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  * Copyright (c) 2012, Intel Corporation.
  27  */
  28 /*
  29  * lustre/lod/lod_lov.c
  30  *
  31  * Author: Alex Zhuravlev <alexey.zhuravlev@intel.com>
  32  */
  33
  34 #ifndef EXPORT_SYMTAB
  35 # define EXPORT_SYMTAB
  36 #endif
  37 #define DEBUG_SUBSYSTEM S_MDS
  38
  39 #include <obd_class.h>
  40 #include <obd_lov.h>
  41
  42 #include "lod_internal.h"
  43
  44 /*
  45  * Keep a refcount of lod->lod_osts usage to prevent racing with
  46  * addition/deletion. Any function that expects lov_tgts to remain stationary
  47  * must take a ref.
  48  *
  49  * \param lod - is the lod device from which we want to grab a reference
  50  */
  51 void lod_getref(struct lod_device *lod)
  52 {
  53         down_read(&lod->lod_rw_sem);
  54         mutex_lock(&lod->lod_mutex);
  55         lod->lod_refcount++;
  56         mutex_unlock(&lod->lod_mutex);
  57 }
  58
  59 /*
  60  * Companion of lod_getref() to release a reference on the lod table.
  61  * If this is the last reference and the ost entry was scheduled for deletion,
  62  * the descriptor is removed from the array.
  63  *
  64  * \param lod - is the lod device from which we release a reference
  65  */
  66 void lod_putref(struct lod_device *lod)
  67 {
  68         mutex_lock(&lod->lod_mutex);
  69         lod->lod_refcount--;
  70         if (lod->lod_refcount == 0 && lod->lod_death_row) {
  71                 struct lod_ost_desc *ost_desc, *tmp;
  72                 int                  idx;
  73                 CFS_LIST_HEAD(kill);
  74
  75                 CDEBUG(D_CONFIG, "destroying %d lod desc\n",
  76                        lod->lod_death_row);
  77
  78                 cfs_foreach_bit(lod->lod_ost_bitmap, idx) {
  79                         ost_desc = OST_TGT(lod, idx);
  80                         LASSERT(ost_desc);
  81
  82                         if (!ost_desc->ltd_reap)
  83                                 continue;
  84
  85                         cfs_list_add(&ost_desc->ltd_kill, &kill);
  86
  87                         lod_ost_pool_remove(&lod->lod_pool_info, idx);
  88                         OST_TGT(lod, idx) = NULL;
  89                         lod->lod_ostnr--;
  90                         cfs_bitmap_clear(lod->lod_ost_bitmap, idx);
  91                         if (ost_desc->ltd_active)
  92                                 lod->lod_desc.ld_active_tgt_count--;
  93                         lod->lod_death_row--;
  94                 }
  95                 mutex_unlock(&lod->lod_mutex);
  96                 up_read(&lod->lod_rw_sem);
  97
  98                 cfs_list_for_each_entry_safe(ost_desc, tmp, &kill, ltd_kill) {
  99                         int rc;
 100                         cfs_list_del(&ost_desc->ltd_kill);
 101                         /* remove from QoS structures */
 102                         rc = qos_del_tgt(lod, ost_desc);
 103                         if (rc)
 104                                 CERROR("%s: qos_del_tgt(%s) failed: rc = %d\n",
 105                                        lod2obd(lod)->obd_name,
 106                                        obd_uuid2str(&ost_desc->ltd_uuid), rc);
 107                         rc = obd_disconnect(ost_desc->ltd_exp);
 108                         if (rc)
 109                                 CERROR("%s: failed to disconnect %s: rc = %d\n",
 110                                        lod2obd(lod)->obd_name,
 111                                        obd_uuid2str(&ost_desc->ltd_uuid), rc);
 112                         OBD_FREE_PTR(ost_desc);
 113                 }
 114         } else {
 115                 mutex_unlock(&lod->lod_mutex);
 116                 up_read(&lod->lod_rw_sem);
 117         }
 118 }
 119
 120 static int lod_bitmap_resize(struct lod_device *lod, __u32 newsize)
 121 {
 122         cfs_bitmap_t *new_bitmap, *old_bitmap = NULL;
 123         int           rc = 0;
 124         ENTRY;
 125
 126         /* grab write reference on the lod. Relocating the array requires
 127          * exclusive access */
 128         down_write(&lod->lod_rw_sem);
 129
 130         if (newsize <= lod->lod_osts_size)
 131                 /* someone else has already resize the array */
 132                 GOTO(out, rc = 0);
 133
 134         /* allocate new bitmap */
 135         new_bitmap = CFS_ALLOCATE_BITMAP(newsize);
 136         if (!new_bitmap)
 137                 GOTO(out, rc = -ENOMEM);
 138
 139         if (lod->lod_osts_size > 0) {
 140                 /* the bitmap already exists, we need
 141                  * to copy data from old one */
 142                 cfs_bitmap_copy(new_bitmap, lod->lod_ost_bitmap);
 143                 old_bitmap = lod->lod_ost_bitmap;
 144         }
 145
 146         lod->lod_osts_size  = newsize;
 147         lod->lod_ost_bitmap = new_bitmap;
 148
 149         if (old_bitmap)
 150                 CFS_FREE_BITMAP(old_bitmap);
 151
 152         CDEBUG(D_CONFIG, "ost size: %d\n", lod->lod_osts_size);
 153
 154         EXIT;
 155 out:
 156         up_write(&lod->lod_rw_sem);
 157         return rc;
 158 }
 159
 160 /*
 161  * Connect LOD to a new OSP and add it to the device table.
