 643                 struct lu_fid           *fid    = &info->lti_fid;
 644                 struct lod_device       *lod;
 645                 __u32                   index;
 646                 int                     type    = LU_SEQ_RANGE_OST;
 647
 648                 lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
 649                 LASSERT(lo->ldo_stripe[i]);
 650
 651                 *fid = *lu_object_fid(&lo->ldo_stripe[i]->do_lu);
 652                 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_MULTIPLE_REF)) {
 653                         if (cfs_fail_val == 0)
 654                                 cfs_fail_val = fid->f_oid;
 655                         else
 656                                 fid->f_oid = cfs_fail_val;
 657                 }
 658
 659                 rc = fid_to_ostid(fid, &info->lti_ostid);
 660                 LASSERT(rc == 0);
 661
 662                 ostid_cpu_to_le(&info->lti_ostid, &objs[i].l_ost_oi);
 663                 objs[i].l_ost_gen    = cpu_to_le32(0);
 664                 rc = lod_fld_lookup(env, lod, fid, &index, &type);
 665                 if (rc < 0) {
 666                         CERROR("%s: Can not locate "DFID": rc = %d\n",
 667                                lod2obd(lod)->obd_name, PFID(fid), rc);
 668                         lod_object_free_striping(env, lo);
 669                         RETURN(rc);
 670                 }
 671                 objs[i].l_ost_idx = cpu_to_le32(index);
 672         }
 673
 674         info->lti_buf.lb_buf = lmm;
 675         info->lti_buf.lb_len = lmm_size;
 676         rc = dt_xattr_set(env, next, &info->lti_buf, XATTR_NAME_LOV, 0,
 677                           th, BYPASS_CAPA);
 678         if (rc < 0)
 679                 lod_object_free_striping(env, lo);
 680
 681         RETURN(rc);
 682 }
 683
 684 /**
 685  * Get LOV EA.
 686  *
 687  * Fill lti_ea_store buffer in the environment with a value for the given
 688  * EA. The buffer is reallocated if the value doesn't fit.
 689  *
 690  * \param[in,out] env           execution environment for this thread
 691  *                              .lti_ea_store buffer is filled with EA's value
 692  * \param[in] lo                LOD object
 693  * \param[in] name              name of the EA
 694  *
 695  * \retval                      0 if EA is fetched successfully
 696  * \retval                      negative error number on failure
 697  */
 698 int lod_get_ea(const struct lu_env *env, struct lod_object *lo,
 699                const char *name)
 700 {
 701         struct lod_thread_info  *info = lod_env_info(env);
 702         struct dt_object        *next = dt_object_child(&lo->ldo_obj);
 703         int                     rc;
 704         ENTRY;
 705
 706         LASSERT(info);
 707
 708         if (unlikely(info->lti_ea_store == NULL)) {
 709                 /* just to enter in allocation block below */
 710                 rc = -ERANGE;
 711         } else {
 712 repeat:
 713                 info->lti_buf.lb_buf = info->lti_ea_store;
 714                 info->lti_buf.lb_len = info->lti_ea_store_size;
 715                 rc = dt_xattr_get(env, next, &info->lti_buf, name, BYPASS_CAPA);
 716         }
 717
 718         /* if object is not striped or inaccessible */
 719         if (rc == -ENODATA || rc == -ENOENT)
 720                 RETURN(0);
 721
 722         if (rc == -ERANGE) {
 723                 /* EA doesn't fit, reallocate new buffer */
 724                 rc = dt_xattr_get(env, next, &LU_BUF_NULL, name,
 725                                   BYPASS_CAPA);
 726                 if (rc == -ENODATA || rc == -ENOENT)
 727                         RETURN(0);
 728                 else if (rc < 0)
 729                         RETURN(rc);
 730
 731                 LASSERT(rc > 0);
 732                 rc = lod_ea_store_resize(info, rc);
 733                 if (rc)
 734                         RETURN(rc);
 735                 goto repeat;
 736         }
 737
 738         RETURN(rc);
 739 }
 740
 741 /**
 742  * Store default striping.
 743  *
 744  * Store default striping for the files in the given directory. The data
 745  * are stored in the LOD-object representing the directory (ldo_def_* fields).
