lustre/ofd/ofd_obd.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, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2009, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2012, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/ofd/ofd_obd.c
  37  *
  38  * Author: Andreas Dilger <adilger@whamcloud.com>
  39  * Author: Alex Zhuravlev <bzzz@whamcloud.com>
  40  * Author: Mike Pershin <tappro@whamcloud.com>
  41  */
  42
  43 #define DEBUG_SUBSYSTEM S_FILTER
  44
  45 #include "ofd_internal.h"
  46 #include <obd_cksum.h>
  47 #include <lustre_quota.h>
  48
  49 static int ofd_export_stats_init(struct ofd_device *ofd,
  50                                  struct obd_export *exp, void *client_nid)
  51 {
  52         struct obd_device       *obd = ofd_obd(ofd);
  53         struct nid_stat         *stats;
  54         int                      num_stats, i;
  55         int                      rc, newnid = 0;
  56
  57         ENTRY;
  58
  59         if (obd_uuid_equals(&exp->exp_client_uuid, &obd->obd_uuid))
  60                 /* Self-export gets no proc entry */
  61                 RETURN(0);
  62
  63         rc = lprocfs_exp_setup(exp, client_nid, &newnid);
  64         if (rc) {
  65                 /* Mask error for already created
  66                  * /proc entries */
  67                 if (rc == -EALREADY)
  68                         rc = 0;
  69                 RETURN(rc);
  70         }
  71
  72         if (newnid == 0)
  73                 RETURN(0);
  74
  75         stats = exp->exp_nid_stats;
  76         LASSERT(stats != NULL);
  77
  78         OBD_ALLOC(stats->nid_brw_stats, sizeof(struct brw_stats));
  79         if (stats->nid_brw_stats == NULL)
  80                 GOTO(clean, rc = -ENOMEM);
  81
  82         for (i = 0; i < BRW_LAST; i++)
  83                 spin_lock_init(&stats->nid_brw_stats->hist[i].oh_lock);
  84
  85         rc = lprocfs_seq_create(stats->nid_proc, "brw_stats", 0644,
  86                                 &ofd_per_nid_stats_fops, stats);
  87         if (rc)
  88                 CWARN("Error adding the brw_stats file\n");
  89
  90         num_stats = (sizeof(*obd->obd_type->typ_dt_ops) / sizeof(void *)) +
  91                      LPROC_OFD_LAST - 1;
  92
  93         stats->nid_stats = lprocfs_alloc_stats(num_stats,
  94                                                LPROCFS_STATS_FLAG_NOPERCPU);
  95         if (stats->nid_stats == NULL)
  96                 return -ENOMEM;
  97
  98         lprocfs_init_ops_stats(LPROC_OFD_LAST, stats->nid_stats);
  99         lprocfs_counter_init(stats->nid_stats, LPROC_OFD_READ_BYTES,
 100                              LPROCFS_CNTR_AVGMINMAX, "read_bytes", "bytes");
 101         lprocfs_counter_init(stats->nid_stats, LPROC_OFD_WRITE_BYTES,
 102                              LPROCFS_CNTR_AVGMINMAX, "write_bytes", "bytes");
 103
 104         rc = lprocfs_register_stats(stats->nid_proc, "stats",
 105                                     stats->nid_stats);
 106         if (rc)
 107                 GOTO(clean, rc);
 108
 109         rc = lprocfs_nid_ldlm_stats_init(stats);
 110         if (rc) {
 111                 lprocfs_free_stats(&stats->nid_stats);
 112                 GOTO(clean, rc);
 113         }
 114
 115         RETURN(0);
 116 clean:
 117         return rc;
 118 }
 119
 120 static int ofd_parse_connect_data(const struct lu_env *env,
 121                                   struct obd_export *exp,
 122                                   struct obd_connect_data *data)
 123 {
 124         struct ofd_device                *ofd = ofd_exp(exp);
 125         struct filter_export_data        *fed = &exp->exp_filter_data;
 126
 127         if (!data)
 128                 RETURN(0);
 129
 130         CDEBUG(D_RPCTRACE, "%s: cli %s/%p ocd_connect_flags: "LPX64
 131                " ocd_version: %x ocd_grant: %d ocd_index: %u"
 132                " ocd_group %u\n",
 133                exp->exp_obd->obd_name, exp->exp_client_uuid.uuid, exp,
 134                data->ocd_connect_flags, data->ocd_version,
 135                data->ocd_grant, data->ocd_index, data->ocd_group);
 136
 137         if (fed->fed_group != 0 && fed->fed_group != data->ocd_group) {
 138                 CWARN("!!! This export (nid %s) used object group %d "
 139                       "earlier; now it's trying to use group %d!  This could "
 140                       "be a bug in the MDS. Please report to "
 141                       "http://bugs.whamcloud.com/\n",
 142                       obd_export_nid2str(exp), fed->fed_group,
 143                       data->ocd_group);
 144                 RETURN(-EPROTO);
 145         }
 146         fed->fed_group = data->ocd_group;
 147
 148         data->ocd_connect_flags &= OST_CONNECT_SUPPORTED;
 149         exp->exp_connect_data = *data;
 150         data->ocd_version = LUSTRE_VERSION_CODE;
 151
 152         /* Kindly make sure the SKIP_ORPHAN flag is from MDS. */
 153         if (data->ocd_connect_flags & OBD_CONNECT_MDS)
 154                 CDEBUG(D_HA, "%s: Received MDS connection for group %u\n",
 155                        exp->exp_obd->obd_name, data->ocd_group);
 156         else if (data->ocd_connect_flags & OBD_CONNECT_SKIP_ORPHAN)
 157                 RETURN(-EPROTO);
 158
 159         if (ofd_grant_param_supp(exp)) {
