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, 2015, 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        *lu_dev;
 214         struct dt_device        *dt_dev;
 215         int                      rc;
 216         struct lod_tgt_desc     *tgt_desc;
 217         struct lod_tgt_descs    *ltd;
 218         struct lustre_cfg       *lcfg;
 219         struct obd_uuid         obd_uuid;
 220         bool                    for_ost;
 221         bool lock = false;
 222         ENTRY;
 223
 224         CDEBUG(D_CONFIG, "osp:%s idx:%d gen:%d\n", osp, index, gen);
 225
 226         if (gen <= 0) {
 227                 CERROR("request to add OBD %s with invalid generation: %d\n",
 228                        osp, gen);
 229                 RETURN(-EINVAL);
 230         }
 231
 232         obd_str2uuid(&obd_uuid, osp);
 233
 234         obd = class_find_client_obd(&obd_uuid, LUSTRE_OSP_NAME,
 235                                 &lod->lod_dt_dev.dd_lu_dev.ld_obd->obd_uuid);
 236         if (obd == NULL) {
 237                 CERROR("can't find %s device\n", osp);
 238                 RETURN(-EINVAL);
 239         }
 240
 241         LASSERT(obd->obd_lu_dev != NULL);
 242         LASSERT(obd->obd_lu_dev->ld_site == lod->lod_dt_dev.dd_lu_dev.ld_site);
 243
 244         lu_dev = obd->obd_lu_dev;
 245         dt_dev = lu2dt_dev(lu_dev);
 246
 247         OBD_ALLOC_PTR(data);
 248         if (data == NULL)
 249                 GOTO(out_cleanup, rc = -ENOMEM);
 250
 251         data->ocd_connect_flags = OBD_CONNECT_INDEX | OBD_CONNECT_VERSION;
 252         data->ocd_version = LUSTRE_VERSION_CODE;
 253         data->ocd_index = index;
 254
 255         if (strcmp(LUSTRE_OSC_NAME, type) == 0) {
 256                 for_ost = true;
 257                 data->ocd_connect_flags |= OBD_CONNECT_AT |
 258                                            OBD_CONNECT_FULL20 |
 259                                            OBD_CONNECT_INDEX |
 260 #ifdef HAVE_LRU_RESIZE_SUPPORT
 261                                            OBD_CONNECT_LRU_RESIZE |
 262 #endif
 263                                            OBD_CONNECT_MDS |
 264                                            OBD_CONNECT_REQPORTAL |
 265                                            OBD_CONNECT_SKIP_ORPHAN |
 266                                            OBD_CONNECT_FID |
 267                                            OBD_CONNECT_LVB_TYPE |
 268                                            OBD_CONNECT_VERSION |
 269                                            OBD_CONNECT_PINGLESS |
 270                                            OBD_CONNECT_LFSCK |
 271                                            OBD_CONNECT_BULK_MBITS;
 272
 273                 data->ocd_group = tgt_index;
 274                 ltd = &lod->lod_ost_descs;
 275         } else {
 276                 struct obd_import *imp = obd->u.cli.cl_import;
 277
 278                 for_ost = false;
 279                 data->ocd_ibits_known = MDS_INODELOCK_UPDATE;
 280                 data->ocd_connect_flags |= OBD_CONNECT_ACL |
 281                                            OBD_CONNECT_IBITS |
 282                                            OBD_CONNECT_MDS_MDS |
 283                                            OBD_CONNECT_FID |
 284                                            OBD_CONNECT_AT |
 285                                            OBD_CONNECT_FULL20 |
 286                                            OBD_CONNECT_LFSCK |
 287                                            OBD_CONNECT_BULK_MBITS;
 288                 spin_lock(&imp->imp_lock);
 289                 imp->imp_server_timeout = 1;
 290                 spin_unlock(&imp->imp_lock);
 291                 imp->imp_client->cli_request_portal = OUT_PORTAL;
 292                 CDEBUG(D_OTHER, "%s: Set 'mds' portal and timeout\n",
 293                       obd->obd_name);
 294                 ltd = &lod->lod_mdt_descs;
 295         }
 296
 297         rc = obd_connect(env, &exp, obd, &obd->obd_uuid, data, NULL);
 298         OBD_FREE_PTR(data);
 299         if (rc) {
 300                 CERROR("%s: cannot connect to next dev %s (%d)\n",
 301                        obd->obd_name, osp, rc);
 302                 GOTO(out_cleanup, rc);
 303         }
 304
 305         /* Allocate ost descriptor and fill it */
 306         OBD_ALLOC_PTR(tgt_desc);
 307         if (!tgt_desc)
 308                 GOTO(out_conn, rc = -ENOMEM);
 309
 310         tgt_desc->ltd_tgt    = dt_dev;
 311         tgt_desc->ltd_exp    = exp;
 312         tgt_desc->ltd_uuid   = obd->u.cli.cl_target_uuid;
 313         tgt_desc->ltd_gen    = gen;
 314         tgt_desc->ltd_index  = index;
 315         tgt_desc->ltd_active = active;
 316
 317         lod_getref(ltd);
 318         if (index >= ltd->ltd_tgts_size) {
 319                 /* we have to increase the size of the lod_osts array */
 320                 __u32  newsize;
 321
 322                 newsize = max(ltd->ltd_tgts_size, (__u32)2);
 323                 while (newsize < index + 1)
 324                         newsize = newsize << 1;
 325
 326                 /* lod_bitmap_resize() needs lod_rw_sem
 327                  * which we hold with th reference */
 328                 lod_putref(lod, ltd);
 329
 330                 rc = ltd_bitmap_resize(ltd, newsize);
 331                 if (rc)
 332                         GOTO(out_desc, rc);
 333
 334                 lod_getref(ltd);
 335         }
 336
 337         mutex_lock(&ltd->ltd_mutex);
 338         lock = true;
 339         if (cfs_bitmap_check(ltd->ltd_tgt_bitmap, index)) {
 340                 CERROR("%s: device %d is registered already\n", obd->obd_name,
 341                        index);
