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 /*
 407  * generate and write LOV EA for given striped object
 408  */
 409 int lod_generate_and_set_lovea(const struct lu_env *env,
 410                                struct lod_object *lo, struct thandle *th)
 411 {
 412         struct lod_thread_info  *info = lod_env_info(env);
 413         struct dt_object        *next = dt_object_child(&lo->ldo_obj);
 414         const struct lu_fid     *fid  = lu_object_fid(&lo->ldo_obj.do_lu);
 415         struct lov_mds_md_v1    *lmm;
 416         struct lov_ost_data_v1  *objs;
 417         __u32                    magic;
 418         int                      i, rc, lmm_size;
 419         ENTRY;
 420
 421         LASSERT(lo);
 422         LASSERT(lo->ldo_stripenr > 0);
 423
 424         magic = lo->ldo_pool ? LOV_MAGIC_V3 : LOV_MAGIC_V1;
 425         lmm_size = lov_mds_md_size(lo->ldo_stripenr, magic);
 426         if (info->lti_ea_store_size < lmm_size) {
 427                 rc = lod_ea_store_resize(info, lmm_size);
 428                 if (rc)
 429                         RETURN(rc);
 430         }
 431
 432         lmm = info->lti_ea_store;
 433
 434         lmm->lmm_magic = cpu_to_le32(magic);
 435         lmm->lmm_pattern = cpu_to_le32(LOV_PATTERN_RAID0);
 436         lmm->lmm_object_id = cpu_to_le64(fid_ver_oid(fid));
 437         lmm->lmm_object_seq = cpu_to_le64(fid_seq(fid));
 438         lmm->lmm_stripe_size = cpu_to_le32(lo->ldo_stripe_size);
 439         lmm->lmm_stripe_count = cpu_to_le16(lo->ldo_stripenr);
 440         lmm->lmm_layout_gen = 0;
 441         if (magic == LOV_MAGIC_V1) {
 442                 objs = &lmm->lmm_objects[0];
 443         } else {
 444                 struct lov_mds_md_v3 *v3 = (struct lov_mds_md_v3 *) lmm;
 445                 strncpy(v3->lmm_pool_name, lo->ldo_pool, LOV_MAXPOOLNAME);
 446                 objs = &v3->lmm_objects[0];
 447         }
 448
 449         for (i = 0; i < lo->ldo_stripenr; i++) {
 450                 const struct lu_fid *fid;
 451
 452                 LASSERT(lo->ldo_stripe[i]);
 453                 fid = lu_object_fid(&lo->ldo_stripe[i]->do_lu);
 454
 455                 rc = fid_ostid_pack(fid, &info->lti_ostid);
 456                 LASSERT(rc == 0);
 457                 LASSERT(info->lti_ostid.oi_seq == FID_SEQ_OST_MDT0);
 458
 459                 objs[i].l_object_id  = cpu_to_le64(info->lti_ostid.oi_id);
 460                 objs[i].l_object_seq = cpu_to_le64(info->lti_ostid.oi_seq);
 461                 objs[i].l_ost_gen    = cpu_to_le32(1);
 462                 objs[i].l_ost_idx    = cpu_to_le32(fid_idif_ost_idx(fid));
 463         }
 464
 465         info->lti_buf.lb_buf = lmm;
 466         info->lti_buf.lb_len = lmm_size;
 467         rc = dt_xattr_set(env, next, &info->lti_buf, XATTR_NAME_LOV, 0,
 468                           th, BYPASS_CAPA);
 469
 470         RETURN(rc);
 471 }
 472
 473 int lod_get_lov_ea(const struct lu_env *env, struct lod_object *lo)
 474 {
 475         struct lod_thread_info *info = lod_env_info(env);
 476         struct dt_object       *next = dt_object_child(&lo->ldo_obj);
 477         int                     rc;
 478         ENTRY;
 479
 480         LASSERT(info);
 481
 482         if (unlikely(info->lti_ea_store_size == 0)) {
 483                 /* just to enter in allocation block below */
 484                 rc = -ERANGE;
 485         } else {
 486 repeat:
 487                 info->lti_buf.lb_buf = info->lti_ea_store;
 488                 info->lti_buf.lb_len = info->lti_ea_store_size;
