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, 2013, 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                 /* XXX set MDS-MDS flags, remove this when running this
 248                  * on client*/
 249                 data->ocd_connect_flags |= OBD_CONNECT_MDS_MDS;
 250                 spin_lock(&imp->imp_lock);
 251                 imp->imp_server_timeout = 1;
 252                 spin_unlock(&imp->imp_lock);
 253                 imp->imp_client->cli_request_portal = MDS_MDS_PORTAL;
 254                 CDEBUG(D_OTHER, "%s: Set 'mds' portal and timeout\n",
 255                       obd->obd_name);
 256                 ltd = &lod->lod_mdt_descs;
 257         }
 258
 259         rc = obd_connect(env, &exp, obd, &obd->obd_uuid, data, NULL);
 260         OBD_FREE_PTR(data);
 261         if (rc) {
 262                 CERROR("%s: cannot connect to next dev %s (%d)\n",
 263                        obd->obd_name, osp, rc);
 264                 GOTO(out_free, rc);
 265         }
 266
 267         LASSERT(obd->obd_lu_dev);
 268         LASSERT(obd->obd_lu_dev->ld_site == lod->lod_dt_dev.dd_lu_dev.ld_site);
 269
 270         ldev = obd->obd_lu_dev;
 271         d = lu2dt_dev(ldev);
 272
 273         /* Allocate ost descriptor and fill it */
 274         OBD_ALLOC_PTR(tgt_desc);
 275         if (!tgt_desc)
 276                 GOTO(out_conn, rc = -ENOMEM);
 277
 278         tgt_desc->ltd_tgt    = d;
 279         tgt_desc->ltd_exp    = exp;
 280         tgt_desc->ltd_uuid   = obd->u.cli.cl_target_uuid;
 281         tgt_desc->ltd_gen    = gen;
 282         tgt_desc->ltd_index  = index;
 283         tgt_desc->ltd_active = active;
 284
 285         lod_getref(ltd);
 286         if (index >= ltd->ltd_tgts_size) {
 287                 /* we have to increase the size of the lod_osts array */
 288                 __u32  newsize;
 289
 290                 newsize = max(ltd->ltd_tgts_size, (__u32)2);
 291                 while (newsize < index + 1)
 292                         newsize = newsize << 1;
 293
 294                 /* lod_bitmap_resize() needs lod_rw_sem
 295                  * which we hold with th reference */
 296                 lod_putref(lod, ltd);
 297
 298                 rc = ltd_bitmap_resize(ltd, newsize);
 299                 if (rc)
 300                         GOTO(out_desc, rc);
 301
 302                 lod_getref(ltd);
 303         }
 304
 305         mutex_lock(&ltd->ltd_mutex);
 306         if (cfs_bitmap_check(ltd->ltd_tgt_bitmap, index)) {
 307                 CERROR("%s: device %d is registered already\n", obd->obd_name,
 308                        index);
 309                 GOTO(out_mutex, rc = -EEXIST);
 310         }
 311
 312         if (ltd->ltd_tgt_idx[index / TGT_PTRS_PER_BLOCK] == NULL) {
 313                 OBD_ALLOC_PTR(ltd->ltd_tgt_idx[index / TGT_PTRS_PER_BLOCK]);
 314                 if (ltd->ltd_tgt_idx[index / TGT_PTRS_PER_BLOCK] == NULL) {
 315                         CERROR("can't allocate index to add %s\n",
 316                                obd->obd_name);
 317                         GOTO(out_mutex, rc = -ENOMEM);
 318                 }
 319         }
 320
 321         if (!strcmp(LUSTRE_OSC_NAME, type)) {
 322                 /* pool and qos are not supported for MDS stack yet */
 323                 rc = lod_ost_pool_add(&lod->lod_pool_info, index,
 324                                       lod->lod_osts_size);
 325                 if (rc) {
 326                         CERROR("%s: can't set up pool, failed with %d\n",
 327                                obd->obd_name, rc);
 328                         GOTO(out_mutex, rc);
 329                 }
 330
 331                 rc = qos_add_tgt(lod, tgt_desc);
 332                 if (rc) {
 333                         CERROR("%s: qos_add_tgt failed with %d\n",
 334                                 obd->obd_name, rc);
 335                         GOTO(out_pool, rc);
 336                 }
 337
 338                 /* The new OST is now a full citizen */
 339                 if (index >= lod->lod_desc.ld_tgt_count)
