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
 108                         rc = obd_disconnect(ost_desc->ltd_exp);
 109                         if (rc)
 110                                 CERROR("%s: failed to disconnect %s: rc = %d\n",
 111                                        lod2obd(lod)->obd_name,
 112                                        obd_uuid2str(&ost_desc->ltd_uuid), rc);
 113                         OBD_FREE_PTR(ost_desc);
 114                 }
 115         } else {
 116                 mutex_unlock(&lod->lod_mutex);
 117                 up_read(&lod->lod_rw_sem);
 118         }
 119 }
 120
 121 static int lod_bitmap_resize(struct lod_device *lod, __u32 newsize)
 122 {
 123         cfs_bitmap_t *new_bitmap, *old_bitmap = NULL;
 124         int           rc = 0;
 125         ENTRY;
 126
 127         /* grab write reference on the lod. Relocating the array requires
 128          * exclusive access */
 129         down_write(&lod->lod_rw_sem);
 130
 131         if (newsize <= lod->lod_osts_size)
 132                 /* someone else has already resize the array */
 133                 GOTO(out, rc = 0);
 134
 135         /* allocate new bitmap */
 136         new_bitmap = CFS_ALLOCATE_BITMAP(newsize);
 137         if (!new_bitmap)
 138                 GOTO(out, rc = -ENOMEM);
 139
 140         if (lod->lod_osts_size > 0) {
 141                 /* the bitmap already exists, we need
 142                  * to copy data from old one */
 143                 cfs_bitmap_copy(new_bitmap, lod->lod_ost_bitmap);
 144                 old_bitmap = lod->lod_ost_bitmap;
 145         }
 146
 147         lod->lod_osts_size  = newsize;
 148         lod->lod_ost_bitmap = new_bitmap;
 149
 150         if (old_bitmap)
 151                 CFS_FREE_BITMAP(old_bitmap);
 152
 153         CDEBUG(D_CONFIG, "ost size: %d\n", lod->lod_osts_size);
 154
 155         EXIT;
 156 out:
 157         up_write(&lod->lod_rw_sem);
 158         return rc;
 159 }
 160
 161 /*
 162  * Connect LOD to a new OSP and add it to the device table.
 163  *
 164  * \param env - is the environment passed by the caller
 165  * \param lod - is the LOD device to be connected to the new OSP
 166  * \param osp - is the name of OSP device name about to be added
 167  * \param index - is the OSP index
 168  * \param gen - is the generation number
 169  */
 170 int lod_add_device(const struct lu_env *env, struct lod_device *lod,
 171                    char *osp, unsigned index, unsigned gen, int active)
 172 {
 173         struct obd_connect_data *data = NULL;
 174         struct obd_export       *exp = NULL;
 175         struct obd_device       *obd;
 176         struct lu_device        *ldev;
 177         struct dt_device        *d;
 178         int                      rc;
 179         struct lod_ost_desc     *ost_desc;
 180         struct obd_uuid          obd_uuid;
 181
 182         ENTRY;
 183
 184         CDEBUG(D_CONFIG, "osp:%s idx:%d gen:%d\n", osp, index, gen);
 185
 186         if (gen <= 0) {
 187                 CERROR("request to add OBD %s with invalid generation: %d\n",
 188                        osp, gen);
 189                 RETURN(-EINVAL);
 190         }
 191
 192         obd_str2uuid(&obd_uuid, osp);
 193
 194         obd = class_find_client_obd(&obd_uuid, LUSTRE_OSP_NAME,
 195                                 &lod->lod_dt_dev.dd_lu_dev.ld_obd->obd_uuid);
 196         if (obd == NULL) {
 197                 CERROR("can't find %s device\n", osp);
 198                 RETURN(-EINVAL);
 199         }
 200
 201         OBD_ALLOC_PTR(data);
 202         if (data == NULL)
 203                 RETURN(-ENOMEM);
 204
 205         data->ocd_connect_flags = OBD_CONNECT_INDEX | OBD_CONNECT_VERSION;
 206         data->ocd_version = LUSTRE_VERSION_CODE;
 207         data->ocd_index = index;
 208
 209         rc = obd_connect(env, &exp, obd, &obd->obd_uuid, data, NULL);
 210         OBD_FREE_PTR(data);
 211         if (rc) {
 212                 CERROR("%s: cannot connect to next dev %s (%d)\n",
 213                        obd->obd_name, osp, rc);
 214                 GOTO(out_free, rc);
 215         }
 216
 217         LASSERT(obd->obd_lu_dev);