 162  *
 163  * \param env - is the environment passed by the caller
 164  * \param lod - is the LOD device to be connected to the new OSP
 165  * \param osp - is the name of OSP device name about to be added
 166  * \param index - is the OSP index
 167  * \param gen - is the generation number
 168  */
 169 int lod_add_device(const struct lu_env *env, struct lod_device *lod,
 170                    char *osp, unsigned index, unsigned gen, int active)
 171 {
 172         struct obd_connect_data *data = NULL;
 173         struct obd_export       *exp = NULL;
 174         struct obd_device       *obd;
 175         struct lu_device        *ldev;
 176         struct dt_device        *d;
 177         int                      rc;
 178         struct lod_ost_desc     *ost_desc;
 179         struct obd_uuid          obd_uuid;
 180
 181         ENTRY;
 182
 183         CDEBUG(D_CONFIG, "osp:%s idx:%d gen:%d\n", osp, index, gen);
 184
 185         if (gen <= 0) {
 186                 CERROR("request to add OBD %s with invalid generation: %d\n",
 187                        osp, gen);
 188                 RETURN(-EINVAL);
 189         }
 190
 191         obd_str2uuid(&obd_uuid, osp);
 192
 193         obd = class_find_client_obd(&obd_uuid, LUSTRE_OSP_NAME,
 194                                 &lod->lod_dt_dev.dd_lu_dev.ld_obd->obd_uuid);
 195         if (obd == NULL) {
 196                 CERROR("can't find %s device\n", osp);
 197                 RETURN(-EINVAL);
 198         }
 199
 200         OBD_ALLOC_PTR(data);
 201         if (data == NULL)
 202                 RETURN(-ENOMEM);
 203
 204         data->ocd_connect_flags = OBD_CONNECT_INDEX | OBD_CONNECT_VERSION;
 205         data->ocd_version = LUSTRE_VERSION_CODE;
 206         data->ocd_index = index;
 207
 208         rc = obd_connect(env, &exp, obd, &obd->obd_uuid, data, NULL);
 209         OBD_FREE_PTR(data);
 210         if (rc) {
 211                 CERROR("%s: cannot connect to next dev %s (%d)\n",
 212                        obd->obd_name, osp, rc);
 213                 GOTO(out_free, rc);
 214         }
 215
 216         LASSERT(obd->obd_lu_dev);
 217         LASSERT(obd->obd_lu_dev->ld_site = lod->lod_dt_dev.dd_lu_dev.ld_site);
 218
 219         ldev = obd->obd_lu_dev;
 220         d = lu2dt_dev(ldev);
 221
 222         /* Allocate ost descriptor and fill it */
 223         OBD_ALLOC_PTR(ost_desc);
 224         if (!ost_desc)
 225                 GOTO(out_conn, rc = -ENOMEM);
 226
 227         ost_desc->ltd_ost    = d;
 228         ost_desc->ltd_exp    = exp;
 229         ost_desc->ltd_uuid   = obd->u.cli.cl_target_uuid;
 230         ost_desc->ltd_gen    = gen;
 231         ost_desc->ltd_index  = index;
 232         ost_desc->ltd_active = active;
 233
 234         lod_getref(lod);
 235         if (index >= lod->lod_osts_size) {
 236                 /* we have to increase the size of the lod_osts array */
 237                 __u32  newsize;
 238
 239                 newsize = max(lod->lod_osts_size, (__u32)2);
 240                 while (newsize < index + 1)
 241                         newsize = newsize << 1;
 242
 243                 /* lod_bitmap_resize() needs lod_rw_sem
 244                  * which we hold with th reference */
 245                 lod_putref(lod);
 246
 247                 rc = lod_bitmap_resize(lod, newsize);
 248                 if (rc)
 249                         GOTO(out_desc, rc);
 250
 251                 lod_getref(lod);
 252         }
 253
 254         mutex_lock(&lod->lod_mutex);
 255         if (cfs_bitmap_check(lod->lod_ost_bitmap, index)) {
 256                 CERROR("%s: device %d is registered already\n", obd->obd_name,
 257                        index);
 258                 GOTO(out_mutex, rc = -EEXIST);
 259         }
 260
 261         if (lod->lod_ost_idx[index / OST_PTRS_PER_BLOCK] == NULL) {
 262                 OBD_ALLOC_PTR(lod->lod_ost_idx[index / OST_PTRS_PER_BLOCK]);
 263                 if (lod->lod_ost_idx[index / OST_PTRS_PER_BLOCK] == NULL) {
 264                         CERROR("can't allocate index to add %s\n",
 265                                obd->obd_name);
 266                         GOTO(out_mutex, rc = -ENOMEM);
 267                 }
 268         }
 269
 270         rc = lod_ost_pool_add(&lod->lod_pool_info, index, lod->lod_osts_size);
 271         if (rc) {
 272                 CERROR("%s: can't set up pool, failed with %d\n",
 273                        obd->obd_name, rc);
 274                 GOTO(out_mutex, rc);
 275         }
 276
 277         rc = qos_add_tgt(lod, ost_desc);
 278         if (rc) {
 279                 CERROR("%s: qos_add_tgt(%s) failed: rc = %d\n", obd->obd_name,
