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  * along with this program; if not, write to the Free Software
  18  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
  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 /*
  27  * lustre/lod/lod_lov.c
  28  *
  29  * Author: Alex Zhuravlev <alexey.zhuravlev@intel.com>
  30  */
  31
  32 #ifndef EXPORT_SYMTAB
  33 # define EXPORT_SYMTAB
  34 #endif
  35 #define DEBUG_SUBSYSTEM S_MDS
  36
  37 #include <obd_class.h>
  38 #include <obd_lov.h>
  39
  40 #include "lod_internal.h"
  41
  42 /*
  43  * Keep a refcount of lod->lod_osts usage to prevent racing with
  44  * addition/deletion. Any function that expects lov_tgts to remain stationary
  45  * must take a ref.
  46  *
  47  * \param lod - is the lod device from which we want to grab a reference
  48  */
  49 void lod_getref(struct lod_device *lod)
  50 {
  51         cfs_down_read(&lod->lod_rw_sem);
  52         cfs_mutex_lock(&lod->lod_mutex);
  53         lod->lod_refcount++;
  54         cfs_mutex_unlock(&lod->lod_mutex);
  55 }
  56
  57 /*
  58  * Companion of lod_getref() to release a reference on the lod table.
  59  * If this is the last reference and the ost entry was scheduled for deletion,
  60  * the descriptor is removed from the array.
  61  *
  62  * \param lod - is the lod device from which we release a reference
  63  */
  64 void lod_putref(struct lod_device *lod)
  65 {
  66         cfs_mutex_lock(&lod->lod_mutex);
  67         lod->lod_refcount--;
  68         if (lod->lod_refcount == 0 && lod->lod_death_row) {
  69                 struct lod_ost_desc *ost_desc, *tmp;
  70                 int                  idx;
  71                 CFS_LIST_HEAD(kill);
  72
  73                 CDEBUG(D_CONFIG, "destroying %d lod desc\n",
  74                        lod->lod_death_row);
  75
  76                 cfs_foreach_bit(lod->lod_ost_bitmap, idx) {
  77                         ost_desc = OST_TGT(lod, idx);
  78                         LASSERT(ost_desc);
  79
  80                         if (!ost_desc->ltd_reap)
  81                                 continue;
  82
  83                         cfs_list_add(&ost_desc->ltd_kill, &kill);
  84
  85                         lod_ost_pool_remove(&lod->lod_pool_info, idx);
  86                         OST_TGT(lod, idx) = NULL;
  87                         lod->lod_ostnr--;
  88                         cfs_bitmap_clear(lod->lod_ost_bitmap, idx);
  89                         if (ost_desc->ltd_active)
  90                                 lod->lod_desc.ld_active_tgt_count--;
  91                         lod->lod_death_row--;
  92                 }
  93                 cfs_mutex_unlock(&lod->lod_mutex);
  94                 cfs_up_read(&lod->lod_rw_sem);
  95
  96                 cfs_list_for_each_entry_safe(ost_desc, tmp, &kill, ltd_kill) {
  97                         int rc;
  98                         cfs_list_del(&ost_desc->ltd_kill);
  99                         /* remove from QoS structures */
 100                         rc = qos_del_tgt(lod, ost_desc);
 101                         if (rc)
 102                                 CERROR("%s: qos_del_tgt(%s) failed: rc = %d\n",
 103                                        lod2obd(lod)->obd_name,
 104                                        obd_uuid2str(&ost_desc->ltd_uuid), rc);
 105                         rc = obd_disconnect(ost_desc->ltd_exp);
 106                         if (rc)
 107                                 CERROR("%s: failed to disconnect %s: rc = %d\n",
 108                                        lod2obd(lod)->obd_name,
 109                                        obd_uuid2str(&ost_desc->ltd_uuid), rc);
 110                         OBD_FREE_PTR(ost_desc);
 111                 }
 112         } else {
 113                 cfs_mutex_unlock(&lod->lod_mutex);
 114                 cfs_up_read(&lod->lod_rw_sem);
 115         }
 116 }
 117
 118 static int lod_bitmap_resize(struct lod_device *lod, __u32 newsize)
 119 {
 120         cfs_bitmap_t *new_bitmap, *old_bitmap = NULL;
 121         int           rc = 0;
 122         ENTRY;
 123
 124         /* grab write reference on the lod. Relocating the array requires
 125          * exclusive access */
 126         cfs_down_write(&lod->lod_rw_sem);
 127
 128         if (newsize <= lod->lod_osts_size)
 129                 /* someone else has already resize the array */
 130                 GOTO(out, rc = 0);
 131
 132         /* allocate new bitmap */
 133         new_bitmap = CFS_ALLOCATE_BITMAP(newsize);
 134         if (!new_bitmap)
 135                 GOTO(out, rc = -ENOMEM);
 136
 137         if (lod->lod_osts_size > 0) {
 138                 /* the bitmap already exists, we need
 139                  * to copy data from old one */
 140                 cfs_bitmap_copy(new_bitmap, lod->lod_ost_bitmap);
 141                 old_bitmap = lod->lod_ost_bitmap;
 142         }
 143
 144         lod->lod_osts_size  = newsize;
 145         lod->lod_ost_bitmap = new_bitmap;
 146
 147         if (old_bitmap)
 148                 CFS_FREE_BITMAP(old_bitmap);
 149
 150         CDEBUG(D_CONFIG, "ost size: %d\n", lod->lod_osts_size);
 151
 152         EXIT;
 153 out:
 154         cfs_up_write(&lod->lod_rw_sem);
 155         return rc;
 156 }
 157
 158 /*
 159  * Connect LOD to a new OSP and add it to the device table.