 340                         lod->lod_desc.ld_tgt_count = index + 1;
 341                 if (active)
 342                         lod->lod_desc.ld_active_tgt_count++;
 343         }
 344
 345         LTD_TGT(ltd, index) = tgt_desc;
 346         cfs_bitmap_set(ltd->ltd_tgt_bitmap, index);
 347         ltd->ltd_tgtnr++;
 348         mutex_unlock(&ltd->ltd_mutex);
 349         lod_putref(lod, ltd);
 350         if (lod->lod_recovery_completed)
 351                 ldev->ld_ops->ldo_recovery_complete(env, ldev);
 352
 353         RETURN(0);
 354
 355 out_pool:
 356         lod_ost_pool_remove(&lod->lod_pool_info, index);
 357 out_mutex:
 358         mutex_unlock(&ltd->ltd_mutex);
 359         lod_putref(lod, ltd);
 360 out_desc:
 361         OBD_FREE_PTR(tgt_desc);
 362 out_conn:
 363         obd_disconnect(exp);
 364 out_free:
 365         return rc;
 366 }
 367
 368 /*
 369  * helper function to schedule OST removal from the device table
 370  */
 371 static void __lod_del_device(struct lod_tgt_descs *ltd,
 372                              unsigned idx)
 373 {
 374         LASSERT(LTD_TGT(ltd, idx));
 375         if (LTD_TGT(ltd, idx)->ltd_reap == 0) {
 376                 LTD_TGT(ltd, idx)->ltd_reap = 1;
 377                 ltd->ltd_death_row++;
 378         }
 379 }
 380
 381 int lod_fini_tgt(struct lod_device *lod, struct lod_tgt_descs *ltd)
 382 {
 383         int idx;
 384
 385         if (ltd->ltd_tgts_size <= 0)
 386                 return 0;
 387         lod_getref(ltd);
 388         mutex_lock(&ltd->ltd_mutex);
 389         cfs_foreach_bit(ltd->ltd_tgt_bitmap, idx)
 390                 __lod_del_device(ltd, idx);
 391         mutex_unlock(&ltd->ltd_mutex);
 392         lod_putref(lod, ltd);
 393         CFS_FREE_BITMAP(ltd->ltd_tgt_bitmap);
 394         for (idx = 0; idx < TGT_PTRS; idx++) {
 395                 if (ltd->ltd_tgt_idx[idx])
 396                         OBD_FREE_PTR(ltd->ltd_tgt_idx[idx]);
 397         }
 398         ltd->ltd_tgts_size = 0;
 399         return 0;
 400 }
 401
 402 /*
 403  * Add support for administratively disabled OST (through the MGS).
 404  * Schedule a target for deletion.  Disconnection and real removal from the
 405  * table takes place in lod_putref() once the last table user release its
 406  * reference.
 407  *
 408  * \param env - is the environment passed by the caller
 409  * \param lod - is the lod device currently connected to the OSP about to be
 410  *              removed
 411  * \param osp - is the name of OSP device about to be removed
 412  * \param idx - is the OSP index
 413  * \param gen - is the generation number, not used currently
 414  */
 415 int lod_del_device(const struct lu_env *env, struct lod_device *lod,
 416                    struct lod_tgt_descs *ltd, char *osp, unsigned idx,
 417                    unsigned gen)
 418 {
 419         struct obd_device *obd;
 420         int                rc = 0;
 421         struct obd_uuid    uuid;
 422         ENTRY;
 423
 424         CDEBUG(D_CONFIG, "osp:%s idx:%d gen:%d\n", osp, idx, gen);
 425
 426         obd_str2uuid(&uuid, osp);
 427
 428         obd = class_find_client_obd(&uuid, LUSTRE_OSP_NAME,
 429                                    &lod->lod_dt_dev.dd_lu_dev.ld_obd->obd_uuid);
 430         if (obd == NULL) {
 431                 CERROR("can't find %s device\n", osp);
 432                 RETURN(-EINVAL);
 433         }
 434
 435         if (gen <= 0) {
 436                 CERROR("%s: request to remove OBD %s with invalid generation %d"
 437                        "\n", obd->obd_name, osp, gen);
 438                 RETURN(-EINVAL);
 439         }
 440
 441         obd_str2uuid(&uuid,  osp);
 442
 443         lod_getref(ltd);
 444         mutex_lock(&ltd->ltd_mutex);
 445         /* check that the index is allocated in the bitmap */
 446         if (!cfs_bitmap_check(ltd->ltd_tgt_bitmap, idx) ||
 447             !LTD_TGT(ltd, idx)) {
 448                 CERROR("%s: device %d is not set up\n", obd->obd_name, idx);
 449                 GOTO(out, rc = -EINVAL);
 450         }
 451
 452         /* check that the UUID matches */