 746  * If default striping matches virtual fs-wide default striping, then we
 747  * store nothing. This mean that the files in the directory will be created
 748  * with filesystem-wide default striping. The transaction must be started.
 749  *
 750  * \param[in] env               execution environment for this thread
 751  * \param[in] dt                dt object representing directory in LOD layer
 752  * \param[in] th                transaction handle
 753  *
 754  * \retval                      0 if stored successfully or no need to store
 755  * \retval                      negative error number on failure
 756  */
 757 int lod_store_def_striping(const struct lu_env *env, struct dt_object *dt,
 758                            struct thandle *th)
 759 {
 760         struct lod_thread_info  *info = lod_env_info(env);
 761         struct lod_object       *lo = lod_dt_obj(dt);
 762         struct dt_object        *next = dt_object_child(dt);
 763         struct lov_user_md_v3   *v3;
 764         int                      rc;
 765         ENTRY;
 766
 767         if (S_ISDIR(dt->do_lu.lo_header->loh_attr))
 768                 RETURN(-ENOTDIR);
 769         /*
 770          * store striping defaults into new directory
 771          * used to implement defaults inheritance
 772          */
 773
 774         /* probably nothing to inherite */
 775         if (lo->ldo_striping_cached == 0)
 776                 RETURN(0);
 777
 778         if (LOVEA_DELETE_VALUES(lo->ldo_def_stripe_size, lo->ldo_def_stripenr,
 779                                 lo->ldo_def_stripe_offset))
 780                 RETURN(0);
 781
 782         v3 = info->lti_ea_store;
 783         if (info->lti_ea_store_size < sizeof(*v3)) {
 784                 rc = lod_ea_store_resize(info, sizeof(*v3));
 785                 if (rc != 0)
 786                         RETURN(rc);
 787                 v3 = info->lti_ea_store;
 788         }
 789         memset(v3, 0, sizeof(*v3));
 790         v3->lmm_magic = cpu_to_le32(LOV_USER_MAGIC_V3);
 791         v3->lmm_stripe_count = cpu_to_le16(lo->ldo_def_stripenr);
 792         v3->lmm_stripe_offset = cpu_to_le16(lo->ldo_def_stripe_offset);
 793         v3->lmm_stripe_size = cpu_to_le32(lo->ldo_def_stripe_size);
 794         if (lo->ldo_pool != NULL)
 795                 strlcpy(v3->lmm_pool_name, lo->ldo_pool,
 796                         sizeof(v3->lmm_pool_name));
 797         info->lti_buf.lb_buf = v3;
 798         info->lti_buf.lb_len = sizeof(*v3);
 799         rc = dt_xattr_set(env, next, &info->lti_buf, XATTR_NAME_LOV, 0, th,
 800                         BYPASS_CAPA);
 801
 802         RETURN(rc);
 803 }
 804
 805 /**
 806  * Verify the target index is present in the current configuration.
 807  *
 808  * \param[in] md                LOD device where the target table is stored
 809  * \param[in] idx               target's index
 810  *
 811  * \retval                      0 if the index is present
 812  * \retval                      -EINVAL if not
 813  */
 814 static int validate_lod_and_idx(struct lod_device *md, __u32 idx)
 815 {
 816         if (unlikely(idx >= md->lod_ost_descs.ltd_tgts_size ||
 817                      !cfs_bitmap_check(md->lod_ost_bitmap, idx))) {
 818                 CERROR("%s: bad idx: %d of %d\n", lod2obd(md)->obd_name, idx,
 819                        md->lod_ost_descs.ltd_tgts_size);
 820                 return -EINVAL;
 821         }
 822
 823         if (unlikely(OST_TGT(md, idx) == NULL)) {
 824                 CERROR("%s: bad lod_tgt_desc for idx: %d\n",
 825                        lod2obd(md)->obd_name, idx);
 826                 return -EINVAL;
 827         }
 828
 829         if (unlikely(OST_TGT(md, idx)->ltd_ost == NULL)) {
 830                 CERROR("%s: invalid lod device, for idx: %d\n",
 831                        lod2obd(md)->obd_name , idx);
 832                 return -EINVAL;
 833         }
 834
 835         return 0;
 836 }
 837
 838 /**
 839  * Instantiate objects for stripes.