 160  *
 161  * \param env - is the environment passed by the caller
 162  * \param lod - is the LOD device to be connected to the new OSP
 163  * \param osp - is the name of OSP device name about to be added
 164  * \param index - is the OSP index
 165  * \param gen - is the generation number
 166  */
 167 int lod_add_device(const struct lu_env *env, struct lod_device *lod,
 168                    char *osp, unsigned index, unsigned gen, int active)
 169 {
 170         struct obd_connect_data *data = NULL;
 171         struct obd_export       *exp = NULL;
 172         struct obd_device       *obd;
 173         struct lu_device        *ldev;
 174         struct dt_device        *d;
 175         int                      rc;
 176         struct lod_ost_desc     *ost_desc;
 177         struct obd_uuid          obd_uuid;
 178
 179         ENTRY;
 180
 181         CDEBUG(D_CONFIG, "osp:%s idx:%d gen:%d\n", osp, index, gen);
 182
 183         if (gen <= 0) {
 184                 CERROR("request to add OBD %s with invalid generation: %d\n",
 185                        osp, gen);
 186                 RETURN(-EINVAL);
 187         }
 188
 189         obd_str2uuid(&obd_uuid, osp);
 190
 191         obd = class_find_client_obd(&obd_uuid, LUSTRE_OSP_NAME,
 192                                 &lod->lod_dt_dev.dd_lu_dev.ld_obd->obd_uuid);
 193         if (obd == NULL) {
 194                 CERROR("can't find %s device\n", osp);
 195                 RETURN(-EINVAL);
 196         }
 197
 198         OBD_ALLOC_PTR(data);
 199         if (data == NULL)
 200                 RETURN(-ENOMEM);
 201
 202         data->ocd_connect_flags = OBD_CONNECT_INDEX | OBD_CONNECT_VERSION;
 203         data->ocd_version = LUSTRE_VERSION_CODE;
 204         data->ocd_index = index;
 205
 206         rc = obd_connect(env, &exp, obd, &obd->obd_uuid, data, NULL);
 207         OBD_FREE_PTR(data);
 208         if (rc) {
 209                 CERROR("%s: cannot connect to next dev %s (%d)\n",
 210                        obd->obd_name, osp, rc);
 211                 GOTO(out_free, rc);
 212         }
 213
 214         LASSERT(obd->obd_lu_dev);
 215         LASSERT(obd->obd_lu_dev->ld_site = lod->lod_dt_dev.dd_lu_dev.ld_site);
 216
 217         ldev = obd->obd_lu_dev;
 218         d = lu2dt_dev(ldev);
 219
 220         /* Allocate ost descriptor and fill it */
 221         OBD_ALLOC_PTR(ost_desc);
 222         if (!ost_desc)
 223                 GOTO(out_conn, rc = -ENOMEM);
 224
 225         ost_desc->ltd_ost    = d;
 226         ost_desc->ltd_exp    = exp;
 227         ost_desc->ltd_uuid   = obd->u.cli.cl_target_uuid;
 228         ost_desc->ltd_gen    = gen;
 229         ost_desc->ltd_index  = index;
 230         ost_desc->ltd_active = active;
 231
 232         lod_getref(lod);
 233         if (index >= lod->lod_osts_size) {
 234                 /* we have to increase the size of the lod_osts array */
 235                 __u32  newsize;
 236
 237                 newsize = max(lod->lod_osts_size, (__u32)2);
 238                 while (newsize < index + 1)
 239                         newsize = newsize << 1;
 240
 241                 /* lod_bitmap_resize() needs lod_rw_sem
 242                  * which we hold with th reference */
 243                 lod_putref(lod);
 244
 245                 rc = lod_bitmap_resize(lod, newsize);
 246                 if (rc)
 247                         GOTO(out_desc, rc);
 248
 249                 lod_getref(lod);
 250         }
 251
 252         cfs_mutex_lock(&lod->lod_mutex);
 253         if (cfs_bitmap_check(lod->lod_ost_bitmap, index)) {
 254                 CERROR("%s: device %d is registered already\n", obd->obd_name,
 255                        index);
 256                 GOTO(out_mutex, rc = -EEXIST);
 257         }
 258
 259         if (lod->lod_ost_idx[index / OST_PTRS_PER_BLOCK] == NULL) {
 260                 OBD_ALLOC_PTR(lod->lod_ost_idx[index / OST_PTRS_PER_BLOCK]);
 261                 if (lod->lod_ost_idx[index / OST_PTRS_PER_BLOCK] == NULL) {
 262                         CERROR("can't allocate index to add %s\n",
 263                                obd->obd_name);
 264                         GOTO(out_mutex, rc = -ENOMEM);
 265                 }
 266         }
 267
 268         rc = lod_ost_pool_add(&lod->lod_pool_info, index, lod->lod_osts_size);
 269         if (rc) {
 270                 CERROR("%s: can't set up pool, failed with %d\n",
 271                        obd->obd_name, rc);
