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->ltd_tgts 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_tgt_descs *ltd)
  52 {
  53         down_read(&ltd->ltd_rw_sem);
  54         mutex_lock(&ltd->ltd_mutex);
  55         ltd->ltd_refcount++;
  56         mutex_unlock(&ltd->ltd_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, struct lod_tgt_descs *ltd)
  67 {
  68         mutex_lock(&ltd->ltd_mutex);
  69         ltd->ltd_refcount--;
  70         if (ltd->ltd_refcount == 0 && ltd->ltd_death_row) {
  71                 struct lod_tgt_desc *tgt_desc, *tmp;
  72                 int                  idx;
  73                 CFS_LIST_HEAD(kill);
  74
  75                 CDEBUG(D_CONFIG, "destroying %d ltd desc\n",
  76                        ltd->ltd_death_row);
  77
  78                 cfs_foreach_bit(ltd->ltd_tgt_bitmap, idx) {
  79                         tgt_desc = LTD_TGT(ltd, idx);
  80                         LASSERT(tgt_desc);
  81
  82                         if (!tgt_desc->ltd_reap)
  83                                 continue;
  84
  85                         cfs_list_add(&tgt_desc->ltd_kill, &kill);
  86                         LTD_TGT(ltd, idx) = NULL;
  87                         /*FIXME: only support ost pool for now */
  88                         if (ltd == &lod->lod_ost_descs) {
  89                                 lod_ost_pool_remove(&lod->lod_pool_info, idx);
  90                                 if (tgt_desc->ltd_active)
  91                                         lod->lod_desc.ld_active_tgt_count--;
  92                         }
  93                         ltd->ltd_tgtnr--;
  94                         cfs_bitmap_clear(ltd->ltd_tgt_bitmap, idx);
  95                         ltd->ltd_death_row--;
  96                 }
  97                 mutex_unlock(&ltd->ltd_mutex);
  98                 up_read(&ltd->ltd_rw_sem);
  99
 100                 cfs_list_for_each_entry_safe(tgt_desc, tmp, &kill, ltd_kill) {
 101                         int rc;
 102                         cfs_list_del(&tgt_desc->ltd_kill);
 103                         if (ltd == &lod->lod_ost_descs) {
 104                                 /* remove from QoS structures */
 105                                 rc = qos_del_tgt(lod, tgt_desc);
 106                                 if (rc)
 107                                         CERROR("%s: qos_del_tgt(%s) failed:"
 108                                                "rc = %d\n",
 109                                                lod2obd(lod)->obd_name,
 110                                               obd_uuid2str(&tgt_desc->ltd_uuid),
 111                                                rc);
 112                         }
 113                         rc = obd_disconnect(tgt_desc->ltd_exp);
 114                         if (rc)
 115                                 CERROR("%s: failed to disconnect %s: rc = %d\n",
 116                                        lod2obd(lod)->obd_name,
 117                                        obd_uuid2str(&tgt_desc->ltd_uuid), rc);
 118                         OBD_FREE_PTR(tgt_desc);
 119                 }
 120         } else {
 121                 mutex_unlock(&ltd->ltd_mutex);
 122                 up_read(&ltd->ltd_rw_sem);
 123         }
 124 }
 125
 126 static int ltd_bitmap_resize(struct lod_tgt_descs *ltd, __u32 newsize)
 127 {
 128         cfs_bitmap_t *new_bitmap, *old_bitmap = NULL;
 129         int           rc = 0;
 130         ENTRY;
 131
 132         /* grab write reference on the lod. Relocating the array requires
 133          * exclusive access */
 134
 135         down_write(&ltd->ltd_rw_sem);
 136         if (newsize <= ltd->ltd_tgts_size)
 137                 /* someone else has already resize the array */
 138                 GOTO(out, rc = 0);
 139
 140         /* allocate new bitmap */
 141         new_bitmap = CFS_ALLOCATE_BITMAP(newsize);
 142         if (!new_bitmap)
 143                 GOTO(out, rc = -ENOMEM);
 144
 145         if (ltd->ltd_tgts_size > 0) {
 146                 /* the bitmap already exists, we need
 147                  * to copy data from old one */
 148                 cfs_bitmap_copy(new_bitmap, ltd->ltd_tgt_bitmap);
 149                 old_bitmap = ltd->ltd_tgt_bitmap;
 150         }
 151
 152         ltd->ltd_tgts_size  = newsize;
 153         ltd->ltd_tgt_bitmap = new_bitmap;
 154
 155         if (old_bitmap)
 156                 CFS_FREE_BITMAP(old_bitmap);
 157
 158         CDEBUG(D_CONFIG, "tgt size: %d\n", ltd->ltd_tgts_size);
 159
 160         EXIT;
 161 out:
 162         up_write(&ltd->ltd_rw_sem);
 163         return rc;
 164 }
 165
 166 /*
 167  * Connect LOD to a new OSP and add it to the device table.
