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, 2016, 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 #include <lustre_swab.h>
  43
  44 #include "lod_internal.h"
  45
  46 /**
  47  * Increase reference count on the target table.
  48  *
  49  * Increase reference count on the target table usage to prevent racing with
  50  * addition/deletion. Any function that expects the table to remain
  51  * stationary must take a ref.
  52  *
  53  * \param[in] ltd       target table (lod_ost_descs or lod_mdt_descs)
  54  */
  55 void lod_getref(struct lod_tgt_descs *ltd)
  56 {
  57         down_read(&ltd->ltd_rw_sem);
  58         mutex_lock(&ltd->ltd_mutex);
  59         ltd->ltd_refcount++;
  60         mutex_unlock(&ltd->ltd_mutex);
  61 }
  62
  63 /**
  64  * Decrease reference count on the target table.
  65  *
  66  * Companion of lod_getref() to release a reference on the target table.
  67  * If this is the last reference and the OST entry was scheduled for deletion,
  68  * the descriptor is removed from the table.
  69  *
  70  * \param[in] lod       LOD device from which we release a reference
  71  * \param[in] ltd       target table (lod_ost_descs or lod_mdt_descs)
  72  */
  73 void lod_putref(struct lod_device *lod, struct lod_tgt_descs *ltd)
  74 {
  75         mutex_lock(&ltd->ltd_mutex);
  76         ltd->ltd_refcount--;
  77         if (ltd->ltd_refcount == 0 && ltd->ltd_death_row) {
  78                 struct lod_tgt_desc *tgt_desc, *tmp;
  79                 struct list_head kill;
  80                 unsigned int idx;
  81
  82                 CDEBUG(D_CONFIG, "destroying %d ltd desc\n",
  83                        ltd->ltd_death_row);
  84
  85                 INIT_LIST_HEAD(&kill);
  86
  87                 cfs_foreach_bit(ltd->ltd_tgt_bitmap, idx) {
  88                         tgt_desc = LTD_TGT(ltd, idx);
  89                         LASSERT(tgt_desc);
  90
  91                         if (!tgt_desc->ltd_reap)
  92                                 continue;
  93
  94                         list_add(&tgt_desc->ltd_kill, &kill);
  95                         LTD_TGT(ltd, idx) = NULL;
  96                         /*FIXME: only support ost pool for now */
  97                         if (ltd == &lod->lod_ost_descs) {
  98                                 lod_ost_pool_remove(&lod->lod_pool_info, idx);
  99                                 if (tgt_desc->ltd_active)
 100                                         lod->lod_desc.ld_active_tgt_count--;
 101                         }
 102                         ltd->ltd_tgtnr--;
 103                         cfs_bitmap_clear(ltd->ltd_tgt_bitmap, idx);
 104                         ltd->ltd_death_row--;
 105                 }
 106                 mutex_unlock(&ltd->ltd_mutex);
 107                 up_read(&ltd->ltd_rw_sem);
 108
 109                 list_for_each_entry_safe(tgt_desc, tmp, &kill, ltd_kill) {
 110                         int rc;
 111                         list_del(&tgt_desc->ltd_kill);
 112                         if (ltd == &lod->lod_ost_descs) {
 113                                 /* remove from QoS structures */
 114                                 rc = qos_del_tgt(lod, tgt_desc);
 115                                 if (rc)
 116                                         CERROR("%s: qos_del_tgt(%s) failed:"
 117                                                "rc = %d\n",
 118                                                lod2obd(lod)->obd_name,
 119                                               obd_uuid2str(&tgt_desc->ltd_uuid),
 120                                                rc);
 121                         }
 122                         rc = obd_disconnect(tgt_desc->ltd_exp);
 123                         if (rc)
 124                                 CERROR("%s: failed to disconnect %s: rc = %d\n",
 125                                        lod2obd(lod)->obd_name,
 126                                        obd_uuid2str(&tgt_desc->ltd_uuid), rc);
 127                         OBD_FREE_PTR(tgt_desc);
 128                 }
 129         } else {
 130                 mutex_unlock(&ltd->ltd_mutex);
 131                 up_read(&ltd->ltd_rw_sem);
 132         }
 133 }
 134
 135 /**
 136  * Expand size of target table.
 137  *
 138  * When the target table is full, we have to extend the table. To do so,
 139  * we allocate new memory with some reserve, move data from the old table
 140  * to the new one and release memory consumed by the old table.
 141  * Notice we take ltd_rw_sem exclusively to ensure atomic switch.
 142  *
 143  * \param[in] ltd               target table
 144  * \param[in] newsize           new size of the table
 145  *
 146  * \retval                      0 on success
 147  * \retval                      -ENOMEM if reallocation failed
 148  */
 149 static int ltd_bitmap_resize(struct lod_tgt_descs *ltd, __u32 newsize)
 150 {
 151         struct cfs_bitmap *new_bitmap, *old_bitmap = NULL;
 152         int           rc = 0;
 153         ENTRY;
 154
 155         /* grab write reference on the lod. Relocating the array requires
 156          * exclusive access */
 157
 158         down_write(&ltd->ltd_rw_sem);
 159         if (newsize <= ltd->ltd_tgts_size)
 160                 /* someone else has already resize the array */
 161                 GOTO(out, rc = 0);
 162
 163         /* allocate new bitmap */
 164         new_bitmap = CFS_ALLOCATE_BITMAP(newsize);
 165         if (!new_bitmap)
 166                 GOTO(out, rc = -ENOMEM);
 167
 168         if (ltd->ltd_tgts_size > 0) {
 169                 /* the bitmap already exists, we need
 170                  * to copy data from old one */
 171                 cfs_bitmap_copy(new_bitmap, ltd->ltd_tgt_bitmap);
 172                 old_bitmap = ltd->ltd_tgt_bitmap;
 173         }
 174
 175         ltd->ltd_tgts_size  = newsize;
 176         ltd->ltd_tgt_bitmap = new_bitmap;
 177
 178         if (old_bitmap)
 179                 CFS_FREE_BITMAP(old_bitmap);
 180
 181         CDEBUG(D_CONFIG, "tgt size: %d\n", ltd->ltd_tgts_size);
 182
 183         EXIT;
 184 out:
 185         up_write(&ltd->ltd_rw_sem);
 186         return rc;
 187 }
 188
 189 /**
 190  * Connect LOD to a new OSP and add it to the target table.
 191  *
 192  * Connect to the OSP device passed, initialize all the internal
 193  * structures related to the device and add it to the target table.
 194  *
 195  * \param[in] env               execution environment for this thread
 196  * \param[in] lod               LOD device to be connected to the new OSP
 197  * \param[in] osp               name of OSP device name to be added
 198  * \param[in] index             index of the new target
 199  * \param[in] gen               target's generation number
 200  * \param[in] tgt_index         OSP's group
 201  * \param[in] type              type of device (mdc or osc)
 202  * \param[in] active            state of OSP: 0 - inactive, 1 - active
 203  *
 204  * \retval                      0 if added successfully
 205  * \retval                      negative error number on failure
 206  */
 207 int lod_add_device(const struct lu_env *env, struct lod_device *lod,
 208                    char *osp, unsigned index, unsigned gen, int tgt_index,
 209                    char *type, int active)
 210 {
 211         struct obd_connect_data *data = NULL;
 212         struct obd_export       *exp = NULL;
 213         struct obd_device       *obd;
 214         struct lu_device        *lu_dev;
 215         struct dt_device        *dt_dev;
 216         int                      rc;
 217         struct lod_tgt_desc     *tgt_desc;
 218         struct lod_tgt_descs    *ltd;
 219         struct lustre_cfg       *lcfg;
 220         struct obd_uuid         obd_uuid;
 221         bool                    for_ost;
 222         bool lock = false;
 223         ENTRY;
 224
 225         CDEBUG(D_CONFIG, "osp:%s idx:%d gen:%d\n", osp, index, gen);
 226
 227         if (gen <= 0) {
 228                 CERROR("request to add OBD %s with invalid generation: %d\n",
 229                        osp, gen);
 230                 RETURN(-EINVAL);
 231         }
 232
 233         obd_str2uuid(&obd_uuid, osp);
 234
 235         obd = class_find_client_obd(&obd_uuid, LUSTRE_OSP_NAME,
 236                                 &lod->lod_dt_dev.dd_lu_dev.ld_obd->obd_uuid);
 237         if (obd == NULL) {
 238                 CERROR("can't find %s device\n", osp);
 239                 RETURN(-EINVAL);
 240         }
 241
 242         LASSERT(obd->obd_lu_dev != NULL);
 243         LASSERT(obd->obd_lu_dev->ld_site == lod->lod_dt_dev.dd_lu_dev.ld_site);
 244
 245         lu_dev = obd->obd_lu_dev;
 246         dt_dev = lu2dt_dev(lu_dev);
 247
 248         OBD_ALLOC_PTR(data);
 249         if (data == NULL)
 250                 GOTO(out_cleanup, rc = -ENOMEM);
 251
 252         data->ocd_connect_flags = OBD_CONNECT_INDEX | OBD_CONNECT_VERSION;
 253         data->ocd_version = LUSTRE_VERSION_CODE;
 254         data->ocd_index = index;
 255
 256         if (strcmp(LUSTRE_OSC_NAME, type) == 0) {
 257                 for_ost = true;
 258                 data->ocd_connect_flags |= OBD_CONNECT_AT |
 259                                            OBD_CONNECT_FULL20 |
 260                                            OBD_CONNECT_INDEX |
 261 #ifdef HAVE_LRU_RESIZE_SUPPORT
