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