 160                 exp->exp_filter_data.fed_pagesize = data->ocd_blocksize;
 161                 /* ocd_{blocksize,inodespace} are log2 values */
 162                 data->ocd_blocksize  = ofd->ofd_blockbits;
 163                 data->ocd_inodespace = ofd->ofd_dt_conf.ddp_inodespace;
 164                 /* ocd_grant_extent is in 1K blocks */
 165                 data->ocd_grant_extent = ofd->ofd_dt_conf.ddp_grant_frag >> 10;
 166         }
 167
 168         if (exp_connect_flags(exp) & OBD_CONNECT_GRANT)
 169                 data->ocd_grant = ofd_grant_connect(env, exp, data->ocd_grant);
 170
 171         if (data->ocd_connect_flags & OBD_CONNECT_INDEX) {
 172                 struct lr_server_data *lsd = &ofd->ofd_lut.lut_lsd;
 173                 int                    index = lsd->lsd_osd_index;
 174
 175                 if (index != data->ocd_index) {
 176                         LCONSOLE_ERROR_MSG(0x136, "Connection from %s to index"
 177                                            " %u doesn't match actual OST index"
 178                                            " %u in last_rcvd file, bad "
 179                                            "configuration?\n",
 180                                            obd_export_nid2str(exp), index,
 181                                            data->ocd_index);
 182                         RETURN(-EBADF);
 183                 }
 184                 if (!(lsd->lsd_feature_compat & OBD_COMPAT_OST)) {
 185                         /* this will only happen on the first connect */
 186                         lsd->lsd_feature_compat |= OBD_COMPAT_OST;
 187                         /* sync is not needed here as lut_client_add will
 188                          * set exp_need_sync flag */
 189                         tgt_server_data_update(env, &ofd->ofd_lut, 0);
 190                 }
 191         }
 192         if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_SIZE)) {
 193                 data->ocd_brw_size = 65536;
 194         } else if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE) {
 195                 data->ocd_brw_size = min(data->ocd_brw_size,
 196                                          (__u32)DT_MAX_BRW_SIZE);
 197                 if (data->ocd_brw_size == 0) {
 198                         CERROR("%s: cli %s/%p ocd_connect_flags: "LPX64
 199                                " ocd_version: %x ocd_grant: %d ocd_index: %u "
 200                                "ocd_brw_size is unexpectedly zero, "
 201                                "network data corruption?"
 202                                "Refusing connection of this client\n",
 203                                exp->exp_obd->obd_name,
 204                                exp->exp_client_uuid.uuid,
 205                                exp, data->ocd_connect_flags, data->ocd_version,
 206                                data->ocd_grant, data->ocd_index);
 207                         RETURN(-EPROTO);
 208                 }
 209         }
 210
 211         if (data->ocd_connect_flags & OBD_CONNECT_CKSUM) {
 212                 __u32 cksum_types = data->ocd_cksum_types;
 213
 214                 /* The client set in ocd_cksum_types the checksum types it
 215                  * supports. We have to mask off the algorithms that we don't
 216                  * support */
 217                 data->ocd_cksum_types &= cksum_types_supported_server();
 218
 219                 if (unlikely(data->ocd_cksum_types == 0)) {
 220                         CERROR("%s: Connect with checksum support but no "
 221                                "ocd_cksum_types is set\n",
 222                                exp->exp_obd->obd_name);
 223                         RETURN(-EPROTO);
 224                 }
 225
 226                 CDEBUG(D_RPCTRACE, "%s: cli %s supports cksum type %x, return "
 227                        "%x\n", exp->exp_obd->obd_name, obd_export_nid2str(exp),
 228                        cksum_types, data->ocd_cksum_types);
 229         } else {
 230                 /* This client does not support OBD_CONNECT_CKSUM
 231                  * fall back to CRC32 */
 232                 CDEBUG(D_RPCTRACE, "%s: cli %s does not support "
 233                        "OBD_CONNECT_CKSUM, CRC32 will be used\n",
 234                        exp->exp_obd->obd_name, obd_export_nid2str(exp));
 235         }
 236
 237         if (data->ocd_connect_flags & OBD_CONNECT_MAXBYTES)
 238                 data->ocd_maxbytes = ofd->ofd_dt_conf.ddp_maxbytes;
 239
 240         if (data->ocd_connect_flags & OBD_CONNECT_PINGLESS) {
 241                 if (suppress_pings) {
 242                         spin_lock(&exp->exp_obd->obd_dev_lock);
 243                         list_del_init(&exp->exp_obd_chain_timed);
 244                         spin_unlock(&exp->exp_obd->obd_dev_lock);
 245                 } else {
 246                         data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
 247                 }
 248         }
 249
 250         RETURN(0);
 251 }
 252
 253 static int ofd_obd_reconnect(const struct lu_env *env, struct obd_export *exp,
 254                              struct obd_device *obd, struct obd_uuid *cluuid,
 255                              struct obd_connect_data *data, void *localdata)
 256 {
 257         struct ofd_device       *ofd;
 258         int                      rc;