 272                 GOTO(out_mutex, rc);
 273         }
 274
 275         rc = qos_add_tgt(lod, ost_desc);
 276         if (rc) {
 277                 CERROR("%s: qos_add_tgt(%s) failed: rc = %d\n", obd->obd_name,
 278                        obd_uuid2str(&ost_desc->ltd_uuid), rc);
 279                 GOTO(out_pool, rc);
 280         }
 281
 282         /* The new OST is now a full citizen */
 283         if (index >= lod->lod_desc.ld_tgt_count)
 284                 lod->lod_desc.ld_tgt_count = index + 1;
 285         if (active)
 286                 lod->lod_desc.ld_active_tgt_count++;
 287         OST_TGT(lod, index) = ost_desc;
 288         cfs_bitmap_set(lod->lod_ost_bitmap, index);
 289         lod->lod_ostnr++;
 290         cfs_mutex_unlock(&lod->lod_mutex);
 291         lod_putref(lod);
 292
 293         if (lod->lod_recovery_completed)
 294                 ldev->ld_ops->ldo_recovery_complete(env, ldev);
 295
 296         RETURN(0);
 297
 298 out_pool:
 299         lod_ost_pool_remove(&lod->lod_pool_info, index);
 300 out_mutex:
 301         cfs_mutex_unlock(&lod->lod_mutex);
 302         lod_putref(lod);
 303 out_desc:
 304         OBD_FREE_PTR(ost_desc);
 305 out_conn:
 306         obd_disconnect(exp);
 307 out_free:
 308         return rc;
 309 }
 310
 311 /*
 312  * helper function to schedule OST removal from the device table
 313  */
 314 static void __lod_del_device(struct lod_device *lod, unsigned idx)
 315 {
 316         LASSERT(OST_TGT(lod,idx));
 317         if (OST_TGT(lod,idx)->ltd_reap == 0) {
 318                 OST_TGT(lod,idx)->ltd_reap = 1;
 319                 lod->lod_death_row++;
 320         }
 321 }
 322
 323 /*
 324  * Add support for administratively disabled OST (through the MGS).
 325  * Schedule a target for deletion.  Disconnection and real removal from the
 326  * table takes place in lod_putref() once the last table user release its
 327  * reference.
 328  *
 329  * \param env - is the environment passed by the caller
 330  * \param lod - is the lod device currently connected to the OSP about to be
 331  *              removed
 332  * \param osp - is the name of OSP device about to be removed
 333  * \param idx - is the OSP index
 334  * \param gen - is the generation number, not used currently
 335  */
 336 int lod_del_device(const struct lu_env *env, struct lod_device *lod,
 337                    char *osp, unsigned idx, unsigned gen)
 338 {
 339         struct obd_device *obd;
 340         int                rc = 0;
 341         struct obd_uuid    uuid;
 342         ENTRY;
 343
 344         CDEBUG(D_CONFIG, "osp:%s idx:%d gen:%d\n", osp, idx, gen);
 345
 346         obd = class_name2obd(osp);
 347         if (obd == NULL) {
 348                 CERROR("can't find %s device\n", osp);
 349                 RETURN(-EINVAL);
 350         }
 351
 352         if (gen <= 0) {
 353                 CERROR("%s: request to remove OBD %s with invalid generation %d"
 354                        "\n", obd->obd_name, osp, gen);
 355                 RETURN(-EINVAL);
 356         }
 357
 358         obd_str2uuid(&uuid,  osp);
 359
 360         lod_getref(lod);
 361         cfs_mutex_lock(&lod->lod_mutex);
 362         /* check that the index is allocated in the bitmap */
 363         if (!cfs_bitmap_check(lod->lod_ost_bitmap, idx) || !OST_TGT(lod,idx)) {
 364                 CERROR("%s: device %d is not set up\n", obd->obd_name, idx);
 365                 GOTO(out, rc = -EINVAL);
 366         }
 367
 368         /* check that the UUID matches */
 369         if (!obd_uuid_equals(&uuid, &OST_TGT(lod,idx)->ltd_uuid)) {
 370                 CERROR("%s: LOD target UUID %s at index %d does not match %s\n",
 371                        obd->obd_name, obd_uuid2str(&OST_TGT(lod,idx)->ltd_uuid),
 372                        idx, osp);
 373                 GOTO(out, rc = -EINVAL);
 374         }
 375
 376         __lod_del_device(lod, idx);
 377         EXIT;
 378 out:
 379         cfs_mutex_unlock(&lod->lod_mutex);
 380         lod_putref(lod);
 381         return(rc);
 382 }
 383
 384 int lod_ea_store_resize(struct lod_thread_info *info, int size)
 385 {
 386         int round = size_roundup_power2(size);
 387
 388         LASSERT(round <= lov_mds_md_size(LOV_MAX_STRIPE_COUNT, LOV_MAGIC_V3));
 389         if (info->lti_ea_store) {
 390                 LASSERT(info->lti_ea_store_size);
 391                 LASSERT(info->lti_ea_store_size < round);