 262                                            OBD_CONNECT_LRU_RESIZE |
 263 #endif
 264                                            OBD_CONNECT_MDS |
 265                                            OBD_CONNECT_REQPORTAL |
 266                                            OBD_CONNECT_SKIP_ORPHAN |
 267                                            OBD_CONNECT_FID |
 268                                            OBD_CONNECT_LVB_TYPE |
 269                                            OBD_CONNECT_VERSION |
 270                                            OBD_CONNECT_PINGLESS |
 271                                            OBD_CONNECT_LFSCK |
 272                                            OBD_CONNECT_BULK_MBITS;
 273
 274                 data->ocd_group = tgt_index;
 275                 ltd = &lod->lod_ost_descs;
 276         } else {
 277                 struct obd_import *imp = obd->u.cli.cl_import;
 278
 279                 for_ost = false;
 280                 data->ocd_ibits_known = MDS_INODELOCK_UPDATE;
 281                 data->ocd_connect_flags |= OBD_CONNECT_ACL |
 282                                            OBD_CONNECT_IBITS |
 283                                            OBD_CONNECT_MDS_MDS |
 284                                            OBD_CONNECT_FID |
 285                                            OBD_CONNECT_AT |
 286                                            OBD_CONNECT_FULL20 |
 287                                            OBD_CONNECT_LFSCK |
 288                                            OBD_CONNECT_BULK_MBITS;
 289                 spin_lock(&imp->imp_lock);
 290                 imp->imp_server_timeout = 1;
 291                 spin_unlock(&imp->imp_lock);
 292                 imp->imp_client->cli_request_portal = OUT_PORTAL;
 293                 CDEBUG(D_OTHER, "%s: Set 'mds' portal and timeout\n",
 294                       obd->obd_name);
 295                 ltd = &lod->lod_mdt_descs;
 296         }
 297
 298         rc = obd_connect(env, &exp, obd, &obd->obd_uuid, data, NULL);
 299         OBD_FREE_PTR(data);
 300         if (rc) {
 301                 CERROR("%s: cannot connect to next dev %s (%d)\n",
 302                        obd->obd_name, osp, rc);
 303                 GOTO(out_cleanup, rc);
 304         }
 305
 306         /* Allocate ost descriptor and fill it */
 307         OBD_ALLOC_PTR(tgt_desc);
 308         if (!tgt_desc)
 309                 GOTO(out_conn, rc = -ENOMEM);
 310
 311         tgt_desc->ltd_tgt    = dt_dev;
 312         tgt_desc->ltd_exp    = exp;
 313         tgt_desc->ltd_uuid   = obd->u.cli.cl_target_uuid;
 314         tgt_desc->ltd_gen    = gen;
 315         tgt_desc->ltd_index  = index;
 316         tgt_desc->ltd_active = active;
 317
 318         lod_getref(ltd);
 319         if (index >= ltd->ltd_tgts_size) {
 320                 /* we have to increase the size of the lod_osts array */
 321                 __u32  newsize;
 322
 323                 newsize = max(ltd->ltd_tgts_size, (__u32)2);
 324                 while (newsize < index + 1)
 325                         newsize = newsize << 1;
 326
 327                 /* lod_bitmap_resize() needs lod_rw_sem
 328                  * which we hold with th reference */
 329                 lod_putref(lod, ltd);
 330
 331                 rc = ltd_bitmap_resize(ltd, newsize);
 332                 if (rc)
 333                         GOTO(out_desc, rc);
 334
 335                 lod_getref(ltd);
 336         }
 337
 338         mutex_lock(&ltd->ltd_mutex);
 339         lock = true;
 340         if (cfs_bitmap_check(ltd->ltd_tgt_bitmap, index)) {
 341                 CERROR("%s: device %d is registered already\n", obd->obd_name,
 342                        index);
 343                 GOTO(out_mutex, rc = -EEXIST);
 344         }
 345
 346         if (ltd->ltd_tgt_idx[index / TGT_PTRS_PER_BLOCK] == NULL) {
 347                 OBD_ALLOC_PTR(ltd->ltd_tgt_idx[index / TGT_PTRS_PER_BLOCK]);
 348                 if (ltd->ltd_tgt_idx[index / TGT_PTRS_PER_BLOCK] == NULL) {
 349                         CERROR("can't allocate index to add %s\n",
 350                                obd->obd_name);
 351                         GOTO(out_mutex, rc = -ENOMEM);
 352                 }
 353         }
 354
 355         if (for_ost) {
 356                 /* pool and qos are not supported for MDS stack yet */
 357                 rc = lod_ost_pool_add(&lod->lod_pool_info, index,
 358                                       lod->lod_osts_size);
 359                 if (rc) {
 360                         CERROR("%s: can't set up pool, failed with %d\n",
 361                                obd->obd_name, rc);
 362                         GOTO(out_mutex, rc);
 363                 }
 364
 365                 rc = qos_add_tgt(lod, tgt_desc);
 366                 if (rc) {
 367                         CERROR("%s: qos_add_tgt failed with %d\n",
 368                                 obd->obd_name, rc);
 369                         GOTO(out_pool, rc);
 370                 }
 371
 372                 /* The new OST is now a full citizen */
 373                 if (index >= lod->lod_desc.ld_tgt_count)
 374                         lod->lod_desc.ld_tgt_count = index + 1;
 375                 if (active)
 376                         lod->lod_desc.ld_active_tgt_count++;
 377         }
 378
 379         LTD_TGT(ltd, index) = tgt_desc;
 380         cfs_bitmap_set(ltd->ltd_tgt_bitmap, index);
 381         ltd->ltd_tgtnr++;
 382         mutex_unlock(&ltd->ltd_mutex);
 383         lod_putref(lod, ltd);
 384         lock = false;
 385         if (lod->lod_recovery_completed)
 386                 lu_dev->ld_ops->ldo_recovery_complete(env, lu_dev);
 387
 388         if (!for_ost && lod->lod_initialized) {
 389                 rc = lod_sub_init_llog(env, lod, tgt_desc->ltd_tgt);
 390                 if (rc != 0) {
 391                         CERROR("%s: cannot start llog on %s:rc = %d\n",
 392                                lod2obd(lod)->obd_name, osp, rc);
 393                         GOTO(out_ltd, rc);
 394                 }
 395         }
 396
 397         rc = lfsck_add_target(env, lod->lod_child, dt_dev, exp, index, for_ost);
 398         if (rc != 0) {
 399                 CERROR("Fail to add LFSCK target: name = %s, type = %s, "
 400                        "index = %u, rc = %d\n", osp, type, index, rc);
 401                 GOTO(out_fini_llog, rc);
 402         }
 403         RETURN(rc);
 404 out_fini_llog:
 405         lod_sub_fini_llog(env, tgt_desc->ltd_tgt,
 406                           tgt_desc->ltd_recovery_thread);
 407 out_ltd:
 408         lod_getref(ltd);
 409         mutex_lock(&ltd->ltd_mutex);
 410         lock = true;
 411         if (!for_ost && LTD_TGT(ltd, index)->ltd_recovery_thread != NULL) {
 412                 struct ptlrpc_thread *thread;
 413
 414                 thread = LTD_TGT(ltd, index)->ltd_recovery_thread;
 415                 OBD_FREE_PTR(thread);
 416         }
 417         ltd->ltd_tgtnr--;
 418         cfs_bitmap_clear(ltd->ltd_tgt_bitmap, index);
 419         LTD_TGT(ltd, index) = NULL;
 420 out_pool:
 421         lod_ost_pool_remove(&lod->lod_pool_info, index);
 422 out_mutex:
 423         if (lock) {
 424                 mutex_unlock(&ltd->ltd_mutex);
 425                 lod_putref(lod, ltd);
 426         }
 427 out_desc:
 428         OBD_FREE_PTR(tgt_desc);
 429 out_conn:
 430         obd_disconnect(exp);
 431 out_cleanup:
 432         /* XXX OSP needs us to send down LCFG_CLEANUP because it uses
 433          * objects from the MDT stack. See LU-7184. */
 434         lcfg = &lod_env_info(env)->lti_lustre_cfg;
 435         memset(lcfg, 0, sizeof(*lcfg));
 436         lcfg->lcfg_version = LUSTRE_CFG_VERSION;
 437         lcfg->lcfg_command = LCFG_CLEANUP;
 438         lu_dev->ld_ops->ldo_process_config(env, lu_dev, lcfg);
 439
 440         return rc;
 441 }
 442
 443 /**
 444  * Schedule target removal from the target table.
 445  *
 446  * Mark the device as dead. The device is not removed here because it may
 447  * still be in use. The device will be removed in lod_putref() when the
 448  * last reference is released.
 449  *
 450  * \param[in] env               execution environment for this thread
 451  * \param[in] lod               LOD device the target table belongs to
 452  * \param[in] ltd               target table
 453  * \param[in] idx               index of the target
 454  * \param[in] for_ost           type of the target: 0 - MDT, 1 - OST
 455  */
 456 static void __lod_del_device(const struct lu_env *env, struct lod_device *lod,
 457                              struct lod_tgt_descs *ltd, unsigned idx,
 458                              bool for_ost)
 459 {
 460         LASSERT(LTD_TGT(ltd, idx));
 461
 462         lfsck_del_target(env, lod->lod_child, LTD_TGT(ltd, idx)->ltd_tgt,
 463                          idx, for_ost);
 464
 465         if (!for_ost && LTD_TGT(ltd, idx)->ltd_recovery_thread != NULL) {
 466                 struct ptlrpc_thread *thread;
 467
 468                 thread = LTD_TGT(ltd, idx)->ltd_recovery_thread;
 469                 OBD_FREE_PTR(thread);
 470         }
 471
 472         if (LTD_TGT(ltd, idx)->ltd_reap == 0) {
 473                 LTD_TGT(ltd, idx)->ltd_reap = 1;
 474                 ltd->ltd_death_row++;
 475         }
 476 }
 477
 478 /**
 479  * Schedule removal of all the targets from the given target table.