 259
 260         ENTRY;
 261
 262         if (exp == NULL || obd == NULL || cluuid == NULL)
 263                 RETURN(-EINVAL);
 264
 265         ofd = ofd_dev(obd->obd_lu_dev);
 266
 267         rc = lu_env_refill((struct lu_env *)env);
 268         if (rc != 0) {
 269                 CERROR("Failure to refill session: '%d'\n", rc);
 270                 RETURN(rc);
 271         }
 272
 273         ofd_info_init(env, exp);
 274         rc = ofd_parse_connect_data(env, exp, data);
 275         if (rc == 0)
 276                 ofd_export_stats_init(ofd, exp, localdata);
 277
 278         RETURN(rc);
 279 }
 280
 281 static int ofd_obd_connect(const struct lu_env *env, struct obd_export **_exp,
 282                            struct obd_device *obd, struct obd_uuid *cluuid,
 283                            struct obd_connect_data *data, void *localdata)
 284 {
 285         struct obd_export       *exp;
 286         struct ofd_device       *ofd;
 287         struct lustre_handle     conn = { 0 };
 288         int                      rc;
 289         ENTRY;
 290
 291         if (_exp == NULL || obd == NULL || cluuid == NULL)
 292                 RETURN(-EINVAL);
 293
 294         ofd = ofd_dev(obd->obd_lu_dev);
 295
 296         rc = class_connect(&conn, obd, cluuid);
 297         if (rc)
 298                 RETURN(rc);
 299
 300         exp = class_conn2export(&conn);
 301         LASSERT(exp != NULL);
 302
 303         rc = lu_env_refill((struct lu_env *)env);
 304         if (rc != 0) {
 305                 CERROR("Failure to refill session: '%d'\n", rc);
 306                 GOTO(out, rc);
 307         }
 308
 309         ofd_info_init(env, exp);
 310
 311         rc = ofd_parse_connect_data(env, exp, data);
 312         if (rc)
 313                 GOTO(out, rc);
 314
 315         if (obd->obd_replayable) {
 316                 struct tg_export_data *ted = &exp->exp_target_data;
 317
 318                 memcpy(ted->ted_lcd->lcd_uuid, cluuid,
 319                        sizeof(ted->ted_lcd->lcd_uuid));
 320                 rc = tgt_client_new(env, exp);
 321                 if (rc != 0)
 322                         GOTO(out, rc);
 323                 ofd_export_stats_init(ofd, exp, localdata);
 324         }
 325
 326         CDEBUG(D_HA, "%s: get connection from MDS %d\n", obd->obd_name,
 327                data->ocd_group);
 328
 329 out:
 330         if (rc != 0) {
 331                 class_disconnect(exp);
 332                 *_exp = NULL;
 333         } else {
 334                 *_exp = exp;
 335         }
 336         RETURN(rc);
 337 }
 338
 339 static int ofd_obd_disconnect(struct obd_export *exp)
 340 {
 341         struct ofd_device       *ofd = ofd_dev(exp->exp_obd->obd_lu_dev);
 342         struct lu_env            env;
 343         int                      rc;
 344
 345         ENTRY;
 346
 347         LASSERT(exp);
 348         class_export_get(exp);
 349
 350         if (!(exp->exp_flags & OBD_OPT_FORCE))
 351                 ofd_grant_sanity_check(ofd_obd(ofd), __FUNCTION__);
 352
 353         rc = server_disconnect_export(exp);
 354
 355         ofd_grant_discard(exp);
 356
 357         rc = lu_env_init(&env, LCT_DT_THREAD);
 358         if (rc)
 359                 RETURN(rc);
 360
 361         /* Do not erase record for recoverable client. */
 362         if (exp->exp_obd->obd_replayable &&
 363             (!exp->exp_obd->obd_fail || exp->exp_failed))
 364                 tgt_client_del(&env, exp);
 365         lu_env_fini(&env);
 366
 367         class_export_put(exp);
 368         RETURN(rc);
 369 }
 370
 371 static int ofd_init_export(struct obd_export *exp)
 372 {
 373         int rc;
 374
 375         spin_lock_init(&exp->exp_filter_data.fed_lock);
 376         CFS_INIT_LIST_HEAD(&exp->exp_filter_data.fed_mod_list);
 377         spin_lock(&exp->exp_lock);
 378         exp->exp_connecting = 1;
 379         spin_unlock(&exp->exp_lock);
 380
 381         /* self-export doesn't need client data and ldlm initialization */
 382         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
 383                                      &exp->exp_client_uuid)))
 384                 return 0;
 385
 386         rc = tgt_client_alloc(exp);
 387         if (rc == 0)
 388                 ldlm_init_export(exp);
 389         if (rc)
 390                 CERROR("%s: Can't initialize export: rc %d\n",
 391                        exp->exp_obd->obd_name, rc);
 392         return rc;
 393 }
 394
 395 static int ofd_destroy_export(struct obd_export *exp)
 396 {
 397         struct ofd_device *ofd = ofd_dev(exp->exp_obd->obd_lu_dev);
 398
 399         if (exp->exp_filter_data.fed_pending)
 400                 CERROR("%s: cli %s/%p has %lu pending on destroyed export"
 401                        "\n", exp->exp_obd->obd_name, exp->exp_client_uuid.uuid,
 402                        exp, exp->exp_filter_data.fed_pending);
 403
 404         target_destroy_export(exp);
 405
 406         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
 407                                      &exp->exp_client_uuid)))
 408                 return 0;
 409
 410         ldlm_destroy_export(exp);
 411         tgt_client_free(exp);
 412
 413         ofd_fmd_cleanup(exp);
 414
 415         /*
 416          * discard grants once we're sure no more