 168  *
 169  * \param env - is the environment passed by the caller
 170  * \param lod - is the LOD device to be connected to the new OSP
 171  * \param osp - is the name of OSP device name about to be added
 172  * \param index - is the OSP index
 173  * \param gen - is the generation number
 174  */
 175 int lod_add_device(const struct lu_env *env, struct lod_device *lod,
 176                    char *osp, unsigned index, unsigned gen, int active)
 177 {
 178         struct obd_connect_data *data = NULL;
 179         struct obd_export       *exp = NULL;
 180         struct obd_device       *obd;
 181         struct lu_device        *ldev;
 182         struct dt_device        *d;
 183         int                      rc;
 184         struct lod_tgt_desc     *tgt_desc;
 185         struct lod_tgt_descs    *ltd = &lod->lod_ost_descs;
 186         struct obd_uuid         obd_uuid;
 187         ENTRY;
 188
 189         CDEBUG(D_CONFIG, "osp:%s idx:%d gen:%d\n", osp, index, gen);
 190
 191         if (gen <= 0) {
 192                 CERROR("request to add OBD %s with invalid generation: %d\n",
 193                        osp, gen);
 194                 RETURN(-EINVAL);
 195         }
 196
 197         obd_str2uuid(&obd_uuid, osp);
 198
 199         obd = class_find_client_obd(&obd_uuid, LUSTRE_OSP_NAME,
 200                                 &lod->lod_dt_dev.dd_lu_dev.ld_obd->obd_uuid);
 201         if (obd == NULL) {
 202                 CERROR("can't find %s device\n", osp);
 203                 RETURN(-EINVAL);
 204         }
 205
 206         OBD_ALLOC_PTR(data);
 207         if (data == NULL)
 208                 RETURN(-ENOMEM);
 209
 210         data->ocd_connect_flags = OBD_CONNECT_INDEX | OBD_CONNECT_VERSION;
 211         data->ocd_version = LUSTRE_VERSION_CODE;
 212         data->ocd_index = index;
 213
 214         rc = obd_connect(env, &exp, obd, &obd->obd_uuid, data, NULL);
 215         OBD_FREE_PTR(data);
 216         if (rc) {
 217                 CERROR("%s: cannot connect to next dev %s (%d)\n",
 218                        obd->obd_name, osp, rc);
 219                 GOTO(out_free, rc);
 220         }
 221
 222         LASSERT(obd->obd_lu_dev);
 223         LASSERT(obd->obd_lu_dev->ld_site = lod->lod_dt_dev.dd_lu_dev.ld_site);
 224
 225         ldev = obd->obd_lu_dev;
 226         d = lu2dt_dev(ldev);
 227
 228         /* Allocate ost descriptor and fill it */
 229         OBD_ALLOC_PTR(tgt_desc);
 230         if (!tgt_desc)
 231                 GOTO(out_conn, rc = -ENOMEM);
 232
 233         tgt_desc->ltd_tgt    = d;
 234         tgt_desc->ltd_exp    = exp;
 235         tgt_desc->ltd_uuid   = obd->u.cli.cl_target_uuid;
 236         tgt_desc->ltd_gen    = gen;
 237         tgt_desc->ltd_index  = index;
 238         tgt_desc->ltd_active = active;
 239
 240         lod_getref(ltd);
 241         if (index >= ltd->ltd_tgts_size) {
 242                 /* we have to increase the size of the lod_osts array */
 243                 __u32  newsize;
 244
 245                 newsize = max(ltd->ltd_tgts_size, (__u32)2);
 246                 while (newsize < index + 1)
 247                         newsize = newsize << 1;
 248
 249                 /* lod_bitmap_resize() needs lod_rw_sem
 250                  * which we hold with th reference */
 251                 lod_putref(lod, ltd);
 252
 253                 rc = ltd_bitmap_resize(ltd, newsize);
 254                 if (rc)
 255                         GOTO(out_desc, rc);
 256
 257                 lod_getref(ltd);
 258         }
 259
 260         mutex_lock(&ltd->ltd_mutex);
 261         if (cfs_bitmap_check(ltd->ltd_tgt_bitmap, index)) {
 262                 CERROR("%s: device %d is registered already\n", obd->obd_name,
 263                        index);
 264                 GOTO(out_mutex, rc = -EEXIST);
 265         }
 266
 267         if (ltd->ltd_tgt_idx[index / TGT_PTRS_PER_BLOCK] == NULL) {
 268                 OBD_ALLOC_PTR(ltd->ltd_tgt_idx[index / TGT_PTRS_PER_BLOCK]);
 269                 if (ltd->ltd_tgt_idx[index / TGT_PTRS_PER_BLOCK] == NULL) {
 270                         CERROR("can't allocate index to add %s\n",
 271                                obd->obd_name);
 272                         GOTO(out_mutex, rc = -ENOMEM);
 273                 }
 274         }
 275
 276         /* pool and qos are not supported for MDS stack yet */
 277         rc = lod_ost_pool_add(&lod->lod_pool_info, index,
 278                               lod->lod_osts_size);
 279         if (rc) {
 280                 CERROR("%s: can't set up pool, failed with %d\n",
 281                        obd->obd_name, rc);
 282                 GOTO(out_mutex, rc);
 283         }
 284