 342                 GOTO(out_mutex, rc = -EEXIST);
 343         }
 344
 345         if (ltd->ltd_tgt_idx[index / TGT_PTRS_PER_BLOCK] == NULL) {
 346                 OBD_ALLOC_PTR(ltd->ltd_tgt_idx[index / TGT_PTRS_PER_BLOCK]);
 347                 if (ltd->ltd_tgt_idx[index / TGT_PTRS_PER_BLOCK] == NULL) {
 348                         CERROR("can't allocate index to add %s\n",
 349                                obd->obd_name);
 350                         GOTO(out_mutex, rc = -ENOMEM);
 351                 }
 352         }
 353
 354         if (for_ost) {
 355                 /* pool and qos are not supported for MDS stack yet */
 356                 rc = lod_ost_pool_add(&lod->lod_pool_info, index,
 357                                       lod->lod_osts_size);
 358                 if (rc) {
 359                         CERROR("%s: can't set up pool, failed with %d\n",
 360                                obd->obd_name, rc);
 361                         GOTO(out_mutex, rc);
 362                 }
 363
 364                 rc = qos_add_tgt(lod, tgt_desc);
 365                 if (rc) {
 366                         CERROR("%s: qos_add_tgt failed with %d\n",
 367                                 obd->obd_name, rc);
 368                         GOTO(out_pool, rc);
 369                 }
 370
 371                 /* The new OST is now a full citizen */
 372                 if (index >= lod->lod_desc.ld_tgt_count)
 373                         lod->lod_desc.ld_tgt_count = index + 1;
 374                 if (active)
 375                         lod->lod_desc.ld_active_tgt_count++;
 376         }
 377
 378         LTD_TGT(ltd, index) = tgt_desc;
 379         cfs_bitmap_set(ltd->ltd_tgt_bitmap, index);
 380         ltd->ltd_tgtnr++;
 381         mutex_unlock(&ltd->ltd_mutex);
 382         lod_putref(lod, ltd);
 383         lock = false;
 384         if (lod->lod_recovery_completed)
 385                 lu_dev->ld_ops->ldo_recovery_complete(env, lu_dev);
 386
 387         if (!for_ost && lod->lod_initialized) {
 388                 rc = lod_sub_init_llog(env, lod, tgt_desc->ltd_tgt);
 389                 if (rc != 0) {
 390                         CERROR("%s: cannot start llog on %s:rc = %d\n",
 391                                lod2obd(lod)->obd_name, osp, rc);
 392                         GOTO(out_ltd, rc);
 393                 }
 394         }
 395
 396         rc = lfsck_add_target(env, lod->lod_child, dt_dev, exp, index, for_ost);
 397         if (rc != 0) {
 398                 CERROR("Fail to add LFSCK target: name = %s, type = %s, "
 399                        "index = %u, rc = %d\n", osp, type, index, rc);
 400                 GOTO(out_fini_llog, rc);
 401         }
 402         RETURN(rc);
 403 out_fini_llog:
 404         lod_sub_fini_llog(env, tgt_desc->ltd_tgt,
 405                           tgt_desc->ltd_recovery_thread);
 406 out_ltd:
 407         lod_getref(ltd);
 408         mutex_lock(&ltd->ltd_mutex);
 409         lock = true;
 410         if (!for_ost && LTD_TGT(ltd, index)->ltd_recovery_thread != NULL) {
 411                 struct ptlrpc_thread *thread;
 412
 413                 thread = LTD_TGT(ltd, index)->ltd_recovery_thread;
 414                 OBD_FREE_PTR(thread);
 415         }
 416         ltd->ltd_tgtnr--;
 417         cfs_bitmap_clear(ltd->ltd_tgt_bitmap, index);
 418         LTD_TGT(ltd, index) = NULL;
 419 out_pool:
 420         lod_ost_pool_remove(&lod->lod_pool_info, index);
 421 out_mutex:
 422         if (lock) {
 423                 mutex_unlock(&ltd->ltd_mutex);
 424                 lod_putref(lod, ltd);
 425         }
 426 out_desc:
 427         OBD_FREE_PTR(tgt_desc);
 428 out_conn:
 429         obd_disconnect(exp);
 430 out_cleanup:
 431         /* XXX OSP needs us to send down LCFG_CLEANUP because it uses
 432          * objects from the MDT stack. See LU-7184. */
 433         lcfg = &lod_env_info(env)->lti_lustre_cfg;
 434         memset(lcfg, 0, sizeof(*lcfg));
 435         lcfg->lcfg_version = LUSTRE_CFG_VERSION;
 436         lcfg->lcfg_command = LCFG_CLEANUP;
 437         lu_dev->ld_ops->ldo_process_config(env, lu_dev, lcfg);
 438
 439         return rc;
 440 }
 441
 442 /**
 443  * Schedule target removal from the target table.
 444  *
 445  * Mark the device as dead. The device is not removed here because it may
 446  * still be in use. The device will be removed in lod_putref() when the
 447  * last reference is released.
 448  *
 449  * \param[in] env               execution environment for this thread
 450  * \param[in] lod               LOD device the target table belongs to
 451  * \param[in] ltd               target table
 452  * \param[in] idx               index of the target
 453  * \param[in] for_ost           type of the target: 0 - MDT, 1 - OST
 454  */
 455 static void __lod_del_device(const struct lu_env *env, struct lod_device *lod,
 456                              struct lod_tgt_descs *ltd, unsigned idx,
 457                              bool for_ost)
 458 {
 459         LASSERT(LTD_TGT(ltd, idx));
 460
 461         lfsck_del_target(env, lod->lod_child, LTD_TGT(ltd, idx)->ltd_tgt,
 462                          idx, for_ost);
 463
 464         if (!for_ost && LTD_TGT(ltd, idx)->ltd_recovery_thread != NULL) {
 465                 struct ptlrpc_thread *thread;
 466
 467                 thread = LTD_TGT(ltd, idx)->ltd_recovery_thread;
 468                 OBD_FREE_PTR(thread);
 469         }
 470
 471         if (LTD_TGT(ltd, idx)->ltd_reap == 0) {
 472                 LTD_TGT(ltd, idx)->ltd_reap = 1;
 473                 ltd->ltd_death_row++;
 474         }
 475 }
 476
 477 /**
 478  * Schedule removal of all the targets from the given target table.