 453         if (!obd_uuid_equals(&uuid, &LTD_TGT(ltd, idx)->ltd_uuid)) {
 454                 CERROR("%s: LOD target UUID %s at index %d does not match %s\n",
 455                        obd->obd_name, obd_uuid2str(&LTD_TGT(ltd,idx)->ltd_uuid),
 456                        idx, osp);
 457                 GOTO(out, rc = -EINVAL);
 458         }
 459
 460         __lod_del_device(ltd, idx);
 461         EXIT;
 462 out:
 463         mutex_unlock(&ltd->ltd_mutex);
 464         lod_putref(lod, ltd);
 465         return(rc);
 466 }
 467
 468 int lod_ea_store_resize(struct lod_thread_info *info, int size)
 469 {
 470         int round = size_roundup_power2(size);
 471
 472         LASSERT(round <= lov_mds_md_size(LOV_MAX_STRIPE_COUNT, LOV_MAGIC_V3));
 473         if (info->lti_ea_store) {
 474                 LASSERT(info->lti_ea_store_size);
 475                 LASSERT(info->lti_ea_store_size < round);
 476                 CDEBUG(D_INFO, "EA store size %d is not enough, need %d\n",
 477                        info->lti_ea_store_size, round);
 478                 OBD_FREE_LARGE(info->lti_ea_store, info->lti_ea_store_size);
 479                 info->lti_ea_store = NULL;
 480                 info->lti_ea_store_size = 0;
 481         }
 482
 483         OBD_ALLOC_LARGE(info->lti_ea_store, round);
 484         if (info->lti_ea_store == NULL)
 485                 RETURN(-ENOMEM);
 486         info->lti_ea_store_size = round;
 487         RETURN(0);
 488 }
 489
 490 /*
 491  * generate and write LOV EA for given striped object
 492  */
 493 int lod_generate_and_set_lovea(const struct lu_env *env,
 494                                struct lod_object *lo, struct thandle *th)
 495 {
 496         struct lod_thread_info  *info = lod_env_info(env);
 497         struct dt_object        *next = dt_object_child(&lo->ldo_obj);
 498         const struct lu_fid     *fid  = lu_object_fid(&lo->ldo_obj.do_lu);
 499         struct lov_mds_md_v1    *lmm;
 500         struct lov_ost_data_v1  *objs;
 501         __u32                    magic;
 502         int                      i, rc, lmm_size;
 503         int                      cplen = 0;
 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         fid_to_lmm_oi(fid, &lmm->lmm_oi);
 522         lmm_oi_cpu_to_le(&lmm->lmm_oi, &lmm->lmm_oi);
 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                 cplen = strlcpy(v3->lmm_pool_name, lo->ldo_pool,
 531                                 sizeof(v3->lmm_pool_name));
 532                 if (cplen >= sizeof(v3->lmm_pool_name))
 533                         RETURN(-E2BIG);
 534                 objs = &v3->lmm_objects[0];
 535         }
 536
 537         for (i = 0; i < lo->ldo_stripenr; i++) {
 538                 const struct lu_fid     *fid;
 539                 struct lod_device       *lod;
 540                 __u32                   index;
 541
 542                 lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
 543                 LASSERT(lo->ldo_stripe[i]);
 544                 fid = lu_object_fid(&lo->ldo_stripe[i]->do_lu);
 545
 546                 rc = fid_to_ostid(fid, &info->lti_ostid);
 547                 LASSERT(rc == 0);
 548
 549                 ostid_cpu_to_le(&info->lti_ostid, &objs[i].l_ost_oi);
 550                 objs[i].l_ost_gen    = cpu_to_le32(0);
 551                 rc = lod_fld_lookup(env, lod, fid, &index, LU_SEQ_RANGE_OST);
 552                 if (rc < 0) {
 553                         CERROR("%s: Can not locate "DFID": rc = %d\n",
 554                                lod2obd(lod)->obd_name, PFID(fid), rc);
 555                         RETURN(rc);
 556                 }
 557                 objs[i].l_ost_idx = cpu_to_le32(index);
 558         }
 559
 560         info->lti_buf.lb_buf = lmm;
 561         info->lti_buf.lb_len = lmm_size;
 562         rc = dt_xattr_set(env, next, &info->lti_buf, XATTR_NAME_LOV, 0,
 563                           th, BYPASS_CAPA);
 564
 565         RETURN(rc);
 566 }
 567
 568 int lod_get_lov_ea(const struct lu_env *env, struct lod_object *lo)
 569 {