 489                 rc = dt_xattr_get(env, next, &info->lti_buf, XATTR_NAME_LOV,
 490                                   BYPASS_CAPA);
 491         }
 492         /* if object is not striped or inaccessible */
 493         if (rc == -ENODATA)
 494                 RETURN(0);
 495
 496         if (rc == -ERANGE) {
 497                 /* EA doesn't fit, reallocate new buffer */
 498                 rc = dt_xattr_get(env, next, &LU_BUF_NULL, XATTR_NAME_LOV,
 499                                   BYPASS_CAPA);
 500                 if (rc == -ENODATA)
 501                         RETURN(0);
 502                 else if (rc < 0)
 503                         RETURN(rc);
 504
 505                 LASSERT(rc > 0);
 506                 rc = lod_ea_store_resize(info, rc);
 507                 if (rc)
 508                         RETURN(rc);
 509                 goto repeat;
 510         }
 511
 512         RETURN(rc);
 513 }
 514
 515 int lod_store_def_striping(const struct lu_env *env, struct dt_object *dt,
 516                            struct thandle *th)
 517 {
 518         struct lod_thread_info  *info = lod_env_info(env);
 519         struct lod_object       *lo = lod_dt_obj(dt);
 520         struct dt_object        *next = dt_object_child(dt);
 521         struct lov_user_md_v3   *v3;
 522         int                      rc;
 523         ENTRY;
 524
 525         LASSERT(S_ISDIR(dt->do_lu.lo_header->loh_attr));
 526
 527         /*
 528          * store striping defaults into new directory
 529          * used to implement defaults inheritance
 530          */
 531
 532         /* probably nothing to inherite */
 533         if (lo->ldo_striping_cached == 0)
 534                 RETURN(0);
 535
 536         if (LOVEA_DELETE_VALUES(lo->ldo_def_stripe_size, lo->ldo_def_stripenr,
 537                                 lo->ldo_def_stripe_offset))
 538                 RETURN(0);
 539
 540         /* XXX: use thread info */
 541         OBD_ALLOC_PTR(v3);
 542         if (v3 == NULL)
 543                 RETURN(-ENOMEM);
 544
 545         v3->lmm_magic = cpu_to_le32(LOV_MAGIC_V3);
 546         v3->lmm_pattern = cpu_to_le32(LOV_PATTERN_RAID0);
 547         v3->lmm_object_id = 0;
 548         v3->lmm_object_seq = 0;
 549         v3->lmm_stripe_size = cpu_to_le32(lo->ldo_def_stripe_size);
 550         v3->lmm_stripe_count = cpu_to_le16(lo->ldo_def_stripenr);
 551         v3->lmm_stripe_offset = cpu_to_le16(lo->ldo_def_stripe_offset);
 552         if (lo->ldo_pool)
 553                 strncpy(v3->lmm_pool_name, lo->ldo_pool, LOV_MAXPOOLNAME);
 554
 555         info->lti_buf.lb_buf = v3;
 556         info->lti_buf.lb_len = sizeof(*v3);
 557         rc = dt_xattr_set(env, next, &info->lti_buf, XATTR_NAME_LOV, 0, th,
 558                         BYPASS_CAPA);
 559
 560         OBD_FREE_PTR(v3);
 561
 562         RETURN(rc);
 563 }
 564
 565 /*
 566  * allocate array of objects pointers, find/create objects
 567  * stripenr and other fields should be initialized by this moment
 568  */
 569 int lod_initialize_objects(const struct lu_env *env, struct lod_object *lo,
 570                            struct lov_ost_data_v1 *objs)
 571 {
 572         struct lod_thread_info  *info = lod_env_info(env);
 573         struct lod_device       *md = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
 574         struct lu_object        *o, *n;
 575         struct lu_device        *nd;
 576         int                     i, idx, rc = 0;
 577         ENTRY;
 578
 579         LASSERT(lo);