 285         rc = qos_add_tgt(lod, tgt_desc);
 286         if (rc) {
 287                 CERROR("%s: qos_add_tgt failed with %d\n",
 288                         obd->obd_name, rc);
 289                 GOTO(out_pool, rc);
 290         }
 291
 292         /* The new OST is now a full citizen */
 293         if (index >= lod->lod_desc.ld_tgt_count)
 294                 lod->lod_desc.ld_tgt_count = index + 1;
 295         if (active)
 296                 lod->lod_desc.ld_active_tgt_count++;
 297
 298         LTD_TGT(ltd, index) = tgt_desc;
 299         cfs_bitmap_set(ltd->ltd_tgt_bitmap, index);
 300         ltd->ltd_tgtnr++;
 301         mutex_unlock(&ltd->ltd_mutex);
 302         lod_putref(lod, ltd);
 303         if (lod->lod_recovery_completed)
 304                 ldev->ld_ops->ldo_recovery_complete(env, ldev);
 305
 306         RETURN(0);
 307
 308 out_pool:
 309         lod_ost_pool_remove(&lod->lod_pool_info, index);
 310 out_mutex:
 311         mutex_unlock(&ltd->ltd_mutex);
 312         lod_putref(lod, ltd);
 313 out_desc:
 314         OBD_FREE_PTR(tgt_desc);
 315 out_conn:
 316         obd_disconnect(exp);
 317 out_free:
 318         return rc;
 319 }
 320
 321 /*
 322  * helper function to schedule OST removal from the device table
 323  */
 324 static void __lod_del_device(struct lod_tgt_descs *ltd,
 325                              unsigned idx)
 326 {
 327         LASSERT(LTD_TGT(ltd, idx));
 328         if (LTD_TGT(ltd, idx)->ltd_reap == 0) {
 329                 LTD_TGT(ltd, idx)->ltd_reap = 1;
 330                 ltd->ltd_death_row++;
 331         }
 332 }
 333
 334 int lod_fini_tgt(struct lod_device *lod, struct lod_tgt_descs *ltd)
 335 {
 336         int idx;
 337
 338         if (ltd->ltd_tgts_size <= 0)
 339                 return 0;
 340         lod_getref(ltd);
 341         mutex_lock(&ltd->ltd_mutex);
 342         cfs_foreach_bit(ltd->ltd_tgt_bitmap, idx)
 343                 __lod_del_device(ltd, idx);
 344         mutex_unlock(&ltd->ltd_mutex);
 345         lod_putref(lod, ltd);
 346         CFS_FREE_BITMAP(ltd->ltd_tgt_bitmap);
 347         for (idx = 0; idx < TGT_PTRS; idx++) {
 348                 if (ltd->ltd_tgt_idx[idx])
 349                         OBD_FREE_PTR(ltd->ltd_tgt_idx[idx]);
 350         }
 351         ltd->ltd_tgts_size = 0;
 352         return 0;
 353 }
 354
 355 /*
 356  * Add support for administratively disabled OST (through the MGS).
 357  * Schedule a target for deletion.  Disconnection and real removal from the
 358  * table takes place in lod_putref() once the last table user release its
 359  * reference.
 360  *
 361  * \param env - is the environment passed by the caller
 362  * \param lod - is the lod device currently connected to the OSP about to be
 363  *              removed
 364  * \param osp - is the name of OSP device about to be removed
 365  * \param idx - is the OSP index
 366  * \param gen - is the generation number, not used currently
 367  */
 368 int lod_del_device(const struct lu_env *env, struct lod_device *lod,
 369                    struct lod_tgt_descs *ltd, char *osp, unsigned idx,
 370                    unsigned gen)
 371 {
 372         struct obd_device *obd;
 373         int                rc = 0;
 374         struct obd_uuid    uuid;
 375         ENTRY;
 376
 377         CDEBUG(D_CONFIG, "osp:%s idx:%d gen:%d\n", osp, idx, gen);
 378
 379         obd = class_name2obd(osp);
 380         if (obd == NULL) {
 381                 CERROR("can't find %s device\n", osp);
 382                 RETURN(-EINVAL);
 383         }
 384
 385         if (gen <= 0) {
 386                 CERROR("%s: request to remove OBD %s with invalid generation %d"
 387                        "\n", obd->obd_name, osp, gen);
 388                 RETURN(-EINVAL);
 389         }
 390
 391         obd_str2uuid(&uuid,  osp);
 392
 393         lod_getref(ltd);
 394         mutex_lock(&ltd->ltd_mutex);
 395         /* check that the index is allocated in the bitmap */
 396         if (!cfs_bitmap_check(ltd->ltd_tgt_bitmap, idx) ||
 397             !LTD_TGT(ltd, idx)) {
 398                 CERROR("%s: device %d is not set up\n", obd->obd_name, idx);
 399                 GOTO(out, rc = -EINVAL);
 400         }
 401
 402         /* check that the UUID matches */
 403         if (!obd_uuid_equals(&uuid, &LTD_TGT(ltd, idx)->ltd_uuid)) {
 404                 CERROR("%s: LOD target UUID %s at index %d does not match %s\n",
 405                        obd->obd_name, obd_uuid2str(&LTD_TGT(ltd,idx)->ltd_uuid),
 406                        idx, osp);
 407                 GOTO(out, rc = -EINVAL);
 408         }
 409
 410         __lod_del_device(ltd, idx);
 411         EXIT;
 412 out:
 413         mutex_unlock(&ltd->ltd_mutex);