 392                 CDEBUG(D_INFO, "EA store size %d is not enough, need %d\n",
 393                        info->lti_ea_store_size, round);
 394                 OBD_FREE_LARGE(info->lti_ea_store, info->lti_ea_store_size);
 395                 info->lti_ea_store = NULL;
 396                 info->lti_ea_store_size = 0;
 397         }
 398
 399         OBD_ALLOC_LARGE(info->lti_ea_store, round);
 400         if (info->lti_ea_store == NULL)
 401                 RETURN(-ENOMEM);
 402         info->lti_ea_store_size = round;
 403         RETURN(0);
 404 }
 405
 406 int lod_get_lov_ea(const struct lu_env *env, struct lod_object *lo)
 407 {
 408         struct lod_thread_info *info = lod_env_info(env);
 409         struct dt_object       *next = dt_object_child(&lo->ldo_obj);
 410         int                     rc;
 411         ENTRY;
 412
 413         LASSERT(info);
 414
 415         if (unlikely(info->lti_ea_store_size == 0)) {
 416                 /* just to enter in allocation block below */
 417                 rc = -ERANGE;
 418         } else {
 419 repeat:
 420                 info->lti_buf.lb_buf = info->lti_ea_store;
 421                 info->lti_buf.lb_len = info->lti_ea_store_size;
 422                 rc = dt_xattr_get(env, next, &info->lti_buf, XATTR_NAME_LOV,
 423                                   BYPASS_CAPA);
 424         }
 425         /* if object is not striped or inaccessible */
 426         if (rc == -ENODATA)
 427                 RETURN(0);
 428
 429         if (rc == -ERANGE) {
 430                 /* EA doesn't fit, reallocate new buffer */
 431                 rc = dt_xattr_get(env, next, &LU_BUF_NULL, XATTR_NAME_LOV,
 432                                   BYPASS_CAPA);
 433                 if (rc == -ENODATA)
 434                         RETURN(0);
 435                 else if (rc < 0)
 436                         RETURN(rc);
 437
 438                 LASSERT(rc > 0);
 439                 rc = lod_ea_store_resize(info, rc);
 440                 if (rc)
 441                         RETURN(rc);
 442                 goto repeat;
 443         }
 444
 445         RETURN(rc);
 446 }
 447
 448 int lod_store_def_striping(const struct lu_env *env, struct dt_object *dt,
 449                            struct thandle *th)
 450 {
 451         struct lod_thread_info  *info = lod_env_info(env);
 452         struct lod_object       *lo = lod_dt_obj(dt);
 453         struct dt_object        *next = dt_object_child(dt);
 454         struct lov_user_md_v3   *v3;
 455         int                      rc;
 456         ENTRY;
 457
 458         LASSERT(S_ISDIR(dt->do_lu.lo_header->loh_attr));
 459
 460         /*
 461          * store striping defaults into new directory
 462          * used to implement defaults inheritance
 463          */
 464
 465         /* probably nothing to inherite */
 466         if (lo->ldo_striping_cached == 0)
 467                 RETURN(0);
 468
 469         if (LOVEA_DELETE_VALUES(lo->ldo_def_stripe_size, lo->ldo_def_stripenr,
 470                                 lo->ldo_def_stripe_offset))
 471                 RETURN(0);
 472
 473         /* XXX: use thread info */
 474         OBD_ALLOC_PTR(v3);
 475         if (v3 == NULL)
 476                 RETURN(-ENOMEM);
 477
 478         v3->lmm_magic = cpu_to_le32(LOV_MAGIC_V3);
 479         v3->lmm_pattern = cpu_to_le32(LOV_PATTERN_RAID0);
 480         v3->lmm_object_id = 0;
 481         v3->lmm_object_seq = 0;
 482         v3->lmm_stripe_size = cpu_to_le32(lo->ldo_def_stripe_size);
 483         v3->lmm_stripe_count = cpu_to_le16(lo->ldo_def_stripenr);
 484         v3->lmm_stripe_offset = cpu_to_le16(lo->ldo_def_stripe_offset);
 485         if (lo->ldo_pool)
 486                 strncpy(v3->lmm_pool_name, lo->ldo_pool, LOV_MAXPOOLNAME);
 487
 488         info->lti_buf.lb_buf = v3;
 489         info->lti_buf.lb_len = sizeof(*v3);
 490         rc = dt_xattr_set(env, next, &info->lti_buf, XATTR_NAME_LOV, 0, th,
 491                         BYPASS_CAPA);
 492
 493         OBD_FREE_PTR(v3);
 494
 495         RETURN(rc);
 496 }
 497
 498 /*
 499  * allocate array of objects pointers, find/create objects
 500  * stripenr and other fields should be initialized by this moment
 501  */
 502 int lod_initialize_objects(const struct lu_env *env, struct lod_object *lo,
 503                            struct lov_ost_data_v1 *objs)
 504 {
 505         struct lod_thread_info  *info = lod_env_info(env);
 506         struct lod_device       *md = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