 280                        obd_uuid2str(&ost_desc->ltd_uuid), rc);
 281                 GOTO(out_pool, rc);
 282         }
 283
 284         /* The new OST is now a full citizen */
 285         if (index >= lod->lod_desc.ld_tgt_count)
 286                 lod->lod_desc.ld_tgt_count = index + 1;
 287         if (active)
 288                 lod->lod_desc.ld_active_tgt_count++;
 289         OST_TGT(lod, index) = ost_desc;
 290         cfs_bitmap_set(lod->lod_ost_bitmap, index);
 291         lod->lod_ostnr++;
 292         mutex_unlock(&lod->lod_mutex);
 293         lod_putref(lod);
 294
 295         if (lod->lod_recovery_completed)
 296                 ldev->ld_ops->ldo_recovery_complete(env, ldev);
 297
 298         RETURN(0);
 299
 300 out_pool:
 301         lod_ost_pool_remove(&lod->lod_pool_info, index);
 302 out_mutex:
 303         mutex_unlock(&lod->lod_mutex);
 304         lod_putref(lod);
 305 out_desc:
 306         OBD_FREE_PTR(ost_desc);
 307 out_conn:
 308         obd_disconnect(exp);
 309 out_free:
 310         return rc;
 311 }
 312
 313 /*
 314  * helper function to schedule OST removal from the device table
 315  */
 316 static void __lod_del_device(struct lod_device *lod, unsigned idx)
 317 {
 318         LASSERT(OST_TGT(lod,idx));
 319         if (OST_TGT(lod,idx)->ltd_reap == 0) {
 320                 OST_TGT(lod,idx)->ltd_reap = 1;
 321                 lod->lod_death_row++;
 322         }
 323 }
 324
 325 /*
 326  * Add support for administratively disabled OST (through the MGS).
 327  * Schedule a target for deletion.  Disconnection and real removal from the
 328  * table takes place in lod_putref() once the last table user release its
 329  * reference.
 330  *
 331  * \param env - is the environment passed by the caller
 332  * \param lod - is the lod device currently connected to the OSP about to be
 333  *              removed
 334  * \param osp - is the name of OSP device about to be removed
 335  * \param idx - is the OSP index
 336  * \param gen - is the generation number, not used currently
 337  */
 338 int lod_del_device(const struct lu_env *env, struct lod_device *lod,
 339                    char *osp, unsigned idx, unsigned gen)
 340 {
 341         struct obd_device *obd;
 342         int                rc = 0;
 343         struct obd_uuid    uuid;
 344         ENTRY;
 345
 346         CDEBUG(D_CONFIG, "osp:%s idx:%d gen:%d\n", osp, idx, gen);
 347
 348         obd = class_name2obd(osp);
 349         if (obd == NULL) {
 350                 CERROR("can't find %s device\n", osp);
 351                 RETURN(-EINVAL);
 352         }
 353
 354         if (gen <= 0) {
 355                 CERROR("%s: request to remove OBD %s with invalid generation %d"
 356                        "\n", obd->obd_name, osp, gen);
 357                 RETURN(-EINVAL);
 358         }
 359
 360         obd_str2uuid(&uuid,  osp);
 361
 362         lod_getref(lod);
 363         mutex_lock(&lod->lod_mutex);
 364         /* check that the index is allocated in the bitmap */
 365         if (!cfs_bitmap_check(lod->lod_ost_bitmap, idx) || !OST_TGT(lod,idx)) {
 366                 CERROR("%s: device %d is not set up\n", obd->obd_name, idx);
 367                 GOTO(out, rc = -EINVAL);
 368         }
 369
 370         /* check that the UUID matches */
 371         if (!obd_uuid_equals(&uuid, &OST_TGT(lod,idx)->ltd_uuid)) {
 372                 CERROR("%s: LOD target UUID %s at index %d does not match %s\n",
 373                        obd->obd_name, obd_uuid2str(&OST_TGT(lod,idx)->ltd_uuid),
 374                        idx, osp);
 375                 GOTO(out, rc = -EINVAL);
 376         }
 377
 378         __lod_del_device(lod, idx);
 379         EXIT;
 380 out:
 381         mutex_unlock(&lod->lod_mutex);
 382         lod_putref(lod);
 383         return(rc);
 384 }
 385
 386 int lod_ea_store_resize(struct lod_thread_info *info, int size)
 387 {
 388         int round = size_roundup_power2(size);
 389
 390         LASSERT(round <= lov_mds_md_size(LOV_MAX_STRIPE_COUNT, LOV_MAGIC_V3));
 391         if (info->lti_ea_store) {
 392                 LASSERT(info->lti_ea_store_size);
 393                 LASSERT(info->lti_ea_store_size < round);
 394                 CDEBUG(D_INFO, "EA store size %d is not enough, need %d\n",
 395                        info->lti_ea_store_size, round);
 396                 OBD_FREE_LARGE(info->lti_ea_store, info->lti_ea_store_size);
 397                 info->lti_ea_store = NULL;
 398                 info->lti_ea_store_size = 0;
 399         }
 400
 401         OBD_ALLOC_LARGE(info->lti_ea_store, round);