 414         lod_putref(lod, ltd);
 415         return(rc);
 416 }
 417
 418 int lod_ea_store_resize(struct lod_thread_info *info, int size)
 419 {
 420         int round = size_roundup_power2(size);
 421
 422         LASSERT(round <= lov_mds_md_size(LOV_MAX_STRIPE_COUNT, LOV_MAGIC_V3));
 423         if (info->lti_ea_store) {
 424                 LASSERT(info->lti_ea_store_size);
 425                 LASSERT(info->lti_ea_store_size < round);
 426                 CDEBUG(D_INFO, "EA store size %d is not enough, need %d\n",
 427                        info->lti_ea_store_size, round);
 428                 OBD_FREE_LARGE(info->lti_ea_store, info->lti_ea_store_size);
 429                 info->lti_ea_store = NULL;
 430                 info->lti_ea_store_size = 0;
 431         }
 432
 433         OBD_ALLOC_LARGE(info->lti_ea_store, round);
 434         if (info->lti_ea_store == NULL)
 435                 RETURN(-ENOMEM);
 436         info->lti_ea_store_size = round;
 437         RETURN(0);
 438 }
 439
 440 /*
 441  * generate and write LOV EA for given striped object
 442  */
 443 int lod_generate_and_set_lovea(const struct lu_env *env,
 444                                struct lod_object *lo, struct thandle *th)
 445 {
 446         struct lod_thread_info  *info = lod_env_info(env);
 447         struct dt_object        *next = dt_object_child(&lo->ldo_obj);
 448         const struct lu_fid     *fid  = lu_object_fid(&lo->ldo_obj.do_lu);
 449         struct lov_mds_md_v1    *lmm;
 450         struct lov_ost_data_v1  *objs;
 451         __u32                    magic;
 452         int                      i, rc, lmm_size;
 453         ENTRY;
 454
 455         LASSERT(lo);
 456         LASSERT(lo->ldo_stripenr > 0);
 457
 458         magic = lo->ldo_pool ? LOV_MAGIC_V3 : LOV_MAGIC_V1;
 459         lmm_size = lov_mds_md_size(lo->ldo_stripenr, magic);
 460         if (info->lti_ea_store_size < lmm_size) {
 461                 rc = lod_ea_store_resize(info, lmm_size);
 462                 if (rc)
 463                         RETURN(rc);
 464         }
 465
 466         lmm = info->lti_ea_store;
 467
 468         lmm->lmm_magic = cpu_to_le32(magic);
 469         lmm->lmm_pattern = cpu_to_le32(LOV_PATTERN_RAID0);
 470         lmm->lmm_object_id = cpu_to_le64(fid_ver_oid(fid));
 471         lmm->lmm_object_seq = cpu_to_le64(fid_seq(fid));
 472         lmm->lmm_stripe_size = cpu_to_le32(lo->ldo_stripe_size);
 473         lmm->lmm_stripe_count = cpu_to_le16(lo->ldo_stripenr);
 474         lmm->lmm_layout_gen = 0;
 475         if (magic == LOV_MAGIC_V1) {
 476                 objs = &lmm->lmm_objects[0];
 477         } else {
 478                 struct lov_mds_md_v3 *v3 = (struct lov_mds_md_v3 *) lmm;
 479                 strncpy(v3->lmm_pool_name, lo->ldo_pool, LOV_MAXPOOLNAME);
 480                 objs = &v3->lmm_objects[0];
 481         }
 482
 483         for (i = 0; i < lo->ldo_stripenr; i++) {
 484                 const struct lu_fid     *fid;
 485                 struct lod_device       *lod;
 486                 __u32                   index;
 487
 488                 lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
 489                 LASSERT(lo->ldo_stripe[i]);
 490                 fid = lu_object_fid(&lo->ldo_stripe[i]->do_lu);
 491
 492                 rc = fid_ostid_pack(fid, &info->lti_ostid);
 493                 LASSERT(rc == 0);
 494
 495                 objs[i].l_object_id  = cpu_to_le64(info->lti_ostid.oi_id);
 496                 objs[i].l_object_seq = cpu_to_le64(info->lti_ostid.oi_seq);
 497                 objs[i].l_ost_gen    = cpu_to_le32(0);
 498                 rc = lod_fld_lookup(env, lod, fid, &index, LU_SEQ_RANGE_OST);
 499                 if (rc < 0) {
 500                         CERROR("%s: Can not locate "DFID": rc = %d\n",
 501                                lod2obd(lod)->obd_name, PFID(fid), rc);
 502                         RETURN(rc);
 503                 }
 504                 objs[i].l_ost_idx = cpu_to_le32(index);
 505         }
 506
 507         info->lti_buf.lb_buf = lmm;
 508         info->lti_buf.lb_len = lmm_size;
 509         rc = dt_xattr_set(env, next, &info->lti_buf, XATTR_NAME_LOV, 0,
 510                           th, BYPASS_CAPA);
 511
 512         RETURN(rc);
 513 }
 514
 515 int lod_get_lov_ea(const struct lu_env *env, struct lod_object *lo)
 516 {
 517         struct lod_thread_info *info = lod_env_info(env);
 518         struct dt_object       *next = dt_object_child(&lo->ldo_obj);
 519         int                     rc;
 520         ENTRY;
 521
 522         LASSERT(info);
 523
 524         if (unlikely(info->lti_ea_store_size == 0)) {
 525                 /* just to enter in allocation block below */