 507         struct lu_object        *o, *n;
 508         struct lu_device        *nd;
 509         int                     i, idx, rc = 0;
 510         ENTRY;
 511
 512         LASSERT(lo);
 513         LASSERT(lo->ldo_stripe == NULL);
 514         LASSERT(lo->ldo_stripenr > 0);
 515         LASSERT(lo->ldo_stripe_size > 0);
 516
 517         i = sizeof(struct dt_object *) * lo->ldo_stripenr;
 518         OBD_ALLOC(lo->ldo_stripe, i);
 519         if (lo->ldo_stripe == NULL)
 520                 GOTO(out, rc = -ENOMEM);
 521         lo->ldo_stripes_allocated = lo->ldo_stripenr;
 522
 523         for (i = 0; i < lo->ldo_stripenr; i++) {
 524
 525                 info->lti_ostid.oi_id = le64_to_cpu(objs[i].l_object_id);
 526                 /* XXX: support for DNE? */
 527                 info->lti_ostid.oi_seq = le64_to_cpu(objs[i].l_object_seq);
 528                 idx = le64_to_cpu(objs[i].l_ost_idx);
 529                 fid_ostid_unpack(&info->lti_fid, &info->lti_ostid, idx);
 530
 531                 /*
 532                  * XXX: assertion is left for testing, to make
 533                  * sure we never process requests till configuration
 534                  * is completed. to be changed to -EINVAL
 535                  */
 536
 537                 lod_getref(md);
 538                 LASSERT(cfs_bitmap_check(md->lod_ost_bitmap, idx));
 539                 LASSERT(OST_TGT(md,idx));
 540                 LASSERTF(OST_TGT(md,idx)->ltd_ost, "idx %d\n", idx);
 541                 nd = &OST_TGT(md,idx)->ltd_ost->dd_lu_dev;
 542                 lod_putref(md);
 543
 544                 o = lu_object_find_at(env, nd, &info->lti_fid, NULL);
 545                 if (IS_ERR(o))
 546                         GOTO(out, rc = PTR_ERR(o));
 547
 548                 n = lu_object_locate(o->lo_header, nd->ld_type);
 549                 LASSERT(n);
 550
 551                 lo->ldo_stripe[i] = container_of(n, struct dt_object, do_lu);
 552         }
 553
 554 out:
 555         RETURN(rc);
 556 }
 557
 558 /*
 559  * Parse striping information stored in lti_ea_store
 560  */
 561 int lod_parse_striping(const struct lu_env *env, struct lod_object *lo,
 562                        const struct lu_buf *buf)
 563 {
 564         struct lov_mds_md_v1    *lmm;
 565         struct lov_ost_data_v1  *objs;
 566         __u32                    magic;
 567         int                      rc = 0;
 568         ENTRY;
 569
 570         LASSERT(buf);
 571         LASSERT(buf->lb_buf);
 572         LASSERT(buf->lb_len);
 573
 574         lmm = (struct lov_mds_md_v1 *) buf->lb_buf;
 575         magic = le32_to_cpu(lmm->lmm_magic);
 576
 577         if (magic != LOV_MAGIC_V1 && magic != LOV_MAGIC_V3)
 578                 GOTO(out, rc = -EINVAL);
 579         if (le32_to_cpu(lmm->lmm_pattern) != LOV_PATTERN_RAID0)
 580                 GOTO(out, rc = -EINVAL);
 581
 582         lo->ldo_stripe_size = le32_to_cpu(lmm->lmm_stripe_size);
 583         lo->ldo_stripenr = le16_to_cpu(lmm->lmm_stripe_count);
 584         lo->ldo_layout_gen = le16_to_cpu(lmm->lmm_layout_gen);
 585
 586         LASSERT(buf->lb_len >= lov_mds_md_size(lo->ldo_stripenr, magic));
 587
 588         if (magic == LOV_MAGIC_V3) {
 589                 struct lov_mds_md_v3 *v3 = (struct lov_mds_md_v3 *) lmm;
 590                 objs = &v3->lmm_objects[0];
 591                 lod_object_set_pool(lo, v3->lmm_pool_name);
 592         } else {
 593                 objs = &lmm->lmm_objects[0];
 594         }
 595
 596         rc = lod_initialize_objects(env, lo, objs);
 597
 598 out:
 599         RETURN(rc);
 600 }
 601
 602 /*
 603  * Load and parse striping information, create in-core representation for the
 604  * stripes
 605  */
 606 int lod_load_striping(const struct lu_env *env, struct lod_object *lo)
 607 {
 608         struct lod_thread_info  *info = lod_env_info(env);
 609         struct dt_object        *next = dt_object_child(&lo->ldo_obj);
 610         int                      rc;
 611         ENTRY;
 612
 613         /*
 614          * currently this code is supposed to be called from declaration
 615          * phase only, thus the object is not expected to be locked by caller
 616          */
 617         dt_write_lock(env, next, 0);
 618         /* already initialized? */
 619         if (lo->ldo_stripe) {
 620                 int i;
 621                 /* check validity */
 622                 for (i = 0; i < lo->ldo_stripenr; i++)
