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  * \param tgt_index - is the group of the OSP.
 175  * \param type - is the type of device (mdc or osc)
 176  */
 177 int lod_add_device(const struct lu_env *env, struct lod_device *lod,
 178                    char *osp, unsigned index, unsigned gen, int tgt_index,
 179                    char *type, int active)
 180 {
 181         struct obd_connect_data *data = NULL;
 182         struct obd_export       *exp = NULL;
 183         struct obd_device       *obd;
 184         struct lu_device        *ldev;
 185         struct dt_device        *d;
 186         int                      rc;
 187         struct lod_tgt_desc     *tgt_desc;
 188         struct lod_tgt_descs    *ltd;
 189         struct obd_uuid         obd_uuid;
 190         ENTRY;
 191
 192         CDEBUG(D_CONFIG, "osp:%s idx:%d gen:%d\n", osp, index, gen);
 193
 194         if (gen <= 0) {
 195                 CERROR("request to add OBD %s with invalid generation: %d\n",
 196                        osp, gen);
 197                 RETURN(-EINVAL);
 198         }
 199
 200         obd_str2uuid(&obd_uuid, osp);
 201
 202         obd = class_find_client_obd(&obd_uuid, LUSTRE_OSP_NAME,
 203                                 &lod->lod_dt_dev.dd_lu_dev.ld_obd->obd_uuid);
 204         if (obd == NULL) {
 205                 CERROR("can't find %s device\n", osp);
 206                 RETURN(-EINVAL);
 207         }
 208
 209         OBD_ALLOC_PTR(data);
 210         if (data == NULL)
 211                 RETURN(-ENOMEM);
 212
 213         data->ocd_connect_flags = OBD_CONNECT_INDEX | OBD_CONNECT_VERSION;
 214         data->ocd_version = LUSTRE_VERSION_CODE;
 215         data->ocd_index = index;
 216
 217         if (strcmp(LUSTRE_OSC_NAME, type) == 0) {
 218                 data->ocd_connect_flags |= OBD_CONNECT_AT |
 219                                            OBD_CONNECT_FULL20 |
 220                                            OBD_CONNECT_INDEX |
 221 #ifdef HAVE_LRU_RESIZE_SUPPORT
 222                                            OBD_CONNECT_LRU_RESIZE |
 223 #endif
 224                                            OBD_CONNECT_MDS |
 225                                            OBD_CONNECT_OSS_CAPA |
 226                                            OBD_CONNECT_REQPORTAL |
 227                                            OBD_CONNECT_SKIP_ORPHAN |
 228                                            OBD_CONNECT_FID |
 229                                            OBD_CONNECT_LVB_TYPE |
 230                                            OBD_CONNECT_VERSION |
 231                                            OBD_CONNECT_PINGLESS;
 232
 233                 data->ocd_group = tgt_index;
 234                 ltd = &lod->lod_ost_descs;
 235         } else {
 236                 struct obd_import *imp = obd->u.cli.cl_import;
 237
 238                 data->ocd_ibits_known = MDS_INODELOCK_UPDATE;
 239                 data->ocd_connect_flags |= OBD_CONNECT_ACL |
 240                                            OBD_CONNECT_MDS_CAPA |
 241                                            OBD_CONNECT_OSS_CAPA |
 242                                            OBD_CONNECT_IBITS |
 243                                            OBD_CONNECT_MDS_MDS |
 244                                            OBD_CONNECT_FID |
 245                                            OBD_CONNECT_AT |
 246                                            OBD_CONNECT_FULL20 |
 247                                            OBD_CONNECT_PINGLESS;
 248                 /* XXX set MDS-MDS flags, remove this when running this
 249                  * on client*/
 250                 data->ocd_connect_flags |= OBD_CONNECT_MDS_MDS;
 251                 spin_lock(&imp->imp_lock);
 252                 imp->imp_server_timeout = 1;
 253                 spin_unlock(&imp->imp_lock);
 254                 imp->imp_client->cli_request_portal = MDS_MDS_PORTAL;
 255                 CDEBUG(D_OTHER, "%s: Set 'mds' portal and timeout\n",
 256                       obd->obd_name);
 257                 ltd = &lod->lod_mdt_descs;
 258         }
 259
 260         rc = obd_connect(env, &exp, obd, &obd->obd_uuid, data, NULL);
 261         OBD_FREE_PTR(data);
 262         if (rc) {
 263                 CERROR("%s: cannot connect to next dev %s (%d)\n",
 264                        obd->obd_name, osp, rc);
 265                 GOTO(out_free, rc);
 266         }
 267
 268         LASSERT(obd->obd_lu_dev);
 269         LASSERT(obd->obd_lu_dev->ld_site = lod->lod_dt_dev.dd_lu_dev.ld_site);
 270
 271         ldev = obd->obd_lu_dev;
 272         d = lu2dt_dev(ldev);
 273
 274         /* Allocate ost descriptor and fill it */
 275         OBD_ALLOC_PTR(tgt_desc);
 276         if (!tgt_desc)
 277                 GOTO(out_conn, rc = -ENOMEM);