 840  *
 841  * Allocate and initialize LU-objects representing the stripes. The number
 842  * of the stripes (ldo_stripenr) must be initialized already. The caller
 843  * must ensure nobody else is calling the function on the object at the same
 844  * time. FLDB service must be running to be able to map a FID to the targets
 845  * and find appropriate device representing that target.
 846  *
 847  * \param[in] env               execution environment for this thread
 848  * \param[in,out] lo            LOD object
 849  * \param[in] objs              an array of IDs to creates the objects from
 850  *
 851  * \retval                      0 if the objects are instantiated successfully
 852  * \retval                      negative error number on failure
 853  */
 854 int lod_initialize_objects(const struct lu_env *env, struct lod_object *lo,
 855                            struct lov_ost_data_v1 *objs)
 856 {
 857         struct lod_thread_info  *info = lod_env_info(env);
 858         struct lod_device       *md;
 859         struct lu_object        *o, *n;
 860         struct lu_device        *nd;
 861         struct dt_object       **stripe;
 862         int                      stripe_len;
 863         int                      i, rc = 0;
 864         __u32                   idx;
 865         ENTRY;
 866
 867         LASSERT(lo != NULL);
 868         md = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
 869         LASSERT(lo->ldo_stripe == NULL);
 870         LASSERT(lo->ldo_stripenr > 0);
 871         LASSERT(lo->ldo_stripe_size > 0);
 872
 873         stripe_len = lo->ldo_stripenr;
 874         OBD_ALLOC(stripe, sizeof(stripe[0]) * stripe_len);
 875         if (stripe == NULL)
 876                 RETURN(-ENOMEM);
 877
 878         for (i = 0; i < lo->ldo_stripenr; i++) {
 879                 if (unlikely(lovea_slot_is_dummy(&objs[i])))
 880                         continue;
 881
 882                 ostid_le_to_cpu(&objs[i].l_ost_oi, &info->lti_ostid);
 883                 idx = le32_to_cpu(objs[i].l_ost_idx);
 884                 rc = ostid_to_fid(&info->lti_fid, &info->lti_ostid, idx);
 885                 if (rc != 0)
 886                         GOTO(out, rc);
 887                 LASSERTF(fid_is_sane(&info->lti_fid), ""DFID" insane!\n",
 888                          PFID(&info->lti_fid));
 889                 lod_getref(&md->lod_ost_descs);
 890
 891                 rc = validate_lod_and_idx(md, idx);
 892                 if (unlikely(rc != 0)) {
 893                         lod_putref(md, &md->lod_ost_descs);
 894                         GOTO(out, rc);
 895                 }
 896
 897                 nd = &OST_TGT(md,idx)->ltd_ost->dd_lu_dev;
 898                 lod_putref(md, &md->lod_ost_descs);
 899
 900                 /* In the function below, .hs_keycmp resolves to
 901                  * u_obj_hop_keycmp() */
 902                 /* coverity[overrun-buffer-val] */
 903                 o = lu_object_find_at(env, nd, &info->lti_fid, NULL);
 904                 if (IS_ERR(o))
 905                         GOTO(out, rc = PTR_ERR(o));
 906
 907                 n = lu_object_locate(o->lo_header, nd->ld_type);
 908                 LASSERT(n);
 909
 910                 stripe[i] = container_of(n, struct dt_object, do_lu);
 911         }
 912
 913 out:
 914         if (rc != 0) {
 915                 for (i = 0; i < stripe_len; i++)
 916                         if (stripe[i] != NULL)
 917                                 lu_object_put(env, &stripe[i]->do_lu);
 918
 919                 OBD_FREE(stripe, sizeof(stripe[0]) * stripe_len);
 920                 lo->ldo_stripenr = 0;
 921         } else {
 922                 lo->ldo_stripe = stripe;
 923                 lo->ldo_stripes_allocated = stripe_len;
 924         }
 925
 926         RETURN(rc);
 927 }
 928
 929 /**
 930  * Instantiate objects for striping.