 479  *
 480  * See more details in the description for __lod_del_device()
 481  *
 482  * \param[in] env               execution environment for this thread
 483  * \param[in] lod               LOD device the target table belongs to
 484  * \param[in] ltd               target table
 485  * \param[in] for_ost           type of the target: MDT or OST
 486  *
 487  * \retval                      0 always
 488  */
 489 int lod_fini_tgt(const struct lu_env *env, struct lod_device *lod,
 490                  struct lod_tgt_descs *ltd, bool for_ost)
 491 {
 492         unsigned int idx;
 493
 494         if (ltd->ltd_tgts_size <= 0)
 495                 return 0;
 496         lod_getref(ltd);
 497         mutex_lock(&ltd->ltd_mutex);
 498         cfs_foreach_bit(ltd->ltd_tgt_bitmap, idx)
 499                 __lod_del_device(env, lod, ltd, idx, for_ost);
 500         mutex_unlock(&ltd->ltd_mutex);
 501         lod_putref(lod, ltd);
 502         CFS_FREE_BITMAP(ltd->ltd_tgt_bitmap);
 503         for (idx = 0; idx < TGT_PTRS; idx++) {
 504                 if (ltd->ltd_tgt_idx[idx])
 505                         OBD_FREE_PTR(ltd->ltd_tgt_idx[idx]);
 506         }
 507         ltd->ltd_tgts_size = 0;
 508         return 0;
 509 }
 510
 511 /**
 512  * Remove device by name.
 513  *
 514  * Remove a device identified by \a osp from the target table. Given
 515  * the device can be in use, the real deletion happens in lod_putref().
 516  *
 517  * \param[in] env               execution environment for this thread
 518  * \param[in] lod               LOD device to be connected to the new OSP
 519  * \param[in] ltd               target table
 520  * \param[in] osp               name of OSP device to be removed
 521  * \param[in] idx               index of the target
 522  * \param[in] gen               generation number, not used currently
 523  * \param[in] for_ost           type of the target: 0 - MDT, 1 - OST
 524  *
 525  * \retval                      0 if the device was scheduled for removal
 526  * \retval                      -EINVAL if no device was found
 527  */
 528 int lod_del_device(const struct lu_env *env, struct lod_device *lod,
 529                    struct lod_tgt_descs *ltd, char *osp, unsigned idx,
 530                    unsigned gen, bool for_ost)
 531 {
 532         struct obd_device *obd;
 533         int                rc = 0;
 534         struct obd_uuid    uuid;
 535         ENTRY;
 536
 537         CDEBUG(D_CONFIG, "osp:%s idx:%d gen:%d\n", osp, idx, gen);
 538
 539         obd_str2uuid(&uuid, osp);
 540
 541         obd = class_find_client_obd(&uuid, LUSTRE_OSP_NAME,
 542                                    &lod->lod_dt_dev.dd_lu_dev.ld_obd->obd_uuid);
 543         if (obd == NULL) {
 544                 CERROR("can't find %s device\n", osp);
 545                 RETURN(-EINVAL);
 546         }
 547
 548         if (gen <= 0) {
 549                 CERROR("%s: request to remove OBD %s with invalid generation %d"
 550                        "\n", obd->obd_name, osp, gen);
 551                 RETURN(-EINVAL);
 552         }
 553
 554         obd_str2uuid(&uuid,  osp);
 555
 556         lod_getref(ltd);
 557         mutex_lock(&ltd->ltd_mutex);
 558         /* check that the index is allocated in the bitmap */
 559         if (!cfs_bitmap_check(ltd->ltd_tgt_bitmap, idx) ||
 560             !LTD_TGT(ltd, idx)) {
 561                 CERROR("%s: device %d is not set up\n", obd->obd_name, idx);
 562                 GOTO(out, rc = -EINVAL);
 563         }
 564
 565         /* check that the UUID matches */
 566         if (!obd_uuid_equals(&uuid, &LTD_TGT(ltd, idx)->ltd_uuid)) {
 567                 CERROR("%s: LOD target UUID %s at index %d does not match %s\n",
 568                        obd->obd_name, obd_uuid2str(&LTD_TGT(ltd,idx)->ltd_uuid),
 569                        idx, osp);
 570                 GOTO(out, rc = -EINVAL);
 571         }
 572
 573         __lod_del_device(env, lod, ltd, idx, for_ost);
 574         EXIT;
 575 out:
 576         mutex_unlock(&ltd->ltd_mutex);
 577         lod_putref(lod, ltd);
 578         return(rc);
 579 }
 580
 581 /**
 582  * Resize per-thread storage to hold specified size.
 583  *
 584  * A helper function to resize per-thread temporary storage. This storage
 585  * is used to process LOV/LVM EAs and may be quite large. We do not want to
 586  * allocate/release it every time, so instead we put it into the env and
 587  * reallocate on demand. The memory is released when the correspondent thread
 588  * is finished.
 589  *
 590  * \param[in] info              LOD-specific storage in the environment
 591  * \param[in] size              new size to grow the buffer to
 592
 593  * \retval                      0 on success, -ENOMEM if reallocation failed
 594  */
 595 int lod_ea_store_resize(struct lod_thread_info *info, size_t size)
 596 {
 597         __u32 round = size_roundup_power2(size);
 598
 599         LASSERT(round <=
 600                 lov_mds_md_size(LOV_MAX_STRIPE_COUNT, LOV_MAGIC_V3));
 601         if (info->lti_ea_store) {
 602                 LASSERT(info->lti_ea_store_size);
 603                 LASSERT(info->lti_ea_store_size < round);
 604                 CDEBUG(D_INFO, "EA store size %d is not enough, need %d\n",
 605                        info->lti_ea_store_size, round);
 606                 OBD_FREE_LARGE(info->lti_ea_store, info->lti_ea_store_size);
 607                 info->lti_ea_store = NULL;
 608                 info->lti_ea_store_size = 0;
 609         }
 610
 611         OBD_ALLOC_LARGE(info->lti_ea_store, round);
 612         if (info->lti_ea_store == NULL)
 613                 RETURN(-ENOMEM);
 614         info->lti_ea_store_size = round;
 615         RETURN(0);
 616 }
 617
 618 /**
 619  * Make LOV EA for striped object.
 620  *
 621  * Generate striping information and store it in the LOV EA of the given
 622  * object. The caller must ensure nobody else is calling the function
 623  * against the object concurrently. The transaction must be started.
 624  * FLDB service must be running as well; it's used to map FID to the target,
 625  * which is stored in LOV EA.