 278
 279         tgt_desc->ltd_tgt    = d;
 280         tgt_desc->ltd_exp    = exp;
 281         tgt_desc->ltd_uuid   = obd->u.cli.cl_target_uuid;
 282         tgt_desc->ltd_gen    = gen;
 283         tgt_desc->ltd_index  = index;
 284         tgt_desc->ltd_active = active;
 285
 286         lod_getref(ltd);
 287         if (index >= ltd->ltd_tgts_size) {
 288                 /* we have to increase the size of the lod_osts array */
 289                 __u32  newsize;
 290
 291                 newsize = max(ltd->ltd_tgts_size, (__u32)2);
 292                 while (newsize < index + 1)
 293                         newsize = newsize << 1;
 294
 295                 /* lod_bitmap_resize() needs lod_rw_sem
 296                  * which we hold with th reference */
 297                 lod_putref(lod, ltd);
 298
 299                 rc = ltd_bitmap_resize(ltd, newsize);
 300                 if (rc)
 301                         GOTO(out_desc, rc);
 302
 303                 lod_getref(ltd);
 304         }
 305
 306         mutex_lock(&ltd->ltd_mutex);
 307         if (cfs_bitmap_check(ltd->ltd_tgt_bitmap, index)) {
 308                 CERROR("%s: device %d is registered already\n", obd->obd_name,
 309                        index);
 310                 GOTO(out_mutex, rc = -EEXIST);
 311         }
 312
 313         if (ltd->ltd_tgt_idx[index / TGT_PTRS_PER_BLOCK] == NULL) {
 314                 OBD_ALLOC_PTR(ltd->ltd_tgt_idx[index / TGT_PTRS_PER_BLOCK]);
 315                 if (ltd->ltd_tgt_idx[index / TGT_PTRS_PER_BLOCK] == NULL) {
 316                         CERROR("can't allocate index to add %s\n",
 317                                obd->obd_name);
 318                         GOTO(out_mutex, rc = -ENOMEM);
 319                 }
 320         }
 321
 322         if (!strcmp(LUSTRE_OSC_NAME, type)) {
 323                 /* pool and qos are not supported for MDS stack yet */
 324                 rc = lod_ost_pool_add(&lod->lod_pool_info, index,
 325                                       lod->lod_osts_size);
 326                 if (rc) {
 327                         CERROR("%s: can't set up pool, failed with %d\n",
 328                                obd->obd_name, rc);
 329                         GOTO(out_mutex, rc);
 330                 }
 331
 332                 rc = qos_add_tgt(lod, tgt_desc);
 333                 if (rc) {
 334                         CERROR("%s: qos_add_tgt failed with %d\n",
 335                                 obd->obd_name, rc);
 336                         GOTO(out_pool, rc);
 337                 }
 338
 339                 /* The new OST is now a full citizen */
 340                 if (index >= lod->lod_desc.ld_tgt_count)
 341                         lod->lod_desc.ld_tgt_count = index + 1;
 342                 if (active)
 343                         lod->lod_desc.ld_active_tgt_count++;
 344         }
 345
 346         LTD_TGT(ltd, index) = tgt_desc;
 347         cfs_bitmap_set(ltd->ltd_tgt_bitmap, index);
 348         ltd->ltd_tgtnr++;
 349         mutex_unlock(&ltd->ltd_mutex);
 350         lod_putref(lod, ltd);
 351         if (lod->lod_recovery_completed)
 352                 ldev->ld_ops->ldo_recovery_complete(env, ldev);
 353
 354         RETURN(0);
 355
 356 out_pool:
 357         lod_ost_pool_remove(&lod->lod_pool_info, index);
 358 out_mutex:
 359         mutex_unlock(&ltd->ltd_mutex);
 360         lod_putref(lod, ltd);
 361 out_desc:
 362         OBD_FREE_PTR(tgt_desc);
 363 out_conn:
 364         obd_disconnect(exp);
 365 out_free:
 366         return rc;
 367 }
 368
 369 /*
 370  * helper function to schedule OST removal from the device table
 371  */
 372 static void __lod_del_device(struct lod_tgt_descs *ltd,
 373                              unsigned idx)
 374 {
 375         LASSERT(LTD_TGT(ltd, idx));
 376         if (LTD_TGT(ltd, idx)->ltd_reap == 0) {
 377                 LTD_TGT(ltd, idx)->ltd_reap = 1;
 378                 ltd->ltd_death_row++;
 379         }
 380 }
 381
 382 int lod_fini_tgt(struct lod_device *lod, struct lod_tgt_descs *ltd)
 383 {
 384         int idx;
 385
 386         if (ltd->ltd_tgts_size <= 0)
 387                 return 0;
 388         lod_getref(ltd);
 389         mutex_lock(&ltd->ltd_mutex);
 390         cfs_foreach_bit(ltd->ltd_tgt_bitmap, idx)
 391                 __lod_del_device(ltd, idx);
 392         mutex_unlock(&ltd->ltd_mutex);
 393         lod_putref(lod, ltd);
 394         CFS_FREE_BITMAP(ltd->ltd_tgt_bitmap);
 395         for (idx = 0; idx < TGT_PTRS; idx++) {
 396                 if (ltd->ltd_tgt_idx[idx])
 397                         OBD_FREE_PTR(ltd->ltd_tgt_idx[idx]);
 398         }
 399         ltd->ltd_tgts_size = 0;
 400         return 0;
 401 }
 402
 403 /*
 404  * Add support for administratively disabled OST (through the MGS).