 570         struct lod_thread_info *info = lod_env_info(env);
 571         struct dt_object       *next = dt_object_child(&lo->ldo_obj);
 572         int                     rc;
 573         ENTRY;
 574
 575         LASSERT(info);
 576
 577         if (unlikely(info->lti_ea_store_size == 0)) {
 578                 /* just to enter in allocation block below */
 579                 rc = -ERANGE;
 580         } else {
 581 repeat:
 582                 info->lti_buf.lb_buf = info->lti_ea_store;
 583                 info->lti_buf.lb_len = info->lti_ea_store_size;
 584                 rc = dt_xattr_get(env, next, &info->lti_buf, XATTR_NAME_LOV,
 585                                   BYPASS_CAPA);
 586         }
 587         /* if object is not striped or inaccessible */
 588         if (rc == -ENODATA)
 589                 RETURN(0);
 590
 591         if (rc == -ERANGE) {
 592                 /* EA doesn't fit, reallocate new buffer */
 593                 rc = dt_xattr_get(env, next, &LU_BUF_NULL, XATTR_NAME_LOV,
 594                                   BYPASS_CAPA);
 595                 if (rc == -ENODATA)
 596                         RETURN(0);
 597                 else if (rc < 0)
 598                         RETURN(rc);
 599
 600                 LASSERT(rc > 0);
 601                 rc = lod_ea_store_resize(info, rc);
 602                 if (rc)
 603                         RETURN(rc);
 604                 goto repeat;
 605         }
 606
 607         RETURN(rc);
 608 }
 609
 610 int lod_store_def_striping(const struct lu_env *env, struct dt_object *dt,
 611                            struct thandle *th)
 612 {
 613         struct lod_thread_info  *info = lod_env_info(env);
 614         struct lod_object       *lo = lod_dt_obj(dt);
 615         struct dt_object        *next = dt_object_child(dt);
 616         struct lov_user_md_v3   *v3;
 617         int                      rc;
 618         int                      cplen = 0;
 619         ENTRY;
 620
 621         LASSERT(S_ISDIR(dt->do_lu.lo_header->loh_attr));
 622
 623         /*
 624          * store striping defaults into new directory
 625          * used to implement defaults inheritance
 626          */
 627
 628         /* probably nothing to inherite */
 629         if (lo->ldo_striping_cached == 0)
 630                 RETURN(0);
 631
 632         if (LOVEA_DELETE_VALUES(lo->ldo_def_stripe_size, lo->ldo_def_stripenr,
 633                                 lo->ldo_def_stripe_offset))
 634                 RETURN(0);
 635
 636         /* XXX: use thread info */
 637         OBD_ALLOC_PTR(v3);
 638         if (v3 == NULL)
 639                 RETURN(-ENOMEM);
 640
 641         v3->lmm_magic = cpu_to_le32(LOV_MAGIC_V3);
 642         v3->lmm_pattern = cpu_to_le32(LOV_PATTERN_RAID0);
 643         v3->lmm_stripe_size = cpu_to_le32(lo->ldo_def_stripe_size);
 644         v3->lmm_stripe_count = cpu_to_le16(lo->ldo_def_stripenr);
 645         v3->lmm_stripe_offset = cpu_to_le16(lo->ldo_def_stripe_offset);
 646         if (lo->ldo_pool) {
 647                 cplen = strlcpy(v3->lmm_pool_name, lo->ldo_pool,
 648                                 sizeof(v3->lmm_pool_name));
 649                 if (cplen >= sizeof(v3->lmm_pool_name)) {
 650                         OBD_FREE_PTR(v3);
 651                         RETURN(-E2BIG);
 652                 }
 653         }
 654
 655         info->lti_buf.lb_buf = v3;
 656         info->lti_buf.lb_len = sizeof(*v3);
 657         rc = dt_xattr_set(env, next, &info->lti_buf, XATTR_NAME_LOV, 0, th,
 658                         BYPASS_CAPA);
 659
 660         OBD_FREE_PTR(v3);
 661
 662         RETURN(rc);
 663 }
 664
 665 /*
 666  * allocate array of objects pointers, find/create objects
 667  * stripenr and other fields should be initialized by this moment
 668  */
 669 int lod_initialize_objects(const struct lu_env *env, struct lod_object *lo,
 670                            struct lov_ost_data_v1 *objs)
 671 {
 672         struct lod_thread_info  *info = lod_env_info(env);
 673         struct lod_device       *md = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