 526                 rc = -ERANGE;
 527         } else {
 528 repeat:
 529                 info->lti_buf.lb_buf = info->lti_ea_store;
 530                 info->lti_buf.lb_len = info->lti_ea_store_size;
 531                 rc = dt_xattr_get(env, next, &info->lti_buf, XATTR_NAME_LOV,
 532                                   BYPASS_CAPA);
 533         }
 534         /* if object is not striped or inaccessible */
 535         if (rc == -ENODATA)
 536                 RETURN(0);
 537
 538         if (rc == -ERANGE) {
 539                 /* EA doesn't fit, reallocate new buffer */
 540                 rc = dt_xattr_get(env, next, &LU_BUF_NULL, XATTR_NAME_LOV,
 541                                   BYPASS_CAPA);
 542                 if (rc == -ENODATA)
 543                         RETURN(0);
 544                 else if (rc < 0)
 545                         RETURN(rc);
 546
 547                 LASSERT(rc > 0);
 548                 rc = lod_ea_store_resize(info, rc);
 549                 if (rc)
 550                         RETURN(rc);
 551                 goto repeat;
 552         }
 553
 554         RETURN(rc);
 555 }
 556
 557 int lod_store_def_striping(const struct lu_env *env, struct dt_object *dt,
 558                            struct thandle *th)
 559 {
 560         struct lod_thread_info  *info = lod_env_info(env);
 561         struct lod_object       *lo = lod_dt_obj(dt);
 562         struct dt_object        *next = dt_object_child(dt);
 563         struct lov_user_md_v3   *v3;
 564         int                      rc;
 565         ENTRY;
 566
 567         LASSERT(S_ISDIR(dt->do_lu.lo_header->loh_attr));
 568
 569         /*
 570          * store striping defaults into new directory
 571          * used to implement defaults inheritance
 572          */
 573
 574         /* probably nothing to inherite */
 575         if (lo->ldo_striping_cached == 0)
 576                 RETURN(0);
 577
 578         if (LOVEA_DELETE_VALUES(lo->ldo_def_stripe_size, lo->ldo_def_stripenr,
 579                                 lo->ldo_def_stripe_offset))
 580                 RETURN(0);
 581
 582         /* XXX: use thread info */
 583         OBD_ALLOC_PTR(v3);
 584         if (v3 == NULL)
 585                 RETURN(-ENOMEM);
 586
 587         v3->lmm_magic = cpu_to_le32(LOV_MAGIC_V3);
 588         v3->lmm_pattern = cpu_to_le32(LOV_PATTERN_RAID0);
 589         v3->lmm_object_id = 0;
 590         v3->lmm_object_seq = 0;
 591         v3->lmm_stripe_size = cpu_to_le32(lo->ldo_def_stripe_size);
 592         v3->lmm_stripe_count = cpu_to_le16(lo->ldo_def_stripenr);
 593         v3->lmm_stripe_offset = cpu_to_le16(lo->ldo_def_stripe_offset);
 594         if (lo->ldo_pool)
 595                 strncpy(v3->lmm_pool_name, lo->ldo_pool, LOV_MAXPOOLNAME);
 596
 597         info->lti_buf.lb_buf = v3;
 598         info->lti_buf.lb_len = sizeof(*v3);
 599         rc = dt_xattr_set(env, next, &info->lti_buf, XATTR_NAME_LOV, 0, th,
 600                         BYPASS_CAPA);
 601
 602         OBD_FREE_PTR(v3);
 603
 604         RETURN(rc);
 605 }
 606
 607 /*
 608  * allocate array of objects pointers, find/create objects
 609  * stripenr and other fields should be initialized by this moment
 610  */
 611 int lod_initialize_objects(const struct lu_env *env, struct lod_object *lo,
 612                            struct lov_ost_data_v1 *objs)
 613 {
 614         struct lod_thread_info  *info = lod_env_info(env);
 615         struct lod_device       *md = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
 616         struct lu_object        *o, *n;
 617         struct lu_device        *nd;
 618         int                     i, idx, rc = 0;
 619         ENTRY;
 620
 621         LASSERT(lo);
 622         LASSERT(lo->ldo_stripe == NULL);
 623         LASSERT(lo->ldo_stripenr > 0);
 624         LASSERT(lo->ldo_stripe_size > 0);
 625
 626         i = sizeof(struct dt_object *) * lo->ldo_stripenr;
 627         OBD_ALLOC(lo->ldo_stripe, i);
 628         if (lo->ldo_stripe == NULL)
 629                 GOTO(out, rc = -ENOMEM);
 630         lo->ldo_stripes_allocated = lo->ldo_stripenr;
 631
 632         for (i = 0; i < lo->ldo_stripenr; i++) {
 633                 info->lti_ostid.oi_id = le64_to_cpu(objs[i].l_object_id);
 634                 info->lti_ostid.oi_seq = le64_to_cpu(objs[i].l_object_seq);
 635                 idx = le64_to_cpu(objs[i].l_ost_idx);
 636                 fid_ostid_unpack(&info->lti_fid, &info->lti_ostid, idx);
 637                 LASSERTF(fid_is_sane(&info->lti_fid), ""DFID" insane!\n",
 638                          PFID(&info->lti_fid));
 639                 /*
 640                  * XXX: assertion is left for testing, to make