 580         LASSERT(lo->ldo_stripe == NULL);
 581         LASSERT(lo->ldo_stripenr > 0);
 582         LASSERT(lo->ldo_stripe_size > 0);
 583
 584         i = sizeof(struct dt_object *) * lo->ldo_stripenr;
 585         OBD_ALLOC(lo->ldo_stripe, i);
 586         if (lo->ldo_stripe == NULL)
 587                 GOTO(out, rc = -ENOMEM);
 588         lo->ldo_stripes_allocated = lo->ldo_stripenr;
 589
 590         for (i = 0; i < lo->ldo_stripenr; i++) {
 591
 592                 info->lti_ostid.oi_id = le64_to_cpu(objs[i].l_object_id);
 593                 /* XXX: support for DNE? */
 594                 info->lti_ostid.oi_seq = le64_to_cpu(objs[i].l_object_seq);
 595                 idx = le64_to_cpu(objs[i].l_ost_idx);
 596                 fid_ostid_unpack(&info->lti_fid, &info->lti_ostid, idx);
 597
 598                 /*
 599                  * XXX: assertion is left for testing, to make
 600                  * sure we never process requests till configuration
 601                  * is completed. to be changed to -EINVAL
 602                  */
 603
 604                 lod_getref(md);
 605                 LASSERT(cfs_bitmap_check(md->lod_ost_bitmap, idx));
 606                 LASSERT(OST_TGT(md,idx));
 607                 LASSERTF(OST_TGT(md,idx)->ltd_ost, "idx %d\n", idx);
 608                 nd = &OST_TGT(md,idx)->ltd_ost->dd_lu_dev;
 609                 lod_putref(md);
 610
 611                 o = lu_object_find_at(env, nd, &info->lti_fid, NULL);
 612                 if (IS_ERR(o))
 613                         GOTO(out, rc = PTR_ERR(o));
 614
 615                 n = lu_object_locate(o->lo_header, nd->ld_type);
 616                 LASSERT(n);
 617
 618                 lo->ldo_stripe[i] = container_of(n, struct dt_object, do_lu);
 619         }
 620
 621 out:
 622         RETURN(rc);
 623 }
 624
 625 /*
 626  * Parse striping information stored in lti_ea_store
 627  */
 628 int lod_parse_striping(const struct lu_env *env, struct lod_object *lo,
 629                        const struct lu_buf *buf)
 630 {
 631         struct lov_mds_md_v1    *lmm;
 632         struct lov_ost_data_v1  *objs;
 633         __u32                    magic;
 634         int                      rc = 0;
 635         ENTRY;
 636
 637         LASSERT(buf);
 638         LASSERT(buf->lb_buf);
 639         LASSERT(buf->lb_len);
 640
 641         lmm = (struct lov_mds_md_v1 *) buf->lb_buf;
 642         magic = le32_to_cpu(lmm->lmm_magic);
 643
 644         if (magic != LOV_MAGIC_V1 && magic != LOV_MAGIC_V3)
 645                 GOTO(out, rc = -EINVAL);
 646         if (le32_to_cpu(lmm->lmm_pattern) != LOV_PATTERN_RAID0)
 647                 GOTO(out, rc = -EINVAL);
 648
 649         lo->ldo_stripe_size = le32_to_cpu(lmm->lmm_stripe_size);
 650         lo->ldo_stripenr = le16_to_cpu(lmm->lmm_stripe_count);
 651         lo->ldo_layout_gen = le16_to_cpu(lmm->lmm_layout_gen);
 652
 653         LASSERT(buf->lb_len >= lov_mds_md_size(lo->ldo_stripenr, magic));
 654
 655         if (magic == LOV_MAGIC_V3) {
 656                 struct lov_mds_md_v3 *v3 = (struct lov_mds_md_v3 *) lmm;
 657                 objs = &v3->lmm_objects[0];
 658                 lod_object_set_pool(lo, v3->lmm_pool_name);
 659         } else {
 660                 objs = &lmm->lmm_objects[0];
 661         }
 662
 663         rc = lod_initialize_objects(env, lo, objs);
 664
 665 out:
 666         RETURN(rc);
 667 }
 668
 669 /*
 670  * Load and parse striping information, create in-core representation for the
 671  * stripes