 931  *
 932  * Parse striping information in \a buf and instantiate the objects
 933  * representing the stripes.
 934  *
 935  * \param[in] env               execution environment for this thread
 936  * \param[in] lo                LOD object
 937  * \param[in] buf               buffer storing LOV EA to parse
 938  *
 939  * \retval                      0 if parsing and objects creation succeed
 940  * \retval                      negative error number on failure
 941  */
 942 int lod_parse_striping(const struct lu_env *env, struct lod_object *lo,
 943                        const struct lu_buf *buf)
 944 {
 945         struct lov_mds_md_v1    *lmm;
 946         struct lov_ost_data_v1  *objs;
 947         __u32                    magic;
 948         __u32                    pattern;
 949         int                      rc = 0;
 950         ENTRY;
 951
 952         LASSERT(buf);
 953         LASSERT(buf->lb_buf);
 954         LASSERT(buf->lb_len);
 955
 956         lmm = (struct lov_mds_md_v1 *) buf->lb_buf;
 957         magic = le32_to_cpu(lmm->lmm_magic);
 958         pattern = le32_to_cpu(lmm->lmm_pattern);
 959
 960         if (magic != LOV_MAGIC_V1 && magic != LOV_MAGIC_V3)
 961                 GOTO(out, rc = -EINVAL);
 962         if (lov_pattern(pattern) != LOV_PATTERN_RAID0)
 963                 GOTO(out, rc = -EINVAL);
 964
 965         lo->ldo_pattern = pattern;
 966         lo->ldo_stripe_size = le32_to_cpu(lmm->lmm_stripe_size);
 967         lo->ldo_layout_gen = le16_to_cpu(lmm->lmm_layout_gen);
 968         lo->ldo_stripenr = le16_to_cpu(lmm->lmm_stripe_count);
 969         /* released file stripenr fixup. */
 970         if (pattern & LOV_PATTERN_F_RELEASED)
 971                 lo->ldo_stripenr = 0;
 972
 973         LASSERT(buf->lb_len >= lov_mds_md_size(lo->ldo_stripenr, magic));
 974
 975         if (magic == LOV_MAGIC_V3) {
 976                 struct lov_mds_md_v3 *v3 = (struct lov_mds_md_v3 *) lmm;
 977                 objs = &v3->lmm_objects[0];
 978                 lod_object_set_pool(lo, v3->lmm_pool_name);
 979         } else {
 980                 objs = &lmm->lmm_objects[0];
 981         }
 982
 983         if (lo->ldo_stripenr > 0)
 984                 rc = lod_initialize_objects(env, lo, objs);
 985
 986 out:
 987         RETURN(rc);
 988 }
 989
 990 /**
 991  * Initialize the object representing the stripes.
 992  *
 993  * Unless the stripes are initialized already, fetch LOV (for regular
 994  * objects) or LMV (for directory objects) EA and call lod_parse_striping()
 995  * to instantiate the objects representing the stripes.