 218         LASSERT(obd->obd_lu_dev->ld_site = lod->lod_dt_dev.dd_lu_dev.ld_site);
 219
 220         ldev = obd->obd_lu_dev;
 221         d = lu2dt_dev(ldev);
 222
 223         /* Allocate ost descriptor and fill it */
 224         OBD_ALLOC_PTR(ost_desc);
 225         if (!ost_desc)
 226                 GOTO(out_conn, rc = -ENOMEM);
 227
 228         ost_desc->ltd_ost    = d;
 229         ost_desc->ltd_exp    = exp;
 230         ost_desc->ltd_uuid   = obd->u.cli.cl_target_uuid;
 231         ost_desc->ltd_gen    = gen;
 232         ost_desc->ltd_index  = index;
 233         ost_desc->ltd_active = active;
 234
 235         lod_getref(lod);
 236         if (index >= lod->lod_osts_size) {
 237                 /* we have to increase the size of the lod_osts array */
 238                 __u32  newsize;
 239
 240                 newsize = max(lod->lod_osts_size, (__u32)2);
 241                 while (newsize < index + 1)
 242                         newsize = newsize << 1;
 243
 244                 /* lod_bitmap_resize() needs lod_rw_sem
 245                  * which we hold with th reference */
 246                 lod_putref(lod);
 247
 248                 rc = lod_bitmap_resize(lod, newsize);
 249                 if (rc)
 250                         GOTO(out_desc, rc);
 251
 252                 lod_getref(lod);
 253         }
 254
 255         mutex_lock(&lod->lod_mutex);
 256         if (cfs_bitmap_check(lod->lod_ost_bitmap, index)) {
 257                 CERROR("%s: device %d is registered already\n", obd->obd_name,
 258                        index);
 259                 GOTO(out_mutex, rc = -EEXIST);
 260         }
 261
 262         if (lod->lod_ost_idx[index / OST_PTRS_PER_BLOCK] == NULL) {
 263                 OBD_ALLOC_PTR(lod->lod_ost_idx[index / OST_PTRS_PER_BLOCK]);
 264                 if (lod->lod_ost_idx[index / OST_PTRS_PER_BLOCK] == NULL) {
 265                         CERROR("can't allocate index to add %s\n",
 266                                obd->obd_name);
 267                         GOTO(out_mutex, rc = -ENOMEM);
 268                 }
 269         }
 270
 271         rc = lod_ost_pool_add(&lod->lod_pool_info, index, lod->lod_osts_size);
 272         if (rc) {
 273                 CERROR("%s: can't set up pool, failed with %d\n",
 274                        obd->obd_name, rc);
 275                 GOTO(out_mutex, rc);
 276         }
 277
 278         rc = qos_add_tgt(lod, ost_desc);
 279         if (rc) {
 280                 CERROR("%s: qos_add_tgt(%s) failed: rc = %d\n", obd->obd_name,
 281                        obd_uuid2str(&ost_desc->ltd_uuid), rc);
 282                 GOTO(out_pool, rc);
 283         }
 284
 285         /* The new OST is now a full citizen */
 286         if (index >= lod->lod_desc.ld_tgt_count)
 287                 lod->lod_desc.ld_tgt_count = index + 1;
 288         if (active)
 289                 lod->lod_desc.ld_active_tgt_count++;
 290         OST_TGT(lod, index) = ost_desc;
 291         cfs_bitmap_set(lod->lod_ost_bitmap, index);
 292         lod->lod_ostnr++;
 293         mutex_unlock(&lod->lod_mutex);
 294         lod_putref(lod);
 295
 296         if (lod->lod_recovery_completed)
 297                 ldev->ld_ops->ldo_recovery_complete(env, ldev);
 298
 299         RETURN(0);
 300
 301 out_pool:
 302         lod_ost_pool_remove(&lod->lod_pool_info, index);
 303 out_mutex:
 304         mutex_unlock(&lod->lod_mutex);
 305         lod_putref(lod);
 306 out_desc:
 307         OBD_FREE_PTR(ost_desc);
 308 out_conn:
 309         obd_disconnect(exp);
 310 out_free:
 311         return rc;
 312 }
 313
 314 /*
 315  * helper function to schedule OST removal from the device table
 316  */
 317 static void __lod_del_device(struct lod_device *lod, unsigned idx)
 318 {
 319         LASSERT(OST_TGT(lod,idx));
 320         if (OST_TGT(lod,idx)->ltd_reap == 0) {
 321                 OST_TGT(lod,idx)->ltd_reap = 1;
 322                 lod->lod_death_row++;
 323         }
 324 }
 325
 326 /*
 327  * Add support for administratively disabled OST (through the MGS).
 328  * Schedule a target for deletion.  Disconnection and real removal from the
 329  * table takes place in lod_putref() once the last table user release its
 330  * reference.