 672  */
 673 int lod_load_striping(const struct lu_env *env, struct lod_object *lo)
 674 {
 675         struct lod_thread_info  *info = lod_env_info(env);
 676         struct dt_object        *next = dt_object_child(&lo->ldo_obj);
 677         int                      rc;
 678         ENTRY;
 679
 680         /*
 681          * currently this code is supposed to be called from declaration
 682          * phase only, thus the object is not expected to be locked by caller
 683          */
 684         dt_write_lock(env, next, 0);
 685         /* already initialized? */
 686         if (lo->ldo_stripe) {
 687                 int i;
 688                 /* check validity */
 689                 for (i = 0; i < lo->ldo_stripenr; i++)
 690                         LASSERTF(lo->ldo_stripe[i], "stripe %d is NULL\n", i);
 691                 GOTO(out, rc = 0);
 692         }
 693
 694         if (!dt_object_exists(next))
 695                 GOTO(out, rc = 0);
 696
 697         /* only regular files can be striped */
 698         if (!(lu_object_attr(lod2lu_obj(lo)) & S_IFREG))
 699                 GOTO(out, rc = 0);
 700
 701         LASSERT(lo->ldo_stripenr == 0);
 702
 703         rc = lod_get_lov_ea(env, lo);
 704         if (rc <= 0)
 705                 GOTO(out, rc);
 706
 707         /*
 708          * there is LOV EA (striping information) in this object
 709          * let's parse it and create in-core objects for the stripes
 710          */
 711         info->lti_buf.lb_buf = info->lti_ea_store;
 712         info->lti_buf.lb_len = info->lti_ea_store_size;
 713         rc = lod_parse_striping(env, lo, &info->lti_buf);
 714 out:
 715         dt_write_unlock(env, next);
 716         RETURN(rc);
 717 }
 718
 719 int lod_verify_striping(struct lod_device *d, const struct lu_buf *buf,
 720                         int specific)
 721 {
 722         struct lov_user_md_v1   *lum;
 723         struct lov_user_md_v3   *v3 = NULL;
 724         struct pool_desc        *pool = NULL;
 725         int                      rc;
 726         ENTRY;
 727
 728         lum = buf->lb_buf;
 729
 730         if (lum->lmm_magic != LOV_USER_MAGIC_V1 &&
 731             lum->lmm_magic != LOV_USER_MAGIC_V3 &&
 732             lum->lmm_magic != LOV_MAGIC_V1_DEF &&
 733             lum->lmm_magic != LOV_MAGIC_V3_DEF) {
 734                 CDEBUG(D_IOCTL, "bad userland LOV MAGIC: %#x\n",
 735                        lum->lmm_magic);
 736                 RETURN(-EINVAL);
 737         }
 738
 739         if ((specific && lum->lmm_pattern != LOV_PATTERN_RAID0) ||
 740             (specific == 0 && lum->lmm_pattern != 0)) {
 741                 CDEBUG(D_IOCTL, "bad userland stripe pattern: %#x\n",
 742                        lum->lmm_pattern);
 743                 RETURN(-EINVAL);
 744         }
 745
 746         /* 64kB is the largest common page size we see (ia64), and matches the
 747          * check in lfs */
 748         if (lum->lmm_stripe_size & (LOV_MIN_STRIPE_SIZE - 1)) {
 749                 CDEBUG(D_IOCTL, "stripe size %u not multiple of %u, fixing\n",
 750                        lum->lmm_stripe_size, LOV_MIN_STRIPE_SIZE);
 751                 RETURN(-EINVAL);
 752         }
 753
 754         /* an offset of -1 is treated as a "special" valid offset */
 755         if (lum->lmm_stripe_offset != (typeof(lum->lmm_stripe_offset))(-1)) {
 756                 /* if offset is not within valid range [0, osts_size) */
 757                 if (lum->lmm_stripe_offset >= d->lod_osts_size) {