 996  *
 997  * \param[in] env               execution environment for this thread
 998  * \param[in,out] lo            LOD object
 999  *
1000  * \retval                      0 if parsing and object creation succeed
1001  * \retval                      negative error number on failure
1002  */
1003 int lod_load_striping_locked(const struct lu_env *env, struct lod_object *lo)
1004 {
1005         struct lod_thread_info  *info = lod_env_info(env);
1006         struct lu_buf           *buf  = &info->lti_buf;
1007         struct dt_object        *next = dt_object_child(&lo->ldo_obj);
1008         int                      rc = 0;
1009         ENTRY;
1010
1011         /* already initialized? */
1012         if (lo->ldo_stripe != NULL)
1013                 GOTO(out, rc = 0);
1014
1015         if (!dt_object_exists(next))
1016                 GOTO(out, rc = 0);
1017
1018         /* Do not load stripe for slaves of striped dir */
1019         if (lo->ldo_dir_slave_stripe)
1020                 GOTO(out, rc = 0);
1021
1022         if (S_ISREG(lu_object_attr(lod2lu_obj(lo)))) {
1023                 rc = lod_get_lov_ea(env, lo);
1024                 if (rc <= 0)
1025                         GOTO(out, rc);
1026                 /*
1027                  * there is LOV EA (striping information) in this object
1028                  * let's parse it and create in-core objects for the stripes
1029                  */
1030                 buf->lb_buf = info->lti_ea_store;
1031                 buf->lb_len = info->lti_ea_store_size;
1032                 rc = lod_parse_striping(env, lo, buf);
1033         } else if (S_ISDIR(lu_object_attr(lod2lu_obj(lo)))) {
1034                 rc = lod_get_lmv_ea(env, lo);
1035                 if (rc < (typeof(rc))sizeof(struct lmv_mds_md_v1))
1036                         GOTO(out, rc = rc > 0 ? -EINVAL : rc);
1037
1038                 buf->lb_buf = info->lti_ea_store;
1039                 buf->lb_len = info->lti_ea_store_size;
1040                 if (rc == sizeof(struct lmv_mds_md_v1)) {
1041                         rc = lod_load_lmv_shards(env, lo, buf, true);
1042                         if (buf->lb_buf != info->lti_ea_store) {
1043                                 OBD_FREE_LARGE(info->lti_ea_store,
1044                                                info->lti_ea_store_size);
1045                                 info->lti_ea_store = buf->lb_buf;
1046                                 info->lti_ea_store_size = buf->lb_len;
1047                         }
1048
1049                         if (rc < 0)
1050                                 GOTO(out, rc);
1051                 }
1052
1053                 /*
1054                  * there is LOV EA (striping information) in this object
1055                  * let's parse it and create in-core objects for the stripes
1056                  */
1057                 rc = lod_parse_dir_striping(env, lo, buf);
1058         }
1059 out:
1060         RETURN(rc);
1061 }
1062
1063 /**
1064  * A generic function to initialize the stripe objects.
1065  *
1066  * A protected version of lod_load_striping_locked() - load the striping
1067  * information from storage, parse that and instantiate LU objects to
1068  * represent the stripes.  The LOD object \a lo supplies a pointer to the
1069  * next sub-object in the LU stack so we can lock it. Also use \a lo to
1070  * return an array of references to the newly instantiated objects.
1071  *
1072  * \param[in] env               execution environment for this thread
1073  * \param[in,out] lo            LOD object, where striping is stored and
1074  *                              which gets an array of references
1075  *
1076  * \retval                      0 if parsing and object creation succeed
1077  * \retval                      negative error number on failure
1078  **/
1079 int lod_load_striping(const struct lu_env *env, struct lod_object *lo)
1080 {
1081         struct dt_object        *next = dt_object_child(&lo->ldo_obj);
1082         int                     rc = 0;
1083
1084         /* currently this code is supposed to be called from declaration
1085          * phase only, thus the object is not expected to be locked by caller */
1086         dt_write_lock(env, next, 0);
1087         rc = lod_load_striping_locked(env, lo);
1088         dt_write_unlock(env, next);
1089         return rc;
1090 }
1091
1092 /**
1093  * Verify striping.
1094  *
1095  * Check the validity of all fields including the magic, stripe size,
1096  * stripe count, stripe offset and that the pool is present.  Also check
1097  * that each target index points to an existing target. The additional
1098  * \a is_from_disk turns additional checks. In some cases zero fields
1099  * are allowed (like pattern=0).