 674         struct lu_object        *o, *n;
 675         struct lu_device        *nd;
 676         struct dt_object       **stripe;
 677         int                      stripe_len;
 678         int                      i, idx, rc = 0;
 679         ENTRY;
 680
 681         LASSERT(lo);
 682         LASSERT(lo->ldo_stripe == NULL);
 683         LASSERT(lo->ldo_stripenr > 0);
 684         LASSERT(lo->ldo_stripe_size > 0);
 685
 686         stripe_len = lo->ldo_stripenr;
 687         OBD_ALLOC(stripe, sizeof(stripe[0]) * stripe_len);
 688         if (stripe == NULL)
 689                 RETURN(-ENOMEM);
 690
 691         for (i = 0; i < lo->ldo_stripenr; i++) {
 692                 ostid_le_to_cpu(&objs[i].l_ost_oi, &info->lti_ostid);
 693                 idx = le64_to_cpu(objs[i].l_ost_idx);
 694                 rc = ostid_to_fid(&info->lti_fid, &info->lti_ostid, idx);
 695                 if (rc != 0)
 696                         GOTO(out, rc);
 697                 LASSERTF(fid_is_sane(&info->lti_fid), ""DFID" insane!\n",
 698                          PFID(&info->lti_fid));
 699                 /*
 700                  * XXX: assertion is left for testing, to make
 701                  * sure we never process requests till configuration
 702                  * is completed. to be changed to -EINVAL
 703                  */
 704
 705                 lod_getref(&md->lod_ost_descs);
 706                 LASSERT(cfs_bitmap_check(md->lod_ost_bitmap, idx));
 707                 LASSERT(OST_TGT(md,idx));
 708                 LASSERTF(OST_TGT(md,idx)->ltd_ost, "idx %d\n", idx);
 709                 nd = &OST_TGT(md,idx)->ltd_ost->dd_lu_dev;
 710                 lod_putref(md, &md->lod_ost_descs);
 711
 712                 /* In the function below, .hs_keycmp resolves to
 713                  * u_obj_hop_keycmp() */
 714                 /* coverity[overrun-buffer-val] */
 715                 o = lu_object_find_at(env, nd, &info->lti_fid, NULL);
 716                 if (IS_ERR(o))
 717                         GOTO(out, rc = PTR_ERR(o));
 718
 719                 n = lu_object_locate(o->lo_header, nd->ld_type);
 720                 LASSERT(n);
 721
 722                 stripe[i] = container_of(n, struct dt_object, do_lu);
 723         }
 724
 725 out:
 726         if (rc != 0) {
 727                 for (i = 0; i < stripe_len; i++)
 728                         if (stripe[i] != NULL)
 729                                 lu_object_put(env, &stripe[i]->do_lu);
 730
 731                 OBD_FREE(stripe, sizeof(stripe[0]) * stripe_len);
 732         } else {
 733                 lo->ldo_stripe = stripe;
 734                 lo->ldo_stripes_allocated = stripe_len;
 735         }
 736
 737         RETURN(rc);
 738 }
 739
 740 /*
 741  * Parse striping information stored in lti_ea_store
 742  */
 743 int lod_parse_striping(const struct lu_env *env, struct lod_object *lo,
 744                        const struct lu_buf *buf)
 745 {
 746         struct lov_mds_md_v1    *lmm;
 747         struct lov_ost_data_v1  *objs;
 748         __u32                    magic;
 749         int                      rc = 0;
 750         ENTRY;
 751
 752         LASSERT(buf);
 753         LASSERT(buf->lb_buf);
 754         LASSERT(buf->lb_len);
 755
 756         lmm = (struct lov_mds_md_v1 *) buf->lb_buf;
 757         magic = le32_to_cpu(lmm->lmm_magic);
 758
 759         if (magic != LOV_MAGIC_V1 && magic != LOV_MAGIC_V3)
 760                 GOTO(out, rc = -EINVAL);
 761         if (le32_to_cpu(lmm->lmm_pattern) != LOV_PATTERN_RAID0)
 762                 GOTO(out, rc = -EINVAL);
 763
 764         lo->ldo_stripe_size = le32_to_cpu(lmm->lmm_stripe_size);
 765         lo->ldo_stripenr = le16_to_cpu(lmm->lmm_stripe_count);
 766         lo->ldo_layout_gen = le16_to_cpu(lmm->lmm_layout_gen);
 767
 768         LASSERT(buf->lb_len >= lov_mds_md_size(lo->ldo_stripenr, magic));
 769
 770         if (magic == LOV_MAGIC_V3) {
 771                 struct lov_mds_md_v3 *v3 = (struct lov_mds_md_v3 *) lmm;
 772                 objs = &v3->lmm_objects[0];
 773                 lod_object_set_pool(lo, v3->lmm_pool_name);