 641                  * sure we never process requests till configuration
 642                  * is completed. to be changed to -EINVAL
 643                  */
 644
 645                 lod_getref(&md->lod_ost_descs);
 646                 LASSERT(cfs_bitmap_check(md->lod_ost_bitmap, idx));
 647                 LASSERT(OST_TGT(md,idx));
 648                 LASSERTF(OST_TGT(md,idx)->ltd_ost, "idx %d\n", idx);
 649                 nd = &OST_TGT(md,idx)->ltd_ost->dd_lu_dev;
 650                 lod_putref(md, &md->lod_ost_descs);
 651
 652                 o = lu_object_find_at(env, nd, &info->lti_fid, NULL);
 653                 if (IS_ERR(o))
 654                         GOTO(out, rc = PTR_ERR(o));
 655
 656                 n = lu_object_locate(o->lo_header, nd->ld_type);
 657                 LASSERT(n);
 658
 659                 lo->ldo_stripe[i] = container_of(n, struct dt_object, do_lu);
 660         }
 661
 662 out:
 663         RETURN(rc);
 664 }
 665
 666 /*
 667  * Parse striping information stored in lti_ea_store
 668  */
 669 int lod_parse_striping(const struct lu_env *env, struct lod_object *lo,
 670                        const struct lu_buf *buf)
 671 {
 672         struct lov_mds_md_v1    *lmm;
 673         struct lov_ost_data_v1  *objs;
 674         __u32                    magic;
 675         int                      rc = 0;
 676         ENTRY;
 677
 678         LASSERT(buf);
 679         LASSERT(buf->lb_buf);
 680         LASSERT(buf->lb_len);
 681
 682         lmm = (struct lov_mds_md_v1 *) buf->lb_buf;
 683         magic = le32_to_cpu(lmm->lmm_magic);
 684
 685         if (magic != LOV_MAGIC_V1 && magic != LOV_MAGIC_V3)
 686                 GOTO(out, rc = -EINVAL);
 687         if (le32_to_cpu(lmm->lmm_pattern) != LOV_PATTERN_RAID0)
 688                 GOTO(out, rc = -EINVAL);
 689
 690         lo->ldo_stripe_size = le32_to_cpu(lmm->lmm_stripe_size);
 691         lo->ldo_stripenr = le16_to_cpu(lmm->lmm_stripe_count);
 692         lo->ldo_layout_gen = le16_to_cpu(lmm->lmm_layout_gen);
 693
 694         LASSERT(buf->lb_len >= lov_mds_md_size(lo->ldo_stripenr, magic));
 695
 696         if (magic == LOV_MAGIC_V3) {
 697                 struct lov_mds_md_v3 *v3 = (struct lov_mds_md_v3 *) lmm;
 698                 objs = &v3->lmm_objects[0];
 699                 lod_object_set_pool(lo, v3->lmm_pool_name);
 700         } else {
 701                 objs = &lmm->lmm_objects[0];
 702         }
 703
 704         rc = lod_initialize_objects(env, lo, objs);
 705
 706 out:
 707         RETURN(rc);
 708 }
 709
 710 /*
 711  * Load and parse striping information, create in-core representation for the
 712  * stripes
 713  */
 714 int lod_load_striping(const struct lu_env *env, struct lod_object *lo)
 715 {
 716         struct lod_thread_info  *info = lod_env_info(env);
 717         struct dt_object        *next = dt_object_child(&lo->ldo_obj);
 718         int                      rc;
 719         ENTRY;
 720
 721         /*
 722          * currently this code is supposed to be called from declaration
 723          * phase only, thus the object is not expected to be locked by caller
 724          */
 725         dt_write_lock(env, next, 0);
 726         /* already initialized? */
 727         if (lo->ldo_stripe) {
 728                 int i;
 729                 /* check validity */
 730                 for (i = 0; i < lo->ldo_stripenr; i++)
 731                         LASSERTF(lo->ldo_stripe[i], "stripe %d is NULL\n", i);
 732                 GOTO(out, rc = 0);
 733         }
 734
 735         if (!dt_object_exists(next))
 736                 GOTO(out, rc = 0);
 737
 738         /* only regular files can be striped */
 739         if (!(lu_object_attr(lod2lu_obj(lo)) & S_IFREG))
 740                 GOTO(out, rc = 0);
 741
 742         LASSERT(lo->ldo_stripenr == 0);
 743
 744         rc = lod_get_lov_ea(env, lo);
 745         if (rc <= 0)
 746                 GOTO(out, rc);
 747
 748         /*
 749          * there is LOV EA (striping information) in this object
 750          * let's parse it and create in-core objects for the stripes
 751          */
 752         info->lti_buf.lb_buf = info->lti_ea_store;
 753         info->lti_buf.lb_len = info->lti_ea_store_size;
 754         rc = lod_parse_striping(env, lo, &info->lti_buf);
 755 out:
 756         dt_write_unlock(env, next);
 757         RETURN(rc);
 758 }
 759
 760 int lod_verify_striping(struct lod_device *d, const struct lu_buf *buf,
 761                         int specific)
 762 {
 763         struct lov_user_md_v1   *lum;
 764         struct lov_user_md_v3   *v3 = NULL;
 765         struct pool_desc        *pool = NULL;
 766         int                      rc;
 767         ENTRY;