1100  *
1101  * \param[in] d                 LOD device
1102  * \param[in] buf               buffer with LOV EA to verify
1103  * \param[in] is_from_disk      0 - from user, allow some fields to be 0
1104  *                              1 - from disk, do not allow
1105  *
1106  * \retval                      0 if the striping is valid
1107  * \retval                      -EINVAL if striping is invalid
1108  */
1109 int lod_verify_striping(struct lod_device *d, const struct lu_buf *buf,
1110                         bool is_from_disk)
1111 {
1112         struct lov_user_md_v1   *lum;
1113         struct lov_user_md_v3   *lum3;
1114         struct pool_desc        *pool = NULL;
1115         __u32                    magic;
1116         __u32                    stripe_size;
1117         __u16                    stripe_count;
1118         __u16                    stripe_offset;
1119         size_t                   lum_size;
1120         int                      rc = 0;
1121         ENTRY;
1122
1123         lum = buf->lb_buf;
1124
1125         LASSERT(sizeof(*lum) < sizeof(*lum3));
1126
1127         if (buf->lb_len < sizeof(*lum)) {
1128                 CDEBUG(D_IOCTL, "buf len %zd too small for lov_user_md\n",
1129                        buf->lb_len);
1130                 GOTO(out, rc = -EINVAL);
1131         }
1132
1133         magic = le32_to_cpu(lum->lmm_magic);
1134         if (magic != LOV_USER_MAGIC_V1 &&
1135             magic != LOV_USER_MAGIC_V3 &&
1136             magic != LOV_MAGIC_V1_DEF &&
1137             magic != LOV_MAGIC_V3_DEF) {
1138                 CDEBUG(D_IOCTL, "bad userland LOV MAGIC: %#x\n", magic);
1139                 GOTO(out, rc = -EINVAL);
1140         }
1141
1142         /* the user uses "0" for default stripe pattern normally. */
1143         if (!is_from_disk && lum->lmm_pattern == 0)
1144                 lum->lmm_pattern = cpu_to_le32(LOV_PATTERN_RAID0);
1145
1146         if (le32_to_cpu(lum->lmm_pattern) != LOV_PATTERN_RAID0) {
1147                 CDEBUG(D_IOCTL, "bad userland stripe pattern: %#x\n",
1148                        le32_to_cpu(lum->lmm_pattern));
1149                 GOTO(out, rc = -EINVAL);
1150         }
1151
1152         /* 64kB is the largest common page size we see (ia64), and matches the
1153          * check in lfs */
1154         stripe_size = le32_to_cpu(lum->lmm_stripe_size);
1155         if (stripe_size & (LOV_MIN_STRIPE_SIZE - 1)) {
1156                 CDEBUG(D_IOCTL, "stripe size %u not a multiple of %u\n",
1157                        stripe_size, LOV_MIN_STRIPE_SIZE);
1158                 GOTO(out, rc = -EINVAL);
1159         }
1160
1161         /* an offset of -1 is treated as a "special" valid offset */
1162         stripe_offset = le16_to_cpu(lum->lmm_stripe_offset);
1163         if (stripe_offset != (typeof(stripe_offset))-1) {
1164                 /* if offset is not within valid range [0, osts_size) */
1165                 if (stripe_offset >= d->lod_osts_size) {
1166                         CDEBUG(D_IOCTL, "stripe offset %u >= bitmap size %u\n",
1167                                stripe_offset, d->lod_osts_size);
1168                         GOTO(out, rc = -EINVAL);
1169                 }
1170
1171                 /* if lmm_stripe_offset is *not* in bitmap */
1172                 if (!cfs_bitmap_check(d->lod_ost_bitmap, stripe_offset)) {
1173                         CDEBUG(D_IOCTL, "stripe offset %u not in bitmap\n",
1174                                stripe_offset);
1175                         GOTO(out, rc = -EINVAL);
1176                 }
1177         }
1178
1179         if (magic == LOV_USER_MAGIC_V1 || magic == LOV_MAGIC_V1_DEF)
1180                 lum_size = offsetof(struct lov_user_md_v1,
1181                                     lmm_objects[0]);
1182         else if (magic == LOV_USER_MAGIC_V3 || magic == LOV_MAGIC_V3_DEF)