 331  *
 332  * \param env - is the environment passed by the caller
 333  * \param lod - is the lod device currently connected to the OSP about to be
 334  *              removed
 335  * \param osp - is the name of OSP device about to be removed
 336  * \param idx - is the OSP index
 337  * \param gen - is the generation number, not used currently
 338  */
 339 int lod_del_device(const struct lu_env *env, struct lod_device *lod,
 340                    char *osp, unsigned idx, unsigned gen)
 341 {
 342         struct obd_device *obd;
 343         int                rc = 0;
 344         struct obd_uuid    uuid;
 345         ENTRY;
 346
 347         CDEBUG(D_CONFIG, "osp:%s idx:%d gen:%d\n", osp, idx, gen);
 348
 349         obd = class_name2obd(osp);
 350         if (obd == NULL) {
 351                 CERROR("can't find %s device\n", osp);
 352                 RETURN(-EINVAL);
 353         }
 354
 355         if (gen <= 0) {
 356                 CERROR("%s: request to remove OBD %s with invalid generation %d"
 357                        "\n", obd->obd_name, osp, gen);
 358                 RETURN(-EINVAL);
 359         }
 360
 361         obd_str2uuid(&uuid,  osp);
 362
 363         lod_getref(lod);
 364         mutex_lock(&lod->lod_mutex);
 365         /* check that the index is allocated in the bitmap */
 366         if (!cfs_bitmap_check(lod->lod_ost_bitmap, idx) || !OST_TGT(lod,idx)) {
 367                 CERROR("%s: device %d is not set up\n", obd->obd_name, idx);
 368                 GOTO(out, rc = -EINVAL);
 369         }
 370
 371         /* check that the UUID matches */
 372         if (!obd_uuid_equals(&uuid, &OST_TGT(lod,idx)->ltd_uuid)) {
 373                 CERROR("%s: LOD target UUID %s at index %d does not match %s\n",
 374                        obd->obd_name, obd_uuid2str(&OST_TGT(lod,idx)->ltd_uuid),
 375                        idx, osp);
 376                 GOTO(out, rc = -EINVAL);
 377         }
 378
 379         __lod_del_device(lod, idx);
 380         EXIT;
 381 out:
 382         mutex_unlock(&lod->lod_mutex);
 383         lod_putref(lod);
 384         return(rc);
 385 }
 386
 387 int lod_ea_store_resize(struct lod_thread_info *info, int size)
 388 {
 389         int round = size_roundup_power2(size);
 390
 391         LASSERT(round <= lov_mds_md_size(LOV_MAX_STRIPE_COUNT, LOV_MAGIC_V3));
 392         if (info->lti_ea_store) {
 393                 LASSERT(info->lti_ea_store_size);
 394                 LASSERT(info->lti_ea_store_size < round);
 395                 CDEBUG(D_INFO, "EA store size %d is not enough, need %d\n",
 396                        info->lti_ea_store_size, round);
 397                 OBD_FREE_LARGE(info->lti_ea_store, info->lti_ea_store_size);
 398                 info->lti_ea_store = NULL;
 399                 info->lti_ea_store_size = 0;
 400         }
 401
 402         OBD_ALLOC_LARGE(info->lti_ea_store, round);
 403         if (info->lti_ea_store == NULL)
 404                 RETURN(-ENOMEM);
 405         info->lti_ea_store_size = round;
 406         RETURN(0);
 407 }
 408
 409 /*
 410  * generate and write LOV EA for given striped object
 411  */
 412 int lod_generate_and_set_lovea(const struct lu_env *env,
 413                                struct lod_object *lo, struct thandle *th)
 414 {
 415         struct lod_thread_info  *info = lod_env_info(env);
 416         struct dt_object        *next = dt_object_child(&lo->ldo_obj);
 417         const struct lu_fid     *fid  = lu_object_fid(&lo->ldo_obj.do_lu);
 418         struct lov_mds_md_v1    *lmm;
 419         struct lov_ost_data_v1  *objs;
 420         __u32                    magic;
 421         int                      i, rc, lmm_size;
 422         ENTRY;
 423
 424         LASSERT(lo);
 425         LASSERT(lo->ldo_stripenr > 0);
 426
 427         magic = lo->ldo_pool ? LOV_MAGIC_V3 : LOV_MAGIC_V1;
 428         lmm_size = lov_mds_md_size(lo->ldo_stripenr, magic);
 429         if (info->lti_ea_store_size < lmm_size) {
 430                 rc = lod_ea_store_resize(info, lmm_size);
 431                 if (rc)
 432                         RETURN(rc);
 433         }
 434
 435         lmm = info->lti_ea_store;
 436
 437         lmm->lmm_magic = cpu_to_le32(magic);
 438         lmm->lmm_pattern = cpu_to_le32(LOV_PATTERN_RAID0);
 439         lmm->lmm_object_id = cpu_to_le64(fid_ver_oid(fid));
 440         lmm->lmm_object_seq = cpu_to_le64(fid_seq(fid));
