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