 417          * interaction with the client is possible
 418          */
 419         ofd_grant_discard(exp);
 420         ofd_fmd_cleanup(exp);
 421
 422         if (exp_connect_flags(exp) & OBD_CONNECT_GRANT_SHRINK) {
 423                 if (ofd->ofd_tot_granted_clients > 0)
 424                         ofd->ofd_tot_granted_clients --;
 425         }
 426
 427         if (!(exp->exp_flags & OBD_OPT_FORCE))
 428                 ofd_grant_sanity_check(exp->exp_obd, __FUNCTION__);
 429
 430         LASSERT(cfs_list_empty(&exp->exp_filter_data.fed_mod_list));
 431         return 0;
 432 }
 433
 434 int ofd_postrecov(const struct lu_env *env, struct ofd_device *ofd)
 435 {
 436         struct lu_device *ldev = &ofd->ofd_dt_dev.dd_lu_dev;
 437
 438         CDEBUG(D_HA, "%s: recovery is over\n", ofd_obd(ofd)->obd_name);
 439         return ldev->ld_ops->ldo_recovery_complete(env, ldev);
 440 }
 441
 442 int ofd_obd_postrecov(struct obd_device *obd)
 443 {
 444         struct lu_env            env;
 445         struct lu_device        *ldev = obd->obd_lu_dev;
 446         int                      rc;
 447
 448         ENTRY;
 449
 450         rc = lu_env_init(&env, LCT_DT_THREAD);
 451         if (rc)
 452                 RETURN(rc);
 453         ofd_info_init(&env, obd->obd_self_export);
 454
 455         rc = ofd_postrecov(&env, ofd_dev(ldev));
 456
 457         lu_env_fini(&env);
 458         RETURN(rc);
 459 }
 460
 461 static int ofd_adapt_sptlrpc_conf(const struct lu_env *env,
 462                                   struct obd_device *obd, int initial)
 463 {
 464         struct filter_obd       *fo = &obd->u.filter;
 465         struct sptlrpc_rule_set  tmp_rset;
 466         int                      rc;
 467
 468         sptlrpc_rule_set_init(&tmp_rset);
 469         rc = sptlrpc_conf_target_get_rules(obd, &tmp_rset, initial);
 470         if (rc) {
 471                 CERROR("%s: failed get sptlrpc rules: rc = %d\n",
 472                        obd->obd_name, rc);
 473                 return rc;
 474         }
 475
 476         sptlrpc_target_update_exp_flavor(obd, &tmp_rset);
 477
 478         write_lock(&fo->fo_sptlrpc_lock);
 479         sptlrpc_rule_set_free(&fo->fo_sptlrpc_rset);
 480         fo->fo_sptlrpc_rset = tmp_rset;
 481         write_unlock(&fo->fo_sptlrpc_lock);
 482
 483         return 0;
 484 }
 485
 486 static int ofd_set_mds_conn(struct obd_export *exp, void *val)
 487 {
 488         int rc = 0;
 489
 490         ENTRY;
 491
 492         LCONSOLE_WARN("%s: received MDS connection from %s\n",
 493                       exp->exp_obd->obd_name, obd_export_nid2str(exp));
 494         RETURN(rc);
 495 }
 496
 497 static int ofd_set_info_async(const struct lu_env *env, struct obd_export *exp,
 498                               __u32 keylen, void *key, __u32 vallen, void *val,
 499                               struct ptlrpc_request_set *set)
 500 {
 501         struct ofd_device       *ofd;
 502         int                      rc = 0;
 503
 504         ENTRY;
 505
 506         if (exp->exp_obd == NULL) {
 507                 CDEBUG(D_IOCTL, "invalid export %p\n", exp);
 508                 RETURN(-EINVAL);
 509         }
 510
 511         ofd = ofd_exp(exp);
 512
 513         if (KEY_IS(KEY_CAPA_KEY)) {
 514                 rc = ofd_update_capa_key(ofd, val);
 515                 if (rc)
 516                         CERROR("%s: update capability key failed: rc = %d\n",
 517                                exp->exp_obd->obd_name, rc);
 518         } else if (KEY_IS(KEY_SPTLRPC_CONF)) {
 519                 ofd_adapt_sptlrpc_conf(env, exp->exp_obd, 0);
 520         } else if (KEY_IS(KEY_MDS_CONN)) {
 521                 rc = ofd_set_mds_conn(exp, val);
 522         } else if (KEY_IS(KEY_GRANT_SHRINK)) {
 523                 struct ost_body *body = val;
 524
 525                 ofd_info_init(env, exp);
 526                 /** handle grant shrink, similar to a read request */
 527                 ofd_grant_prepare_read(env, exp, &body->oa);
 528         } else {
 529                 CERROR("%s: Unsupported key %s\n",
 530                        exp->exp_obd->obd_name, (char*)key);
 531                 rc = -EOPNOTSUPP;
 532         }
 533
 534         RETURN(rc);
 535 }
 536
 537 static int ofd_get_info(const struct lu_env *env, struct obd_export *exp,
 538                         __u32 keylen, void *key, __u32 *vallen, void *val,
 539                         struct lov_stripe_md *lsm)
 540 {
 541         struct ofd_device       *ofd;
 542         int                      rc = 0;
 543
 544         ENTRY;
 545
 546         if (exp->exp_obd == NULL) {
 547                 CDEBUG(D_IOCTL, "invalid client export %p\n", exp);
 548                 RETURN(-EINVAL);
 549         }
 550
 551         ofd = ofd_exp(exp);
 552
 553         if (KEY_IS(KEY_BLOCKSIZE)) {
 554                 __u32 *blocksize = val;
 555                 if (blocksize) {
 556                         if (*vallen < sizeof(*blocksize))
 557                                 RETURN(-EOVERFLOW);
 558                         *blocksize = 1 << ofd->ofd_dt_conf.ddp_block_shift;
 559                 }
 560                 *vallen = sizeof(*blocksize);
 561         } else if (KEY_IS(KEY_BLOCKSIZE_BITS)) {
 562                 __u32 *blocksize_bits = val;
 563                 if (blocksize_bits) {