 441         lmm->lmm_stripe_size = cpu_to_le32(lo->ldo_stripe_size);
 442         lmm->lmm_stripe_count = cpu_to_le16(lo->ldo_stripenr);
 443         lmm->lmm_layout_gen = 0;
 444         if (magic == LOV_MAGIC_V1) {
 445                 objs = &lmm->lmm_objects[0];
 446         } else {
 447                 struct lov_mds_md_v3 *v3 = (struct lov_mds_md_v3 *) lmm;
 448                 strncpy(v3->lmm_pool_name, lo->ldo_pool, LOV_MAXPOOLNAME);
 449                 objs = &v3->lmm_objects[0];
 450         }
 451
 452         for (i = 0; i < lo->ldo_stripenr; i++) {
 453                 const struct lu_fid     *fid;
 454                 struct lod_device       *lod;
 455                 __u32                   index;
 456
 457                 lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
 458                 LASSERT(lo->ldo_stripe[i]);
 459                 fid = lu_object_fid(&lo->ldo_stripe[i]->do_lu);
 460
 461                 rc = fid_ostid_pack(fid, &info->lti_ostid);
 462                 LASSERT(rc == 0);
 463
 464                 objs[i].l_object_id  = cpu_to_le64(info->lti_ostid.oi_id);
 465                 objs[i].l_object_seq = cpu_to_le64(info->lti_ostid.oi_seq);
 466                 objs[i].l_ost_gen    = cpu_to_le32(0);
 467                 rc = lod_fld_lookup(env, lod, fid, &index, LU_SEQ_RANGE_OST);
 468                 if (rc < 0) {
 469                         CERROR("%s: Can not locate "DFID": rc = %d\n",
 470                                lod2obd(lod)->obd_name, PFID(fid), rc);
 471                         RETURN(rc);
 472                 }
 473                 objs[i].l_ost_idx = cpu_to_le32(index);
 474         }
 475
 476         info->lti_buf.lb_buf = lmm;
 477         info->lti_buf.lb_len = lmm_size;
 478         rc = dt_xattr_set(env, next, &info->lti_buf, XATTR_NAME_LOV, 0,
 479                           th, BYPASS_CAPA);
 480
 481         RETURN(rc);
 482 }
 483
 484 int lod_get_lov_ea(const struct lu_env *env, struct lod_object *lo)
 485 {
 486         struct lod_thread_info *info = lod_env_info(env);
 487         struct dt_object       *next = dt_object_child(&lo->ldo_obj);
 488         int                     rc;
 489         ENTRY;
 490
 491         LASSERT(info);
 492
 493         if (unlikely(info->lti_ea_store_size == 0)) {
 494                 /* just to enter in allocation block below */
 495                 rc = -ERANGE;
 496         } else {
 497 repeat:
 498                 info->lti_buf.lb_buf = info->lti_ea_store;
 499                 info->lti_buf.lb_len = info->lti_ea_store_size;
 500                 rc = dt_xattr_get(env, next, &info->lti_buf, XATTR_NAME_LOV,
 501                                   BYPASS_CAPA);
 502         }
 503         /* if object is not striped or inaccessible */
 504         if (rc == -ENODATA)
 505                 RETURN(0);
 506
 507         if (rc == -ERANGE) {
 508                 /* EA doesn't fit, reallocate new buffer */
 509                 rc = dt_xattr_get(env, next, &LU_BUF_NULL, XATTR_NAME_LOV,
 510                                   BYPASS_CAPA);
 511                 if (rc == -ENODATA)
 512                         RETURN(0);
 513                 else if (rc < 0)
 514                         RETURN(rc);
 515
 516                 LASSERT(rc > 0);
 517                 rc = lod_ea_store_resize(info, rc);
 518                 if (rc)
 519                         RETURN(rc);
 520                 goto repeat;
 521         }
 522
 523         RETURN(rc);
 524 }
 525
 526 int lod_store_def_striping(const struct lu_env *env, struct dt_object *dt,
 527                            struct thandle *th)
 528 {
 529         struct lod_thread_info  *info = lod_env_info(env);
 530         struct lod_object       *lo = lod_dt_obj(dt);
 531         struct dt_object        *next = dt_object_child(dt);
 532         struct lov_user_md_v3   *v3;
 533         int                      rc;
 534         ENTRY;
 535
 536         LASSERT(S_ISDIR(dt->do_lu.lo_header->loh_attr));
 537
 538         /*
 539          * store striping defaults into new directory
 540          * used to implement defaults inheritance
 541          */
 542
 543         /* probably nothing to inherite */
 544         if (lo->ldo_striping_cached == 0)
 545                 RETURN(0);
 546
 547         if (LOVEA_DELETE_VALUES(lo->ldo_def_stripe_size, lo->ldo_def_stripenr,
 548                                 lo->ldo_def_stripe_offset))
 549                 RETURN(0);