 480  *
 481  * See more details in the description for __lod_del_device()
 482  *
 483  * \param[in] env               execution environment for this thread
 484  * \param[in] lod               LOD device the target table belongs to
 485  * \param[in] ltd               target table
 486  * \param[in] for_ost           type of the target: MDT or OST
 487  *
 488  * \retval                      0 always
 489  */
 490 int lod_fini_tgt(const struct lu_env *env, struct lod_device *lod,
 491                  struct lod_tgt_descs *ltd, bool for_ost)
 492 {
 493         unsigned int idx;
 494
 495         if (ltd->ltd_tgts_size <= 0)
 496                 return 0;
 497         lod_getref(ltd);
 498         mutex_lock(&ltd->ltd_mutex);
 499         cfs_foreach_bit(ltd->ltd_tgt_bitmap, idx)
 500                 __lod_del_device(env, lod, ltd, idx, for_ost);
 501         mutex_unlock(&ltd->ltd_mutex);
 502         lod_putref(lod, ltd);
 503         CFS_FREE_BITMAP(ltd->ltd_tgt_bitmap);
 504         for (idx = 0; idx < TGT_PTRS; idx++) {
 505                 if (ltd->ltd_tgt_idx[idx])
 506                         OBD_FREE_PTR(ltd->ltd_tgt_idx[idx]);
 507         }
 508         ltd->ltd_tgts_size = 0;
 509         return 0;
 510 }
 511
 512 /**
 513  * Remove device by name.
 514  *
 515  * Remove a device identified by \a osp from the target table. Given
 516  * the device can be in use, the real deletion happens in lod_putref().
 517  *
 518  * \param[in] env               execution environment for this thread
 519  * \param[in] lod               LOD device to be connected to the new OSP
 520  * \param[in] ltd               target table
 521  * \param[in] osp               name of OSP device to be removed
 522  * \param[in] idx               index of the target
 523  * \param[in] gen               generation number, not used currently
 524  * \param[in] for_ost           type of the target: 0 - MDT, 1 - OST
 525  *
 526  * \retval                      0 if the device was scheduled for removal
 527  * \retval                      -EINVAL if no device was found
 528  */
 529 int lod_del_device(const struct lu_env *env, struct lod_device *lod,
 530                    struct lod_tgt_descs *ltd, char *osp, unsigned idx,
 531                    unsigned gen, bool for_ost)
 532 {
 533         struct obd_device *obd;
 534         int                rc = 0;
 535         struct obd_uuid    uuid;
 536         ENTRY;
 537
 538         CDEBUG(D_CONFIG, "osp:%s idx:%d gen:%d\n", osp, idx, gen);
 539
 540         obd_str2uuid(&uuid, osp);
 541
 542         obd = class_find_client_obd(&uuid, LUSTRE_OSP_NAME,
 543                                    &lod->lod_dt_dev.dd_lu_dev.ld_obd->obd_uuid);
 544         if (obd == NULL) {
 545                 CERROR("can't find %s device\n", osp);
 546                 RETURN(-EINVAL);
 547         }
 548
 549         if (gen <= 0) {
 550                 CERROR("%s: request to remove OBD %s with invalid generation %d"
 551                        "\n", obd->obd_name, osp, gen);
 552                 RETURN(-EINVAL);
 553         }
 554
 555         obd_str2uuid(&uuid,  osp);
 556
 557         lod_getref(ltd);
 558         mutex_lock(&ltd->ltd_mutex);
 559         /* check that the index is allocated in the bitmap */
 560         if (!cfs_bitmap_check(ltd->ltd_tgt_bitmap, idx) ||
 561             !LTD_TGT(ltd, idx)) {
 562                 CERROR("%s: device %d is not set up\n", obd->obd_name, idx);
 563                 GOTO(out, rc = -EINVAL);
 564         }
 565
 566         /* check that the UUID matches */
 567         if (!obd_uuid_equals(&uuid, &LTD_TGT(ltd, idx)->ltd_uuid)) {
 568                 CERROR("%s: LOD target UUID %s at index %d does not match %s\n",
 569                        obd->obd_name, obd_uuid2str(&LTD_TGT(ltd,idx)->ltd_uuid),
 570                        idx, osp);
 571                 GOTO(out, rc = -EINVAL);
 572         }
 573
 574         __lod_del_device(env, lod, ltd, idx, for_ost);
 575         EXIT;
 576 out:
 577         mutex_unlock(&ltd->ltd_mutex);
 578         lod_putref(lod, ltd);
 579         return(rc);
 580 }
 581
 582 /**
 583  * Resize per-thread storage to hold specified size.
 584  *
 585  * A helper function to resize per-thread temporary storage. This storage
 586  * is used to process LOV/LVM EAs and may be quite large. We do not want to
 587  * allocate/release it every time, so instead we put it into the env and
 588  * reallocate on demand. The memory is released when the correspondent thread
 589  * is finished.
 590  *
 591  * \param[in] info              LOD-specific storage in the environment
 592  * \param[in] size              new size to grow the buffer to
 593
 594  * \retval                      0 on success, -ENOMEM if reallocation failed
 595  */
 596 int lod_ea_store_resize(struct lod_thread_info *info, size_t size)
 597 {
 598         __u32 round = size_roundup_power2(size);
 599
 600         LASSERT(round <=
 601                 lov_mds_md_size(LOV_MAX_STRIPE_COUNT, LOV_MAGIC_V3));
 602         if (info->lti_ea_store) {
 603                 LASSERT(info->lti_ea_store_size);
 604                 LASSERT(info->lti_ea_store_size < round);
 605                 CDEBUG(D_INFO, "EA store size %d is not enough, need %d\n",
 606                        info->lti_ea_store_size, round);
 607                 OBD_FREE_LARGE(info->lti_ea_store, info->lti_ea_store_size);
 608                 info->lti_ea_store = NULL;
 609                 info->lti_ea_store_size = 0;
 610         }
 611
 612         OBD_ALLOC_LARGE(info->lti_ea_store, round);
 613         if (info->lti_ea_store == NULL)
 614                 RETURN(-ENOMEM);
 615         info->lti_ea_store_size = round;
 616
 617         RETURN(0);
 618 }
 619
 620 static void lod_free_comp_buffer(struct lod_layout_component *entries,
 621                                  __u16 count, __u32 bufsize)
 622 {
 623         struct lod_layout_component *entry;
 624         int i;
 625
 626         for (i = 0; i < count; i++) {
 627                 entry = &entries[i];
 628                 if (entry->llc_pool != NULL)
 629                         lod_set_pool(&entry->llc_pool, NULL);
 630                 LASSERT(entry->llc_stripe == NULL);
 631                 LASSERT(entry->llc_stripes_allocated == 0);
 632         }
 633
 634         if (bufsize != 0)
 635                 OBD_FREE_LARGE(entries, bufsize);
 636 }
 637
 638 void lod_free_def_comp_entries(struct lod_default_striping *lds)
 639 {
 640         lod_free_comp_buffer(lds->lds_def_comp_entries,
 641                              lds->lds_def_comp_size_cnt,
 642                              size_roundup_power2(
 643                                      sizeof(*lds->lds_def_comp_entries) *
 644                                      lds->lds_def_comp_size_cnt));
 645         lds->lds_def_comp_entries = NULL;
 646         lds->lds_def_comp_cnt = 0;
 647         lds->lds_def_striping_is_composite = 0;
 648         lds->lds_def_comp_size_cnt = 0;
 649 }
 650
 651 /**
 652  * Resize per-thread storage to hold default striping component entries
 653  *
 654  * A helper function to resize per-thread temporary storage. This storage
 655  * is used to hold default LOV/LVM EAs and may be quite large. We do not want
 656  * to allocate/release it every time, so instead we put it into the env and
 657  * reallocate it on demand. The memory is released when the correspondent
 658  * thread is finished.
 659  *
 660  * \param[in,out] lds           default striping
 661  * \param[in] count             new component count to grow the buffer to
 662
 663  * \retval                      0 on success, -ENOMEM if reallocation failed
 664  */
 665 int lod_def_striping_comp_resize(struct lod_default_striping *lds, __u16 count)
 666 {
 667         struct lod_layout_component *entries;
 668         __u32 new = size_roundup_power2(sizeof(*lds->lds_def_comp_entries) *
 669                                         count);
 670         __u32 old = size_roundup_power2(sizeof(*lds->lds_def_comp_entries) *
 671                                         lds->lds_def_comp_size_cnt);
 672
 673         if (new <= old)
 674                 return 0;
 675
 676         OBD_ALLOC_LARGE(entries, new);
 677         if (entries == NULL)
 678                 return -ENOMEM;
 679
 680         if (lds->lds_def_comp_entries != NULL) {
 681                 CDEBUG(D_INFO, "default striping component size %d is not "
 682                        "enough, need %d\n", old, new);
 683                 lod_free_def_comp_entries(lds);
 684         }
 685
 686         lds->lds_def_comp_entries = entries;
 687         lds->lds_def_comp_size_cnt = count;
 688
 689         RETURN(0);
 690 }
 691
 692 void lod_free_comp_entries(struct lod_object *lo)
 693 {
 694         lod_free_comp_buffer(lo->ldo_comp_entries,
 695                              lo->ldo_comp_cnt,
 696                              sizeof(*lo->ldo_comp_entries) * lo->ldo_comp_cnt);
 697         lo->ldo_comp_entries = NULL;
 698         lo->ldo_comp_cnt = 0;
 699         lo->ldo_is_composite = 0;
 700 }
 701
 702 int lod_alloc_comp_entries(struct lod_object *lo, int cnt)
 703 {
 704         LASSERT(cnt != 0);
 705         LASSERT(lo->ldo_comp_cnt == 0 && lo->ldo_comp_entries == NULL);
 706
 707         OBD_ALLOC_LARGE(lo->ldo_comp_entries,
 708                         sizeof(*lo->ldo_comp_entries) * cnt);
 709         if (lo->ldo_comp_entries == NULL)
 710                 return -ENOMEM;
 711         lo->ldo_comp_cnt = cnt;
 712         return 0;
 713 }
 714
 715 /**
 716  * Generate on-disk lov_mds_md structure for each layout component based on
 717  * the information in lod_object->ldo_comp_entries[i].