 564                         if (*vallen < sizeof(*blocksize_bits))
 565                                 RETURN(-EOVERFLOW);
 566                         *blocksize_bits = ofd->ofd_dt_conf.ddp_block_shift;
 567                 }
 568                 *vallen = sizeof(*blocksize_bits);
 569         } else if (KEY_IS(KEY_LAST_ID)) {
 570                 obd_id *last_id = val;
 571                 struct ofd_seq *oseq;
 572
 573                 if (val == NULL) {
 574                         *vallen = sizeof(obd_id);
 575                         RETURN(0);
 576                 }
 577                 ofd_info_init(env, exp);
 578                 oseq = ofd_seq_load(env, ofd,
 579                                     (obd_seq)exp->exp_filter_data.fed_group);
 580                 LASSERT(!IS_ERR(oseq));
 581                 if (last_id) {
 582                         if (*vallen < sizeof(*last_id)) {
 583                                 ofd_seq_put(env, oseq);
 584                                 RETURN(-EOVERFLOW);
 585                         }
 586                         *last_id = ofd_seq_last_oid(oseq);
 587                 }
 588                 ofd_seq_put(env, oseq);
 589                 *vallen = sizeof(*last_id);
 590         } else if (KEY_IS(KEY_FIEMAP)) {
 591                 struct ofd_thread_info          *info;
 592                 struct ofd_device               *ofd = ofd_exp(exp);
 593                 struct ofd_object               *fo;
 594                 struct ll_fiemap_info_key       *fm_key = key;
 595
 596                 if (val == NULL) {
 597                         *vallen = fiemap_count_to_size(
 598                                                fm_key->fiemap.fm_extent_count);
 599                         RETURN(0);
 600                 }
 601
 602                 info = ofd_info_init(env, exp);
 603
 604                 fid_ostid_unpack(&info->fti_fid, &fm_key->oa.o_oi, 0);
 605
 606                 CDEBUG(D_INODE, "get FIEMAP of object "DFID"\n",
 607                        PFID(&info->fti_fid));
 608
 609                 fo = ofd_object_find(env, ofd, &info->fti_fid);
 610                 if (IS_ERR(fo)) {
 611                         CERROR("%s: error finding object "DFID"\n",
 612                                exp->exp_obd->obd_name, PFID(&info->fti_fid));
 613                         rc = PTR_ERR(fo);
 614                 } else {
 615                         struct ll_user_fiemap *fiemap = val;
 616
 617                         ofd_read_lock(env, fo);
 618                         if (ofd_object_exists(fo)) {
 619                                 *fiemap = fm_key->fiemap;
 620                                 rc = dt_fiemap_get(env,
 621                                                    ofd_object_child(fo),
 622                                                    fiemap);
 623                         } else {
 624                                 rc = -ENOENT;
 625                         }
 626                         ofd_read_unlock(env, fo);
 627                         ofd_object_put(env, fo);
 628                 }
 629         } else if (KEY_IS(KEY_SYNC_LOCK_CANCEL)) {
 630                 *((__u32 *) val) = ofd->ofd_sync_lock_cancel;
 631                 *vallen = sizeof(__u32);
 632         } else if (KEY_IS(KEY_LAST_FID)) {
 633                 struct lu_env           env;
 634                 struct ofd_device       *ofd = ofd_exp(exp);
 635                 struct ofd_seq          *oseq;
 636                 struct ost_id           *oid = val;
 637                 int                     rc;
 638
 639                 if (oid == NULL) {
 640                         *vallen = sizeof(struct ost_id);
 641                         RETURN(0);
 642                 }
 643
 644                 if (*vallen < sizeof(*oid))
 645                         RETURN(-EOVERFLOW);
 646
 647                 rc = lu_env_init(&env, LCT_DT_THREAD);
 648                 if (rc != 0)
 649                         RETURN(rc);
 650                 ofd_info_init(&env, exp);
 651
 652                 ostid_le_to_cpu(oid, oid);
 653                 CDEBUG(D_HA, "Get LAST FID for seq "LPX64"\n", oid->oi_seq);
 654
 655                 oseq = ofd_seq_load(&env, ofd, oid->oi_seq);
 656                 if (IS_ERR(oseq))
 657                         GOTO(out_fini, rc = PTR_ERR(oseq));
 658
 659                 CDEBUG(D_HA, "LAST FID is "POSTID"\n", oseq->os_last_oid,
 660                        oseq->os_seq);
 661
 662                 *oid = oseq->os_oi;
 663                 *vallen = sizeof(*oid);
 664                 ofd_seq_put(&env, oseq);
 665 out_fini:
 666                 lu_env_fini(&env);
 667         } else {
 668                 CERROR("Not supported key %s\n", (char*)key);
 669                 rc = -EOPNOTSUPP;
 670         }
 671
 672         RETURN(rc);
 673 }
 674
 675 /** helper function for statfs, also used by grant code */
 676 int ofd_statfs_internal(const struct lu_env *env, struct ofd_device *ofd,
 677                         struct obd_statfs *osfs, __u64 max_age, int *from_cache)
 678 {
 679         int rc;
 680
 681         spin_lock(&ofd->ofd_osfs_lock);
 682         if (cfs_time_before_64(ofd->ofd_osfs_age, max_age) || max_age == 0) {
 683                 obd_size unstable;
 684
 685                 /* statfs data are too old, get up-to-date one.