 550
 551         /* XXX: use thread info */
 552         OBD_ALLOC_PTR(v3);
 553         if (v3 == NULL)
 554                 RETURN(-ENOMEM);
 555
 556         v3->lmm_magic = cpu_to_le32(LOV_MAGIC_V3);
 557         v3->lmm_pattern = cpu_to_le32(LOV_PATTERN_RAID0);
 558         v3->lmm_object_id = 0;
 559         v3->lmm_object_seq = 0;
 560         v3->lmm_stripe_size = cpu_to_le32(lo->ldo_def_stripe_size);
 561         v3->lmm_stripe_count = cpu_to_le16(lo->ldo_def_stripenr);
 562         v3->lmm_stripe_offset = cpu_to_le16(lo->ldo_def_stripe_offset);
 563         if (lo->ldo_pool)
 564                 strncpy(v3->lmm_pool_name, lo->ldo_pool, LOV_MAXPOOLNAME);
 565
 566         info->lti_buf.lb_buf = v3;
 567         info->lti_buf.lb_len = sizeof(*v3);
 568         rc = dt_xattr_set(env, next, &info->lti_buf, XATTR_NAME_LOV, 0, th,
 569                         BYPASS_CAPA);
 570
 571         OBD_FREE_PTR(v3);
 572
 573         RETURN(rc);
 574 }
 575
 576 /*
 577  * allocate array of objects pointers, find/create objects
 578  * stripenr and other fields should be initialized by this moment
 579  */
 580 int lod_initialize_objects(const struct lu_env *env, struct lod_object *lo,
 581                            struct lov_ost_data_v1 *objs)
 582 {
 583         struct lod_thread_info  *info = lod_env_info(env);
 584         struct lod_device       *md = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
 585         struct lu_object        *o, *n;
 586         struct lu_device        *nd;
 587         int                     i, idx, rc = 0;
 588         ENTRY;
 589
 590         LASSERT(lo);
 591         LASSERT(lo->ldo_stripe == NULL);
 592         LASSERT(lo->ldo_stripenr > 0);
 593         LASSERT(lo->ldo_stripe_size > 0);
 594
 595         i = sizeof(struct dt_object *) * lo->ldo_stripenr;
 596         OBD_ALLOC(lo->ldo_stripe, i);
 597         if (lo->ldo_stripe == NULL)
 598                 GOTO(out, rc = -ENOMEM);
 599         lo->ldo_stripes_allocated = lo->ldo_stripenr;
 600
 601         for (i = 0; i < lo->ldo_stripenr; i++) {
 602                 info->lti_ostid.oi_id = le64_to_cpu(objs[i].l_object_id);
 603                 info->lti_ostid.oi_seq = le64_to_cpu(objs[i].l_object_seq);
 604                 idx = le64_to_cpu(objs[i].l_ost_idx);
 605                 fid_ostid_unpack(&info->lti_fid, &info->lti_ostid, idx);
 606                 LASSERTF(fid_is_sane(&info->lti_fid), ""DFID" insane!\n",
 607                          PFID(&info->lti_fid));
 608                 /*
 609                  * XXX: assertion is left for testing, to make
 610                  * sure we never process requests till configuration
 611                  * is completed. to be changed to -EINVAL
 612                  */
 613
 614                 lod_getref(md);
 615                 LASSERT(cfs_bitmap_check(md->lod_ost_bitmap, idx));
 616                 LASSERT(OST_TGT(md,idx));
 617                 LASSERTF(OST_TGT(md,idx)->ltd_ost, "idx %d\n", idx);
 618                 nd = &OST_TGT(md,idx)->ltd_ost->dd_lu_dev;
 619                 lod_putref(md);
 620
 621                 o = lu_object_find_at(env, nd, &info->lti_fid, NULL);
 622                 if (IS_ERR(o))
 623                         GOTO(out, rc = PTR_ERR(o));
 624
 625                 n = lu_object_locate(o->lo_header, nd->ld_type);
 626                 LASSERT(n);
 627
 628                 lo->ldo_stripe[i] = container_of(n, struct dt_object, do_lu);
 629         }
 630
 631 out:
 632         RETURN(rc);
 633 }
 634
 635 /*
 636  * Parse striping information stored in lti_ea_store
 637  */
 638 int lod_parse_striping(const struct lu_env *env, struct lod_object *lo,
 639                        const struct lu_buf *buf)
 640 {
 641         struct lov_mds_md_v1    *lmm;
 642         struct lov_ost_data_v1  *objs;
 643         __u32                    magic;
 644         int                      rc = 0;
 645         ENTRY;
 646
 647         LASSERT(buf);
 648         LASSERT(buf->lb_buf);
 649         LASSERT(buf->lb_len);
 650
 651         lmm = (struct lov_mds_md_v1 *) buf->lb_buf;
 652         magic = le32_to_cpu(lmm->lmm_magic);
 653
 654         if (magic != LOV_MAGIC_V1 && magic != LOV_MAGIC_V3)
 655                 GOTO(out, rc = -EINVAL);
 656         if (le32_to_cpu(lmm->lmm_pattern) != LOV_PATTERN_RAID0)
 657                 GOTO(out, rc = -EINVAL);
 658
 659         lo->ldo_stripe_size = le32_to_cpu(lmm->lmm_stripe_size);