1183                 lum_size = offsetof(struct lov_user_md_v3,
1184                                     lmm_objects[0]);
1185         else
1186                 GOTO(out, rc = -EINVAL);
1187
1188         stripe_count = le16_to_cpu(lum->lmm_stripe_count);
1189         if (buf->lb_len != lum_size) {
1190                 CDEBUG(D_IOCTL, "invalid buf len %zd for lov_user_md with "
1191                        "magic %#x and stripe_count %u\n",
1192                        buf->lb_len, magic, stripe_count);
1193                 GOTO(out, rc = -EINVAL);
1194         }
1195
1196         if (!(magic == LOV_USER_MAGIC_V3 || magic == LOV_MAGIC_V3_DEF))
1197                 goto out;
1198
1199         lum3 = buf->lb_buf;
1200         if (buf->lb_len < sizeof(*lum3)) {
1201                 CDEBUG(D_IOCTL, "buf len %zd too small for lov_user_md_v3\n",
1202                        buf->lb_len);
1203                 GOTO(out, rc = -EINVAL);
1204         }
1205
1206         /* In the function below, .hs_keycmp resolves to
1207          * pool_hashkey_keycmp() */
1208         /* coverity[overrun-buffer-val] */
1209         pool = lod_find_pool(d, lum3->lmm_pool_name);
1210         if (pool == NULL)
1211                 goto out;
1212
1213         if (stripe_offset != (typeof(stripe_offset))-1) {
1214                 rc = lod_check_index_in_pool(stripe_offset, pool);
1215                 if (rc < 0)
1216                         GOTO(out, rc = -EINVAL);
1217         }
1218
1219         if (is_from_disk && stripe_count > pool_tgt_count(pool)) {
1220                 CDEBUG(D_IOCTL,
1221                        "stripe count %u > # OSTs %u in the pool\n",
1222                        stripe_count, pool_tgt_count(pool));
1223                 GOTO(out, rc = -EINVAL);
1224         }
1225
1226 out:
1227         if (pool != NULL)
1228                 lod_pool_putref(pool);
1229
1230         RETURN(rc);
1231 }
1232
1233 void lod_fix_desc_stripe_size(__u64 *val)
1234 {
1235         if (*val < LOV_MIN_STRIPE_SIZE) {
1236                 if (*val != 0)
1237                         LCONSOLE_INFO("Increasing default stripe size to "
1238                                       "minimum value %u\n",
1239                                       LOV_DESC_STRIPE_SIZE_DEFAULT);
1240                 *val = LOV_DESC_STRIPE_SIZE_DEFAULT;
1241         } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) {
1242                 *val &= ~(LOV_MIN_STRIPE_SIZE - 1);
1243                 LCONSOLE_WARN("Changing default stripe size to "LPU64" (a "
1244                               "multiple of %u)\n",
1245                               *val, LOV_MIN_STRIPE_SIZE);
1246         }
1247 }
1248
1249 void lod_fix_desc_stripe_count(__u32 *val)
1250 {
1251         if (*val == 0)
1252                 *val = 1;
1253 }
1254
1255 void lod_fix_desc_pattern(__u32 *val)
1256 {
1257         /* from lov_setstripe */
1258         if ((*val != 0) && (*val != LOV_PATTERN_RAID0)) {
1259                 LCONSOLE_WARN("Unknown stripe pattern: %#x\n", *val);
1260                 *val = 0;
1261         }
1262 }
1263
1264 void lod_fix_desc_qos_maxage(__u32 *val)
1265 {
1266         /* fix qos_maxage */
1267         if (*val == 0)
1268                 *val = LOV_DESC_QOS_MAXAGE_DEFAULT;
1269 }
1270
1271 /**
1272  * Used to fix insane default striping.
1273  *
1274  * \param[in] desc      striping description
1275  */
1276 void lod_fix_desc(struct lov_desc *desc)
1277 {
1278         lod_fix_desc_stripe_size(&desc->ld_default_stripe_size);
1279         lod_fix_desc_stripe_count(&desc->ld_default_stripe_count);
1280         lod_fix_desc_pattern(&desc->ld_pattern);
1281         lod_fix_desc_qos_maxage(&desc->ld_qos_maxage);
1282 }
1283
1284 /**
1285  * Initialize the structures used to store pools and default striping.
1286  *
1287  * \param[in] lod       LOD device
1288  * \param[in] lcfg      configuration structure storing default striping.