 626  *
 627  * \param[in] env               execution environment for this thread
 628  * \param[in] lo                LOD object
 629  * \param[in] th                transaction handle
 630  *
 631  * \retval                      0 if LOV EA is stored successfully
 632  * \retval                      negative error number on failure
 633  */
 634 int lod_generate_and_set_lovea(const struct lu_env *env,
 635                                struct lod_object *lo, struct thandle *th)
 636 {
 637         struct lod_thread_info  *info = lod_env_info(env);
 638         struct dt_object        *next = dt_object_child(&lo->ldo_obj);
 639         const struct lu_fid     *fid  = lu_object_fid(&lo->ldo_obj.do_lu);
 640         struct lov_mds_md_v1    *lmm;
 641         struct lov_ost_data_v1  *objs;
 642         __u32                    magic;
 643         int                      i, rc;
 644         size_t                   lmm_size;
 645         ENTRY;
 646
 647         LASSERT(lo);
 648
 649         magic = lo->ldo_pool != NULL ? LOV_MAGIC_V3 : LOV_MAGIC_V1;
 650         lmm_size = lov_mds_md_size(lo->ldo_stripenr, magic);
 651         if (info->lti_ea_store_size < lmm_size) {
 652                 rc = lod_ea_store_resize(info, lmm_size);
 653                 if (rc)
 654                         RETURN(rc);
 655         }
 656
 657         if (lo->ldo_pattern == 0) /* default striping */
 658                 lo->ldo_pattern = LOV_PATTERN_RAID0;
 659
 660         lmm = info->lti_ea_store;
 661
 662         lmm->lmm_magic = cpu_to_le32(magic);
 663         lmm->lmm_pattern = cpu_to_le32(lo->ldo_pattern);
 664         fid_to_lmm_oi(fid, &lmm->lmm_oi);
 665         if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_LMMOI))
 666                 lmm->lmm_oi.oi.oi_id++;
 667         lmm_oi_cpu_to_le(&lmm->lmm_oi, &lmm->lmm_oi);
 668         lmm->lmm_stripe_size = cpu_to_le32(lo->ldo_stripe_size);
 669         lmm->lmm_stripe_count = cpu_to_le16(lo->ldo_stripenr);
 670         if (lo->ldo_pattern & LOV_PATTERN_F_RELEASED)
 671                 lmm->lmm_stripe_count = cpu_to_le16(lo->ldo_released_stripenr);
 672         lmm->lmm_layout_gen = 0;
 673         if (magic == LOV_MAGIC_V1) {
 674                 objs = &lmm->lmm_objects[0];
 675         } else {
 676                 struct lov_mds_md_v3 *v3 = (struct lov_mds_md_v3 *) lmm;
 677                 size_t cplen = strlcpy(v3->lmm_pool_name, lo->ldo_pool,
 678                                 sizeof(v3->lmm_pool_name));
 679                 if (cplen >= sizeof(v3->lmm_pool_name))
 680                         RETURN(-E2BIG);
 681                 objs = &v3->lmm_objects[0];
 682         }
 683
 684         for (i = 0; i < lo->ldo_stripenr; i++) {
 685                 struct lu_fid           *fid    = &info->lti_fid;
 686                 struct lod_device       *lod;
 687                 __u32                   index;
 688                 int                     type    = LU_SEQ_RANGE_OST;
 689
 690                 lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
 691                 LASSERT(lo->ldo_stripe[i]);
 692
 693                 *fid = *lu_object_fid(&lo->ldo_stripe[i]->do_lu);
 694                 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_MULTIPLE_REF)) {
 695                         if (cfs_fail_val == 0)
 696                                 cfs_fail_val = fid->f_oid;
 697                         else
 698                                 fid->f_oid = cfs_fail_val;
 699                 }
 700
 701                 rc = fid_to_ostid(fid, &info->lti_ostid);
 702                 LASSERT(rc == 0);
 703
 704                 ostid_cpu_to_le(&info->lti_ostid, &objs[i].l_ost_oi);
 705                 objs[i].l_ost_gen    = cpu_to_le32(0);
 706                 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_FLD_LOOKUP))
 707                         rc = -ENOENT;
 708                 else
 709                         rc = lod_fld_lookup(env, lod, fid,
 710                                             &index, &type);
 711                 if (rc < 0) {
 712                         CERROR("%s: Can not locate "DFID": rc = %d\n",
 713                                lod2obd(lod)->obd_name, PFID(fid), rc);
 714                         lod_object_free_striping(env, lo);
 715                         RETURN(rc);
 716                 }
 717                 objs[i].l_ost_idx = cpu_to_le32(index);
 718         }
 719
 720         info->lti_buf.lb_buf = lmm;
 721         info->lti_buf.lb_len = lmm_size;
 722         rc = lod_sub_object_xattr_set(env, next, &info->lti_buf, XATTR_NAME_LOV,
 723                                       0, th);
 724         if (rc < 0) {
 725                 lod_object_free_striping(env, lo);
 726                 RETURN(rc);
 727         }
 728
 729         RETURN(rc);
 730 }
 731
 732 /**
 733  * Get LOV EA.
 734  *
 735  * Fill lti_ea_store buffer in the environment with a value for the given
 736  * EA. The buffer is reallocated if the value doesn't fit.
 737  *
 738  * \param[in,out] env           execution environment for this thread
 739  *                              .lti_ea_store buffer is filled with EA's value
 740  * \param[in] lo                LOD object
 741  * \param[in] name              name of the EA
 742  *
 743  * \retval                      0 if EA is fetched successfully
 744  * \retval                      negative error number on failure
 745  */
 746 int lod_get_ea(const struct lu_env *env, struct lod_object *lo,
 747                const char *name)
 748 {
 749         struct lod_thread_info  *info = lod_env_info(env);
 750         struct dt_object        *next = dt_object_child(&lo->ldo_obj);
 751         int                     rc;
 752         ENTRY;
 753
 754         LASSERT(info);
 755
 756         if (unlikely(info->lti_ea_store == NULL)) {
 757                 /* just to enter in allocation block below */
 758                 rc = -ERANGE;
 759         } else {
 760 repeat:
 761                 info->lti_buf.lb_buf = info->lti_ea_store;
 762                 info->lti_buf.lb_len = info->lti_ea_store_size;
 763                 rc = dt_xattr_get(env, next, &info->lti_buf, name);
 764         }
 765
 766         /* if object is not striped or inaccessible */
 767         if (rc == -ENODATA || rc == -ENOENT)
 768                 RETURN(0);
 769
 770         if (rc == -ERANGE) {
 771                 /* EA doesn't fit, reallocate new buffer */
 772                 rc = dt_xattr_get(env, next, &LU_BUF_NULL, name);
 773                 if (rc == -ENODATA || rc == -ENOENT)
 774                         RETURN(0);
 775                 else if (rc < 0)
 776                         RETURN(rc);
 777
 778                 LASSERT(rc > 0);
 779                 rc = lod_ea_store_resize(info, rc);
 780                 if (rc)
 781                         RETURN(rc);
 782                 goto repeat;
 783         }
 784
 785         RETURN(rc);
 786 }
 787
 788 /**
 789  * Verify the target index is present in the current configuration.