 758                         CDEBUG(D_IOCTL, "stripe offset %u >= bitmap size %u\n",
 759                                lum->lmm_stripe_offset, d->lod_osts_size);
 760                         RETURN(-EINVAL);
 761                 }
 762
 763                 /* if lmm_stripe_offset is *not* in bitmap */
 764                 if (!cfs_bitmap_check(d->lod_ost_bitmap,
 765                                       lum->lmm_stripe_offset)) {
 766                         CDEBUG(D_IOCTL, "stripe offset %u not in bitmap\n",
 767                                lum->lmm_stripe_offset);
 768                         RETURN(-EINVAL);
 769                 }
 770         }
 771
 772         if (lum->lmm_magic == LOV_USER_MAGIC_V3)
 773                 v3 = buf->lb_buf;
 774
 775         if (v3)
 776                 pool = lod_find_pool(d, v3->lmm_pool_name);
 777
 778         if (pool != NULL) {
 779                 __u16 offs = v3->lmm_stripe_offset;
 780
 781                 if (offs != (typeof(v3->lmm_stripe_offset))(-1)) {
 782                         rc = lod_check_index_in_pool(offs, pool);
 783                         if (rc < 0) {
 784                                 lod_pool_putref(pool);
 785                                 RETURN(-EINVAL);
 786                         }
 787                 }
 788
 789                 if (specific && lum->lmm_stripe_count > pool_tgt_count(pool)) {
 790                         CDEBUG(D_IOCTL,
 791                                "stripe count %u > # OSTs %u in the pool\n",
 792                                lum->lmm_stripe_count, pool_tgt_count(pool));
 793                         lod_pool_putref(pool);
 794                         RETURN(-EINVAL);
 795                 }
 796
 797                 lod_pool_putref(pool);
 798         }
 799
 800         RETURN(0);
 801 }
 802
 803 void lod_fix_desc_stripe_size(__u64 *val)
 804 {
 805         if (*val < PTLRPC_MAX_BRW_SIZE) {
 806                 LCONSOLE_WARN("Increasing default stripe size to min %u\n",
 807                               PTLRPC_MAX_BRW_SIZE);
 808                 *val = PTLRPC_MAX_BRW_SIZE;
 809         } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) {
 810                 *val &= ~(LOV_MIN_STRIPE_SIZE - 1);
 811                 LCONSOLE_WARN("Changing default stripe size to "LPU64" (a "
 812                               "multiple of %u)\n",
 813                               *val, LOV_MIN_STRIPE_SIZE);
 814         }
 815 }
 816
 817 void lod_fix_desc_stripe_count(__u32 *val)
 818 {
 819         if (*val == 0)
 820                 *val = 1;
 821 }
 822
 823 void lod_fix_desc_pattern(__u32 *val)
 824 {
 825         /* from lov_setstripe */
 826         if ((*val != 0) && (*val != LOV_PATTERN_RAID0)) {
 827                 LCONSOLE_WARN("Unknown stripe pattern: %#x\n", *val);
 828                 *val = 0;
 829         }
 830 }
 831
 832 void lod_fix_desc_qos_maxage(__u32 *val)
 833 {
 834         /* fix qos_maxage */
 835         if (*val == 0)
 836                 *val = QOS_DEFAULT_MAXAGE;
 837 }
 838
 839 void lod_fix_desc(struct lov_desc *desc)
 840 {
 841         lod_fix_desc_stripe_size(&desc->ld_default_stripe_size);
 842         lod_fix_desc_stripe_count(&desc->ld_default_stripe_count);
 843         lod_fix_desc_pattern(&desc->ld_pattern);
 844         lod_fix_desc_qos_maxage(&desc->ld_qos_maxage);
 845 }
 846
 847 int lod_pools_init(struct lod_device *lod, struct lustre_cfg *lcfg)
 848 {
 849         struct lprocfs_static_vars  lvars = { 0 };
 850         struct obd_device          *obd;