 718  *
 719  * \param[in] env               execution environment for this thread
 720  * \param[in] lo                LOD object
 721  * \param[in] comp_idx          index of ldo_comp_entries
 722  * \param[in] lmm               buffer to cotain the on-disk lov_mds_md
 723  * \param[in|out] lmm_size      buffer size/lmm size
 724  * \param[in] is_dir            generate lov ea for dir or file? For dir case,
 725  *                              the stripe info is from the default stripe
 726  *                              template, which is collected in lod_ah_init(),
 727  *                              either from parent object or root object; for
 728  *                              file case, it's from the @lo object
 729  *
 730  * \retval                      0 if on disk structure is created successfully
 731  * \retval                      negative error number on failure
 732  */
 733 static int lod_gen_component_ea(const struct lu_env *env,
 734                                 struct lod_object *lo, int comp_idx,
 735                                 struct lov_mds_md *lmm, int *lmm_size,
 736                                 bool is_dir)
 737 {
 738         struct lod_thread_info  *info = lod_env_info(env);
 739         const struct lu_fid     *fid  = lu_object_fid(&lo->ldo_obj.do_lu);
 740         struct lod_device       *lod;
 741         struct lov_ost_data_v1  *objs;
 742         struct lod_layout_component *lod_comp;
 743         __u32   magic;
 744         int     i, rc = 0;
 745         ENTRY;
 746
 747         LASSERT(lo);
 748         if (is_dir)
 749                 lod_comp =
 750                         &lo->ldo_def_striping->lds_def_comp_entries[comp_idx];
 751         else
 752                 lod_comp = &lo->ldo_comp_entries[comp_idx];
 753
 754         magic = lod_comp->llc_pool != NULL ? LOV_MAGIC_V3 : LOV_MAGIC_V1;
 755         if (lod_comp->llc_pattern == 0) /* default striping */
 756                 lod_comp->llc_pattern = LOV_PATTERN_RAID0;
 757
 758         lmm->lmm_magic = cpu_to_le32(magic);
 759         lmm->lmm_pattern = cpu_to_le32(lod_comp->llc_pattern);
 760         fid_to_lmm_oi(fid, &lmm->lmm_oi);
 761         if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_LMMOI))
 762                 lmm->lmm_oi.oi.oi_id++;
 763         lmm_oi_cpu_to_le(&lmm->lmm_oi, &lmm->lmm_oi);
 764
 765         lmm->lmm_stripe_size = cpu_to_le32(lod_comp->llc_stripe_size);
 766         lmm->lmm_stripe_count = cpu_to_le16(lod_comp->llc_stripenr);
 767         /* for dir, lmm_layout_gen stores default stripe offset. */
 768         lmm->lmm_layout_gen = is_dir ?
 769                         cpu_to_le16(lod_comp->llc_stripe_offset) :
 770                         cpu_to_le16(lod_comp->llc_layout_gen);
 771
 772         if (magic == LOV_MAGIC_V1) {
 773                 objs = &lmm->lmm_objects[0];
 774         } else {
 775                 struct lov_mds_md_v3 *v3 = (struct lov_mds_md_v3 *)lmm;
 776                 size_t cplen = strlcpy(v3->lmm_pool_name,
 777                                        lod_comp->llc_pool,
 778                                        sizeof(v3->lmm_pool_name));
 779                 if (cplen >= sizeof(v3->lmm_pool_name))
 780                         RETURN(-E2BIG);
 781                 objs = &v3->lmm_objects[0];
 782         }
 783
 784         if (is_dir || lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED)
 785                 GOTO(done, rc = 0);
 786
 787         lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
 788         for (i = 0; i < lod_comp->llc_stripenr; i++) {
 789                 struct dt_object        *object;
 790                 __u32   ost_idx;
 791                 int     type = LU_SEQ_RANGE_OST;
 792
 793                 object = lod_comp->llc_stripe[i];
 794                 LASSERT(object != NULL);
 795                 info->lti_fid = *lu_object_fid(&object->do_lu);
 796
 797                 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_MULTIPLE_REF)) {
 798                         if (cfs_fail_val == 0)
 799                                 cfs_fail_val = info->lti_fid.f_oid;
 800                         else
 801                                 info->lti_fid.f_oid = cfs_fail_val;
 802                 }
 803
 804                 rc = fid_to_ostid(&info->lti_fid, &info->lti_ostid);
 805                 LASSERT(rc == 0);
 806
 807                 ostid_cpu_to_le(&info->lti_ostid, &objs[i].l_ost_oi);
 808                 objs[i].l_ost_gen = cpu_to_le32(0);
 809                 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_FLD_LOOKUP))
 810                         rc = -ENOENT;
 811                 else
 812                         rc = lod_fld_lookup(env, lod, &info->lti_fid,
 813                                             &ost_idx, &type);
 814                 if (rc < 0) {
 815                         CERROR("%s: Can not locate "DFID": rc = %d\n",
 816                                lod2obd(lod)->obd_name, PFID(&info->lti_fid),
 817                                rc);
 818                         RETURN(rc);
 819                 }
 820                 objs[i].l_ost_idx = cpu_to_le32(ost_idx);
 821         }
 822 done:
 823         if (lmm_size != NULL)
 824                 *lmm_size = lov_mds_md_size(is_dir ?
 825                                 0 : lod_comp->llc_stripenr, magic);
 826         RETURN(rc);
 827 }
 828
 829 /**
 830  * Generate component ID for new created component.
 831  *
 832  * \param[in] lo                LOD object
 833  * \param[in] comp_idx          index of ldo_comp_entries
 834  *
 835  * \retval                      component ID on success
 836  * \retval                      LCME_ID_INVAL on failure
 837  */
 838 static __u32 lod_gen_component_id(struct lod_object *lo, int comp_idx)
 839 {
 840         struct lod_layout_component *lod_comp;
 841         __u32   id, start, end;
 842         int     i;
 843
 844         LASSERT(lo->ldo_comp_entries[comp_idx].llc_id == LCME_ID_INVAL);
 845
 846         lod_obj_inc_layout_gen(lo);
 847         id = lo->ldo_layout_gen;
 848         if (likely(id <= LCME_ID_MAX))
 849                 return id;
 850
 851         /* Layout generation wraps, need to check collisions. */
 852         start = id & LCME_ID_MASK;
 853         end = (__u32)LCME_ID_MAX + 1;
 854 again:
 855         for (id = start; id < end; id++) {
 856                 for (i = 0; i < lo->ldo_comp_cnt; i++) {
 857                         lod_comp = &lo->ldo_comp_entries[i];
 858                         if (id == lod_comp->llc_id)
 859                                 break;
 860                 }
 861                 /* Found the ununsed ID */
 862                 if (i == lo->ldo_comp_cnt)
 863                         return id;
 864         }
 865         if (end == (__u32)LCME_ID_MAX + 1) {
 866                 start = 0;
 867                 end = lo->ldo_layout_gen & LCME_ID_MASK;
 868                 goto again;
 869         }
 870
 871         return LCME_ID_INVAL;
 872 }
 873
 874 /**
 875  * Generate on-disk lov_mds_md structure based on the information in
 876  * the lod_object->ldo_comp_entries.
 877  *
 878  * \param[in] env               execution environment for this thread
 879  * \param[in] lo                LOD object
 880  * \param[in] lmm               buffer to cotain the on-disk lov_mds_md
 881  * \param[in|out] lmm_size      buffer size/lmm size
 882  * \param[in] is_dir            generate lov ea for dir or file? For dir case,
 883  *                              the stripe info is from the default stripe
 884  *                              template, which is collected in lod_ah_init(),
 885  *                              either from parent object or root object; for
 886  *                              file case, it's from the @lo object
 887  *
 888  * \retval                      0 if on disk structure is created successfully
 889  * \retval                      negative error number on failure
 890  */
 891 int lod_generate_lovea(const struct lu_env *env, struct lod_object *lo,
 892                        struct lov_mds_md *lmm, int *lmm_size, bool is_dir)
 893 {
 894         struct lov_comp_md_entry_v1 *lcme;
 895         struct lov_comp_md_v1 *lcm;
 896         struct lod_layout_component *comp_entries;
 897         __u16 comp_cnt;
 898         bool is_composite;
 899         int i, rc = 0, offset;
 900         ENTRY;
 901
 902         if (is_dir) {
 903                 comp_cnt = lo->ldo_def_striping->lds_def_comp_cnt;
 904                 comp_entries = lo->ldo_def_striping->lds_def_comp_entries;
 905                 is_composite =
 906                         lo->ldo_def_striping->lds_def_striping_is_composite;
 907         } else {
 908                 comp_cnt = lo->ldo_comp_cnt;
 909                 comp_entries = lo->ldo_comp_entries;
 910                 is_composite = lo->ldo_is_composite;
 911         }
 912
 913         LASSERT(lmm_size != NULL);
 914         LASSERT(comp_cnt != 0 && comp_entries != NULL);
 915
 916         if (!is_composite) {
 917                 rc = lod_gen_component_ea(env, lo, 0, lmm, lmm_size, is_dir);
 918                 RETURN(rc);
 919         }
 920
 921         lcm = (struct lov_comp_md_v1 *)lmm;
 922         lcm->lcm_magic = cpu_to_le32(LOV_MAGIC_COMP_V1);
 923         lcm->lcm_entry_count = cpu_to_le16(comp_cnt);
 924
 925         offset = sizeof(*lcm) + sizeof(*lcme) * comp_cnt;
 926         LASSERT(offset % sizeof(__u64) == 0);
 927
 928         for (i = 0; i < comp_cnt; i++) {
 929                 struct lod_layout_component *lod_comp;
 930                 struct lov_mds_md *sub_md;
 931                 int size;
 932
 933                 lod_comp = &comp_entries[i];
 934                 lcme = &lcm->lcm_entries[i];
 935
 936                 if (lod_comp->llc_id == LCME_ID_INVAL && !is_dir) {
 937                         lod_comp->llc_id = lod_gen_component_id(lo, i);
 938                         if (lod_comp->llc_id == LCME_ID_INVAL)
 939                                 GOTO(out, rc = -ERANGE);
 940                 }
 941                 lcme->lcme_id = cpu_to_le32(lod_comp->llc_id);
 942                 /* component must has been inistantiated */
 943                 LASSERT(ergo(!is_dir, lod_comp->llc_flags & LCME_FL_INIT));
 944                 lcme->lcme_flags = cpu_to_le32(lod_comp->llc_flags);
 945                 lcme->lcme_extent.e_start =
 946                         cpu_to_le64(lod_comp->llc_extent.e_start);
 947                 lcme->lcme_extent.e_end =
 948                         cpu_to_le64(lod_comp->llc_extent.e_end);
 949                 lcme->lcme_offset = cpu_to_le32(offset);
 950
 951                 sub_md = (struct lov_mds_md *)((char *)lcm + offset);
 952                 rc = lod_gen_component_ea(env, lo, i, sub_md, &size, is_dir);
 953                 if (rc)
 954                         GOTO(out, rc);
 955                 lcme->lcme_size = cpu_to_le32(size);
 956                 offset += size;
 957                 LASSERTF((offset <= *lmm_size) && (offset % sizeof(__u64) == 0),
 958                          "offset:%d lmm_size:%d\n", offset, *lmm_size);
 959         }
 960         lcm->lcm_size = cpu_to_le32(offset);
 961         lcm->lcm_layout_gen = cpu_to_le32(is_dir ? 0 : lo->ldo_layout_gen);
 962
 963         lustre_print_user_md(D_LAYOUT, (struct lov_user_md *)lmm,
 964                              "generate lum");
 965 out:
 966         if (rc == 0)
 967                 *lmm_size = offset;
 968         RETURN(rc);
 969 }
 970
 971 /**
 972  * Get LOV EA.