 660         lo->ldo_stripenr = le16_to_cpu(lmm->lmm_stripe_count);
 661         lo->ldo_layout_gen = le16_to_cpu(lmm->lmm_layout_gen);
 662
 663         LASSERT(buf->lb_len >= lov_mds_md_size(lo->ldo_stripenr, magic));
 664
 665         if (magic == LOV_MAGIC_V3) {
 666                 struct lov_mds_md_v3 *v3 = (struct lov_mds_md_v3 *) lmm;
 667                 objs = &v3->lmm_objects[0];
 668                 lod_object_set_pool(lo, v3->lmm_pool_name);
 669         } else {
 670                 objs = &lmm->lmm_objects[0];
 671         }
 672
 673         rc = lod_initialize_objects(env, lo, objs);
 674
 675 out:
 676         RETURN(rc);
 677 }
 678
 679 /*
 680  * Load and parse striping information, create in-core representation for the
 681  * stripes
 682  */
 683 int lod_load_striping(const struct lu_env *env, struct lod_object *lo)
 684 {
 685         struct lod_thread_info  *info = lod_env_info(env);
 686         struct dt_object        *next = dt_object_child(&lo->ldo_obj);
 687         int                      rc;
 688         ENTRY;
 689
 690         /*
 691          * currently this code is supposed to be called from declaration
 692          * phase only, thus the object is not expected to be locked by caller
 693          */
 694         dt_write_lock(env, next, 0);
 695         /* already initialized? */
 696         if (lo->ldo_stripe) {
 697                 int i;
 698                 /* check validity */
 699                 for (i = 0; i < lo->ldo_stripenr; i++)
 700                         LASSERTF(lo->ldo_stripe[i], "stripe %d is NULL\n", i);
 701                 GOTO(out, rc = 0);
 702         }
 703
 704         if (!dt_object_exists(next))
 705                 GOTO(out, rc = 0);
 706
 707         /* only regular files can be striped */
 708         if (!(lu_object_attr(lod2lu_obj(lo)) & S_IFREG))
 709                 GOTO(out, rc = 0);
 710
 711         LASSERT(lo->ldo_stripenr == 0);
 712
 713         rc = lod_get_lov_ea(env, lo);
 714         if (rc <= 0)
 715                 GOTO(out, rc);
 716
 717         /*
 718          * there is LOV EA (striping information) in this object
 719          * let's parse it and create in-core objects for the stripes
 720          */
 721         info->lti_buf.lb_buf = info->lti_ea_store;
 722         info->lti_buf.lb_len = info->lti_ea_store_size;
 723         rc = lod_parse_striping(env, lo, &info->lti_buf);
 724 out:
 725         dt_write_unlock(env, next);
 726         RETURN(rc);
 727 }
 728
 729 int lod_verify_striping(struct lod_device *d, const struct lu_buf *buf,
 730                         int specific)
 731 {
 732         struct lov_user_md_v1   *lum;
 733         struct lov_user_md_v3   *v3 = NULL;
 734         struct pool_desc        *pool = NULL;
 735         int                      rc;
 736         ENTRY;
 737
 738         lum = buf->lb_buf;
 739
 740         if (lum->lmm_magic != LOV_USER_MAGIC_V1 &&
 741             lum->lmm_magic != LOV_USER_MAGIC_V3 &&
 742             lum->lmm_magic != LOV_MAGIC_V1_DEF &&
 743             lum->lmm_magic != LOV_MAGIC_V3_DEF) {
 744                 CDEBUG(D_IOCTL, "bad userland LOV MAGIC: %#x\n",
 745                        lum->lmm_magic);
 746                 RETURN(-EINVAL);
 747         }
 748
 749         if ((specific && lum->lmm_pattern != LOV_PATTERN_RAID0) ||
 750             (specific == 0 && lum->lmm_pattern != 0)) {
 751                 CDEBUG(D_IOCTL, "bad userland stripe pattern: %#x\n",
 752                        lum->lmm_pattern);
 753                 RETURN(-EINVAL);
 754         }
 755
 756         /* 64kB is the largest common page size we see (ia64), and matches the
 757          * check in lfs */
 758         if (lum->lmm_stripe_size & (LOV_MIN_STRIPE_SIZE - 1)) {
 759                 CDEBUG(D_IOCTL, "stripe size %u not multiple of %u, fixing\n",
 760                        lum->lmm_stripe_size, LOV_MIN_STRIPE_SIZE);
 761                 RETURN(-EINVAL);
 762         }
 763
 764         /* an offset of -1 is treated as a "special" valid offset */
 765         if (lum->lmm_stripe_offset != (typeof(lum->lmm_stripe_offset))(-1)) {
 766                 /* if offset is not within valid range [0, osts_size) */
 767                 if (lum->lmm_stripe_offset >= d->lod_osts_size) {
 768                         CDEBUG(D_IOCTL, "stripe offset %u >= bitmap size %u\n",
 769                                lum->lmm_stripe_offset, d->lod_osts_size);