 686                  * we must be cautious here since multiple threads might be
 687                  * willing to update statfs data concurrently and we must
 688                  * grant that cached statfs data are always consistent */
 689
 690                 if (ofd->ofd_statfs_inflight == 0)
 691                         /* clear inflight counter if no users, although it would
 692                          * take a while to overflow this 64-bit counter ... */
 693                         ofd->ofd_osfs_inflight = 0;
 694                 /* notify ofd_grant_commit() that we want to track writes
 695                  * completed as of now */
 696                 ofd->ofd_statfs_inflight++;
 697                 /* record value of inflight counter before running statfs to
 698                  * compute the diff once statfs is completed */
 699                 unstable = ofd->ofd_osfs_inflight;
 700                 spin_unlock(&ofd->ofd_osfs_lock);
 701
 702                 /* statfs can sleep ... hopefully not for too long since we can
 703                  * call it fairly often as space fills up */
 704                 rc = dt_statfs(env, ofd->ofd_osd, osfs);
 705                 if (unlikely(rc))
 706                         return rc;
 707
 708                 spin_lock(&ofd->ofd_grant_lock);
 709                 spin_lock(&ofd->ofd_osfs_lock);
 710                 /* calculate how much space was written while we released the
 711                  * ofd_osfs_lock */
 712                 unstable = ofd->ofd_osfs_inflight - unstable;
 713                 ofd->ofd_osfs_unstable = 0;
 714                 if (unstable) {
 715                         /* some writes completed while we were running statfs
 716                          * w/o the ofd_osfs_lock. Those ones got added to
 717                          * the cached statfs data that we are about to crunch.
 718                          * Take them into account in the new statfs data */
 719                         osfs->os_bavail -= min_t(obd_size, osfs->os_bavail,
 720                                                unstable >> ofd->ofd_blockbits);
 721                         /* However, we don't really know if those writes got
 722                          * accounted in the statfs call, so tell
 723                          * ofd_grant_space_left() there is some uncertainty
 724                          * on the accounting of those writes.
 725                          * The purpose is to prevent spurious error messages in
 726                          * ofd_grant_space_left() since those writes might be
 727                          * accounted twice. */
 728                         ofd->ofd_osfs_unstable += unstable;
 729                 }
 730                 /* similarly, there is some uncertainty on write requests
 731                  * between prepare & commit */
 732                 ofd->ofd_osfs_unstable += ofd->ofd_tot_pending;
 733                 spin_unlock(&ofd->ofd_grant_lock);
 734
 735                 /* finally udpate cached statfs data */
 736                 ofd->ofd_osfs = *osfs;
 737                 ofd->ofd_osfs_age = cfs_time_current_64();
 738
 739                 ofd->ofd_statfs_inflight--; /* stop tracking */
 740                 if (ofd->ofd_statfs_inflight == 0)
 741                         ofd->ofd_osfs_inflight = 0;
 742                 spin_unlock(&ofd->ofd_osfs_lock);
 743
 744                 if (from_cache)
 745                         *from_cache = 0;
 746         } else {
 747                 /* use cached statfs data */
 748                 *osfs = ofd->ofd_osfs;
 749                 spin_unlock(&ofd->ofd_osfs_lock);
 750                 if (from_cache)
 751                         *from_cache = 1;
 752         }
 753         return 0;
 754 }
 755
 756 static int ofd_statfs(const struct lu_env *env,  struct obd_export *exp,
 757                       struct obd_statfs *osfs, __u64 max_age, __u32 flags)
 758 {
 759         struct obd_device       *obd = class_exp2obd(exp);
 760         struct ofd_device       *ofd = ofd_dev(exp->exp_obd->obd_lu_dev);
 761         int                      rc;
 762
 763         ENTRY;
 764
 765         rc = ofd_statfs_internal(env, ofd, osfs, max_age, NULL);
 766         if (unlikely(rc))
 767                 GOTO(out, rc);
 768
 769         /* at least try to account for cached pages.  its still racy and
 770          * might be under-reporting if clients haven't announced their
 771          * caches with brw recently */
 772
 773         CDEBUG(D_SUPER | D_CACHE, "blocks cached "LPU64" granted "LPU64
 774                " pending "LPU64" free "LPU64" avail "LPU64"\n",
 775                ofd->ofd_tot_dirty, ofd->ofd_tot_granted, ofd->ofd_tot_pending,
 776                osfs->os_bfree << ofd->ofd_blockbits,
 777                osfs->os_bavail << ofd->ofd_blockbits);
 778