 623                         LASSERTF(lo->ldo_stripe[i], "stripe %d is NULL\n", i);
 624                 GOTO(out, rc = 0);
 625         }
 626
 627         if (!dt_object_exists(next))
 628                 GOTO(out, rc = 0);
 629
 630         /* only regular files can be striped */
 631         if (!(lu_object_attr(lod2lu_obj(lo)) & S_IFREG))
 632                 GOTO(out, rc = 0);
 633
 634         LASSERT(lo->ldo_stripenr == 0);
 635
 636         rc = lod_get_lov_ea(env, lo);
 637         if (rc <= 0)
 638                 GOTO(out, rc);
 639
 640         /*
 641          * there is LOV EA (striping information) in this object
 642          * let's parse it and create in-core objects for the stripes
 643          */
 644         info->lti_buf.lb_buf = info->lti_ea_store;
 645         info->lti_buf.lb_len = info->lti_ea_store_size;
 646         rc = lod_parse_striping(env, lo, &info->lti_buf);
 647 out:
 648         dt_write_unlock(env, next);
 649         RETURN(rc);
 650 }
 651
 652 int lod_verify_striping(struct lod_device *d, const struct lu_buf *buf,
 653                         int specific)
 654 {
 655         struct lov_user_md_v1   *lum;
 656         struct lov_user_md_v3   *v3 = NULL;
 657         struct pool_desc        *pool = NULL;
 658         int                      rc;
 659         ENTRY;
 660
 661         lum = buf->lb_buf;
 662
 663         if (lum->lmm_magic != LOV_USER_MAGIC_V1 &&
 664             lum->lmm_magic != LOV_USER_MAGIC_V3 &&
 665             lum->lmm_magic != LOV_MAGIC_V1_DEF &&
 666             lum->lmm_magic != LOV_MAGIC_V3_DEF) {
 667                 CDEBUG(D_IOCTL, "bad userland LOV MAGIC: %#x\n",
 668                        lum->lmm_magic);
 669                 RETURN(-EINVAL);
 670         }
 671
 672         if ((specific && lum->lmm_pattern != LOV_PATTERN_RAID0) ||
 673             (specific == 0 && lum->lmm_pattern != 0)) {
 674                 CDEBUG(D_IOCTL, "bad userland stripe pattern: %#x\n",
 675                        lum->lmm_pattern);
 676                 RETURN(-EINVAL);
 677         }
 678
 679         /* 64kB is the largest common page size we see (ia64), and matches the
 680          * check in lfs */
 681         if (lum->lmm_stripe_size & (LOV_MIN_STRIPE_SIZE - 1)) {
 682                 CDEBUG(D_IOCTL, "stripe size %u not multiple of %u, fixing\n",
 683                        lum->lmm_stripe_size, LOV_MIN_STRIPE_SIZE);
 684                 RETURN(-EINVAL);
 685         }
 686
 687         /* an offset of -1 is treated as a "special" valid offset */
 688         if (lum->lmm_stripe_offset != (typeof(lum->lmm_stripe_offset))(-1)) {
 689                 /* if offset is not within valid range [0, osts_size) */
 690                 if (lum->lmm_stripe_offset >= d->lod_osts_size) {
 691                         CDEBUG(D_IOCTL, "stripe offset %u >= bitmap size %u\n",
 692                                lum->lmm_stripe_offset, d->lod_osts_size);
 693                         RETURN(-EINVAL);
 694                 }
 695
 696                 /* if lmm_stripe_offset is *not* in bitmap */
 697                 if (!cfs_bitmap_check(d->lod_ost_bitmap,
 698                                       lum->lmm_stripe_offset)) {
 699                         CDEBUG(D_IOCTL, "stripe offset %u not in bitmap\n",
 700                                lum->lmm_stripe_offset);
 701                         RETURN(-EINVAL);
 702                 }
 703         }
 704
 705         if (lum->lmm_magic == LOV_USER_MAGIC_V3)
 706                 v3 = buf->lb_buf;
 707
 708         if (v3)
 709                 pool = lod_find_pool(d, v3->lmm_pool_name);
 710
 711         if (pool != NULL) {
 712                 __u16 offs = v3->lmm_stripe_offset;
 713
 714                 if (offs != (typeof(v3->lmm_stripe_offset))(-1)) {
 715                         rc = lod_check_index_in_pool(offs, pool);
 716                         if (rc < 0) {
 717                                 lod_pool_putref(pool);
 718                                 RETURN(-EINVAL);
 719                         }
 720                 }
 721
 722                 if (specific && lum->lmm_stripe_count > pool_tgt_count(pool)) {
 723                         CDEBUG(D_IOCTL,
 724                                "stripe count %u > # OSTs %u in the pool\n",
 725                                lum->lmm_stripe_count, pool_tgt_count(pool));
 726                         lod_pool_putref(pool);