 973  *
 974  * Fill lti_ea_store buffer in the environment with a value for the given
 975  * EA. The buffer is reallocated if the value doesn't fit.
 976  *
 977  * \param[in,out] env           execution environment for this thread
 978  *                              .lti_ea_store buffer is filled with EA's value
 979  * \param[in] lo                LOD object
 980  * \param[in] name              name of the EA
 981  *
 982  * \retval                      0 if EA is fetched successfully
 983  * \retval                      negative error number on failure
 984  */
 985 int lod_get_ea(const struct lu_env *env, struct lod_object *lo,
 986                const char *name)
 987 {
 988         struct lod_thread_info  *info = lod_env_info(env);
 989         struct dt_object        *next = dt_object_child(&lo->ldo_obj);
 990         int                     rc;
 991         ENTRY;
 992
 993         LASSERT(info);
 994
 995         if (unlikely(info->lti_ea_store == NULL)) {
 996                 /* just to enter in allocation block below */
 997                 rc = -ERANGE;
 998         } else {
 999 repeat:
1000                 info->lti_buf.lb_buf = info->lti_ea_store;
1001                 info->lti_buf.lb_len = info->lti_ea_store_size;
1002                 rc = dt_xattr_get(env, next, &info->lti_buf, name);
1003         }
1004
1005         /* if object is not striped or inaccessible */
1006         if (rc == -ENODATA || rc == -ENOENT)
1007                 RETURN(0);
1008
1009         if (rc == -ERANGE) {
1010                 /* EA doesn't fit, reallocate new buffer */
1011                 rc = dt_xattr_get(env, next, &LU_BUF_NULL, name);
1012                 if (rc == -ENODATA || rc == -ENOENT)
1013                         RETURN(0);
1014                 else if (rc < 0)
1015                         RETURN(rc);
1016
1017                 LASSERT(rc > 0);
1018                 rc = lod_ea_store_resize(info, rc);
1019                 if (rc)
1020                         RETURN(rc);
1021                 goto repeat;
1022         }
1023
1024         RETURN(rc);
1025 }
1026
1027 /**
1028  * Verify the target index is present in the current configuration.
1029  *
1030  * \param[in] md                LOD device where the target table is stored
1031  * \param[in] idx               target's index
1032  *
1033  * \retval                      0 if the index is present
1034  * \retval                      -EINVAL if not
1035  */
1036 static int validate_lod_and_idx(struct lod_device *md, __u32 idx)
1037 {
1038         if (unlikely(idx >= md->lod_ost_descs.ltd_tgts_size ||
1039                      !cfs_bitmap_check(md->lod_ost_bitmap, idx))) {
1040                 CERROR("%s: bad idx: %d of %d\n", lod2obd(md)->obd_name, idx,
1041                        md->lod_ost_descs.ltd_tgts_size);
1042                 return -EINVAL;
1043         }
1044
1045         if (unlikely(OST_TGT(md, idx) == NULL)) {
1046                 CERROR("%s: bad lod_tgt_desc for idx: %d\n",
1047                        lod2obd(md)->obd_name, idx);
1048                 return -EINVAL;
1049         }
1050
1051         if (unlikely(OST_TGT(md, idx)->ltd_ost == NULL)) {
1052                 CERROR("%s: invalid lod device, for idx: %d\n",
1053                        lod2obd(md)->obd_name , idx);
1054                 return -EINVAL;
1055         }
1056
1057         return 0;
1058 }
1059
1060 /**
1061  * Instantiate objects for stripes.
1062  *
1063  * Allocate and initialize LU-objects representing the stripes. The number
1064  * of the stripes (ldo_stripenr) must be initialized already. The caller
1065  * must ensure nobody else is calling the function on the object at the same
1066  * time. FLDB service must be running to be able to map a FID to the targets
1067  * and find appropriate device representing that target.
1068  *
1069  * \param[in] env               execution environment for this thread
1070  * \param[in,out] lo            LOD object
1071  * \param[in] objs              an array of IDs to creates the objects from
1072  * \param[in] comp_idx          index of ldo_comp_entries
1073  *
1074  * \retval                      0 if the objects are instantiated successfully
1075  * \retval                      negative error number on failure
1076  */
1077 int lod_initialize_objects(const struct lu_env *env, struct lod_object *lo,
1078                            struct lov_ost_data_v1 *objs, int comp_idx)
1079 {
1080         struct lod_layout_component     *lod_comp;
1081         struct lod_thread_info  *info = lod_env_info(env);
1082         struct lod_device       *md;
1083         struct lu_object        *o, *n;
1084         struct lu_device        *nd;
1085         struct dt_object       **stripe;
1086         int                      stripe_len;
1087         int                      i, rc = 0;
1088         __u32                   idx;
1089         ENTRY;
1090
1091         LASSERT(lo != NULL);
1092         md = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
1093
1094         LASSERT(lo->ldo_comp_cnt != 0 && lo->ldo_comp_entries != NULL);
1095         lod_comp = &lo->ldo_comp_entries[comp_idx];
1096
1097         LASSERT(lod_comp->llc_stripe == NULL);
1098         LASSERT(lod_comp->llc_stripenr > 0);
1099         LASSERT(lod_comp->llc_stripe_size > 0);
1100
1101         stripe_len = lod_comp->llc_stripenr;
1102         OBD_ALLOC(stripe, sizeof(stripe[0]) * stripe_len);
1103         if (stripe == NULL)
1104                 RETURN(-ENOMEM);
1105
1106         for (i = 0; i < lod_comp->llc_stripenr; i++) {
1107                 if (unlikely(lovea_slot_is_dummy(&objs[i])))
1108                         continue;
1109
1110                 ostid_le_to_cpu(&objs[i].l_ost_oi, &info->lti_ostid);
1111                 idx = le32_to_cpu(objs[i].l_ost_idx);
1112                 rc = ostid_to_fid(&info->lti_fid, &info->lti_ostid, idx);
1113                 if (rc != 0)
1114                         GOTO(out, rc);
1115                 LASSERTF(fid_is_sane(&info->lti_fid), ""DFID" insane!\n",
1116                          PFID(&info->lti_fid));
1117                 lod_getref(&md->lod_ost_descs);
1118
1119                 rc = validate_lod_and_idx(md, idx);
1120                 if (unlikely(rc != 0)) {
1121                         lod_putref(md, &md->lod_ost_descs);
1122                         GOTO(out, rc);
1123                 }
1124
1125                 nd = &OST_TGT(md,idx)->ltd_ost->dd_lu_dev;
1126                 lod_putref(md, &md->lod_ost_descs);
1127
1128                 /* In the function below, .hs_keycmp resolves to
1129                  * u_obj_hop_keycmp() */
1130                 /* coverity[overrun-buffer-val] */
1131                 o = lu_object_find_at(env, nd, &info->lti_fid, NULL);
1132                 if (IS_ERR(o))
1133                         GOTO(out, rc = PTR_ERR(o));
1134
1135                 n = lu_object_locate(o->lo_header, nd->ld_type);
1136                 LASSERT(n);
1137
1138                 stripe[i] = container_of(n, struct dt_object, do_lu);
1139         }
1140
1141 out:
1142         if (rc != 0) {
1143                 for (i = 0; i < stripe_len; i++)
1144                         if (stripe[i] != NULL)
1145                                 dt_object_put(env, stripe[i]);
1146
1147                 OBD_FREE(stripe, sizeof(stripe[0]) * stripe_len);
1148                 lod_comp->llc_stripenr = 0;
1149         } else {
1150                 lod_comp->llc_stripe = stripe;
1151                 lod_comp->llc_stripes_allocated = stripe_len;
1152         }
1153
1154         RETURN(rc);
1155 }
1156
1157 /**
1158  * Instantiate objects for striping.