 405  * Schedule a target for deletion.  Disconnection and real removal from the
 406  * table takes place in lod_putref() once the last table user release its
 407  * reference.
 408  *
 409  * \param env - is the environment passed by the caller
 410  * \param lod - is the lod device currently connected to the OSP about to be
 411  *              removed
 412  * \param osp - is the name of OSP device about to be removed
 413  * \param idx - is the OSP index
 414  * \param gen - is the generation number, not used currently
 415  */
 416 int lod_del_device(const struct lu_env *env, struct lod_device *lod,
 417                    struct lod_tgt_descs *ltd, char *osp, unsigned idx,
 418                    unsigned gen)
 419 {
 420         struct obd_device *obd;
 421         int                rc = 0;
 422         struct obd_uuid    uuid;
 423         ENTRY;
 424
 425         CDEBUG(D_CONFIG, "osp:%s idx:%d gen:%d\n", osp, idx, gen);
 426
 427         obd_str2uuid(&uuid, osp);
 428
 429         obd = class_find_client_obd(&uuid, LUSTRE_OSP_NAME,
 430                                    &lod->lod_dt_dev.dd_lu_dev.ld_obd->obd_uuid);
 431         if (obd == NULL) {
 432                 CERROR("can't find %s device\n", osp);
 433                 RETURN(-EINVAL);
 434         }
 435
 436         if (gen <= 0) {
 437                 CERROR("%s: request to remove OBD %s with invalid generation %d"
 438                        "\n", obd->obd_name, osp, gen);
 439                 RETURN(-EINVAL);
 440         }
 441
 442         obd_str2uuid(&uuid,  osp);
 443
 444         lod_getref(ltd);
 445         mutex_lock(&ltd->ltd_mutex);
 446         /* check that the index is allocated in the bitmap */
 447         if (!cfs_bitmap_check(ltd->ltd_tgt_bitmap, idx) ||
 448             !LTD_TGT(ltd, idx)) {
 449                 CERROR("%s: device %d is not set up\n", obd->obd_name, idx);
 450                 GOTO(out, rc = -EINVAL);
 451         }
 452
 453         /* check that the UUID matches */
 454         if (!obd_uuid_equals(&uuid, &LTD_TGT(ltd, idx)->ltd_uuid)) {
 455                 CERROR("%s: LOD target UUID %s at index %d does not match %s\n",
 456                        obd->obd_name, obd_uuid2str(&LTD_TGT(ltd,idx)->ltd_uuid),
 457                        idx, osp);
 458                 GOTO(out, rc = -EINVAL);
 459         }
 460
 461         __lod_del_device(ltd, idx);
 462         EXIT;
 463 out:
 464         mutex_unlock(&ltd->ltd_mutex);
 465         lod_putref(lod, ltd);
 466         return(rc);
 467 }
 468
 469 int lod_ea_store_resize(struct lod_thread_info *info, int size)
 470 {
 471         int round = size_roundup_power2(size);
 472
 473         LASSERT(round <= lov_mds_md_size(LOV_MAX_STRIPE_COUNT, LOV_MAGIC_V3));
 474         if (info->lti_ea_store) {
 475                 LASSERT(info->lti_ea_store_size);
 476                 LASSERT(info->lti_ea_store_size < round);
 477                 CDEBUG(D_INFO, "EA store size %d is not enough, need %d\n",
 478                        info->lti_ea_store_size, round);
 479                 OBD_FREE_LARGE(info->lti_ea_store, info->lti_ea_store_size);
 480                 info->lti_ea_store = NULL;
 481                 info->lti_ea_store_size = 0;
 482         }
 483
 484         OBD_ALLOC_LARGE(info->lti_ea_store, round);
 485         if (info->lti_ea_store == NULL)
 486                 RETURN(-ENOMEM);
 487         info->lti_ea_store_size = round;
 488         RETURN(0);
 489 }
 490
 491 /*
 492  * generate and write LOV EA for given striped object
 493  */
 494 int lod_generate_and_set_lovea(const struct lu_env *env,
 495                                struct lod_object *lo, struct thandle *th)
 496 {
 497         struct lod_thread_info  *info = lod_env_info(env);
 498         struct dt_object        *next = dt_object_child(&lo->ldo_obj);
 499         const struct lu_fid     *fid  = lu_object_fid(&lo->ldo_obj.do_lu);
 500         struct lov_mds_md_v1    *lmm;
 501         struct lov_ost_data_v1  *objs;
 502         __u32                    magic;
 503         int                      i, rc, lmm_size;
 504         ENTRY;
 505
 506         LASSERT(lo);
 507         LASSERT(lo->ldo_stripenr > 0);
 508
 509         magic = lo->ldo_pool ? LOV_MAGIC_V3 : LOV_MAGIC_V1;
 510         lmm_size = lov_mds_md_size(lo->ldo_stripenr, magic);