 790  *
 791  * \param[in] md                LOD device where the target table is stored
 792  * \param[in] idx               target's index
 793  *
 794  * \retval                      0 if the index is present
 795  * \retval                      -EINVAL if not
 796  */
 797 static int validate_lod_and_idx(struct lod_device *md, __u32 idx)
 798 {
 799         if (unlikely(idx >= md->lod_ost_descs.ltd_tgts_size ||
 800                      !cfs_bitmap_check(md->lod_ost_bitmap, idx))) {
 801                 CERROR("%s: bad idx: %d of %d\n", lod2obd(md)->obd_name, idx,
 802                        md->lod_ost_descs.ltd_tgts_size);
 803                 return -EINVAL;
 804         }
 805
 806         if (unlikely(OST_TGT(md, idx) == NULL)) {
 807                 CERROR("%s: bad lod_tgt_desc for idx: %d\n",
 808                        lod2obd(md)->obd_name, idx);
 809                 return -EINVAL;
 810         }
 811
 812         if (unlikely(OST_TGT(md, idx)->ltd_ost == NULL)) {
 813                 CERROR("%s: invalid lod device, for idx: %d\n",
 814                        lod2obd(md)->obd_name , idx);
 815                 return -EINVAL;
 816         }
 817
 818         return 0;
 819 }
 820
 821 /**
 822  * Instantiate objects for stripes.
 823  *
 824  * Allocate and initialize LU-objects representing the stripes. The number
 825  * of the stripes (ldo_stripenr) must be initialized already. The caller
 826  * must ensure nobody else is calling the function on the object at the same
 827  * time. FLDB service must be running to be able to map a FID to the targets
 828  * and find appropriate device representing that target.
 829  *
 830  * \param[in] env               execution environment for this thread
 831  * \param[in,out] lo            LOD object
 832  * \param[in] objs              an array of IDs to creates the objects from
 833  *
 834  * \retval                      0 if the objects are instantiated successfully
 835  * \retval                      negative error number on failure
 836  */
 837 int lod_initialize_objects(const struct lu_env *env, struct lod_object *lo,
 838                            struct lov_ost_data_v1 *objs)
 839 {
 840         struct lod_thread_info  *info = lod_env_info(env);
 841         struct lod_device       *md;
 842         struct lu_object        *o, *n;
 843         struct lu_device        *nd;
 844         struct dt_object       **stripe;
 845         int                      stripe_len;
 846         int                      i, rc = 0;
 847         __u32                   idx;
 848         ENTRY;
 849
 850         LASSERT(lo != NULL);
 851         md = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
 852         LASSERT(lo->ldo_stripe == NULL);
 853         LASSERT(lo->ldo_stripenr > 0);
 854         LASSERT(lo->ldo_stripe_size > 0);
 855
 856         stripe_len = lo->ldo_stripenr;
 857         OBD_ALLOC(stripe, sizeof(stripe[0]) * stripe_len);
 858         if (stripe == NULL)
 859                 RETURN(-ENOMEM);
 860
 861         for (i = 0; i < lo->ldo_stripenr; i++) {
 862                 if (unlikely(lovea_slot_is_dummy(&objs[i])))
 863                         continue;
 864
 865                 ostid_le_to_cpu(&objs[i].l_ost_oi, &info->lti_ostid);
 866                 idx = le32_to_cpu(objs[i].l_ost_idx);
 867                 rc = ostid_to_fid(&info->lti_fid, &info->lti_ostid, idx);
 868                 if (rc != 0)
 869                         GOTO(out, rc);
 870                 LASSERTF(fid_is_sane(&info->lti_fid), ""DFID" insane!\n",
 871                          PFID(&info->lti_fid));
 872                 lod_getref(&md->lod_ost_descs);
 873
 874                 rc = validate_lod_and_idx(md, idx);
 875                 if (unlikely(rc != 0)) {
 876                         lod_putref(md, &md->lod_ost_descs);
 877                         GOTO(out, rc);
 878                 }
 879
 880                 nd = &OST_TGT(md,idx)->ltd_ost->dd_lu_dev;
 881                 lod_putref(md, &md->lod_ost_descs);
 882
 883                 /* In the function below, .hs_keycmp resolves to
 884                  * u_obj_hop_keycmp() */
 885                 /* coverity[overrun-buffer-val] */
 886                 o = lu_object_find_at(env, nd, &info->lti_fid, NULL);
 887                 if (IS_ERR(o))
 888                         GOTO(out, rc = PTR_ERR(o));
 889
 890                 n = lu_object_locate(o->lo_header, nd->ld_type);
 891                 LASSERT(n);
 892
 893                 stripe[i] = container_of(n, struct dt_object, do_lu);
 894         }
 895
 896 out:
 897         if (rc != 0) {
 898                 for (i = 0; i < stripe_len; i++)
 899                         if (stripe[i] != NULL)
 900                                 lu_object_put(env, &stripe[i]->do_lu);
 901
 902                 OBD_FREE(stripe, sizeof(stripe[0]) * stripe_len);
 903                 lo->ldo_stripenr = 0;
 904         } else {
 905                 lo->ldo_stripe = stripe;
 906                 lo->ldo_stripes_allocated = stripe_len;
 907         }
 908
 909         RETURN(rc);
 910 }
 911
 912 /**
 913  * Instantiate objects for striping.
 914  *
 915  * Parse striping information in \a buf and instantiate the objects
 916  * representing the stripes.