 774         } else {
 775                 objs = &lmm->lmm_objects[0];
 776         }
 777
 778         rc = lod_initialize_objects(env, lo, objs);
 779
 780 out:
 781         RETURN(rc);
 782 }
 783
 784 /*
 785  * Load and parse striping information, create in-core representation for the
 786  * stripes
 787  */
 788 int lod_load_striping(const struct lu_env *env, struct lod_object *lo)
 789 {
 790         struct lod_thread_info  *info = lod_env_info(env);
 791         struct dt_object        *next = dt_object_child(&lo->ldo_obj);
 792         int                      rc;
 793         ENTRY;
 794
 795         /*
 796          * currently this code is supposed to be called from declaration
 797          * phase only, thus the object is not expected to be locked by caller
 798          */
 799         dt_write_lock(env, next, 0);
 800         /* already initialized? */
 801         if (lo->ldo_stripe) {
 802                 int i;
 803                 /* check validity */
 804                 for (i = 0; i < lo->ldo_stripenr; i++)
 805                         LASSERTF(lo->ldo_stripe[i], "stripe %d is NULL\n", i);
 806                 GOTO(out, rc = 0);
 807         }
 808
 809         if (!dt_object_exists(next))
 810                 GOTO(out, rc = 0);
 811
 812         /* only regular files can be striped */
 813         if (!(lu_object_attr(lod2lu_obj(lo)) & S_IFREG))
 814                 GOTO(out, rc = 0);
 815
 816         LASSERT(lo->ldo_stripenr == 0);
 817
 818         rc = lod_get_lov_ea(env, lo);
 819         if (rc <= 0)
 820                 GOTO(out, rc);
 821
 822         /*
 823          * there is LOV EA (striping information) in this object
 824          * let's parse it and create in-core objects for the stripes
 825          */
 826         info->lti_buf.lb_buf = info->lti_ea_store;
 827         info->lti_buf.lb_len = info->lti_ea_store_size;
 828         rc = lod_parse_striping(env, lo, &info->lti_buf);
 829 out:
 830         dt_write_unlock(env, next);
 831         RETURN(rc);
 832 }
 833
 834 int lod_verify_striping(struct lod_device *d, const struct lu_buf *buf,
 835                         int specific)
 836 {
 837         struct lov_user_md_v1   *lum;
 838         struct lov_user_md_v3   *v3 = NULL;
 839         struct pool_desc        *pool = NULL;
 840         int                      rc;
 841         ENTRY;
 842
 843         lum = buf->lb_buf;
 844
 845         if (lum->lmm_magic != LOV_USER_MAGIC_V1 &&
 846             lum->lmm_magic != LOV_USER_MAGIC_V3 &&
 847             lum->lmm_magic != LOV_MAGIC_V1_DEF &&
 848             lum->lmm_magic != LOV_MAGIC_V3_DEF) {
 849                 CDEBUG(D_IOCTL, "bad userland LOV MAGIC: %#x\n",
 850                        lum->lmm_magic);
 851                 RETURN(-EINVAL);
 852         }
 853
 854         if ((specific && lum->lmm_pattern != LOV_PATTERN_RAID0) ||
 855             (specific == 0 && lum->lmm_pattern != 0)) {
 856                 CDEBUG(D_IOCTL, "bad userland stripe pattern: %#x\n",
 857                        lum->lmm_pattern);
 858                 RETURN(-EINVAL);
 859         }
 860
 861         /* 64kB is the largest common page size we see (ia64), and matches the
 862          * check in lfs */
 863         if (lum->lmm_stripe_size & (LOV_MIN_STRIPE_SIZE - 1)) {
 864                 CDEBUG(D_IOCTL, "stripe size %u not multiple of %u, fixing\n",
 865                        lum->lmm_stripe_size, LOV_MIN_STRIPE_SIZE);
 866                 RETURN(-EINVAL);
 867         }
 868
 869         /* an offset of -1 is treated as a "special" valid offset */
 870         if (lum->lmm_stripe_offset != (typeof(lum->lmm_stripe_offset))(-1)) {
 871                 /* if offset is not within valid range [0, osts_size) */
 872                 if (lum->lmm_stripe_offset >= d->lod_osts_size) {
 873                         CDEBUG(D_IOCTL, "stripe offset %u >= bitmap size %u\n",
 874                                lum->lmm_stripe_offset, d->lod_osts_size);
 875                         RETURN(-EINVAL);
 876                 }
 877