 727                         RETURN(-EINVAL);
 728                 }
 729
 730                 lod_pool_putref(pool);
 731         }
 732
 733         RETURN(0);
 734 }
 735
 736 void lod_fix_desc_stripe_size(__u64 *val)
 737 {
 738         if (*val < PTLRPC_MAX_BRW_SIZE) {
 739                 LCONSOLE_WARN("Increasing default stripe size to min %u\n",
 740                               PTLRPC_MAX_BRW_SIZE);
 741                 *val = PTLRPC_MAX_BRW_SIZE;
 742         } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) {
 743                 *val &= ~(LOV_MIN_STRIPE_SIZE - 1);
 744                 LCONSOLE_WARN("Changing default stripe size to "LPU64" (a "
 745                               "multiple of %u)\n",
 746                               *val, LOV_MIN_STRIPE_SIZE);
 747         }
 748 }
 749
 750 void lod_fix_desc_stripe_count(__u32 *val)
 751 {
 752         if (*val == 0)
 753                 *val = 1;
 754 }
 755
 756 void lod_fix_desc_pattern(__u32 *val)
 757 {
 758         /* from lov_setstripe */
 759         if ((*val != 0) && (*val != LOV_PATTERN_RAID0)) {
 760                 LCONSOLE_WARN("Unknown stripe pattern: %#x\n", *val);
 761                 *val = 0;
 762         }
 763 }
 764
 765 void lod_fix_desc_qos_maxage(__u32 *val)
 766 {
 767         /* fix qos_maxage */
 768         if (*val == 0)
 769                 *val = QOS_DEFAULT_MAXAGE;
 770 }
 771
 772 void lod_fix_desc(struct lov_desc *desc)
 773 {
 774         lod_fix_desc_stripe_size(&desc->ld_default_stripe_size);
 775         lod_fix_desc_stripe_count(&desc->ld_default_stripe_count);
 776         lod_fix_desc_pattern(&desc->ld_pattern);
 777         lod_fix_desc_qos_maxage(&desc->ld_qos_maxage);
 778 }
 779
 780 int lod_pools_init(struct lod_device *lod, struct lustre_cfg *lcfg)
 781 {
 782         struct lprocfs_static_vars  lvars = { 0 };
 783         struct obd_device          *obd;
 784         struct lov_desc            *desc;
 785         int                         rc;
 786         ENTRY;
 787
 788         obd = class_name2obd(lustre_cfg_string(lcfg, 0));
 789         LASSERT(obd != NULL);
 790         obd->obd_lu_dev = &lod->lod_dt_dev.dd_lu_dev;
 791
 792         if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
 793                 CERROR("LOD setup requires a descriptor\n");
 794                 RETURN(-EINVAL);
 795         }
 796
 797         desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
 798
 799         if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
 800                 CERROR("descriptor size wrong: %d > %d\n",
 801                        (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
 802                 RETURN(-EINVAL);
 803         }
 804
 805         if (desc->ld_magic != LOV_DESC_MAGIC) {
 806                 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
 807                         CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
 808                                obd->obd_name, desc);
 809                         lustre_swab_lov_desc(desc);
 810                 } else {
 811                         CERROR("%s: Bad lov desc magic: %#x\n",
 812                                obd->obd_name, desc->ld_magic);
 813                         RETURN(-EINVAL);
 814                 }
 815         }
 816
 817         lod_fix_desc(desc);
 818
 819         desc->ld_active_tgt_count = 0;
 820         lod->lod_desc = *desc;
 821
 822         lod->lod_sp_me = LUSTRE_SP_CLI;
 823
 824         /* Set up allocation policy (QoS and RR) */
 825         CFS_INIT_LIST_HEAD(&lod->lod_qos.lq_oss_list);
 826         cfs_init_rwsem(&lod->lod_qos.lq_rw_sem);
 827         lod->lod_qos.lq_dirty = 1;
 828         lod->lod_qos.lq_rr.lqr_dirty = 1;
 829         lod->lod_qos.lq_reset = 1;
 830         /* Default priority is toward free space balance */
 831         lod->lod_qos.lq_prio_free = 232;
 832         /* Default threshold for rr (roughly 17%) */
 833         lod->lod_qos.lq_threshold_rr = 43;
 834         /* Init statfs fields */
 835         OBD_ALLOC_PTR(lod->lod_qos.lq_statfs_data);
 836         if (NULL == lod->lod_qos.lq_statfs_data)
 837                 RETURN(-ENOMEM);
 838         cfs_waitq_init(&lod->lod_qos.lq_statfs_waitq);
 839
 840         /* Set up OST pool environment */
 841         lod->lod_pools_hash_body = cfs_hash_create("POOLS", HASH_POOLS_CUR_BITS,
 842                                                    HASH_POOLS_MAX_BITS,