 917  *
 918  * \param[in] env               execution environment for this thread
 919  * \param[in] lo                LOD object
 920  * \param[in] buf               buffer storing LOV EA to parse
 921  *
 922  * \retval                      0 if parsing and objects creation succeed
 923  * \retval                      negative error number on failure
 924  */
 925 int lod_parse_striping(const struct lu_env *env, struct lod_object *lo,
 926                        const struct lu_buf *buf)
 927 {
 928         struct lov_mds_md_v1    *lmm;
 929         struct lov_ost_data_v1  *objs;
 930         __u32                    magic;
 931         __u32                    pattern;
 932         int                      rc = 0;
 933         ENTRY;
 934
 935         LASSERT(buf);
 936         LASSERT(buf->lb_buf);
 937         LASSERT(buf->lb_len);
 938
 939         lmm = (struct lov_mds_md_v1 *) buf->lb_buf;
 940         magic = le32_to_cpu(lmm->lmm_magic);
 941         pattern = le32_to_cpu(lmm->lmm_pattern);
 942
 943         if (magic != LOV_MAGIC_V1 && magic != LOV_MAGIC_V3)
 944                 GOTO(out, rc = -EINVAL);
 945         if (lov_pattern(pattern) != LOV_PATTERN_RAID0)
 946                 GOTO(out, rc = -EINVAL);
 947
 948         lo->ldo_pattern = pattern;
 949         lo->ldo_stripe_size = le32_to_cpu(lmm->lmm_stripe_size);
 950         lo->ldo_layout_gen = le16_to_cpu(lmm->lmm_layout_gen);
 951         lo->ldo_stripenr = le16_to_cpu(lmm->lmm_stripe_count);
 952         /* released file stripenr fixup. */
 953         if (pattern & LOV_PATTERN_F_RELEASED)
 954                 lo->ldo_stripenr = 0;
 955
 956         LASSERT(buf->lb_len >= lov_mds_md_size(lo->ldo_stripenr, magic));
 957
 958         if (magic == LOV_MAGIC_V3) {
 959                 struct lov_mds_md_v3 *v3 = (struct lov_mds_md_v3 *) lmm;
 960                 objs = &v3->lmm_objects[0];
 961                 lod_object_set_pool(lo, v3->lmm_pool_name);
 962         } else {
 963                 objs = &lmm->lmm_objects[0];
 964         }
 965
 966         if (lo->ldo_stripenr > 0)
 967                 rc = lod_initialize_objects(env, lo, objs);
 968
 969 out:
 970         RETURN(rc);
 971 }
 972
 973 /**
 974  * Initialize the object representing the stripes.
 975  *
 976  * Unless the stripes are initialized already, fetch LOV (for regular
 977  * objects) or LMV (for directory objects) EA and call lod_parse_striping()
 978  * to instantiate the objects representing the stripes.
 979  *
 980  * \param[in] env               execution environment for this thread
 981  * \param[in,out] lo            LOD object
 982  *
 983  * \retval                      0 if parsing and object creation succeed
 984  * \retval                      negative error number on failure
 985  */
 986 int lod_load_striping_locked(const struct lu_env *env, struct lod_object *lo)
 987 {
 988         struct lod_thread_info  *info = lod_env_info(env);
 989         struct lu_buf           *buf  = &info->lti_buf;
 990         struct dt_object        *next = dt_object_child(&lo->ldo_obj);
 991         int                      rc = 0;
 992         ENTRY;
 993
 994         /* already initialized? */
 995         if (lo->ldo_stripe != NULL)
 996                 GOTO(out, rc = 0);
 997
 998         if (!dt_object_exists(next))
 999                 GOTO(out, rc = 0);
1000
1001         /* Do not load stripe for slaves of striped dir */
1002         if (lo->ldo_dir_slave_stripe)
1003                 GOTO(out, rc = 0);
1004
1005         if (S_ISREG(lu_object_attr(lod2lu_obj(lo)))) {
1006                 rc = lod_get_lov_ea(env, lo);
1007                 if (rc <= 0)
1008                         GOTO(out, rc);
1009                 /*
1010                  * there is LOV EA (striping information) in this object
1011                  * let's parse it and create in-core objects for the stripes
1012                  */
1013                 buf->lb_buf = info->lti_ea_store;
1014                 buf->lb_len = info->lti_ea_store_size;
1015                 rc = lod_parse_striping(env, lo, buf);
1016         } else if (S_ISDIR(lu_object_attr(lod2lu_obj(lo)))) {
1017                 rc = lod_get_lmv_ea(env, lo);
1018                 if (rc < (typeof(rc))sizeof(struct lmv_mds_md_v1))
1019                         GOTO(out, rc = rc > 0 ? -EINVAL : rc);
1020
1021                 buf->lb_buf = info->lti_ea_store;
1022                 buf->lb_len = info->lti_ea_store_size;
1023                 if (rc == sizeof(struct lmv_mds_md_v1)) {
1024                         rc = lod_load_lmv_shards(env, lo, buf, true);
1025                         if (buf->lb_buf != info->lti_ea_store) {
1026                                 OBD_FREE_LARGE(info->lti_ea_store,
1027                                                info->lti_ea_store_size);
1028                                 info->lti_ea_store = buf->lb_buf;
1029                                 info->lti_ea_store_size = buf->lb_len;
1030                         }
1031
1032                         if (rc < 0)
1033                                 GOTO(out, rc);
1034                 }
1035
1036                 /*
1037                  * there is LOV EA (striping information) in this object
1038                  * let's parse it and create in-core objects for the stripes
1039                  */
1040                 rc = lod_parse_dir_striping(env, lo, buf);
1041         }
1042
1043         if (rc == 0)
1044                 lo->ldo_striping_cached = 1;
1045 out:
1046         RETURN(rc);
1047 }
1048
1049 /**
1050  * A generic function to initialize the stripe objects.
1051  *
1052  * A protected version of lod_load_striping_locked() - load the striping
1053  * information from storage, parse that and instantiate LU objects to
1054  * represent the stripes.  The LOD object \a lo supplies a pointer to the
1055  * next sub-object in the LU stack so we can lock it. Also use \a lo to
1056  * return an array of references to the newly instantiated objects.
1057  *
1058  * \param[in] env               execution environment for this thread
1059  * \param[in,out] lo            LOD object, where striping is stored and
1060  *                              which gets an array of references
1061  *
1062  * \retval                      0 if parsing and object creation succeed
1063  * \retval                      negative error number on failure
1064  **/
1065 int lod_load_striping(const struct lu_env *env, struct lod_object *lo)
1066 {
1067         struct dt_object        *next = dt_object_child(&lo->ldo_obj);
1068         int                     rc = 0;
1069
1070         /* currently this code is supposed to be called from declaration
1071          * phase only, thus the object is not expected to be locked by caller */
1072         dt_write_lock(env, next, 0);
1073         rc = lod_load_striping_locked(env, lo);
1074         dt_write_unlock(env, next);
1075         return rc;
1076 }
1077
1078 /**
1079  * Verify striping.
1080  *
1081  * Check the validity of all fields including the magic, stripe size,
1082  * stripe count, stripe offset and that the pool is present.  Also check
1083  * that each target index points to an existing target. The additional
1084  * \a is_from_disk turns additional checks. In some cases zero fields
1085  * are allowed (like pattern=0).