 878                 /* if lmm_stripe_offset is *not* in bitmap */
 879                 if (!cfs_bitmap_check(d->lod_ost_bitmap,
 880                                       lum->lmm_stripe_offset)) {
 881                         CDEBUG(D_IOCTL, "stripe offset %u not in bitmap\n",
 882                                lum->lmm_stripe_offset);
 883                         RETURN(-EINVAL);
 884                 }
 885         }
 886
 887         if (lum->lmm_magic == LOV_USER_MAGIC_V3)
 888                 v3 = buf->lb_buf;
 889
 890         if (v3)
 891                 /* In the function below, .hs_keycmp resolves to
 892                  * pool_hashkey_keycmp() */
 893                 /* coverity[overrun-buffer-val] */
 894                 pool = lod_find_pool(d, v3->lmm_pool_name);
 895
 896         if (pool != NULL) {
 897                 __u16 offs = v3->lmm_stripe_offset;
 898
 899                 if (offs != (typeof(v3->lmm_stripe_offset))(-1)) {
 900                         rc = lod_check_index_in_pool(offs, pool);
 901                         if (rc < 0) {
 902                                 lod_pool_putref(pool);
 903                                 RETURN(-EINVAL);
 904                         }
 905                 }
 906
 907                 if (specific && lum->lmm_stripe_count > pool_tgt_count(pool)) {
 908                         CDEBUG(D_IOCTL,
 909                                "stripe count %u > # OSTs %u in the pool\n",
 910                                lum->lmm_stripe_count, pool_tgt_count(pool));
 911                         lod_pool_putref(pool);
 912                         RETURN(-EINVAL);
 913                 }
 914
 915                 lod_pool_putref(pool);
 916         }
 917
 918         RETURN(0);
 919 }
 920
 921 void lod_fix_desc_stripe_size(__u64 *val)
 922 {
 923         if (*val < LOV_MIN_STRIPE_SIZE) {
 924                 if (*val != 0)
 925                         LCONSOLE_INFO("Increasing default stripe size to "
 926                                       "minimum value %u\n",
 927                                       LOV_DEFAULT_STRIPE_SIZE);
 928                 *val = LOV_DEFAULT_STRIPE_SIZE;
 929         } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) {
 930                 *val &= ~(LOV_MIN_STRIPE_SIZE - 1);
 931                 LCONSOLE_WARN("Changing default stripe size to "LPU64" (a "
 932                               "multiple of %u)\n",
 933                               *val, LOV_MIN_STRIPE_SIZE);
 934         }
 935 }
 936
 937 void lod_fix_desc_stripe_count(__u32 *val)
 938 {
 939         if (*val == 0)
 940                 *val = 1;
 941 }
 942
 943 void lod_fix_desc_pattern(__u32 *val)
 944 {
 945         /* from lov_setstripe */
 946         if ((*val != 0) && (*val != LOV_PATTERN_RAID0)) {
 947                 LCONSOLE_WARN("Unknown stripe pattern: %#x\n", *val);
 948                 *val = 0;
 949         }
 950 }
 951
 952 void lod_fix_desc_qos_maxage(__u32 *val)
 953 {
 954         /* fix qos_maxage */
 955         if (*val == 0)
 956                 *val = QOS_DEFAULT_MAXAGE;
 957 }
 958
 959 void lod_fix_desc(struct lov_desc *desc)
 960 {
 961         lod_fix_desc_stripe_size(&desc->ld_default_stripe_size);
 962         lod_fix_desc_stripe_count(&desc->ld_default_stripe_count);
 963         lod_fix_desc_pattern(&desc->ld_pattern);
 964         lod_fix_desc_qos_maxage(&desc->ld_qos_maxage);
 965 }
 966
 967 int lod_pools_init(struct lod_device *lod, struct lustre_cfg *lcfg)
 968 {
 969         struct obd_device          *obd;
 970         struct lov_desc            *desc;
 971         int                         rc;
 972         ENTRY;
 973
 974         obd = class_name2obd(lustre_cfg_string(lcfg, 0));
 975         LASSERT(obd != NULL);
 976         obd->obd_lu_dev = &lod->lod_dt_dev.dd_lu_dev;
 977
 978         if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
 979                 CERROR("LOD setup requires a descriptor\n");
 980                 RETURN(-EINVAL);
 981         }
 982
 983         desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
 984