 768
 769         lum = buf->lb_buf;
 770
 771         if (lum->lmm_magic != LOV_USER_MAGIC_V1 &&
 772             lum->lmm_magic != LOV_USER_MAGIC_V3 &&
 773             lum->lmm_magic != LOV_MAGIC_V1_DEF &&
 774             lum->lmm_magic != LOV_MAGIC_V3_DEF) {
 775                 CDEBUG(D_IOCTL, "bad userland LOV MAGIC: %#x\n",
 776                        lum->lmm_magic);
 777                 RETURN(-EINVAL);
 778         }
 779
 780         if ((specific && lum->lmm_pattern != LOV_PATTERN_RAID0) ||
 781             (specific == 0 && lum->lmm_pattern != 0)) {
 782                 CDEBUG(D_IOCTL, "bad userland stripe pattern: %#x\n",
 783                        lum->lmm_pattern);
 784                 RETURN(-EINVAL);
 785         }
 786
 787         /* 64kB is the largest common page size we see (ia64), and matches the
 788          * check in lfs */
 789         if (lum->lmm_stripe_size & (LOV_MIN_STRIPE_SIZE - 1)) {
 790                 CDEBUG(D_IOCTL, "stripe size %u not multiple of %u, fixing\n",
 791                        lum->lmm_stripe_size, LOV_MIN_STRIPE_SIZE);
 792                 RETURN(-EINVAL);
 793         }
 794
 795         /* an offset of -1 is treated as a "special" valid offset */
 796         if (lum->lmm_stripe_offset != (typeof(lum->lmm_stripe_offset))(-1)) {
 797                 /* if offset is not within valid range [0, osts_size) */
 798                 if (lum->lmm_stripe_offset >= d->lod_osts_size) {
 799                         CDEBUG(D_IOCTL, "stripe offset %u >= bitmap size %u\n",
 800                                lum->lmm_stripe_offset, d->lod_osts_size);
 801                         RETURN(-EINVAL);
 802                 }
 803
 804                 /* if lmm_stripe_offset is *not* in bitmap */
 805                 if (!cfs_bitmap_check(d->lod_ost_bitmap,
 806                                       lum->lmm_stripe_offset)) {
 807                         CDEBUG(D_IOCTL, "stripe offset %u not in bitmap\n",
 808                                lum->lmm_stripe_offset);
 809                         RETURN(-EINVAL);
 810                 }
 811         }
 812
 813         if (lum->lmm_magic == LOV_USER_MAGIC_V3)
 814                 v3 = buf->lb_buf;
 815
 816         if (v3)
 817                 pool = lod_find_pool(d, v3->lmm_pool_name);
 818
 819         if (pool != NULL) {
 820                 __u16 offs = v3->lmm_stripe_offset;
 821
 822                 if (offs != (typeof(v3->lmm_stripe_offset))(-1)) {
 823                         rc = lod_check_index_in_pool(offs, pool);
 824                         if (rc < 0) {
 825                                 lod_pool_putref(pool);
 826                                 RETURN(-EINVAL);
 827                         }
 828                 }
 829
 830                 if (specific && lum->lmm_stripe_count > pool_tgt_count(pool)) {
 831                         CDEBUG(D_IOCTL,
 832                                "stripe count %u > # OSTs %u in the pool\n",
 833                                lum->lmm_stripe_count, pool_tgt_count(pool));
 834                         lod_pool_putref(pool);
 835                         RETURN(-EINVAL);
 836                 }
 837
 838                 lod_pool_putref(pool);
 839         }
 840
 841         RETURN(0);
 842 }
 843
 844 void lod_fix_desc_stripe_size(__u64 *val)
 845 {
 846         if (*val < PTLRPC_MAX_BRW_SIZE) {
 847                 LCONSOLE_WARN("Increasing default stripe size to min %u\n",
 848                               PTLRPC_MAX_BRW_SIZE);
 849                 *val = PTLRPC_MAX_BRW_SIZE;
 850         } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) {
 851                 *val &= ~(LOV_MIN_STRIPE_SIZE - 1);
 852                 LCONSOLE_WARN("Changing default stripe size to "LPU64" (a "
 853                               "multiple of %u)\n",
 854                               *val, LOV_MIN_STRIPE_SIZE);
 855         }
 856 }
 857
 858 void lod_fix_desc_stripe_count(__u32 *val)
 859 {
 860         if (*val == 0)
 861                 *val = 1;
 862 }
 863
 864 void lod_fix_desc_pattern(__u32 *val)
 865 {
 866         /* from lov_setstripe */
 867         if ((*val != 0) && (*val != LOV_PATTERN_RAID0)) {
 868                 LCONSOLE_WARN("Unknown stripe pattern: %#x\n", *val);
 869                 *val = 0;
 870         }
 871 }
 872
 873 void lod_fix_desc_qos_maxage(__u32 *val)
 874 {
 875         /* fix qos_maxage */
 876         if (*val == 0)
 877                 *val = QOS_DEFAULT_MAXAGE;
 878 }
 879
 880 void lod_fix_desc(struct lov_desc *desc)
 881 {
 882         lod_fix_desc_stripe_size(&desc->ld_default_stripe_size);
 883         lod_fix_desc_stripe_count(&desc->ld_default_stripe_count);