1086  *
1087  * \param[in] d                 LOD device
1088  * \param[in] buf               buffer with LOV EA to verify
1089  * \param[in] is_from_disk      0 - from user, allow some fields to be 0
1090  *                              1 - from disk, do not allow
1091  *
1092  * \retval                      0 if the striping is valid
1093  * \retval                      -EINVAL if striping is invalid
1094  */
1095 int lod_verify_striping(struct lod_device *d, const struct lu_buf *buf,
1096                         bool is_from_disk)
1097 {
1098         struct lov_user_md_v1   *lum;
1099         struct lov_user_md_v3   *lum3;
1100         struct pool_desc        *pool = NULL;
1101         __u32                    magic;
1102         __u32                    stripe_size;
1103         __u16                    stripe_count;
1104         __u16                    stripe_offset;
1105         size_t                   lum_size;
1106         int                      rc = 0;
1107         ENTRY;
1108
1109         lum = buf->lb_buf;
1110
1111         LASSERT(sizeof(*lum) < sizeof(*lum3));
1112
1113         if (buf->lb_len < sizeof(*lum)) {
1114                 CDEBUG(D_IOCTL, "buf len %zu too small for lov_user_md\n",
1115                        buf->lb_len);
1116                 GOTO(out, rc = -EINVAL);
1117         }
1118
1119         magic = le32_to_cpu(lum->lmm_magic);
1120         if (magic != LOV_USER_MAGIC_V1 &&
1121             magic != LOV_USER_MAGIC_V3 &&
1122             magic != LOV_MAGIC_V1_DEF &&
1123             magic != LOV_MAGIC_V3_DEF) {
1124                 CDEBUG(D_IOCTL, "bad userland LOV MAGIC: %#x\n", magic);
1125                 GOTO(out, rc = -EINVAL);
1126         }
1127
1128         /* the user uses "0" for default stripe pattern normally. */
1129         if (!is_from_disk && lum->lmm_pattern == 0)
1130                 lum->lmm_pattern = cpu_to_le32(LOV_PATTERN_RAID0);
1131
1132         if (le32_to_cpu(lum->lmm_pattern) != LOV_PATTERN_RAID0) {
1133                 CDEBUG(D_IOCTL, "bad userland stripe pattern: %#x\n",
1134                        le32_to_cpu(lum->lmm_pattern));
1135                 GOTO(out, rc = -EINVAL);
1136         }
1137
1138         /* 64kB is the largest common page size we see (ia64), and matches the
1139          * check in lfs */
1140         stripe_size = le32_to_cpu(lum->lmm_stripe_size);
1141         if (stripe_size & (LOV_MIN_STRIPE_SIZE - 1)) {
1142                 CDEBUG(D_IOCTL, "stripe size %u not a multiple of %u\n",
1143                        stripe_size, LOV_MIN_STRIPE_SIZE);
1144                 GOTO(out, rc = -EINVAL);
1145         }
1146
1147         stripe_offset = le16_to_cpu(lum->lmm_stripe_offset);
1148         if (stripe_offset != LOV_OFFSET_DEFAULT) {
1149                 /* if offset is not within valid range [0, osts_size) */
1150                 if (stripe_offset >= d->lod_osts_size) {
1151                         CDEBUG(D_IOCTL, "stripe offset %u >= bitmap size %u\n",
1152                                stripe_offset, d->lod_osts_size);
1153                         GOTO(out, rc = -EINVAL);
1154                 }
1155
1156                 /* if lmm_stripe_offset is *not* in bitmap */
1157                 if (!cfs_bitmap_check(d->lod_ost_bitmap, stripe_offset)) {
1158                         CDEBUG(D_IOCTL, "stripe offset %u not in bitmap\n",
1159                                stripe_offset);
1160                         GOTO(out, rc = -EINVAL);
1161                 }
1162         }
1163
1164         if (magic == LOV_USER_MAGIC_V1 || magic == LOV_MAGIC_V1_DEF)
1165                 lum_size = offsetof(struct lov_user_md_v1,
1166                                     lmm_objects[0]);
1167         else if (magic == LOV_USER_MAGIC_V3 || magic == LOV_MAGIC_V3_DEF)
1168                 lum_size = offsetof(struct lov_user_md_v3,
1169                                     lmm_objects[0]);
1170         else
1171                 GOTO(out, rc = -EINVAL);
1172
1173         stripe_count = le16_to_cpu(lum->lmm_stripe_count);
1174         if (buf->lb_len != lum_size) {
1175                 CDEBUG(D_IOCTL, "invalid buf len %zu for lov_user_md with "
1176                        "magic %#x and stripe_count %u\n",
1177                        buf->lb_len, magic, stripe_count);
1178                 GOTO(out, rc = -EINVAL);
1179         }
1180
1181         if (!(magic == LOV_USER_MAGIC_V3 || magic == LOV_MAGIC_V3_DEF))
1182                 goto out;
1183
1184         lum3 = buf->lb_buf;
1185         if (buf->lb_len < sizeof(*lum3)) {
1186                 CDEBUG(D_IOCTL, "buf len %zu too small for lov_user_md_v3\n",
1187                        buf->lb_len);
1188                 GOTO(out, rc = -EINVAL);
1189         }
1190
1191         /* In the function below, .hs_keycmp resolves to
1192          * pool_hashkey_keycmp() */
1193         /* coverity[overrun-buffer-val] */
1194         pool = lod_find_pool(d, lum3->lmm_pool_name);
1195         if (pool == NULL)
1196                 goto out;
1197
1198         if (stripe_offset != LOV_OFFSET_DEFAULT) {
1199                 rc = lod_check_index_in_pool(stripe_offset, pool);
1200                 if (rc < 0)
1201                         GOTO(out, rc = -EINVAL);
1202         }
1203
1204         if (is_from_disk && stripe_count > pool_tgt_count(pool)) {
1205                 CDEBUG(D_IOCTL,
1206                        "stripe count %u > # OSTs %u in the pool\n",
1207                        stripe_count, pool_tgt_count(pool));
1208                 GOTO(out, rc = -EINVAL);
1209         }
1210
1211 out:
1212         if (pool != NULL)
1213                 lod_pool_putref(pool);
1214
1215         RETURN(rc);
1216 }
1217
1218 void lod_fix_desc_stripe_size(__u64 *val)
1219 {
1220         if (*val < LOV_MIN_STRIPE_SIZE) {
1221                 if (*val != 0)
1222                         LCONSOLE_INFO("Increasing default stripe size to "
1223                                       "minimum value %u\n",
1224                                       LOV_DESC_STRIPE_SIZE_DEFAULT);
1225                 *val = LOV_DESC_STRIPE_SIZE_DEFAULT;
1226         } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) {
1227                 *val &= ~(LOV_MIN_STRIPE_SIZE - 1);
1228                 LCONSOLE_WARN("Changing default stripe size to "LPU64" (a "
1229                               "multiple of %u)\n",
1230                               *val, LOV_MIN_STRIPE_SIZE);
1231         }
1232 }
1233
1234 void lod_fix_desc_stripe_count(__u32 *val)
1235 {
1236         if (*val == 0)
1237                 *val = 1;
1238 }
1239
1240 void lod_fix_desc_pattern(__u32 *val)
1241 {
1242         /* from lov_setstripe */
1243         if ((*val != 0) && (*val != LOV_PATTERN_RAID0)) {
1244                 LCONSOLE_WARN("Unknown stripe pattern: %#x\n", *val);
1245                 *val = 0;
1246         }
1247 }
1248
1249 void lod_fix_desc_qos_maxage(__u32 *val)
1250 {
1251         /* fix qos_maxage */
1252         if (*val == 0)
1253                 *val = LOV_DESC_QOS_MAXAGE_DEFAULT;
1254 }
1255
1256 /**
1257  * Used to fix insane default striping.