 851         struct lov_desc            *desc;
 852         int                         rc;
 853         ENTRY;
 854
 855         obd = class_name2obd(lustre_cfg_string(lcfg, 0));
 856         LASSERT(obd != NULL);
 857         obd->obd_lu_dev = &lod->lod_dt_dev.dd_lu_dev;
 858
 859         if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
 860                 CERROR("LOD setup requires a descriptor\n");
 861                 RETURN(-EINVAL);
 862         }
 863
 864         desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
 865
 866         if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
 867                 CERROR("descriptor size wrong: %d > %d\n",
 868                        (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
 869                 RETURN(-EINVAL);
 870         }
 871
 872         if (desc->ld_magic != LOV_DESC_MAGIC) {
 873                 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
 874                         CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
 875                                obd->obd_name, desc);
 876                         lustre_swab_lov_desc(desc);
 877                 } else {
 878                         CERROR("%s: Bad lov desc magic: %#x\n",
 879                                obd->obd_name, desc->ld_magic);
 880                         RETURN(-EINVAL);
 881                 }
 882         }
 883
 884         lod_fix_desc(desc);
 885
 886         desc->ld_active_tgt_count = 0;
 887         lod->lod_desc = *desc;
 888
 889         lod->lod_sp_me = LUSTRE_SP_CLI;
 890
 891         /* Set up allocation policy (QoS and RR) */
 892         CFS_INIT_LIST_HEAD(&lod->lod_qos.lq_oss_list);
 893         cfs_init_rwsem(&lod->lod_qos.lq_rw_sem);
 894         lod->lod_qos.lq_dirty = 1;
 895         lod->lod_qos.lq_rr.lqr_dirty = 1;
 896         lod->lod_qos.lq_reset = 1;
 897         /* Default priority is toward free space balance */
 898         lod->lod_qos.lq_prio_free = 232;
 899         /* Default threshold for rr (roughly 17%) */
 900         lod->lod_qos.lq_threshold_rr = 43;
 901         /* Init statfs fields */
 902         OBD_ALLOC_PTR(lod->lod_qos.lq_statfs_data);
 903         if (NULL == lod->lod_qos.lq_statfs_data)
 904                 RETURN(-ENOMEM);
 905         cfs_waitq_init(&lod->lod_qos.lq_statfs_waitq);
 906
 907         /* Set up OST pool environment */
 908         lod->lod_pools_hash_body = cfs_hash_create("POOLS", HASH_POOLS_CUR_BITS,
 909                                                    HASH_POOLS_MAX_BITS,
 910                                                    HASH_POOLS_BKT_BITS, 0,
 911                                                    CFS_HASH_MIN_THETA,
 912                                                    CFS_HASH_MAX_THETA,
 913                                                    &pool_hash_operations,
 914                                                    CFS_HASH_DEFAULT);
 915         if (!lod->lod_pools_hash_body)
 916                 GOTO(out_statfs, rc = -ENOMEM);
 917         CFS_INIT_LIST_HEAD(&lod->lod_pool_list);
 918         lod->lod_pool_count = 0;
 919         rc = lod_ost_pool_init(&lod->lod_pool_info, 0);
 920         if (rc)
 921                 GOTO(out_hash, rc);
 922         rc = lod_ost_pool_init(&lod->lod_qos.lq_rr.lqr_pool, 0);
 923         if (rc)
 924                 GOTO(out_pool_info, rc);
 925
 926         /* the OST array and bitmap are allocated/grown dynamically as OSTs are
 927          * added to the LOD, see lod_add_device() */