 843                                                    HASH_POOLS_BKT_BITS, 0,
 844                                                    CFS_HASH_MIN_THETA,
 845                                                    CFS_HASH_MAX_THETA,
 846                                                    &pool_hash_operations,
 847                                                    CFS_HASH_DEFAULT);
 848         if (!lod->lod_pools_hash_body)
 849                 GOTO(out_statfs, rc = -ENOMEM);
 850         CFS_INIT_LIST_HEAD(&lod->lod_pool_list);
 851         lod->lod_pool_count = 0;
 852         rc = lod_ost_pool_init(&lod->lod_pool_info, 0);
 853         if (rc)
 854                 GOTO(out_hash, rc);
 855         rc = lod_ost_pool_init(&lod->lod_qos.lq_rr.lqr_pool, 0);
 856         if (rc)
 857                 GOTO(out_pool_info, rc);
 858
 859         /* the OST array and bitmap are allocated/grown dynamically as OSTs are
 860          * added to the LOD, see lod_add_device() */
 861         lod->lod_ost_bitmap = NULL;
 862         lod->lod_osts_size  = 0;
 863         lod->lod_ostnr      = 0;
 864
 865         lod->lod_death_row = 0;
 866         lod->lod_refcount  = 0;
 867
 868         lprocfs_lod_init_vars(&lvars);
 869         lprocfs_obd_setup(obd, lvars.obd_vars);
 870
 871 #ifdef LPROCFS
 872         rc = lprocfs_seq_create(obd->obd_proc_entry, "target_obd",
 873                                 0444, &lod_proc_target_fops, obd);
 874         if (rc) {
 875                 CWARN("%s: Error adding the target_obd file %d\n",
 876                       obd->obd_name, rc);
 877                 GOTO(out_lproc, rc);
 878         }
 879         lod->lod_pool_proc_entry = lprocfs_register("pools",
 880                                                     obd->obd_proc_entry,
 881                                                     NULL, NULL);
 882         if (IS_ERR(lod->lod_pool_proc_entry)) {
 883                 int ret = PTR_ERR(lod->lod_pool_proc_entry);
 884                 lod->lod_pool_proc_entry = NULL;
 885                 CWARN("%s: Failed to create pool proc file %d\n",
 886                       obd->obd_name, ret);
 887                 rc = lod_pools_fini(lod);
 888                 RETURN(ret);
 889         }
 890 #endif
 891
 892         RETURN(0);
 893
 894 out_lproc:
 895         lprocfs_obd_cleanup(obd);
 896         lod_ost_pool_free(&lod->lod_qos.lq_rr.lqr_pool);
 897 out_pool_info:
 898         lod_ost_pool_free(&lod->lod_pool_info);
 899 out_hash:
 900         cfs_hash_putref(lod->lod_pools_hash_body);
 901 out_statfs:
 902         OBD_FREE_PTR(lod->lod_qos.lq_statfs_data);
 903         return rc;
 904 }
 905
 906 int lod_pools_fini(struct lod_device *lod)
 907 {
 908         struct obd_device   *obd = lod2obd(lod);
 909         cfs_list_t          *pos, *tmp;
 910         struct pool_desc    *pool;
 911         ENTRY;
 912
 913         cfs_list_for_each_safe(pos, tmp, &lod->lod_pool_list) {
 914                 pool = cfs_list_entry(pos, struct pool_desc, pool_list);
 915                 /* free pool structs */
 916                 CDEBUG(D_INFO, "delete pool %p\n", pool);
 917                 lod_pool_del(obd, pool->pool_name);
 918         }
 919
 920         if (lod->lod_osts_size > 0) {
 921                 int idx;
 922                 lod_getref(lod);
 923                 cfs_mutex_lock(&lod->lod_mutex);
 924                 cfs_foreach_bit(lod->lod_ost_bitmap, idx)
 925                         __lod_del_device(lod, idx);
 926                 cfs_mutex_unlock(&lod->lod_mutex);
 927                 lod_putref(lod);
 928                 CFS_FREE_BITMAP(lod->lod_ost_bitmap);
 929                 for (idx = 0; idx < OST_PTRS; idx++) {
 930                         if (lod->lod_ost_idx[idx])
 931                                 OBD_FREE_PTR(lod->lod_ost_idx[idx]);
 932                 }
 933                 lod->lod_osts_size = 0;
 934         }
 935
 936         cfs_hash_putref(lod->lod_pools_hash_body);
 937         lod_ost_pool_free(&(lod->lod_qos.lq_rr.lqr_pool));
 938         lod_ost_pool_free(&lod->lod_pool_info);
 939
 940         /* clear pools parent proc entry only after all pools are killed */
 941         if (lod->lod_pool_proc_entry) {
 942                 lprocfs_remove(&lod->lod_pool_proc_entry);
 943                 lod->lod_pool_proc_entry = NULL;
 944         }
 945
 946         lprocfs_obd_cleanup(obd);
 947
 948         OBD_FREE_PTR(lod->lod_qos.lq_statfs_data);
 949         RETURN(0);
 950 }
 951