 402         if (info->lti_ea_store == NULL)
 403                 RETURN(-ENOMEM);
 404         info->lti_ea_store_size = round;
 405         RETURN(0);
 406 }
 407
 408 /*
 409  * generate and write LOV EA for given striped object
 410  */
 411 int lod_generate_and_set_lovea(const struct lu_env *env,
 412                                struct lod_object *lo, struct thandle *th)
 413 {
 414         struct lod_thread_info  *info = lod_env_info(env);
 415         struct dt_object        *next = dt_object_child(&lo->ldo_obj);
 416         const struct lu_fid     *fid  = lu_object_fid(&lo->ldo_obj.do_lu);
 417         struct lov_mds_md_v1    *lmm;
 418         struct lov_ost_data_v1  *objs;
 419         __u32                    magic;
 420         int                      i, rc, lmm_size;
 421         ENTRY;
 422
 423         LASSERT(lo);
 424         LASSERT(lo->ldo_stripenr > 0);
 425
 426         magic = lo->ldo_pool ? LOV_MAGIC_V3 : LOV_MAGIC_V1;
 427         lmm_size = lov_mds_md_size(lo->ldo_stripenr, magic);
 428         if (info->lti_ea_store_size < lmm_size) {
 429                 rc = lod_ea_store_resize(info, lmm_size);
 430                 if (rc)
 431                         RETURN(rc);
 432         }
 433
 434         lmm = info->lti_ea_store;
 435
 436         lmm->lmm_magic = cpu_to_le32(magic);
 437         lmm->lmm_pattern = cpu_to_le32(LOV_PATTERN_RAID0);
 438         lmm->lmm_object_id = cpu_to_le64(fid_ver_oid(fid));
 439         lmm->lmm_object_seq = cpu_to_le64(fid_seq(fid));
 440         lmm->lmm_stripe_size = cpu_to_le32(lo->ldo_stripe_size);
 441         lmm->lmm_stripe_count = cpu_to_le16(lo->ldo_stripenr);
 442         lmm->lmm_layout_gen = 0;
 443         if (magic == LOV_MAGIC_V1) {
 444                 objs = &lmm->lmm_objects[0];
 445         } else {
 446                 struct lov_mds_md_v3 *v3 = (struct lov_mds_md_v3 *) lmm;
 447                 strncpy(v3->lmm_pool_name, lo->ldo_pool, LOV_MAXPOOLNAME);
 448                 objs = &v3->lmm_objects[0];
 449         }
 450
 451         for (i = 0; i < lo->ldo_stripenr; i++) {
 452                 const struct lu_fid *fid;
 453
 454                 LASSERT(lo->ldo_stripe[i]);
 455                 fid = lu_object_fid(&lo->ldo_stripe[i]->do_lu);
 456
 457                 rc = fid_ostid_pack(fid, &info->lti_ostid);
 458                 LASSERT(rc == 0);
 459                 LASSERT(info->lti_ostid.oi_seq == FID_SEQ_OST_MDT0);
 460
 461                 objs[i].l_object_id  = cpu_to_le64(info->lti_ostid.oi_id);
 462                 objs[i].l_object_seq = cpu_to_le64(info->lti_ostid.oi_seq);
 463                 objs[i].l_ost_gen    = cpu_to_le32(0);
 464                 objs[i].l_ost_idx    = cpu_to_le32(fid_idif_ost_idx(fid));
 465         }
 466
 467         info->lti_buf.lb_buf = lmm;
 468         info->lti_buf.lb_len = lmm_size;
 469         rc = dt_xattr_set(env, next, &info->lti_buf, XATTR_NAME_LOV, 0,
 470                           th, BYPASS_CAPA);
 471
 472         RETURN(rc);
 473 }
 474
 475 int lod_get_lov_ea(const struct lu_env *env, struct lod_object *lo)
 476 {
 477         struct lod_thread_info *info = lod_env_info(env);
 478         struct dt_object       *next = dt_object_child(&lo->ldo_obj);
 479         int                     rc;
 480         ENTRY;
 481
 482         LASSERT(info);
 483
 484         if (unlikely(info->lti_ea_store_size == 0)) {
 485                 /* just to enter in allocation block below */
 486                 rc = -ERANGE;
 487         } else {
 488 repeat:
 489                 info->lti_buf.lb_buf = info->lti_ea_store;
 490                 info->lti_buf.lb_len = info->lti_ea_store_size;
 491                 rc = dt_xattr_get(env, next, &info->lti_buf, XATTR_NAME_LOV,
 492                                   BYPASS_CAPA);
 493         }
 494         /* if object is not striped or inaccessible */
 495         if (rc == -ENODATA)
 496                 RETURN(0);
 497
 498         if (rc == -ERANGE) {
 499                 /* EA doesn't fit, reallocate new buffer */
 500                 rc = dt_xattr_get(env, next, &LU_BUF_NULL, XATTR_NAME_LOV,
 501                                   BYPASS_CAPA);
 502                 if (rc == -ENODATA)
 503                         RETURN(0);
 504                 else if (rc < 0)
 505                         RETURN(rc);
 506
 507                 LASSERT(rc > 0);
 508                 rc = lod_ea_store_resize(info, rc);
 509                 if (rc)
 510                         RETURN(rc);
 511                 goto repeat;
 512         }
 513
 514         RETURN(rc);
 515 }
 516
 517 int lod_store_def_striping(const struct lu_env *env, struct dt_object *dt,