 511         if (info->lti_ea_store_size < lmm_size) {
 512                 rc = lod_ea_store_resize(info, lmm_size);
 513                 if (rc)
 514                         RETURN(rc);
 515         }
 516
 517         lmm = info->lti_ea_store;
 518
 519         lmm->lmm_magic = cpu_to_le32(magic);
 520         lmm->lmm_pattern = cpu_to_le32(LOV_PATTERN_RAID0);
 521         lmm->lmm_object_id = cpu_to_le64(fid_ver_oid(fid));
 522         lmm->lmm_object_seq = cpu_to_le64(fid_seq(fid));
 523         lmm->lmm_stripe_size = cpu_to_le32(lo->ldo_stripe_size);
 524         lmm->lmm_stripe_count = cpu_to_le16(lo->ldo_stripenr);
 525         lmm->lmm_layout_gen = 0;
 526         if (magic == LOV_MAGIC_V1) {
 527                 objs = &lmm->lmm_objects[0];
 528         } else {
 529                 struct lov_mds_md_v3 *v3 = (struct lov_mds_md_v3 *) lmm;
 530                 strncpy(v3->lmm_pool_name, lo->ldo_pool, LOV_MAXPOOLNAME);
 531                 objs = &v3->lmm_objects[0];
 532         }
 533
 534         for (i = 0; i < lo->ldo_stripenr; i++) {
 535                 const struct lu_fid     *fid;
 536                 struct lod_device       *lod;
 537                 __u32                   index;
 538
 539                 lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
 540                 LASSERT(lo->ldo_stripe[i]);
 541                 fid = lu_object_fid(&lo->ldo_stripe[i]->do_lu);
 542
 543                 rc = fid_ostid_pack(fid, &info->lti_ostid);
 544                 LASSERT(rc == 0);
 545
 546                 objs[i].l_object_id  = cpu_to_le64(info->lti_ostid.oi_id);
 547                 objs[i].l_object_seq = cpu_to_le64(info->lti_ostid.oi_seq);
 548                 objs[i].l_ost_gen    = cpu_to_le32(0);
 549                 rc = lod_fld_lookup(env, lod, fid, &index, LU_SEQ_RANGE_OST);
 550                 if (rc < 0) {
 551                         CERROR("%s: Can not locate "DFID": rc = %d\n",
 552                                lod2obd(lod)->obd_name, PFID(fid), rc);
 553                         RETURN(rc);
 554                 }
 555                 objs[i].l_ost_idx = cpu_to_le32(index);
 556         }
 557
 558         info->lti_buf.lb_buf = lmm;
 559         info->lti_buf.lb_len = lmm_size;
 560         rc = dt_xattr_set(env, next, &info->lti_buf, XATTR_NAME_LOV, 0,
 561                           th, BYPASS_CAPA);
 562
 563         RETURN(rc);
 564 }
 565
 566 int lod_get_lov_ea(const struct lu_env *env, struct lod_object *lo)
 567 {
 568         struct lod_thread_info *info = lod_env_info(env);
 569         struct dt_object       *next = dt_object_child(&lo->ldo_obj);
 570         int                     rc;
 571         ENTRY;
 572
 573         LASSERT(info);
 574
 575         if (unlikely(info->lti_ea_store_size == 0)) {
 576                 /* just to enter in allocation block below */
 577                 rc = -ERANGE;
 578         } else {
 579 repeat:
 580                 info->lti_buf.lb_buf = info->lti_ea_store;
 581                 info->lti_buf.lb_len = info->lti_ea_store_size;
 582                 rc = dt_xattr_get(env, next, &info->lti_buf, XATTR_NAME_LOV,
 583                                   BYPASS_CAPA);
 584         }
 585         /* if object is not striped or inaccessible */
 586         if (rc == -ENODATA)
 587                 RETURN(0);
 588
 589         if (rc == -ERANGE) {
 590                 /* EA doesn't fit, reallocate new buffer */
 591                 rc = dt_xattr_get(env, next, &LU_BUF_NULL, XATTR_NAME_LOV,
 592                                   BYPASS_CAPA);
 593                 if (rc == -ENODATA)
 594                         RETURN(0);
 595                 else if (rc < 0)
 596                         RETURN(rc);
 597
 598                 LASSERT(rc > 0);
 599                 rc = lod_ea_store_resize(info, rc);
 600                 if (rc)
 601                         RETURN(rc);
 602                 goto repeat;
 603         }
 604
 605         RETURN(rc);
 606 }
 607
 608 int lod_store_def_striping(const struct lu_env *env, struct dt_object *dt,
 609                            struct thandle *th)
 610 {
 611         struct lod_thread_info  *info = lod_env_info(env);
 612         struct lod_object       *lo = lod_dt_obj(dt);
 613         struct dt_object        *next = dt_object_child(dt);
 614         struct lov_user_md_v3   *v3;