1258  *
1259  * \param[in] desc      striping description
1260  */
1261 void lod_fix_desc(struct lov_desc *desc)
1262 {
1263         lod_fix_desc_stripe_size(&desc->ld_default_stripe_size);
1264         lod_fix_desc_stripe_count(&desc->ld_default_stripe_count);
1265         lod_fix_desc_pattern(&desc->ld_pattern);
1266         lod_fix_desc_qos_maxage(&desc->ld_qos_maxage);
1267 }
1268
1269 /**
1270  * Initialize the structures used to store pools and default striping.
1271  *
1272  * \param[in] lod       LOD device
1273  * \param[in] lcfg      configuration structure storing default striping.
1274  *
1275  * \retval              0 if initialization succeeds
1276  * \retval              negative error number on failure
1277  */
1278 int lod_pools_init(struct lod_device *lod, struct lustre_cfg *lcfg)
1279 {
1280         struct obd_device          *obd;
1281         struct lov_desc            *desc;
1282         int                         rc;
1283         ENTRY;
1284
1285         obd = class_name2obd(lustre_cfg_string(lcfg, 0));
1286         LASSERT(obd != NULL);
1287         obd->obd_lu_dev = &lod->lod_dt_dev.dd_lu_dev;
1288
1289         if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
1290                 CERROR("LOD setup requires a descriptor\n");
1291                 RETURN(-EINVAL);
1292         }
1293
1294         desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
1295
1296         if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
1297                 CERROR("descriptor size wrong: %d > %d\n",
1298                        (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
1299                 RETURN(-EINVAL);
1300         }
1301
1302         if (desc->ld_magic != LOV_DESC_MAGIC) {
1303                 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
1304                         CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
1305                                obd->obd_name, desc);
1306                         lustre_swab_lov_desc(desc);
1307                 } else {
1308                         CERROR("%s: Bad lov desc magic: %#x\n",
1309                                obd->obd_name, desc->ld_magic);
1310                         RETURN(-EINVAL);
1311                 }
1312         }
1313
1314         lod_fix_desc(desc);
1315
1316         desc->ld_active_tgt_count = 0;
1317         lod->lod_desc = *desc;
1318
1319         lod->lod_sp_me = LUSTRE_SP_CLI;
1320
1321         /* Set up allocation policy (QoS and RR) */
1322         INIT_LIST_HEAD(&lod->lod_qos.lq_oss_list);
1323         init_rwsem(&lod->lod_qos.lq_rw_sem);
1324         lod->lod_qos.lq_dirty = 1;
1325         lod->lod_qos.lq_rr.lqr_dirty = 1;
1326         lod->lod_qos.lq_reset = 1;
1327         /* Default priority is toward free space balance */
1328         lod->lod_qos.lq_prio_free = 232;
1329         /* Default threshold for rr (roughly 17%) */
1330         lod->lod_qos.lq_threshold_rr = 43;
1331
1332         /* Set up OST pool environment */
1333         lod->lod_pools_hash_body = cfs_hash_create("POOLS", HASH_POOLS_CUR_BITS,
1334                                                    HASH_POOLS_MAX_BITS,
1335                                                    HASH_POOLS_BKT_BITS, 0,
1336                                                    CFS_HASH_MIN_THETA,
1337                                                    CFS_HASH_MAX_THETA,
1338                                                    &pool_hash_operations,
1339                                                    CFS_HASH_DEFAULT);
1340         if (lod->lod_pools_hash_body == NULL)
1341                 RETURN(-ENOMEM);
1342
1343         INIT_LIST_HEAD(&lod->lod_pool_list);
1344         lod->lod_pool_count = 0;
1345         rc = lod_ost_pool_init(&lod->lod_pool_info, 0);
1346         if (rc)
1347                 GOTO(out_hash, rc);
1348         lod_qos_rr_init(&lod->lod_qos.lq_rr);
1349         rc = lod_ost_pool_init(&lod->lod_qos.lq_rr.lqr_pool, 0);
1350         if (rc)
1351                 GOTO(out_pool_info, rc);
1352
1353         RETURN(0);
1354
1355 out_pool_info:
1356         lod_ost_pool_free(&lod->lod_pool_info);
1357 out_hash:
1358         cfs_hash_putref(lod->lod_pools_hash_body);
1359
1360         return rc;
1361 }
1362
1363 /**
1364  * Release the structures describing the pools.
1365  *
1366  * \param[in] lod       LOD device from which we release the structures
1367  *
1368  * \retval              0 always
1369  */
1370 int lod_pools_fini(struct lod_device *lod)
1371 {
1372         struct obd_device   *obd = lod2obd(lod);
1373         struct pool_desc    *pool, *tmp;
1374         ENTRY;
1375
1376         list_for_each_entry_safe(pool, tmp, &lod->lod_pool_list, pool_list) {
1377                 /* free pool structs */
1378                 CDEBUG(D_INFO, "delete pool %p\n", pool);
1379                 /* In the function below, .hs_keycmp resolves to
1380                  * pool_hashkey_keycmp() */
1381                 /* coverity[overrun-buffer-val] */
1382                 lod_pool_del(obd, pool->pool_name);
1383         }
1384
1385         cfs_hash_putref(lod->lod_pools_hash_body);
1386         lod_ost_pool_free(&(lod->lod_qos.lq_rr.lqr_pool));
1387         lod_ost_pool_free(&lod->lod_pool_info);
1388
1389         RETURN(0);
1390 }