 779         osfs->os_bavail -= min_t(obd_size, osfs->os_bavail,
 780                                  ((ofd->ofd_tot_dirty + ofd->ofd_tot_pending +
 781                                    osfs->os_bsize - 1) >> ofd->ofd_blockbits));
 782
 783         /* The QoS code on the MDS does not care about space reserved for
 784          * precreate, so take it out. */
 785         if (exp_connect_flags(exp) & OBD_CONNECT_MDS) {
 786                 struct filter_export_data *fed;
 787
 788                 fed = &obd->obd_self_export->exp_filter_data;
 789                 osfs->os_bavail -= min_t(obd_size, osfs->os_bavail,
 790                                          fed->fed_grant >> ofd->ofd_blockbits);
 791         }
 792
 793         ofd_grant_sanity_check(obd, __FUNCTION__);
 794         CDEBUG(D_CACHE, LPU64" blocks: "LPU64" free, "LPU64" avail; "
 795                LPU64" objects: "LPU64" free; state %x\n",
 796                osfs->os_blocks, osfs->os_bfree, osfs->os_bavail,
 797                osfs->os_files, osfs->os_ffree, osfs->os_state);
 798
 799         if (OBD_FAIL_CHECK_VALUE(OBD_FAIL_OST_ENOINO,
 800                                  ofd->ofd_lut.lut_lsd.lsd_osd_index))
 801                 osfs->os_ffree = 0;
 802
 803         /* OS_STATE_READONLY can be set by OSD already */
 804         if (ofd->ofd_raid_degraded)
 805                 osfs->os_state |= OS_STATE_DEGRADED;
 806
 807         if (obd->obd_self_export != exp && ofd_grant_compat(exp, ofd)) {
 808                 /* clients which don't support OBD_CONNECT_GRANT_PARAM
 809                  * should not see a block size > page size, otherwise
 810                  * cl_lost_grant goes mad. Therefore, we emulate a 4KB (=2^12)
 811                  * block size which is the biggest block size known to work
 812                  * with all client's page size. */
 813                 osfs->os_blocks <<= ofd->ofd_blockbits - COMPAT_BSIZE_SHIFT;
 814                 osfs->os_bfree  <<= ofd->ofd_blockbits - COMPAT_BSIZE_SHIFT;
 815                 osfs->os_bavail <<= ofd->ofd_blockbits - COMPAT_BSIZE_SHIFT;
 816                 osfs->os_bsize    = 1 << COMPAT_BSIZE_SHIFT;
 817         }
 818
 819         if (OBD_FAIL_CHECK_VALUE(OBD_FAIL_OST_ENOSPC,
 820                                  ofd->ofd_lut.lut_lsd.lsd_osd_index))
 821                 osfs->os_bfree = osfs->os_bavail = 2;
 822
 823         EXIT;
 824 out:
 825         return rc;
 826 }
 827
 828 int ofd_setattr(const struct lu_env *env, struct obd_export *exp,
 829                 struct obd_info *oinfo, struct obd_trans_info *oti)
 830 {
 831         struct ofd_thread_info  *info;
 832         struct ofd_device       *ofd = ofd_exp(exp);
 833         struct ldlm_namespace   *ns = ofd->ofd_namespace;
 834         struct ldlm_resource    *res;
 835         struct ofd_object       *fo;
 836         struct obdo             *oa = oinfo->oi_oa;
 837         struct filter_fid       *ff = NULL;
 838         int                      rc = 0;
 839
 840         ENTRY;
 841
 842         info = ofd_info_init(env, exp);
 843         ofd_oti2info(info, oti);
 844
 845         fid_ostid_unpack(&info->fti_fid, &oinfo->oi_oa->o_oi, 0);
 846         ofd_build_resid(&info->fti_fid, &info->fti_resid);
 847
 848         rc = ofd_auth_capa(exp, &info->fti_fid, oa->o_seq,
 849                            oinfo_capa(oinfo), CAPA_OPC_META_WRITE);
 850         if (rc)
 851                 GOTO(out, rc);
 852
 853         /* This would be very bad - accidentally truncating a file when
 854          * changing the time or similar - bug 12203. */
 855         if (oinfo->oi_oa->o_valid & OBD_MD_FLSIZE &&
 856             oinfo->oi_policy.l_extent.end != OBD_OBJECT_EOF) {
 857                 static char mdsinum[48];
 858
 859                 if (oinfo->oi_oa->o_valid & OBD_MD_FLFID)
 860                         snprintf(mdsinum, sizeof(mdsinum) - 1,
 861                                  "of parent "DFID, oinfo->oi_oa->o_parent_seq,
 862                                  oinfo->oi_oa->o_parent_oid, 0);
 863                 else
 864                         mdsinum[0] = '\0';
 865
 866                 CERROR("%s: setattr from %s trying to truncate object "DFID
 867                        " %s\n", exp->exp_obd->obd_name,
 868                        obd_export_nid2str(exp), PFID(&info->fti_fid), mdsinum);
 869                 GOTO(out, rc = -EPERM);
 870         }
 871
 872         fo = ofd_object_find(env, ofd, &info->fti_fid);
 873         if (IS_ERR(fo)) {
 874                 CERROR("%s: can't find object "DFID"\n",
 875                        exp->exp_obd->obd_name, PFID(&info->fti_fid));
 876                 GOTO(out, rc = PTR_ERR(fo));
 877         }
 878
 879         la_from_obdo(&info->fti_attr, oinfo->oi_oa, oinfo->oi_oa->o_valid);
 880         info->fti_attr.la_valid &= ~LA_TYPE;
 881
 882         if (oa->o_valid & OBD_MD_FLFID) {
 883                 ff = &info->fti_mds_fid;