 615         int                      rc;
 616         ENTRY;
 617
 618         LASSERT(S_ISDIR(dt->do_lu.lo_header->loh_attr));
 619
 620         /*
 621          * store striping defaults into new directory
 622          * used to implement defaults inheritance
 623          */
 624
 625         /* probably nothing to inherite */
 626         if (lo->ldo_striping_cached == 0)
 627                 RETURN(0);
 628
 629         if (LOVEA_DELETE_VALUES(lo->ldo_def_stripe_size, lo->ldo_def_stripenr,
 630                                 lo->ldo_def_stripe_offset))
 631                 RETURN(0);
 632
 633         /* XXX: use thread info */
 634         OBD_ALLOC_PTR(v3);
 635         if (v3 == NULL)
 636                 RETURN(-ENOMEM);
 637
 638         v3->lmm_magic = cpu_to_le32(LOV_MAGIC_V3);
 639         v3->lmm_pattern = cpu_to_le32(LOV_PATTERN_RAID0);
 640         v3->lmm_object_id = 0;
 641         v3->lmm_object_seq = 0;
 642         v3->lmm_stripe_size = cpu_to_le32(lo->ldo_def_stripe_size);
 643         v3->lmm_stripe_count = cpu_to_le16(lo->ldo_def_stripenr);
 644         v3->lmm_stripe_offset = cpu_to_le16(lo->ldo_def_stripe_offset);
 645         if (lo->ldo_pool)
 646                 strncpy(v3->lmm_pool_name, lo->ldo_pool, LOV_MAXPOOLNAME);
 647
 648         info->lti_buf.lb_buf = v3;
 649         info->lti_buf.lb_len = sizeof(*v3);
 650         rc = dt_xattr_set(env, next, &info->lti_buf, XATTR_NAME_LOV, 0, th,
 651                         BYPASS_CAPA);
 652
 653         OBD_FREE_PTR(v3);
 654
 655         RETURN(rc);
 656 }
 657
 658 /*
 659  * allocate array of objects pointers, find/create objects
 660  * stripenr and other fields should be initialized by this moment
 661  */
 662 int lod_initialize_objects(const struct lu_env *env, struct lod_object *lo,
 663                            struct lov_ost_data_v1 *objs)
 664 {
 665         struct lod_thread_info  *info = lod_env_info(env);
 666         struct lod_device       *md = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
 667         struct lu_object        *o, *n;
 668         struct lu_device        *nd;
 669         int                     i, idx, rc = 0;
 670         ENTRY;
 671
 672         LASSERT(lo);
 673         LASSERT(lo->ldo_stripe == NULL);
 674         LASSERT(lo->ldo_stripenr > 0);
 675         LASSERT(lo->ldo_stripe_size > 0);
 676
 677         i = sizeof(struct dt_object *) * lo->ldo_stripenr;
 678         OBD_ALLOC(lo->ldo_stripe, i);
 679         if (lo->ldo_stripe == NULL)
 680                 GOTO(out, rc = -ENOMEM);
 681         lo->ldo_stripes_allocated = lo->ldo_stripenr;
 682
 683         for (i = 0; i < lo->ldo_stripenr; i++) {
 684                 info->lti_ostid.oi_id = le64_to_cpu(objs[i].l_object_id);
 685                 info->lti_ostid.oi_seq = le64_to_cpu(objs[i].l_object_seq);
 686                 idx = le64_to_cpu(objs[i].l_ost_idx);
 687                 fid_ostid_unpack(&info->lti_fid, &info->lti_ostid, idx);
 688                 LASSERTF(fid_is_sane(&info->lti_fid), ""DFID" insane!\n",
 689                          PFID(&info->lti_fid));
 690                 /*
 691                  * XXX: assertion is left for testing, to make
 692                  * sure we never process requests till configuration
 693                  * is completed. to be changed to -EINVAL
 694                  */
 695
 696                 lod_getref(&md->lod_ost_descs);
 697                 LASSERT(cfs_bitmap_check(md->lod_ost_bitmap, idx));
 698                 LASSERT(OST_TGT(md,idx));
 699                 LASSERTF(OST_TGT(md,idx)->ltd_ost, "idx %d\n", idx);
 700                 nd = &OST_TGT(md,idx)->ltd_ost->dd_lu_dev;
 701                 lod_putref(md, &md->lod_ost_descs);
 702
 703                 o = lu_object_find_at(env, nd, &info->lti_fid, NULL);
 704                 if (IS_ERR(o))
 705                         GOTO(out, rc = PTR_ERR(o));
 706
 707                 n = lu_object_locate(o->lo_header, nd->ld_type);
 708                 LASSERT(n);
 709
 710                 lo->ldo_stripe[i] = container_of(n, struct dt_object, do_lu);
 711         }
 712
 713 out:
 714         RETURN(rc);
 715 }
 716
 717 /*
 718  * Parse striping information stored in lti_ea_store
 719  */
 720 int lod_parse_striping(const struct lu_env *env, struct lod_object *lo,