 770                         RETURN(-EINVAL);
 771                 }
 772
 773                 /* if lmm_stripe_offset is *not* in bitmap */
 774                 if (!cfs_bitmap_check(d->lod_ost_bitmap,
 775                                       lum->lmm_stripe_offset)) {
 776                         CDEBUG(D_IOCTL, "stripe offset %u not in bitmap\n",
 777                                lum->lmm_stripe_offset);
 778                         RETURN(-EINVAL);
 779                 }
 780         }
 781
 782         if (lum->lmm_magic == LOV_USER_MAGIC_V3)
 783                 v3 = buf->lb_buf;
 784
 785         if (v3)
 786                 pool = lod_find_pool(d, v3->lmm_pool_name);
 787
 788         if (pool != NULL) {
 789                 __u16 offs = v3->lmm_stripe_offset;
 790
 791                 if (offs != (typeof(v3->lmm_stripe_offset))(-1)) {
 792                         rc = lod_check_index_in_pool(offs, pool);
 793                         if (rc < 0) {
 794                                 lod_pool_putref(pool);
 795                                 RETURN(-EINVAL);
 796                         }
 797                 }
 798
 799                 if (specific && lum->lmm_stripe_count > pool_tgt_count(pool)) {
 800                         CDEBUG(D_IOCTL,
 801                                "stripe count %u > # OSTs %u in the pool\n",
 802                                lum->lmm_stripe_count, pool_tgt_count(pool));
 803                         lod_pool_putref(pool);
 804                         RETURN(-EINVAL);
 805                 }
 806
 807                 lod_pool_putref(pool);
 808         }
 809
 810         RETURN(0);
 811 }
 812
 813 void lod_fix_desc_stripe_size(__u64 *val)
 814 {
 815         if (*val < PTLRPC_MAX_BRW_SIZE) {
 816                 LCONSOLE_WARN("Increasing default stripe size to min %u\n",
 817                               PTLRPC_MAX_BRW_SIZE);
 818                 *val = PTLRPC_MAX_BRW_SIZE;
 819         } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) {
 820                 *val &= ~(LOV_MIN_STRIPE_SIZE - 1);
 821                 LCONSOLE_WARN("Changing default stripe size to "LPU64" (a "
 822                               "multiple of %u)\n",
 823                               *val, LOV_MIN_STRIPE_SIZE);
 824         }
 825 }
 826
 827 void lod_fix_desc_stripe_count(__u32 *val)
 828 {
 829         if (*val == 0)
 830                 *val = 1;
 831 }
 832
 833 void lod_fix_desc_pattern(__u32 *val)
 834 {
 835         /* from lov_setstripe */
 836         if ((*val != 0) && (*val != LOV_PATTERN_RAID0)) {
 837                 LCONSOLE_WARN("Unknown stripe pattern: %#x\n", *val);
 838                 *val = 0;
 839         }
 840 }
 841
 842 void lod_fix_desc_qos_maxage(__u32 *val)
 843 {
 844         /* fix qos_maxage */
 845         if (*val == 0)
 846                 *val = QOS_DEFAULT_MAXAGE;
 847 }
 848
 849 void lod_fix_desc(struct lov_desc *desc)
 850 {
 851         lod_fix_desc_stripe_size(&desc->ld_default_stripe_size);
 852         lod_fix_desc_stripe_count(&desc->ld_default_stripe_count);
 853         lod_fix_desc_pattern(&desc->ld_pattern);
 854         lod_fix_desc_qos_maxage(&desc->ld_qos_maxage);
 855 }
 856
 857 int lod_pools_init(struct lod_device *lod, struct lustre_cfg *lcfg)
 858 {
 859         struct obd_device          *obd;
 860         struct lov_desc            *desc;
 861         int                         rc;
 862         ENTRY;
 863
 864         obd = class_name2obd(lustre_cfg_string(lcfg, 0));
 865         LASSERT(obd != NULL);
 866         obd->obd_lu_dev = &lod->lod_dt_dev.dd_lu_dev;
 867
 868         if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
 869                 CERROR("LOD setup requires a descriptor\n");
 870                 RETURN(-EINVAL);
 871         }
 872
 873         desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
 874
 875         if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
 876                 CERROR("descriptor size wrong: %d > %d\n",
 877                        (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
 878                 RETURN(-EINVAL);
 879         }
 880
 881         if (desc->ld_magic != LOV_DESC_MAGIC) {
 882                 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
 883                         CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
 884                                obd->obd_name, desc);