 985         if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
 986                 CERROR("descriptor size wrong: %d > %d\n",
 987                        (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
 988                 RETURN(-EINVAL);
 989         }
 990
 991         if (desc->ld_magic != LOV_DESC_MAGIC) {
 992                 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
 993                         CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
 994                                obd->obd_name, desc);
 995                         lustre_swab_lov_desc(desc);
 996                 } else {
 997                         CERROR("%s: Bad lov desc magic: %#x\n",
 998                                obd->obd_name, desc->ld_magic);
 999                         RETURN(-EINVAL);
1000                 }
1001         }
1002
1003         lod_fix_desc(desc);
1004
1005         desc->ld_active_tgt_count = 0;
1006         lod->lod_desc = *desc;
1007
1008         lod->lod_sp_me = LUSTRE_SP_CLI;
1009
1010         /* Set up allocation policy (QoS and RR) */
1011         CFS_INIT_LIST_HEAD(&lod->lod_qos.lq_oss_list);
1012         init_rwsem(&lod->lod_qos.lq_rw_sem);
1013         lod->lod_qos.lq_dirty = 1;
1014         lod->lod_qos.lq_rr.lqr_dirty = 1;
1015         lod->lod_qos.lq_reset = 1;
1016         /* Default priority is toward free space balance */
1017         lod->lod_qos.lq_prio_free = 232;
1018         /* Default threshold for rr (roughly 17%) */
1019         lod->lod_qos.lq_threshold_rr = 43;
1020         /* Init statfs fields */
1021         OBD_ALLOC_PTR(lod->lod_qos.lq_statfs_data);
1022         if (NULL == lod->lod_qos.lq_statfs_data)
1023                 RETURN(-ENOMEM);
1024         cfs_waitq_init(&lod->lod_qos.lq_statfs_waitq);
1025
1026         /* Set up OST pool environment */
1027         lod->lod_pools_hash_body = cfs_hash_create("POOLS", HASH_POOLS_CUR_BITS,
1028                                                    HASH_POOLS_MAX_BITS,
1029                                                    HASH_POOLS_BKT_BITS, 0,
1030                                                    CFS_HASH_MIN_THETA,
1031                                                    CFS_HASH_MAX_THETA,
1032                                                    &pool_hash_operations,
1033                                                    CFS_HASH_DEFAULT);
1034         if (!lod->lod_pools_hash_body)
1035                 GOTO(out_statfs, rc = -ENOMEM);
1036         CFS_INIT_LIST_HEAD(&lod->lod_pool_list);
1037         lod->lod_pool_count = 0;
1038         rc = lod_ost_pool_init(&lod->lod_pool_info, 0);
1039         if (rc)
1040                 GOTO(out_hash, rc);
1041         rc = lod_ost_pool_init(&lod->lod_qos.lq_rr.lqr_pool, 0);
1042         if (rc)
1043                 GOTO(out_pool_info, rc);
1044
1045         RETURN(0);
1046
1047 out_pool_info:
1048         lod_ost_pool_free(&lod->lod_pool_info);
1049 out_hash:
1050         cfs_hash_putref(lod->lod_pools_hash_body);
1051 out_statfs:
1052         OBD_FREE_PTR(lod->lod_qos.lq_statfs_data);
1053         return rc;
1054 }
1055
1056 int lod_pools_fini(struct lod_device *lod)
1057 {
1058         struct obd_device   *obd = lod2obd(lod);
1059         cfs_list_t          *pos, *tmp;
1060         struct pool_desc    *pool;
1061         ENTRY;
1062
1063         cfs_list_for_each_safe(pos, tmp, &lod->lod_pool_list) {
1064                 pool = cfs_list_entry(pos, struct pool_desc, pool_list);
1065                 /* free pool structs */
1066                 CDEBUG(D_INFO, "delete pool %p\n", pool);
1067                 /* In the function below, .hs_keycmp resolves to
1068                  * pool_hashkey_keycmp() */
1069                 /* coverity[overrun-buffer-val] */
1070                 lod_pool_del(obd, pool->pool_name);
1071         }
1072
1073         cfs_hash_putref(lod->lod_pools_hash_body);
1074         lod_ost_pool_free(&(lod->lod_qos.lq_rr.lqr_pool));
1075         lod_ost_pool_free(&lod->lod_pool_info);
1076         OBD_FREE_PTR(lod->lod_qos.lq_statfs_data);
1077         RETURN(0);
1078 }
1079