 721                        const struct lu_buf *buf)
 722 {
 723         struct lov_mds_md_v1    *lmm;
 724         struct lov_ost_data_v1  *objs;
 725         __u32                    magic;
 726         int                      rc = 0;
 727         ENTRY;
 728
 729         LASSERT(buf);
 730         LASSERT(buf->lb_buf);
 731         LASSERT(buf->lb_len);
 732
 733         lmm = (struct lov_mds_md_v1 *) buf->lb_buf;
 734         magic = le32_to_cpu(lmm->lmm_magic);
 735
 736         if (magic != LOV_MAGIC_V1 && magic != LOV_MAGIC_V3)
 737                 GOTO(out, rc = -EINVAL);
 738         if (le32_to_cpu(lmm->lmm_pattern) != LOV_PATTERN_RAID0)
 739                 GOTO(out, rc = -EINVAL);
 740
 741         lo->ldo_stripe_size = le32_to_cpu(lmm->lmm_stripe_size);
 742         lo->ldo_stripenr = le16_to_cpu(lmm->lmm_stripe_count);
 743         lo->ldo_layout_gen = le16_to_cpu(lmm->lmm_layout_gen);
 744
 745         LASSERT(buf->lb_len >= lov_mds_md_size(lo->ldo_stripenr, magic));
 746
 747         if (magic == LOV_MAGIC_V3) {
 748                 struct lov_mds_md_v3 *v3 = (struct lov_mds_md_v3 *) lmm;
 749                 objs = &v3->lmm_objects[0];
 750                 lod_object_set_pool(lo, v3->lmm_pool_name);
 751         } else {
 752                 objs = &lmm->lmm_objects[0];
 753         }
 754
 755         rc = lod_initialize_objects(env, lo, objs);
 756
 757 out:
 758         RETURN(rc);
 759 }
 760
 761 /*
 762  * Load and parse striping information, create in-core representation for the
 763  * stripes
 764  */
 765 int lod_load_striping(const struct lu_env *env, struct lod_object *lo)
 766 {
 767         struct lod_thread_info  *info = lod_env_info(env);
 768         struct dt_object        *next = dt_object_child(&lo->ldo_obj);
 769         int                      rc;
 770         ENTRY;
 771
 772         /*
 773          * currently this code is supposed to be called from declaration
 774          * phase only, thus the object is not expected to be locked by caller
 775          */
 776         dt_write_lock(env, next, 0);
 777         /* already initialized? */
 778         if (lo->ldo_stripe) {
 779                 int i;
 780                 /* check validity */
 781                 for (i = 0; i < lo->ldo_stripenr; i++)
 782                         LASSERTF(lo->ldo_stripe[i], "stripe %d is NULL\n", i);
 783                 GOTO(out, rc = 0);
 784         }
 785
 786         if (!dt_object_exists(next))
 787                 GOTO(out, rc = 0);
 788
 789         /* only regular files can be striped */
 790         if (!(lu_object_attr(lod2lu_obj(lo)) & S_IFREG))
 791                 GOTO(out, rc = 0);
 792
 793         LASSERT(lo->ldo_stripenr == 0);
 794
 795         rc = lod_get_lov_ea(env, lo);
 796         if (rc <= 0)
 797                 GOTO(out, rc);
 798
 799         /*
 800          * there is LOV EA (striping information) in this object
 801          * let's parse it and create in-core objects for the stripes
 802          */
 803         info->lti_buf.lb_buf = info->lti_ea_store;
 804         info->lti_buf.lb_len = info->lti_ea_store_size;
 805         rc = lod_parse_striping(env, lo, &info->lti_buf);
 806 out:
 807         dt_write_unlock(env, next);
 808         RETURN(rc);
 809 }
 810
 811 int lod_verify_striping(struct lod_device *d, const struct lu_buf *buf,
 812                         int specific)
 813 {
 814         struct lov_user_md_v1   *lum;
 815         struct lov_user_md_v3   *v3 = NULL;
 816         struct pool_desc        *pool = NULL;
 817         int                      rc;
 818         ENTRY;
 819
 820         lum = buf->lb_buf;
 821
 822         if (lum->lmm_magic != LOV_USER_MAGIC_V1 &&
 823             lum->lmm_magic != LOV_USER_MAGIC_V3 &&
 824             lum->lmm_magic != LOV_MAGIC_V1_DEF &&
 825             lum->lmm_magic != LOV_MAGIC_V3_DEF) {
 826                 CDEBUG(D_IOCTL, "bad userland LOV MAGIC: %#x\n",
 827                        lum->lmm_magic);
 828                 RETURN(-EINVAL);
 829         }
 830
 831         if ((specific && lum->lmm_pattern != LOV_PATTERN_RAID0) ||
 832             (specific == 0 && lum->lmm_pattern != 0)) {
 833                 CDEBUG(D_IOCTL, "bad userland stripe pattern: %#x\n",
 834                        lum->lmm_pattern);
 835                 RETURN(-EINVAL);
 836         }
 837