 885                         lustre_swab_lov_desc(desc);
 886                 } else {
 887                         CERROR("%s: Bad lov desc magic: %#x\n",
 888                                obd->obd_name, desc->ld_magic);
 889                         RETURN(-EINVAL);
 890                 }
 891         }
 892
 893         lod_fix_desc(desc);
 894
 895         desc->ld_active_tgt_count = 0;
 896         lod->lod_desc = *desc;
 897
 898         lod->lod_sp_me = LUSTRE_SP_CLI;
 899
 900         /* Set up allocation policy (QoS and RR) */
 901         CFS_INIT_LIST_HEAD(&lod->lod_qos.lq_oss_list);
 902         init_rwsem(&lod->lod_qos.lq_rw_sem);
 903         lod->lod_qos.lq_dirty = 1;
 904         lod->lod_qos.lq_rr.lqr_dirty = 1;
 905         lod->lod_qos.lq_reset = 1;
 906         /* Default priority is toward free space balance */
 907         lod->lod_qos.lq_prio_free = 232;
 908         /* Default threshold for rr (roughly 17%) */
 909         lod->lod_qos.lq_threshold_rr = 43;
 910         /* Init statfs fields */
 911         OBD_ALLOC_PTR(lod->lod_qos.lq_statfs_data);
 912         if (NULL == lod->lod_qos.lq_statfs_data)
 913                 RETURN(-ENOMEM);
 914         cfs_waitq_init(&lod->lod_qos.lq_statfs_waitq);
 915
 916         /* Set up OST pool environment */
 917         lod->lod_pools_hash_body = cfs_hash_create("POOLS", HASH_POOLS_CUR_BITS,
 918                                                    HASH_POOLS_MAX_BITS,
 919                                                    HASH_POOLS_BKT_BITS, 0,
 920                                                    CFS_HASH_MIN_THETA,
 921                                                    CFS_HASH_MAX_THETA,
 922                                                    &pool_hash_operations,
 923                                                    CFS_HASH_DEFAULT);
 924         if (!lod->lod_pools_hash_body)
 925                 GOTO(out_statfs, rc = -ENOMEM);
 926         CFS_INIT_LIST_HEAD(&lod->lod_pool_list);
 927         lod->lod_pool_count = 0;
 928         rc = lod_ost_pool_init(&lod->lod_pool_info, 0);
 929         if (rc)
 930                 GOTO(out_hash, rc);
 931         rc = lod_ost_pool_init(&lod->lod_qos.lq_rr.lqr_pool, 0);
 932         if (rc)
 933                 GOTO(out_pool_info, rc);
 934
 935         /* the OST array and bitmap are allocated/grown dynamically as OSTs are
 936          * added to the LOD, see lod_add_device() */
 937         lod->lod_ost_bitmap = NULL;
 938         lod->lod_osts_size  = 0;
 939         lod->lod_ostnr      = 0;
 940
 941         lod->lod_death_row = 0;
 942         lod->lod_refcount  = 0;
 943
 944         RETURN(0);
 945
 946 out_pool_info:
 947         lod_ost_pool_free(&lod->lod_pool_info);
 948 out_hash:
 949         cfs_hash_putref(lod->lod_pools_hash_body);
 950 out_statfs:
 951         OBD_FREE_PTR(lod->lod_qos.lq_statfs_data);
 952         return rc;
 953 }
 954
 955 int lod_pools_fini(struct lod_device *lod)
 956 {
 957         struct obd_device   *obd = lod2obd(lod);
 958         cfs_list_t          *pos, *tmp;
 959         struct pool_desc    *pool;
 960         ENTRY;
 961
 962         cfs_list_for_each_safe(pos, tmp, &lod->lod_pool_list) {
 963                 pool = cfs_list_entry(pos, struct pool_desc, pool_list);
 964                 /* free pool structs */
 965                 CDEBUG(D_INFO, "delete pool %p\n", pool);
 966                 lod_pool_del(obd, pool->pool_name);
 967         }
 968
 969         if (lod->lod_osts_size > 0) {
 970                 int idx;
 971                 lod_getref(lod);
 972                 mutex_lock(&lod->lod_mutex);
 973                 cfs_foreach_bit(lod->lod_ost_bitmap, idx)
 974                         __lod_del_device(lod, idx);
 975                 mutex_unlock(&lod->lod_mutex);
 976                 lod_putref(lod);
 977                 CFS_FREE_BITMAP(lod->lod_ost_bitmap);
 978                 for (idx = 0; idx < OST_PTRS; idx++) {
 979                         if (lod->lod_ost_idx[idx])
 980                                 OBD_FREE_PTR(lod->lod_ost_idx[idx]);
 981                 }
 982                 lod->lod_osts_size = 0;
 983         }
 984
 985         cfs_hash_putref(lod->lod_pools_hash_body);
 986         lod_ost_pool_free(&(lod->lod_qos.lq_rr.lqr_pool));
 987         lod_ost_pool_free(&lod->lod_pool_info);
 988         OBD_FREE_PTR(lod->lod_qos.lq_statfs_data);
 989         RETURN(0);
 990 }
 991