1159  *
1160  * Parse striping information in \a buf and instantiate the objects
1161  * representing the stripes.
1162  *
1163  * \param[in] env               execution environment for this thread
1164  * \param[in] lo                LOD object
1165  * \param[in] buf               buffer storing LOV EA to parse
1166  *
1167  * \retval                      0 if parsing and objects creation succeed
1168  * \retval                      negative error number on failure
1169  */
1170 int lod_parse_striping(const struct lu_env *env, struct lod_object *lo,
1171                        const struct lu_buf *buf)
1172 {
1173         struct lov_mds_md_v1    *lmm;
1174         struct lov_comp_md_v1   *comp_v1 = NULL;
1175         struct lov_ost_data_v1  *objs;
1176         __u32   magic, pattern;
1177         int     i, rc = 0;
1178         __u16   comp_cnt;
1179         ENTRY;
1180
1181         LASSERT(buf);
1182         LASSERT(buf->lb_buf);
1183         LASSERT(buf->lb_len);
1184
1185         lmm = (struct lov_mds_md_v1 *)buf->lb_buf;
1186         magic = le32_to_cpu(lmm->lmm_magic);
1187
1188         if (magic != LOV_MAGIC_V1 && magic != LOV_MAGIC_V3 &&
1189             magic != LOV_MAGIC_COMP_V1)
1190                 GOTO(out, rc = -EINVAL);
1191
1192         lod_free_comp_entries(lo);
1193
1194         if (magic == LOV_MAGIC_COMP_V1) {
1195                 comp_v1 = (struct lov_comp_md_v1 *)lmm;
1196                 comp_cnt = le16_to_cpu(comp_v1->lcm_entry_count);
1197                 if (comp_cnt == 0)
1198                         GOTO(out, rc = -EINVAL);
1199                 lo->ldo_layout_gen = le32_to_cpu(comp_v1->lcm_layout_gen);
1200                 lo->ldo_is_composite = 1;
1201         } else {
1202                 comp_cnt = 1;
1203                 lo->ldo_layout_gen = le16_to_cpu(lmm->lmm_layout_gen);
1204                 lo->ldo_is_composite = 0;
1205         }
1206
1207         rc = lod_alloc_comp_entries(lo, comp_cnt);
1208         if (rc)
1209                 GOTO(out, rc);
1210
1211         for (i = 0; i < comp_cnt; i++) {
1212                 struct lod_layout_component     *lod_comp;
1213                 struct lu_extent        *ext;
1214                 __u32   offs;
1215
1216                 lod_comp = &lo->ldo_comp_entries[i];
1217                 if (lo->ldo_is_composite) {
1218                         offs = le32_to_cpu(comp_v1->lcm_entries[i].lcme_offset);
1219                         lmm = (struct lov_mds_md_v1 *)((char *)comp_v1 + offs);
1220                         magic = le32_to_cpu(lmm->lmm_magic);
1221
1222                         ext = &comp_v1->lcm_entries[i].lcme_extent;
1223                         lod_comp->llc_extent.e_start =
1224                                 le64_to_cpu(ext->e_start);
1225                         lod_comp->llc_extent.e_end = le64_to_cpu(ext->e_end);
1226                         lod_comp->llc_flags =
1227                                 le32_to_cpu(comp_v1->lcm_entries[i].lcme_flags);
1228                         lod_comp->llc_id =
1229                                 le32_to_cpu(comp_v1->lcm_entries[i].lcme_id);
1230                         if (lod_comp->llc_id == LCME_ID_INVAL)
1231                                 GOTO(out, rc = -EINVAL);
1232                 } else {
1233                         lod_comp->llc_flags = LCME_FL_INIT;
1234                 }
1235
1236                 pattern = le32_to_cpu(lmm->lmm_pattern);
1237                 if (lov_pattern(pattern) != LOV_PATTERN_RAID0)
1238                         GOTO(out, rc = -EINVAL);
1239
1240                 lod_comp->llc_pattern = pattern;
1241                 lod_comp->llc_stripe_size = le32_to_cpu(lmm->lmm_stripe_size);
1242                 lod_comp->llc_stripenr = le16_to_cpu(lmm->lmm_stripe_count);
1243                 lod_comp->llc_layout_gen = le16_to_cpu(lmm->lmm_layout_gen);
1244
1245                 if (magic == LOV_MAGIC_V3) {
1246                         struct lov_mds_md_v3 *v3 = (struct lov_mds_md_v3 *)lmm;
1247                         objs = &v3->lmm_objects[0];
1248                         /* no need to set pool, which is used in create only */
1249                 } else {
1250                         objs = &lmm->lmm_objects[0];
1251                 }
1252
1253                 if (!(lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED)) {
1254                         rc = lod_initialize_objects(env, lo, objs, i);
1255                         if (rc)
1256                                 GOTO(out, rc);
1257                 }
1258         }
1259 out:
1260         if (rc)
1261                 lod_object_free_striping(env, lo);
1262         RETURN(rc);
1263 }
1264
1265 /**
1266  * Check whether the striping (LOVEA for regular file, LMVEA for directory)
1267  * is already cached.
1268  *
1269  * \param[in] lo        LOD object
1270  *
1271  * \retval              True if the striping is cached, otherwise
1272  *                      return false.
1273  */
1274 static bool lod_striping_loaded(struct lod_object *lo)
1275 {
1276         if (S_ISREG(lod2lu_obj(lo)->lo_header->loh_attr) &&
1277             lo->ldo_comp_cached)
1278                 return true;
1279
1280         if (S_ISDIR(lod2lu_obj(lo)->lo_header->loh_attr)) {
1281                 if (lo->ldo_stripe != NULL)
1282                         return true;
1283
1284                 /* Never load LMV stripe for slaves of striped dir */
1285                 if (lo->ldo_dir_slave_stripe)
1286                         return true;
1287         }
1288
1289         return false;
1290 }
1291
1292 /**
1293  * Initialize the object representing the stripes.
1294  *
1295  * Unless the stripes are initialized already, fetch LOV (for regular
1296  * objects) or LMV (for directory objects) EA and call lod_parse_striping()
1297  * to instantiate the objects representing the stripes. Caller should
1298  * hold the dt_write_lock(next).
1299  *
1300  * \param[in] env               execution environment for this thread
1301  * \param[in,out] lo            LOD object
1302  *
1303  * \retval                      0 if parsing and object creation succeed
1304  * \retval                      negative error number on failure
1305  */
1306 int lod_load_striping_locked(const struct lu_env *env, struct lod_object *lo)
1307 {
1308         struct lod_thread_info  *info = lod_env_info(env);
1309         struct lu_buf           *buf  = &info->lti_buf;
1310         struct dt_object        *next = dt_object_child(&lo->ldo_obj);
1311         int                      rc = 0;
1312         ENTRY;
1313
1314         if (!dt_object_exists(next))
1315                 GOTO(out, rc = 0);
1316
1317         if (lod_striping_loaded(lo))
1318                 GOTO(out, rc = 0);
1319
1320         if (S_ISREG(lod2lu_obj(lo)->lo_header->loh_attr)) {
1321                 rc = lod_get_lov_ea(env, lo);
1322                 if (rc <= 0)
1323                         GOTO(out, rc);
1324                 /*
1325                  * there is LOV EA (striping information) in this object
1326                  * let's parse it and create in-core objects for the stripes
1327                  */
1328                 buf->lb_buf = info->lti_ea_store;
1329                 buf->lb_len = info->lti_ea_store_size;
1330                 rc = lod_parse_striping(env, lo, buf);
1331                 if (rc == 0)
1332                         lo->ldo_comp_cached = 1;
1333         } else if (S_ISDIR(lod2lu_obj(lo)->lo_header->loh_attr)) {
1334                 rc = lod_get_lmv_ea(env, lo);
1335                 if (rc < (typeof(rc))sizeof(struct lmv_mds_md_v1))
1336                         GOTO(out, rc = rc > 0 ? -EINVAL : rc);
1337
1338                 buf->lb_buf = info->lti_ea_store;
1339                 buf->lb_len = info->lti_ea_store_size;
1340                 if (rc == sizeof(struct lmv_mds_md_v1)) {
1341                         rc = lod_load_lmv_shards(env, lo, buf, true);
1342                         if (buf->lb_buf != info->lti_ea_store) {
1343                                 OBD_FREE_LARGE(info->lti_ea_store,
1344                                                info->lti_ea_store_size);
1345                                 info->lti_ea_store = buf->lb_buf;
1346                                 info->lti_ea_store_size = buf->lb_len;
1347                         }
1348
1349                         if (rc < 0)
1350                                 GOTO(out, rc);
1351                 }
1352
1353                 /*
1354                  * there is LMV EA (striping information) in this object
1355                  * let's parse it and create in-core objects for the stripes
1356                  */
1357                 rc = lod_parse_dir_striping(env, lo, buf);
1358         }
1359 out:
1360         RETURN(rc);
1361 }
1362
1363 /**
1364  * A generic function to initialize the stripe objects.
1365  *
1366  * A protected version of lod_load_striping_locked() - load the striping
1367  * information from storage, parse that and instantiate LU objects to
1368  * represent the stripes.  The LOD object \a lo supplies a pointer to the
1369  * next sub-object in the LU stack so we can lock it. Also use \a lo to
1370  * return an array of references to the newly instantiated objects.