 518                            struct thandle *th)
 519 {
 520         struct lod_thread_info  *info = lod_env_info(env);
 521         struct lod_object       *lo = lod_dt_obj(dt);
 522         struct dt_object        *next = dt_object_child(dt);
 523         struct lov_user_md_v3   *v3;
 524         int                      rc;
 525         ENTRY;
 526
 527         LASSERT(S_ISDIR(dt->do_lu.lo_header->loh_attr));
 528
 529         /*
 530          * store striping defaults into new directory
 531          * used to implement defaults inheritance
 532          */
 533
 534         /* probably nothing to inherite */
 535         if (lo->ldo_striping_cached == 0)
 536                 RETURN(0);
 537
 538         if (LOVEA_DELETE_VALUES(lo->ldo_def_stripe_size, lo->ldo_def_stripenr,
 539                                 lo->ldo_def_stripe_offset))
 540                 RETURN(0);
 541
 542         /* XXX: use thread info */
 543         OBD_ALLOC_PTR(v3);
 544         if (v3 == NULL)
 545                 RETURN(-ENOMEM);
 546
 547         v3->lmm_magic = cpu_to_le32(LOV_MAGIC_V3);
 548         v3->lmm_pattern = cpu_to_le32(LOV_PATTERN_RAID0);
 549         v3->lmm_object_id = 0;
 550         v3->lmm_object_seq = 0;
 551         v3->lmm_stripe_size = cpu_to_le32(lo->ldo_def_stripe_size);
 552         v3->lmm_stripe_count = cpu_to_le16(lo->ldo_def_stripenr);
 553         v3->lmm_stripe_offset = cpu_to_le16(lo->ldo_def_stripe_offset);
 554         if (lo->ldo_pool)
 555                 strncpy(v3->lmm_pool_name, lo->ldo_pool, LOV_MAXPOOLNAME);
 556
 557         info->lti_buf.lb_buf = v3;
 558         info->lti_buf.lb_len = sizeof(*v3);
 559         rc = dt_xattr_set(env, next, &info->lti_buf, XATTR_NAME_LOV, 0, th,
 560                         BYPASS_CAPA);
 561
 562         OBD_FREE_PTR(v3);
 563
 564         RETURN(rc);
 565 }
 566
 567 /*
 568  * allocate array of objects pointers, find/create objects
 569  * stripenr and other fields should be initialized by this moment
 570  */
 571 int lod_initialize_objects(const struct lu_env *env, struct lod_object *lo,
 572                            struct lov_ost_data_v1 *objs)
 573 {
 574         struct lod_thread_info  *info = lod_env_info(env);
 575         struct lod_device       *md = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
 576         struct lu_object        *o, *n;
 577         struct lu_device        *nd;
 578         int                     i, idx, rc = 0;
 579         ENTRY;
 580
 581         LASSERT(lo);
 582         LASSERT(lo->ldo_stripe == NULL);
 583         LASSERT(lo->ldo_stripenr > 0);
 584         LASSERT(lo->ldo_stripe_size > 0);
 585
 586         i = sizeof(struct dt_object *) * lo->ldo_stripenr;
 587         OBD_ALLOC(lo->ldo_stripe, i);
 588         if (lo->ldo_stripe == NULL)
 589                 GOTO(out, rc = -ENOMEM);
 590         lo->ldo_stripes_allocated = lo->ldo_stripenr;
 591
 592         for (i = 0; i < lo->ldo_stripenr; i++) {
 593
 594                 info->lti_ostid.oi_id = le64_to_cpu(objs[i].l_object_id);
 595                 /* XXX: support for DNE? */
 596                 info->lti_ostid.oi_seq = le64_to_cpu(objs[i].l_object_seq);
 597                 idx = le64_to_cpu(objs[i].l_ost_idx);
 598                 fid_ostid_unpack(&info->lti_fid, &info->lti_ostid, idx);
 599
 600                 /*
 601                  * XXX: assertion is left for testing, to make
 602                  * sure we never process requests till configuration
 603                  * is completed. to be changed to -EINVAL
 604                  */
 605
 606                 lod_getref(md);
 607                 LASSERT(cfs_bitmap_check(md->lod_ost_bitmap, idx));
 608                 LASSERT(OST_TGT(md,idx));
 609                 LASSERTF(OST_TGT(md,idx)->ltd_ost, "idx %d\n", idx);
 610                 nd = &OST_TGT(md,idx)->ltd_ost->dd_lu_dev;
 611                 lod_putref(md);
 612
 613                 o = lu_object_find_at(env, nd, &info->lti_fid, NULL);
 614                 if (IS_ERR(o))
 615                         GOTO(out, rc = PTR_ERR(o));
 616
 617                 n = lu_object_locate(o->lo_header, nd->ld_type);
 618                 LASSERT(n);
 619
 620                 lo->ldo_stripe[i] = container_of(n, struct dt_object, do_lu);
 621         }
 622
 623 out:
 624         RETURN(rc);
 625 }
 626
 627 /*
 628  * Parse striping information stored in lti_ea_store
 629  */
 630 int lod_parse_striping(const struct lu_env *env, struct lod_object *lo,
 631                        const struct lu_buf *buf)
 632 {
 633         struct lov_mds_md_v1    *lmm;
 634         struct lov_ost_data_v1  *objs;