 884         lod_fix_desc_pattern(&desc->ld_pattern);
 885         lod_fix_desc_qos_maxage(&desc->ld_qos_maxage);
 886 }
 887
 888 int lod_pools_init(struct lod_device *lod, struct lustre_cfg *lcfg)
 889 {
 890         struct obd_device          *obd;
 891         struct lov_desc            *desc;
 892         int                         rc;
 893         ENTRY;
 894
 895         obd = class_name2obd(lustre_cfg_string(lcfg, 0));
 896         LASSERT(obd != NULL);
 897         obd->obd_lu_dev = &lod->lod_dt_dev.dd_lu_dev;
 898
 899         if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
 900                 CERROR("LOD setup requires a descriptor\n");
 901                 RETURN(-EINVAL);
 902         }
 903
 904         desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
 905
 906         if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
 907                 CERROR("descriptor size wrong: %d > %d\n",
 908                        (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
 909                 RETURN(-EINVAL);
 910         }
 911
 912         if (desc->ld_magic != LOV_DESC_MAGIC) {
 913                 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
 914                         CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
 915                                obd->obd_name, desc);
 916                         lustre_swab_lov_desc(desc);
 917                 } else {
 918                         CERROR("%s: Bad lov desc magic: %#x\n",
 919                                obd->obd_name, desc->ld_magic);
 920                         RETURN(-EINVAL);
 921                 }
 922         }
 923
 924         lod_fix_desc(desc);
 925
 926         desc->ld_active_tgt_count = 0;
 927         lod->lod_desc = *desc;
 928
 929         lod->lod_sp_me = LUSTRE_SP_CLI;
 930
 931         /* Set up allocation policy (QoS and RR) */
 932         CFS_INIT_LIST_HEAD(&lod->lod_qos.lq_oss_list);
 933         init_rwsem(&lod->lod_qos.lq_rw_sem);
 934         lod->lod_qos.lq_dirty = 1;
 935         lod->lod_qos.lq_rr.lqr_dirty = 1;
 936         lod->lod_qos.lq_reset = 1;
 937         /* Default priority is toward free space balance */
 938         lod->lod_qos.lq_prio_free = 232;
 939         /* Default threshold for rr (roughly 17%) */
 940         lod->lod_qos.lq_threshold_rr = 43;
 941         /* Init statfs fields */
 942         OBD_ALLOC_PTR(lod->lod_qos.lq_statfs_data);
 943         if (NULL == lod->lod_qos.lq_statfs_data)
 944                 RETURN(-ENOMEM);
 945         cfs_waitq_init(&lod->lod_qos.lq_statfs_waitq);
 946
 947         /* Set up OST pool environment */
 948         lod->lod_pools_hash_body = cfs_hash_create("POOLS", HASH_POOLS_CUR_BITS,
 949                                                    HASH_POOLS_MAX_BITS,
 950                                                    HASH_POOLS_BKT_BITS, 0,
 951                                                    CFS_HASH_MIN_THETA,
 952                                                    CFS_HASH_MAX_THETA,
 953                                                    &pool_hash_operations,
 954                                                    CFS_HASH_DEFAULT);
 955         if (!lod->lod_pools_hash_body)
 956                 GOTO(out_statfs, rc = -ENOMEM);
 957         CFS_INIT_LIST_HEAD(&lod->lod_pool_list);
 958         lod->lod_pool_count = 0;
 959         rc = lod_ost_pool_init(&lod->lod_pool_info, 0);
 960         if (rc)
 961                 GOTO(out_hash, rc);
 962         rc = lod_ost_pool_init(&lod->lod_qos.lq_rr.lqr_pool, 0);
 963         if (rc)
 964                 GOTO(out_pool_info, rc);
 965
 966         RETURN(0);
 967
 968 out_pool_info:
 969         lod_ost_pool_free(&lod->lod_pool_info);
 970 out_hash:
 971         cfs_hash_putref(lod->lod_pools_hash_body);
 972 out_statfs:
 973         OBD_FREE_PTR(lod->lod_qos.lq_statfs_data);
 974         return rc;
 975 }
 976
 977 int lod_pools_fini(struct lod_device *lod)
 978 {
 979         struct obd_device   *obd = lod2obd(lod);
 980         cfs_list_t          *pos, *tmp;
 981         struct pool_desc    *pool;
 982         ENTRY;
 983
 984         cfs_list_for_each_safe(pos, tmp, &lod->lod_pool_list) {
 985                 pool = cfs_list_entry(pos, struct pool_desc, pool_list);
 986                 /* free pool structs */
 987                 CDEBUG(D_INFO, "delete pool %p\n", pool);
 988                 lod_pool_del(obd, pool->pool_name);
 989         }
 990
 991         cfs_hash_putref(lod->lod_pools_hash_body);
 992         lod_ost_pool_free(&(lod->lod_qos.lq_rr.lqr_pool));
 993         lod_ost_pool_free(&lod->lod_pool_info);
 994         OBD_FREE_PTR(lod->lod_qos.lq_statfs_data);
 995         RETURN(0);
 996 }
 997