1371  *
1372  * \param[in] env               execution environment for this thread
1373  * \param[in,out] lo            LOD object, where striping is stored and
1374  *                              which gets an array of references
1375  *
1376  * \retval                      0 if parsing and object creation succeed
1377  * \retval                      negative error number on failure
1378  **/
1379 int lod_load_striping(const struct lu_env *env, struct lod_object *lo)
1380 {
1381         struct dt_object        *next = dt_object_child(&lo->ldo_obj);
1382         int                     rc;
1383
1384         if (!dt_object_exists(next))
1385                 return 0;
1386
1387         /* Check without locking first */
1388         if (lod_striping_loaded(lo))
1389                 return 0;
1390
1391         /* currently this code is supposed to be called from declaration
1392          * phase only, thus the object is not expected to be locked by caller */
1393         dt_write_lock(env, next, 0);
1394         rc = lod_load_striping_locked(env, lo);
1395         dt_write_unlock(env, next);
1396         return rc;
1397 }
1398
1399 /**
1400  * Verify lov_user_md_v1/v3 striping.
1401  *
1402  * Check the validity of all fields including the magic, stripe size,
1403  * stripe count, stripe offset and that the pool is present.  Also check
1404  * that each target index points to an existing target. The additional
1405  * \a is_from_disk turns additional checks. In some cases zero fields
1406  * are allowed (like pattern=0).
1407  *
1408  * \param[in] d                 LOD device
1409  * \param[in] buf               buffer with LOV EA to verify
1410  * \param[in] is_from_disk      0 - from user, allow some fields to be 0
1411  *                              1 - from disk, do not allow
1412  *
1413  * \retval                      0 if the striping is valid
1414  * \retval                      -EINVAL if striping is invalid
1415  */
1416 static int lod_verify_v1v3(struct lod_device *d, const struct lu_buf *buf,
1417                            bool is_from_disk)
1418 {
1419         struct lov_user_md_v1   *lum;
1420         struct lov_user_md_v3   *lum3;
1421         struct pool_desc        *pool = NULL;
1422         __u32                    magic;
1423         __u32                    stripe_size;
1424         __u16                    stripe_count;
1425         __u16                    stripe_offset;
1426         size_t                   lum_size;
1427         int                      rc = 0;
1428         ENTRY;
1429
1430         lum = buf->lb_buf;
1431
1432         if (buf->lb_len < sizeof(*lum)) {
1433                 CDEBUG(D_LAYOUT, "buf len %zu too small for lov_user_md\n",
1434                        buf->lb_len);
1435                 GOTO(out, rc = -EINVAL);
1436         }
1437
1438         magic = le32_to_cpu(lum->lmm_magic) & ~LOV_MAGIC_DEF;
1439         if (magic != LOV_USER_MAGIC_V1 &&
1440             magic != LOV_USER_MAGIC_V3 &&
1441             magic != LOV_USER_MAGIC_SPECIFIC) {
1442                 CDEBUG(D_LAYOUT, "bad userland LOV MAGIC: %#x\n",
1443                        le32_to_cpu(lum->lmm_magic));
1444                 GOTO(out, rc = -EINVAL);
1445         }
1446
1447         /* the user uses "0" for default stripe pattern normally. */
1448         if (!is_from_disk && lum->lmm_pattern == 0)
1449                 lum->lmm_pattern = cpu_to_le32(LOV_PATTERN_RAID0);
1450
1451         if (!lov_pattern_supported(le32_to_cpu(lum->lmm_pattern))) {
1452                 CDEBUG(D_LAYOUT, "bad userland stripe pattern: %#x\n",
1453                        le32_to_cpu(lum->lmm_pattern));
1454                 GOTO(out, rc = -EINVAL);
1455         }
1456
1457         /* a released lum comes from creating orphan on hsm release,
1458          * doesn't make sense to verify it. */
1459         if (le32_to_cpu(lum->lmm_pattern) & LOV_PATTERN_F_RELEASED)
1460                 GOTO(out, rc = 0);
1461
1462         /* 64kB is the largest common page size we see (ia64), and matches the
1463          * check in lfs */
1464         stripe_size = le32_to_cpu(lum->lmm_stripe_size);
1465         if (stripe_size & (LOV_MIN_STRIPE_SIZE - 1)) {
1466                 CDEBUG(D_LAYOUT, "stripe size %u not a multiple of %u\n",
1467                        stripe_size, LOV_MIN_STRIPE_SIZE);
1468                 GOTO(out, rc = -EINVAL);
1469         }
1470
1471         stripe_offset = le16_to_cpu(lum->lmm_stripe_offset);
1472         if (!is_from_disk && stripe_offset != LOV_OFFSET_DEFAULT) {
1473                 /* if offset is not within valid range [0, osts_size) */
1474                 if (stripe_offset >= d->lod_osts_size) {
1475                         CDEBUG(D_LAYOUT, "stripe offset %u >= bitmap size %u\n",
1476                                stripe_offset, d->lod_osts_size);
1477                         GOTO(out, rc = -EINVAL);
1478                 }
1479
1480                 /* if lmm_stripe_offset is *not* in bitmap */
1481                 if (!cfs_bitmap_check(d->lod_ost_bitmap, stripe_offset)) {
1482                         CDEBUG(D_LAYOUT, "stripe offset %u not in bitmap\n",
1483                                stripe_offset);
1484                         GOTO(out, rc = -EINVAL);
1485                 }
1486         }
1487
1488         if (magic == LOV_USER_MAGIC_V1)
1489                 lum_size = offsetof(struct lov_user_md_v1,
1490                                     lmm_objects[0]);
1491         else if (magic == LOV_USER_MAGIC_V3 || magic == LOV_USER_MAGIC_SPECIFIC)
1492                 lum_size = offsetof(struct lov_user_md_v3,
1493                                     lmm_objects[0]);
1494         else
1495                 GOTO(out, rc = -EINVAL);
1496
1497         stripe_count = le16_to_cpu(lum->lmm_stripe_count);
1498         if (buf->lb_len < lum_size) {
1499                 CDEBUG(D_LAYOUT, "invalid buf len %zu/%zu for lov_user_md with "
1500                        "magic %#x and stripe_count %u\n",
1501                        buf->lb_len, lum_size, magic, stripe_count);
1502                 GOTO(out, rc = -EINVAL);
1503         }
1504
1505         if (!(magic == LOV_USER_MAGIC_V3 || magic == LOV_USER_MAGIC_SPECIFIC))
1506                 goto out;
1507
1508         lum3 = buf->lb_buf;
1509         /* In the function below, .hs_keycmp resolves to
1510          * pool_hashkey_keycmp() */
1511         /* coverity[overrun-buffer-val] */
1512         pool = lod_find_pool(d, lum3->lmm_pool_name);
1513         if (pool == NULL)
1514                 goto out;
1515
1516         if (!is_from_disk && stripe_offset != LOV_OFFSET_DEFAULT) {
1517                 rc = lod_check_index_in_pool(stripe_offset, pool);
1518                 if (rc < 0)
1519                         GOTO(out, rc = -EINVAL);
1520         }
1521
1522         if (is_from_disk && stripe_count > pool_tgt_count(pool)) {
1523                 CDEBUG(D_LAYOUT, "stripe count %u > # OSTs %u in the pool\n",
1524                        stripe_count, pool_tgt_count(pool));
1525                 GOTO(out, rc = -EINVAL);
1526         }
1527
1528 out:
1529         if (pool != NULL)
1530                 lod_pool_putref(pool);
1531
1532         RETURN(rc);
1533 }
1534
1535 /**
1536  * Verify LOV striping.