 635         __u32                    magic;
 636         int                      rc = 0;
 637         ENTRY;
 638
 639         LASSERT(buf);
 640         LASSERT(buf->lb_buf);
 641         LASSERT(buf->lb_len);
 642
 643         lmm = (struct lov_mds_md_v1 *) buf->lb_buf;
 644         magic = le32_to_cpu(lmm->lmm_magic);
 645
 646         if (magic != LOV_MAGIC_V1 && magic != LOV_MAGIC_V3)
 647                 GOTO(out, rc = -EINVAL);
 648         if (le32_to_cpu(lmm->lmm_pattern) != LOV_PATTERN_RAID0)
 649                 GOTO(out, rc = -EINVAL);
 650
 651         lo->ldo_stripe_size = le32_to_cpu(lmm->lmm_stripe_size);
 652         lo->ldo_stripenr = le16_to_cpu(lmm->lmm_stripe_count);
 653         lo->ldo_layout_gen = le16_to_cpu(lmm->lmm_layout_gen);
 654
 655         LASSERT(buf->lb_len >= lov_mds_md_size(lo->ldo_stripenr, magic));
 656
 657         if (magic == LOV_MAGIC_V3) {
 658                 struct lov_mds_md_v3 *v3 = (struct lov_mds_md_v3 *) lmm;
 659                 objs = &v3->lmm_objects[0];
 660                 lod_object_set_pool(lo, v3->lmm_pool_name);
 661         } else {
 662                 objs = &lmm->lmm_objects[0];
 663         }
 664
 665         rc = lod_initialize_objects(env, lo, objs);
 666
 667 out:
 668         RETURN(rc);
 669 }
 670
 671 /*
 672  * Load and parse striping information, create in-core representation for the
 673  * stripes
 674  */
 675 int lod_load_striping(const struct lu_env *env, struct lod_object *lo)
 676 {
 677         struct lod_thread_info  *info = lod_env_info(env);
 678         struct dt_object        *next = dt_object_child(&lo->ldo_obj);
 679         int                      rc;
 680         ENTRY;
 681
 682         /*
 683          * currently this code is supposed to be called from declaration
 684          * phase only, thus the object is not expected to be locked by caller
 685          */
 686         dt_write_lock(env, next, 0);
 687         /* already initialized? */
 688         if (lo->ldo_stripe) {
 689                 int i;
 690                 /* check validity */
 691                 for (i = 0; i < lo->ldo_stripenr; i++)
 692                         LASSERTF(lo->ldo_stripe[i], "stripe %d is NULL\n", i);
 693                 GOTO(out, rc = 0);
 694         }
 695
 696         if (!dt_object_exists(next))
 697                 GOTO(out, rc = 0);
 698
 699         /* only regular files can be striped */
 700         if (!(lu_object_attr(lod2lu_obj(lo)) & S_IFREG))
 701                 GOTO(out, rc = 0);
 702
 703         LASSERT(lo->ldo_stripenr == 0);
 704
 705         rc = lod_get_lov_ea(env, lo);
 706         if (rc <= 0)
 707                 GOTO(out, rc);
 708
 709         /*
 710          * there is LOV EA (striping information) in this object
 711          * let's parse it and create in-core objects for the stripes
 712          */
 713         info->lti_buf.lb_buf = info->lti_ea_store;
 714         info->lti_buf.lb_len = info->lti_ea_store_size;
 715         rc = lod_parse_striping(env, lo, &info->lti_buf);
 716 out:
 717         dt_write_unlock(env, next);
 718         RETURN(rc);
 719 }
 720
 721 int lod_verify_striping(struct lod_device *d, const struct lu_buf *buf,
 722                         int specific)
 723 {
 724         struct lov_user_md_v1   *lum;
 725         struct lov_user_md_v3   *v3 = NULL;
 726         struct pool_desc        *pool = NULL;
 727         int                      rc;
 728         ENTRY;
 729
 730         lum = buf->lb_buf;
 731
 732         if (lum->lmm_magic != LOV_USER_MAGIC_V1 &&
 733             lum->lmm_magic != LOV_USER_MAGIC_V3 &&
 734             lum->lmm_magic != LOV_MAGIC_V1_DEF &&
 735             lum->lmm_magic != LOV_MAGIC_V3_DEF) {
 736                 CDEBUG(D_IOCTL, "bad userland LOV MAGIC: %#x\n",
 737                        lum->lmm_magic);
 738                 RETURN(-EINVAL);
 739         }
 740
 741         if ((specific && lum->lmm_pattern != LOV_PATTERN_RAID0) ||
 742             (specific == 0 && lum->lmm_pattern != 0)) {
 743                 CDEBUG(D_IOCTL, "bad userland stripe pattern: %#x\n",
 744                        lum->lmm_pattern);
 745                 RETURN(-EINVAL);
 746         }
 747
 748         /* 64kB is the largest common page size we see (ia64), and matches the
 749          * check in lfs */
 750         if (lum->lmm_stripe_size & (LOV_MIN_STRIPE_SIZE - 1)) {
 751                 CDEBUG(D_IOCTL, "stripe size %u not multiple of %u, fixing\n",
 752                        lum->lmm_stripe_size, LOV_MIN_STRIPE_SIZE);
 753                 RETURN(-EINVAL);
 754         }
 755