 838         /* 64kB is the largest common page size we see (ia64), and matches the
 839          * check in lfs */
 840         if (lum->lmm_stripe_size & (LOV_MIN_STRIPE_SIZE - 1)) {
 841                 CDEBUG(D_IOCTL, "stripe size %u not multiple of %u, fixing\n",
 842                        lum->lmm_stripe_size, LOV_MIN_STRIPE_SIZE);
 843                 RETURN(-EINVAL);
 844         }
 845
 846         /* an offset of -1 is treated as a "special" valid offset */
 847         if (lum->lmm_stripe_offset != (typeof(lum->lmm_stripe_offset))(-1)) {
 848                 /* if offset is not within valid range [0, osts_size) */
 849                 if (lum->lmm_stripe_offset >= d->lod_osts_size) {
 850                         CDEBUG(D_IOCTL, "stripe offset %u >= bitmap size %u\n",
 851                                lum->lmm_stripe_offset, d->lod_osts_size);
 852                         RETURN(-EINVAL);
 853                 }
 854
 855                 /* if lmm_stripe_offset is *not* in bitmap */
 856                 if (!cfs_bitmap_check(d->lod_ost_bitmap,
 857                                       lum->lmm_stripe_offset)) {
 858                         CDEBUG(D_IOCTL, "stripe offset %u not in bitmap\n",
 859                                lum->lmm_stripe_offset);
 860                         RETURN(-EINVAL);
 861                 }
 862         }
 863
 864         if (lum->lmm_magic == LOV_USER_MAGIC_V3)
 865                 v3 = buf->lb_buf;
 866
 867         if (v3)
 868                 pool = lod_find_pool(d, v3->lmm_pool_name);
 869
 870         if (pool != NULL) {
 871                 __u16 offs = v3->lmm_stripe_offset;
 872
 873                 if (offs != (typeof(v3->lmm_stripe_offset))(-1)) {
 874                         rc = lod_check_index_in_pool(offs, pool);
 875                         if (rc < 0) {
 876                                 lod_pool_putref(pool);
 877                                 RETURN(-EINVAL);
 878                         }
 879                 }
 880
 881                 if (specific && lum->lmm_stripe_count > pool_tgt_count(pool)) {
 882                         CDEBUG(D_IOCTL,
 883                                "stripe count %u > # OSTs %u in the pool\n",
 884                                lum->lmm_stripe_count, pool_tgt_count(pool));
 885                         lod_pool_putref(pool);
 886                         RETURN(-EINVAL);
 887                 }
 888
 889                 lod_pool_putref(pool);
 890         }
 891
 892         RETURN(0);
 893 }
 894
 895 void lod_fix_desc_stripe_size(__u64 *val)
 896 {
 897         if (*val < LOV_DEFAULT_STRIPE_SIZE) {
 898                 LCONSOLE_WARN("Increasing default stripe size to min %u\n",
 899                               LOV_DEFAULT_STRIPE_SIZE);
 900                 *val = LOV_DEFAULT_STRIPE_SIZE;
 901         } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) {
 902                 *val &= ~(LOV_MIN_STRIPE_SIZE - 1);
 903                 LCONSOLE_WARN("Changing default stripe size to "LPU64" (a "
 904                               "multiple of %u)\n",
 905                               *val, LOV_MIN_STRIPE_SIZE);
 906         }
 907 }
 908
 909 void lod_fix_desc_stripe_count(__u32 *val)
 910 {
 911         if (*val == 0)
 912                 *val = 1;
 913 }
 914
 915 void lod_fix_desc_pattern(__u32 *val)
 916 {
 917         /* from lov_setstripe */
 918         if ((*val != 0) && (*val != LOV_PATTERN_RAID0)) {
 919                 LCONSOLE_WARN("Unknown stripe pattern: %#x\n", *val);
 920                 *val = 0;
 921         }
 922 }
 923
 924 void lod_fix_desc_qos_maxage(__u32 *val)
 925 {
 926         /* fix qos_maxage */
 927         if (*val == 0)
 928                 *val = QOS_DEFAULT_MAXAGE;
 929 }
 930
 931 void lod_fix_desc(struct lov_desc *desc)
 932 {
 933         lod_fix_desc_stripe_size(&desc->ld_default_stripe_size);
 934         lod_fix_desc_stripe_count(&desc->ld_default_stripe_count);
 935         lod_fix_desc_pattern(&desc->ld_pattern);
 936         lod_fix_desc_qos_maxage(&desc->ld_qos_maxage);
 937 }
 938
 939 int lod_pools_init(struct lod_device *lod, struct lustre_cfg *lcfg)
 940 {
 941         struct obd_device          *obd;
 942         struct lov_desc            *desc;
 943         int                         rc;
 944         ENTRY;
 945