1537  *
1538  * \param[in] d                 LOD device
1539  * \param[in] buf               buffer with LOV EA to verify
1540  * \param[in] is_from_disk      0 - from user, allow some fields to be 0
1541  *                              1 - from disk, do not allow
1542  * \param[in] start             extent start for composite layout
1543  *
1544  * \retval                      0 if the striping is valid
1545  * \retval                      -EINVAL if striping is invalid
1546  */
1547 int lod_verify_striping(struct lod_device *d, const struct lu_buf *buf,
1548                         bool is_from_disk, __u64 start)
1549 {
1550         struct lov_user_md_v1   *lum;
1551         struct lov_comp_md_v1   *comp_v1;
1552         __u32   magic;
1553         int     rc = 0, i;
1554         ENTRY;
1555
1556         lum = buf->lb_buf;
1557
1558         if (buf->lb_len < sizeof(*lum)) {
1559                 CDEBUG(D_LAYOUT, "buf len %zu too small for lov_user_md\n",
1560                        buf->lb_len);
1561                 RETURN(-EINVAL);
1562         }
1563
1564         magic = le32_to_cpu(lum->lmm_magic) & ~LOV_MAGIC_DEF;
1565         if (magic != LOV_USER_MAGIC_V1 &&
1566             magic != LOV_USER_MAGIC_V3 &&
1567             magic != LOV_USER_MAGIC_SPECIFIC &&
1568             magic != LOV_USER_MAGIC_COMP_V1) {
1569                 CDEBUG(D_LAYOUT, "bad userland LOV MAGIC: %#x\n",
1570                        le32_to_cpu(lum->lmm_magic));
1571                 RETURN(-EINVAL);
1572         }
1573
1574         if (magic == LOV_USER_MAGIC_COMP_V1) {
1575                 struct lov_comp_md_entry_v1     *ent;
1576                 struct lu_extent        *ext;
1577                 struct lov_desc *desc = &d->lod_desc;
1578                 struct lu_buf   tmp;
1579                 __u32   stripe_size = 0;
1580                 __u64   prev_end = start;
1581
1582                 comp_v1 = buf->lb_buf;
1583                 if (buf->lb_len < le32_to_cpu(comp_v1->lcm_size)) {
1584                         CDEBUG(D_LAYOUT, "buf len %zu is less than %u\n",
1585                                buf->lb_len, le32_to_cpu(comp_v1->lcm_size));
1586                         RETURN(-EINVAL);
1587                 }
1588
1589                 if (le32_to_cpu(comp_v1->lcm_entry_count) == 0) {
1590                         CDEBUG(D_LAYOUT, "entry count is zero\n");
1591                         RETURN(-EINVAL);
1592                 }
1593
1594                 for (i = 0; i < le32_to_cpu(comp_v1->lcm_entry_count); i++) {
1595                         ent = &comp_v1->lcm_entries[i];
1596                         ext = &ent->lcme_extent;
1597
1598                         if (is_from_disk &&
1599                             (le32_to_cpu(ent->lcme_id) == 0 ||
1600                              le32_to_cpu(ent->lcme_id) > LCME_ID_MAX)) {
1601                                 CDEBUG(D_LAYOUT, "invalid id %u\n",
1602                                        le32_to_cpu(ent->lcme_id));
1603                                 RETURN(-EINVAL);
1604                         }
1605
1606                         if (le64_to_cpu(ext->e_start) >=
1607                             le64_to_cpu(ext->e_end)) {
1608                                 CDEBUG(D_LAYOUT, "invalid extent "
1609                                        "[%llu, %llu)\n",
1610                                        le64_to_cpu(ext->e_start),
1611                                        le64_to_cpu(ext->e_end));
1612                                 RETURN(-EINVAL);
1613                         }
1614
1615                         /* first component must start with 0, and the next
1616                          * must be adjacent with the previous one */
1617                         if (le64_to_cpu(ext->e_start) != prev_end) {
1618                                 CDEBUG(D_LAYOUT, "invalid start "
1619                                        "actual:%llu, expect:%llu\n",
1620                                        le64_to_cpu(ext->e_start), prev_end);
1621                                 RETURN(-EINVAL);
1622                         }
1623                         prev_end = le64_to_cpu(ext->e_end);
1624
1625                         tmp.lb_buf = (char *)comp_v1 +
1626                                      le32_to_cpu(ent->lcme_offset);
1627                         tmp.lb_len = le32_to_cpu(ent->lcme_size);
1628                         rc = lod_verify_v1v3(d, &tmp, is_from_disk);
1629                         if (rc)
1630                                 break;
1631
1632                         lum = tmp.lb_buf;
1633
1634                         /* extent end must be aligned with the stripe_size */
1635                         stripe_size = le32_to_cpu(lum->lmm_stripe_size);
1636                         if (stripe_size == 0)
1637                                 stripe_size = desc->ld_default_stripe_size;
1638                         if (stripe_size == 0 ||
1639                             (prev_end != LUSTRE_EOF &&
1640                              (prev_end & (stripe_size - 1)))) {
1641                                 CDEBUG(D_LAYOUT, "stripe size isn't aligned. "
1642                                        " stripe_sz: %u, [%llu, %llu)\n",
1643                                        stripe_size, ext->e_start, prev_end);
1644                                 RETURN(-EINVAL);
1645                         }
1646                 }
1647         } else {
1648                 rc = lod_verify_v1v3(d, buf, is_from_disk);
1649         }
1650
1651         RETURN(rc);
1652 }
1653
1654 void lod_fix_desc_stripe_size(__u64 *val)
1655 {
1656         if (*val < LOV_MIN_STRIPE_SIZE) {
1657                 if (*val != 0)
1658                         LCONSOLE_INFO("Increasing default stripe size to "
1659                                       "minimum value %u\n",
1660                                       LOV_DESC_STRIPE_SIZE_DEFAULT);
1661                 *val = LOV_DESC_STRIPE_SIZE_DEFAULT;
1662         } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) {
1663                 *val &= ~(LOV_MIN_STRIPE_SIZE - 1);
1664                 LCONSOLE_WARN("Changing default stripe size to %llu (a "
1665                               "multiple of %u)\n",
1666                               *val, LOV_MIN_STRIPE_SIZE);
1667         }
1668 }
1669
1670 void lod_fix_desc_stripe_count(__u32 *val)
1671 {
1672         if (*val == 0)
1673                 *val = 1;
1674 }
1675
1676 void lod_fix_desc_pattern(__u32 *val)
1677 {
1678         /* from lov_setstripe */
1679         if ((*val != 0) && (*val != LOV_PATTERN_RAID0)) {
1680                 LCONSOLE_WARN("Unknown stripe pattern: %#x\n", *val);
1681                 *val = 0;
1682         }
1683 }
1684
1685 void lod_fix_desc_qos_maxage(__u32 *val)
1686 {
1687         /* fix qos_maxage */
1688         if (*val == 0)
1689                 *val = LOV_DESC_QOS_MAXAGE_DEFAULT;
1690 }
1691
1692 /**
1693  * Used to fix insane default striping.
1694  *
1695  * \param[in] desc      striping description
1696  */
1697 void lod_fix_desc(struct lov_desc *desc)
1698 {
1699         lod_fix_desc_stripe_size(&desc->ld_default_stripe_size);
1700         lod_fix_desc_stripe_count(&desc->ld_default_stripe_count);
1701         lod_fix_desc_pattern(&desc->ld_pattern);
1702         lod_fix_desc_qos_maxage(&desc->ld_qos_maxage);
1703 }
1704
1705 /**
1706  * Initialize the structures used to store pools and default striping.
1707  *
1708  * \param[in] lod       LOD device
1709  * \param[in] lcfg      configuration structure storing default striping.
1710  *
1711  * \retval              0 if initialization succeeds
1712  * \retval              negative error number on failure
1713  */
1714 int lod_pools_init(struct lod_device *lod, struct lustre_cfg *lcfg)
1715 {
1716         struct obd_device          *obd;
1717         struct lov_desc            *desc;
1718         int                         rc;
1719         ENTRY;
1720
1721         obd = class_name2obd(lustre_cfg_string(lcfg, 0));
1722         LASSERT(obd != NULL);
1723         obd->obd_lu_dev = &lod->lod_dt_dev.dd_lu_dev;
1724
1725         if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
1726                 CERROR("LOD setup requires a descriptor\n");
1727                 RETURN(-EINVAL);
1728         }
1729
1730         desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
1731
1732         if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
1733                 CERROR("descriptor size wrong: %d > %d\n",
1734                        (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
1735                 RETURN(-EINVAL);
1736         }
1737
1738         if (desc->ld_magic != LOV_DESC_MAGIC) {
1739                 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
1740                         CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
1741                                obd->obd_name, desc);
1742                         lustre_swab_lov_desc(desc);
1743                 } else {
1744                         CERROR("%s: Bad lov desc magic: %#x\n",
1745                                obd->obd_name, desc->ld_magic);
1746                         RETURN(-EINVAL);
1747                 }
1748         }
1749
1750         lod_fix_desc(desc);
1751
1752         desc->ld_active_tgt_count = 0;
1753         lod->lod_desc = *desc;
1754
1755         lod->lod_sp_me = LUSTRE_SP_CLI;
1756
1757         /* Set up allocation policy (QoS and RR) */
1758         INIT_LIST_HEAD(&lod->lod_qos.lq_oss_list);
1759         init_rwsem(&lod->lod_qos.lq_rw_sem);
1760         lod->lod_qos.lq_dirty = 1;
1761         lod->lod_qos.lq_rr.lqr_dirty = 1;
1762         lod->lod_qos.lq_reset = 1;
1763         /* Default priority is toward free space balance */
1764         lod->lod_qos.lq_prio_free = 232;
1765         /* Default threshold for rr (roughly 17%) */
1766         lod->lod_qos.lq_threshold_rr = 43;
1767
1768         /* Set up OST pool environment */
1769         lod->lod_pools_hash_body = cfs_hash_create("POOLS", HASH_POOLS_CUR_BITS,
1770                                                    HASH_POOLS_MAX_BITS,
1771                                                    HASH_POOLS_BKT_BITS, 0,
1772                                                    CFS_HASH_MIN_THETA,
1773                                                    CFS_HASH_MAX_THETA,
1774                                                    &pool_hash_operations,
1775                                                    CFS_HASH_DEFAULT);
1776         if (lod->lod_pools_hash_body == NULL)
1777                 RETURN(-ENOMEM);
1778
1779         INIT_LIST_HEAD(&lod->lod_pool_list);
1780         lod->lod_pool_count = 0;
1781         rc = lod_ost_pool_init(&lod->lod_pool_info, 0);
1782         if (rc)
1783                 GOTO(out_hash, rc);
1784         lod_qos_rr_init(&lod->lod_qos.lq_rr);
1785         rc = lod_ost_pool_init(&lod->lod_qos.lq_rr.lqr_pool, 0);
1786         if (rc)
1787                 GOTO(out_pool_info, rc);
1788
1789         RETURN(0);
1790
1791 out_pool_info:
1792         lod_ost_pool_free(&lod->lod_pool_info);
1793 out_hash:
1794         cfs_hash_putref(lod->lod_pools_hash_body);
1795
1796         return rc;
1797 }
1798
1799 /**
1800  * Release the structures describing the pools.
1801  *
1802  * \param[in] lod       LOD device from which we release the structures
1803  *
1804  * \retval              0 always
1805  */
1806 int lod_pools_fini(struct lod_device *lod)
1807 {
1808         struct obd_device   *obd = lod2obd(lod);
1809         struct pool_desc    *pool, *tmp;
1810         ENTRY;
1811
1812         list_for_each_entry_safe(pool, tmp, &lod->lod_pool_list, pool_list) {
1813                 /* free pool structs */
1814                 CDEBUG(D_INFO, "delete pool %p\n", pool);
1815                 /* In the function below, .hs_keycmp resolves to
1816                  * pool_hashkey_keycmp() */
1817                 /* coverity[overrun-buffer-val] */
1818                 lod_pool_del(obd, pool->pool_name);
1819         }
1820
1821         cfs_hash_putref(lod->lod_pools_hash_body);
1822         lod_ost_pool_free(&(lod->lod_qos.lq_rr.lqr_pool));
1823         lod_ost_pool_free(&lod->lod_pool_info);
1824
1825         RETURN(0);
1826 }