 756         /* an offset of -1 is treated as a "special" valid offset */
 757         if (lum->lmm_stripe_offset != (typeof(lum->lmm_stripe_offset))(-1)) {
 758                 /* if offset is not within valid range [0, osts_size) */
 759                 if (lum->lmm_stripe_offset >= d->lod_osts_size) {
 760                         CDEBUG(D_IOCTL, "stripe offset %u >= bitmap size %u\n",
 761                                lum->lmm_stripe_offset, d->lod_osts_size);
 762                         RETURN(-EINVAL);
 763                 }
 764
 765                 /* if lmm_stripe_offset is *not* in bitmap */
 766                 if (!cfs_bitmap_check(d->lod_ost_bitmap,
 767                                       lum->lmm_stripe_offset)) {
 768                         CDEBUG(D_IOCTL, "stripe offset %u not in bitmap\n",
 769                                lum->lmm_stripe_offset);
 770                         RETURN(-EINVAL);
 771                 }
 772         }
 773
 774         if (lum->lmm_magic == LOV_USER_MAGIC_V3)
 775                 v3 = buf->lb_buf;
 776
 777         if (v3)
 778                 pool = lod_find_pool(d, v3->lmm_pool_name);
 779
 780         if (pool != NULL) {
 781                 __u16 offs = v3->lmm_stripe_offset;
 782
 783                 if (offs != (typeof(v3->lmm_stripe_offset))(-1)) {
 784                         rc = lod_check_index_in_pool(offs, pool);
 785                         if (rc < 0) {
 786                                 lod_pool_putref(pool);
 787                                 RETURN(-EINVAL);
 788                         }
 789                 }
 790
 791                 if (specific && lum->lmm_stripe_count > pool_tgt_count(pool)) {
 792                         CDEBUG(D_IOCTL,
 793                                "stripe count %u > # OSTs %u in the pool\n",
 794                                lum->lmm_stripe_count, pool_tgt_count(pool));
 795                         lod_pool_putref(pool);
 796                         RETURN(-EINVAL);
 797                 }
 798
 799                 lod_pool_putref(pool);
 800         }
 801
 802         RETURN(0);
 803 }
 804
 805 void lod_fix_desc_stripe_size(__u64 *val)
 806 {
 807         if (*val < PTLRPC_MAX_BRW_SIZE) {
 808                 LCONSOLE_WARN("Increasing default stripe size to min %u\n",
 809                               PTLRPC_MAX_BRW_SIZE);
 810                 *val = PTLRPC_MAX_BRW_SIZE;
 811         } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) {
 812                 *val &= ~(LOV_MIN_STRIPE_SIZE - 1);
 813                 LCONSOLE_WARN("Changing default stripe size to "LPU64" (a "
 814                               "multiple of %u)\n",
 815                               *val, LOV_MIN_STRIPE_SIZE);
 816         }
 817 }
 818
 819 void lod_fix_desc_stripe_count(__u32 *val)
 820 {
 821         if (*val == 0)
 822                 *val = 1;
 823 }
 824
 825 void lod_fix_desc_pattern(__u32 *val)
 826 {
 827         /* from lov_setstripe */
 828         if ((*val != 0) && (*val != LOV_PATTERN_RAID0)) {
 829                 LCONSOLE_WARN("Unknown stripe pattern: %#x\n", *val);
 830                 *val = 0;
 831         }
 832 }
 833
 834 void lod_fix_desc_qos_maxage(__u32 *val)
 835 {
 836         /* fix qos_maxage */
 837         if (*val == 0)
 838                 *val = QOS_DEFAULT_MAXAGE;
 839 }
 840
 841 void lod_fix_desc(struct lov_desc *desc)
 842 {
 843         lod_fix_desc_stripe_size(&desc->ld_default_stripe_size);
 844         lod_fix_desc_stripe_count(&desc->ld_default_stripe_count);
 845         lod_fix_desc_pattern(&desc->ld_pattern);
 846         lod_fix_desc_qos_maxage(&desc->ld_qos_maxage);
 847 }
 848
 849 int lod_pools_init(struct lod_device *lod, struct lustre_cfg *lcfg)
 850 {
 851         struct obd_device          *obd;
 852         struct lov_desc            *desc;
 853         int                         rc;
 854         ENTRY;
 855
 856         obd = class_name2obd(lustre_cfg_string(lcfg, 0));
 857         LASSERT(obd != NULL);
 858         obd->obd_lu_dev = &lod->lod_dt_dev.dd_lu_dev;
 859
 860         if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
 861                 CERROR("LOD setup requires a descriptor\n");
 862                 RETURN(-EINVAL);
 863         }
 864
 865         desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
 866
 867         if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
 868                 CERROR("descriptor size wrong: %d > %d\n",
 869                        (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
 870                 RETURN(-EINVAL);
 871         }
 872
 873         if (desc->ld_magic != LOV_DESC_MAGIC) {