 946         obd = class_name2obd(lustre_cfg_string(lcfg, 0));
 947         LASSERT(obd != NULL);
 948         obd->obd_lu_dev = &lod->lod_dt_dev.dd_lu_dev;
 949
 950         if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
 951                 CERROR("LOD setup requires a descriptor\n");
 952                 RETURN(-EINVAL);
 953         }
 954
 955         desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
 956
 957         if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
 958                 CERROR("descriptor size wrong: %d > %d\n",
 959                        (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
 960                 RETURN(-EINVAL);
 961         }
 962
 963         if (desc->ld_magic != LOV_DESC_MAGIC) {
 964                 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
 965                         CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
 966                                obd->obd_name, desc);
 967                         lustre_swab_lov_desc(desc);
 968                 } else {
 969                         CERROR("%s: Bad lov desc magic: %#x\n",
 970                                obd->obd_name, desc->ld_magic);
 971                         RETURN(-EINVAL);
 972                 }
 973         }
 974
 975         lod_fix_desc(desc);
 976
 977         desc->ld_active_tgt_count = 0;
 978         lod->lod_desc = *desc;
 979
 980         lod->lod_sp_me = LUSTRE_SP_CLI;
 981
 982         /* Set up allocation policy (QoS and RR) */
 983         CFS_INIT_LIST_HEAD(&lod->lod_qos.lq_oss_list);
 984         init_rwsem(&lod->lod_qos.lq_rw_sem);
 985         lod->lod_qos.lq_dirty = 1;
 986         lod->lod_qos.lq_rr.lqr_dirty = 1;
 987         lod->lod_qos.lq_reset = 1;
 988         /* Default priority is toward free space balance */
 989         lod->lod_qos.lq_prio_free = 232;
 990         /* Default threshold for rr (roughly 17%) */
 991         lod->lod_qos.lq_threshold_rr = 43;
 992         /* Init statfs fields */
 993         OBD_ALLOC_PTR(lod->lod_qos.lq_statfs_data);
 994         if (NULL == lod->lod_qos.lq_statfs_data)
 995                 RETURN(-ENOMEM);
 996         cfs_waitq_init(&lod->lod_qos.lq_statfs_waitq);
 997
 998         /* Set up OST pool environment */
 999         lod->lod_pools_hash_body = cfs_hash_create("POOLS", HASH_POOLS_CUR_BITS,
1000                                                    HASH_POOLS_MAX_BITS,
1001                                                    HASH_POOLS_BKT_BITS, 0,
1002                                                    CFS_HASH_MIN_THETA,
1003                                                    CFS_HASH_MAX_THETA,
1004                                                    &pool_hash_operations,
1005                                                    CFS_HASH_DEFAULT);
1006         if (!lod->lod_pools_hash_body)
1007                 GOTO(out_statfs, rc = -ENOMEM);
1008         CFS_INIT_LIST_HEAD(&lod->lod_pool_list);
1009         lod->lod_pool_count = 0;
1010         rc = lod_ost_pool_init(&lod->lod_pool_info, 0);
1011         if (rc)
1012                 GOTO(out_hash, rc);
1013         rc = lod_ost_pool_init(&lod->lod_qos.lq_rr.lqr_pool, 0);
1014         if (rc)
1015                 GOTO(out_pool_info, rc);
1016
1017         RETURN(0);
1018
1019 out_pool_info:
1020         lod_ost_pool_free(&lod->lod_pool_info);
1021 out_hash:
1022         cfs_hash_putref(lod->lod_pools_hash_body);
1023 out_statfs:
1024         OBD_FREE_PTR(lod->lod_qos.lq_statfs_data);
1025         return rc;
1026 }
1027
1028 int lod_pools_fini(struct lod_device *lod)
1029 {
1030         struct obd_device   *obd = lod2obd(lod);
1031         cfs_list_t          *pos, *tmp;
1032         struct pool_desc    *pool;
1033         ENTRY;
1034
1035         cfs_list_for_each_safe(pos, tmp, &lod->lod_pool_list) {
1036                 pool = cfs_list_entry(pos, struct pool_desc, pool_list);
1037                 /* free pool structs */
1038                 CDEBUG(D_INFO, "delete pool %p\n", pool);
1039                 lod_pool_del(obd, pool->pool_name);
1040         }
1041
1042         cfs_hash_putref(lod->lod_pools_hash_body);
1043         lod_ost_pool_free(&(lod->lod_qos.lq_rr.lqr_pool));
1044         lod_ost_pool_free(&lod->lod_pool_info);
1045         OBD_FREE_PTR(lod->lod_qos.lq_statfs_data);
1046         RETURN(0);
1047 }
1048