1289  *
1290  * \retval              0 if initialization succeeds
1291  * \retval              negative error number on failure
1292  */
1293 int lod_pools_init(struct lod_device *lod, struct lustre_cfg *lcfg)
1294 {
1295         struct obd_device          *obd;
1296         struct lov_desc            *desc;
1297         int                         rc;
1298         ENTRY;
1299
1300         obd = class_name2obd(lustre_cfg_string(lcfg, 0));
1301         LASSERT(obd != NULL);
1302         obd->obd_lu_dev = &lod->lod_dt_dev.dd_lu_dev;
1303
1304         if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
1305                 CERROR("LOD setup requires a descriptor\n");
1306                 RETURN(-EINVAL);
1307         }
1308
1309         desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
1310
1311         if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
1312                 CERROR("descriptor size wrong: %d > %d\n",
1313                        (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
1314                 RETURN(-EINVAL);
1315         }
1316
1317         if (desc->ld_magic != LOV_DESC_MAGIC) {
1318                 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
1319                         CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
1320                                obd->obd_name, desc);
1321                         lustre_swab_lov_desc(desc);
1322                 } else {
1323                         CERROR("%s: Bad lov desc magic: %#x\n",
1324                                obd->obd_name, desc->ld_magic);
1325                         RETURN(-EINVAL);
1326                 }
1327         }
1328
1329         lod_fix_desc(desc);
1330
1331         desc->ld_active_tgt_count = 0;
1332         lod->lod_desc = *desc;
1333
1334         lod->lod_sp_me = LUSTRE_SP_CLI;
1335
1336         /* Set up allocation policy (QoS and RR) */
1337         INIT_LIST_HEAD(&lod->lod_qos.lq_oss_list);
1338         init_rwsem(&lod->lod_qos.lq_rw_sem);
1339         lod->lod_qos.lq_dirty = 1;
1340         lod->lod_qos.lq_rr.lqr_dirty = 1;
1341         lod->lod_qos.lq_reset = 1;
1342         /* Default priority is toward free space balance */
1343         lod->lod_qos.lq_prio_free = 232;
1344         /* Default threshold for rr (roughly 17%) */
1345         lod->lod_qos.lq_threshold_rr = 43;
1346
1347         /* Set up OST pool environment */
1348         lod->lod_pools_hash_body = cfs_hash_create("POOLS", HASH_POOLS_CUR_BITS,
1349                                                    HASH_POOLS_MAX_BITS,
1350                                                    HASH_POOLS_BKT_BITS, 0,
1351                                                    CFS_HASH_MIN_THETA,
1352                                                    CFS_HASH_MAX_THETA,
1353                                                    &pool_hash_operations,
1354                                                    CFS_HASH_DEFAULT);
1355         if (lod->lod_pools_hash_body == NULL)
1356                 RETURN(-ENOMEM);
1357
1358         INIT_LIST_HEAD(&lod->lod_pool_list);
1359         lod->lod_pool_count = 0;
1360         rc = lod_ost_pool_init(&lod->lod_pool_info, 0);
1361         if (rc)
1362                 GOTO(out_hash, rc);
1363         rc = lod_ost_pool_init(&lod->lod_qos.lq_rr.lqr_pool, 0);
1364         if (rc)
1365                 GOTO(out_pool_info, rc);
1366
1367         RETURN(0);
1368
1369 out_pool_info:
1370         lod_ost_pool_free(&lod->lod_pool_info);
1371 out_hash:
1372         cfs_hash_putref(lod->lod_pools_hash_body);
1373
1374         return rc;
1375 }
1376
1377 /**
1378  * Release the structures describing the pools.
1379  *
1380  * \param[in] lod       LOD device from which we release the structures
1381  *
1382  * \retval              0 always
1383  */
1384 int lod_pools_fini(struct lod_device *lod)
1385 {
1386         struct obd_device   *obd = lod2obd(lod);
1387         struct pool_desc    *pool, *tmp;
1388         ENTRY;
1389
1390         list_for_each_entry_safe(pool, tmp, &lod->lod_pool_list, pool_list) {
1391                 /* free pool structs */
1392                 CDEBUG(D_INFO, "delete pool %p\n", pool);
1393                 /* In the function below, .hs_keycmp resolves to
1394                  * pool_hashkey_keycmp() */
1395                 /* coverity[overrun-buffer-val] */
1396                 lod_pool_del(obd, pool->pool_name);
1397         }
1398
1399         cfs_hash_putref(lod->lod_pools_hash_body);
1400         lod_ost_pool_free(&(lod->lod_qos.lq_rr.lqr_pool));
1401         lod_ost_pool_free(&lod->lod_pool_info);
1402
1403         RETURN(0);
1404 }