 874                 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
 875                         CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
 876                                obd->obd_name, desc);
 877                         lustre_swab_lov_desc(desc);
 878                 } else {
 879                         CERROR("%s: Bad lov desc magic: %#x\n",
 880                                obd->obd_name, desc->ld_magic);
 881                         RETURN(-EINVAL);
 882                 }
 883         }
 884
 885         lod_fix_desc(desc);
 886
 887         desc->ld_active_tgt_count = 0;
 888         lod->lod_desc = *desc;
 889
 890         lod->lod_sp_me = LUSTRE_SP_CLI;
 891
 892         /* Set up allocation policy (QoS and RR) */
 893         CFS_INIT_LIST_HEAD(&lod->lod_qos.lq_oss_list);
 894         init_rwsem(&lod->lod_qos.lq_rw_sem);
 895         lod->lod_qos.lq_dirty = 1;
 896         lod->lod_qos.lq_rr.lqr_dirty = 1;
 897         lod->lod_qos.lq_reset = 1;
 898         /* Default priority is toward free space balance */
 899         lod->lod_qos.lq_prio_free = 232;
 900         /* Default threshold for rr (roughly 17%) */
 901         lod->lod_qos.lq_threshold_rr = 43;
 902         /* Init statfs fields */
 903         OBD_ALLOC_PTR(lod->lod_qos.lq_statfs_data);
 904         if (NULL == lod->lod_qos.lq_statfs_data)
 905                 RETURN(-ENOMEM);
 906         cfs_waitq_init(&lod->lod_qos.lq_statfs_waitq);
 907
 908         /* Set up OST pool environment */
 909         lod->lod_pools_hash_body = cfs_hash_create("POOLS", HASH_POOLS_CUR_BITS,
 910                                                    HASH_POOLS_MAX_BITS,
 911                                                    HASH_POOLS_BKT_BITS, 0,
 912                                                    CFS_HASH_MIN_THETA,
 913                                                    CFS_HASH_MAX_THETA,
 914                                                    &pool_hash_operations,
 915                                                    CFS_HASH_DEFAULT);
 916         if (!lod->lod_pools_hash_body)
 917                 GOTO(out_statfs, rc = -ENOMEM);
 918         CFS_INIT_LIST_HEAD(&lod->lod_pool_list);
 919         lod->lod_pool_count = 0;
 920         rc = lod_ost_pool_init(&lod->lod_pool_info, 0);
 921         if (rc)
 922                 GOTO(out_hash, rc);
 923         rc = lod_ost_pool_init(&lod->lod_qos.lq_rr.lqr_pool, 0);
 924         if (rc)
 925                 GOTO(out_pool_info, rc);
 926
 927         /* the OST array and bitmap are allocated/grown dynamically as OSTs are
 928          * added to the LOD, see lod_add_device() */
 929         lod->lod_ost_bitmap = NULL;
 930         lod->lod_osts_size  = 0;
 931         lod->lod_ostnr      = 0;
 932
 933         lod->lod_death_row = 0;
 934         lod->lod_refcount  = 0;
 935
 936         RETURN(0);
 937
 938 out_pool_info:
 939         lod_ost_pool_free(&lod->lod_pool_info);
 940 out_hash:
 941         cfs_hash_putref(lod->lod_pools_hash_body);
 942 out_statfs:
 943         OBD_FREE_PTR(lod->lod_qos.lq_statfs_data);
 944         return rc;
 945 }
 946
 947 int lod_pools_fini(struct lod_device *lod)
 948 {
 949         struct obd_device   *obd = lod2obd(lod);
 950         cfs_list_t          *pos, *tmp;
 951         struct pool_desc    *pool;
 952         ENTRY;
 953
 954         cfs_list_for_each_safe(pos, tmp, &lod->lod_pool_list) {
 955                 pool = cfs_list_entry(pos, struct pool_desc, pool_list);
 956                 /* free pool structs */
 957                 CDEBUG(D_INFO, "delete pool %p\n", pool);
 958                 lod_pool_del(obd, pool->pool_name);
 959         }
 960
 961         if (lod->lod_osts_size > 0) {
 962                 int idx;
 963                 lod_getref(lod);
 964                 mutex_lock(&lod->lod_mutex);
 965                 cfs_foreach_bit(lod->lod_ost_bitmap, idx)
 966                         __lod_del_device(lod, idx);
 967                 mutex_unlock(&lod->lod_mutex);
 968                 lod_putref(lod);
 969                 CFS_FREE_BITMAP(lod->lod_ost_bitmap);
 970                 for (idx = 0; idx < OST_PTRS; idx++) {
 971                         if (lod->lod_ost_idx[idx])
 972                                 OBD_FREE_PTR(lod->lod_ost_idx[idx]);
 973                 }
 974                 lod->lod_osts_size = 0;
 975         }
 976
 977         cfs_hash_putref(lod->lod_pools_hash_body);
 978         lod_ost_pool_free(&(lod->lod_qos.lq_rr.lqr_pool));
 979         lod_ost_pool_free(&lod->lod_pool_info);
 980         OBD_FREE_PTR(lod->lod_qos.lq_statfs_data);
 981         RETURN(0);
 982 }
 983