 928         lod->lod_ost_bitmap = NULL;
 929         lod->lod_osts_size  = 0;
 930         lod->lod_ostnr      = 0;
 931
 932         lod->lod_death_row = 0;
 933         lod->lod_refcount  = 0;
 934
 935         lprocfs_lod_init_vars(&lvars);
 936         lprocfs_obd_setup(obd, lvars.obd_vars);
 937
 938 #ifdef LPROCFS
 939         rc = lprocfs_seq_create(obd->obd_proc_entry, "target_obd",
 940                                 0444, &lod_proc_target_fops, obd);
 941         if (rc) {
 942                 CWARN("%s: Error adding the target_obd file %d\n",
 943                       obd->obd_name, rc);
 944                 GOTO(out_lproc, rc);
 945         }
 946         lod->lod_pool_proc_entry = lprocfs_register("pools",
 947                                                     obd->obd_proc_entry,
 948                                                     NULL, NULL);
 949         if (IS_ERR(lod->lod_pool_proc_entry)) {
 950                 int ret = PTR_ERR(lod->lod_pool_proc_entry);
 951                 lod->lod_pool_proc_entry = NULL;
 952                 CWARN("%s: Failed to create pool proc file %d\n",
 953                       obd->obd_name, ret);
 954                 rc = lod_pools_fini(lod);
 955                 RETURN(ret);
 956         }
 957 #endif
 958
 959         RETURN(0);
 960
 961 out_lproc:
 962         lprocfs_obd_cleanup(obd);
 963         lod_ost_pool_free(&lod->lod_qos.lq_rr.lqr_pool);
 964 out_pool_info:
 965         lod_ost_pool_free(&lod->lod_pool_info);
 966 out_hash:
 967         cfs_hash_putref(lod->lod_pools_hash_body);
 968 out_statfs:
 969         OBD_FREE_PTR(lod->lod_qos.lq_statfs_data);
 970         return rc;
 971 }
 972
 973 int lod_pools_fini(struct lod_device *lod)
 974 {
 975         struct obd_device   *obd = lod2obd(lod);
 976         cfs_list_t          *pos, *tmp;
 977         struct pool_desc    *pool;
 978         ENTRY;
 979
 980         cfs_list_for_each_safe(pos, tmp, &lod->lod_pool_list) {
 981                 pool = cfs_list_entry(pos, struct pool_desc, pool_list);
 982                 /* free pool structs */
 983                 CDEBUG(D_INFO, "delete pool %p\n", pool);
 984                 lod_pool_del(obd, pool->pool_name);
 985         }
 986
 987         if (lod->lod_osts_size > 0) {
 988                 int idx;
 989                 lod_getref(lod);
 990                 cfs_mutex_lock(&lod->lod_mutex);
 991                 cfs_foreach_bit(lod->lod_ost_bitmap, idx)
 992                         __lod_del_device(lod, idx);
 993                 cfs_mutex_unlock(&lod->lod_mutex);
 994                 lod_putref(lod);
 995                 CFS_FREE_BITMAP(lod->lod_ost_bitmap);
 996                 for (idx = 0; idx < OST_PTRS; idx++) {
 997                         if (lod->lod_ost_idx[idx])
 998                                 OBD_FREE_PTR(lod->lod_ost_idx[idx]);
 999                 }
1000                 lod->lod_osts_size = 0;
1001         }
1002
1003         cfs_hash_putref(lod->lod_pools_hash_body);
1004         lod_ost_pool_free(&(lod->lod_qos.lq_rr.lqr_pool));
1005         lod_ost_pool_free(&lod->lod_pool_info);
1006
1007         /* clear pools parent proc entry only after all pools are killed */
1008         if (lod->lod_pool_proc_entry) {
1009                 lprocfs_remove(&lod->lod_pool_proc_entry);
1010                 lod->lod_pool_proc_entry = NULL;
1011         }
1012
1013         lprocfs_obd_cleanup(obd);
1014
1015         OBD_FREE_PTR(lod->lod_qos.lq_statfs_data);
1016         RETURN(0);
1017 }
1018