 884                 ofd_prepare_fidea(ff, oa);
 885         }
 886
 887         /* setting objects attributes (including owner/group) */
 888         rc = ofd_attr_set(env, fo, &info->fti_attr, ff);
 889         if (rc)
 890                 GOTO(out_unlock, rc);
 891
 892         res = ldlm_resource_get(ns, NULL, &info->fti_resid, LDLM_EXTENT, 0);
 893         if (res != NULL) {
 894                 ldlm_res_lvbo_update(res, NULL, 0);
 895                 ldlm_resource_putref(res);
 896         }
 897
 898         oinfo->oi_oa->o_valid = OBD_MD_FLID;
 899
 900         /* Quota release needs uid/gid info */
 901         rc = ofd_attr_get(env, fo, &info->fti_attr);
 902         obdo_from_la(oinfo->oi_oa, &info->fti_attr,
 903                      OFD_VALID_FLAGS | LA_UID | LA_GID);
 904         ofd_info2oti(info, oti);
 905
 906         ofd_counter_incr(exp, LPROC_OFD_STATS_SETATTR, oti->oti_jobid, 1);
 907         EXIT;
 908 out_unlock:
 909         ofd_object_put(env, fo);
 910 out:
 911         return rc;
 912 }
 913
 914 static int ofd_punch(const struct lu_env *env, struct obd_export *exp,
 915                      struct obd_info *oinfo, struct obd_trans_info *oti,
 916                      struct ptlrpc_request_set *rqset)
 917 {
 918         struct ofd_thread_info  *info;
 919         struct ofd_device       *ofd = ofd_exp(exp);
 920         struct ldlm_namespace   *ns = ofd->ofd_namespace;
 921         struct ldlm_resource    *res;
 922         struct ofd_object       *fo;
 923         struct filter_fid       *ff = NULL;
 924         int                      rc = 0;
 925
 926         ENTRY;
 927
 928         info = ofd_info_init(env, exp);
 929         ofd_oti2info(info, oti);
 930
 931         fid_ostid_unpack(&info->fti_fid, &oinfo->oi_oa->o_oi, 0);
 932         ofd_build_resid(&info->fti_fid, &info->fti_resid);
 933
 934         CDEBUG(D_INODE, "calling punch for object "DFID", valid = "LPX64
 935                ", start = "LPD64", end = "LPD64"\n", PFID(&info->fti_fid),
 936                oinfo->oi_oa->o_valid, oinfo->oi_policy.l_extent.start,
 937                oinfo->oi_policy.l_extent.end);
 938
 939         rc = ofd_auth_capa(exp, &info->fti_fid, oinfo->oi_oa->o_seq,
 940                            oinfo_capa(oinfo), CAPA_OPC_OSS_TRUNC);
 941         if (rc)
 942                 GOTO(out_env, rc);
 943
 944         fo = ofd_object_find(env, ofd, &info->fti_fid);
 945         if (IS_ERR(fo)) {
 946                 CERROR("%s: error finding object "DFID": rc = %ld\n",
 947                        exp->exp_obd->obd_name, PFID(&info->fti_fid),
 948                        PTR_ERR(fo));
 949                 GOTO(out_env, rc = PTR_ERR(fo));
 950         }
 951
 952         LASSERT(oinfo->oi_policy.l_extent.end == OBD_OBJECT_EOF);
 953         if (oinfo->oi_policy.l_extent.end == OBD_OBJECT_EOF) {
 954                 /* Truncate case */
 955                 oinfo->oi_oa->o_size = oinfo->oi_policy.l_extent.start;
 956         } else if (oinfo->oi_policy.l_extent.end >= oinfo->oi_oa->o_size) {
 957                 oinfo->oi_oa->o_size = oinfo->oi_policy.l_extent.end;
 958         }
 959
 960         la_from_obdo(&info->fti_attr, oinfo->oi_oa,
 961                      OBD_MD_FLMTIME | OBD_MD_FLATIME | OBD_MD_FLCTIME);
 962         info->fti_attr.la_valid &= ~LA_TYPE;
 963         info->fti_attr.la_size = oinfo->oi_policy.l_extent.start;
 964         info->fti_attr.la_valid |= LA_SIZE;
 965
 966         if (oinfo->oi_oa->o_valid & OBD_MD_FLFID) {
 967                 ff = &info->fti_mds_fid;
 968                 ofd_prepare_fidea(ff, oinfo->oi_oa);
 969         }
 970
 971         rc = ofd_object_punch(env, fo, oinfo->oi_policy.l_extent.start,
 972                               oinfo->oi_policy.l_extent.end, &info->fti_attr,
 973                               ff);
 974         if (rc)
 975                 GOTO(out, rc);
 976
 977         res = ldlm_resource_get(ns, NULL, &info->fti_resid, LDLM_EXTENT, 0);
 978         if (res != NULL) {
 979                 ldlm_res_lvbo_update(res, NULL, 0);
 980                 ldlm_resource_putref(res);
 981         }
 982
 983         oinfo->oi_oa->o_valid = OBD_MD_FLID;
 984         /* Quota release needs uid/gid info */
 985         rc = ofd_attr_get(env, fo, &info->fti_attr);
 986         obdo_from_la(oinfo->oi_oa, &info->fti_attr,
 987                      OFD_VALID_FLAGS | LA_UID | LA_GID);
 988         ofd_info2oti(info, oti);
 989
 990         ofd_counter_incr(exp, LPROC_OFD_STATS_PUNCH, oti->oti_jobid, 1);
 991         EXIT;
 992 out:
 993         ofd_object_put(env, fo);
 994 out_env:
 995         return rc;
 996 }
 997
 998 static int ofd_destroy_by_fid(const struct lu_env *env,
 999                               struct ofd_device *ofd,
1000                               const struct lu_fid *fid, int orphan)
1001 {
1002         struct ofd_thread_info  *info = ofd_info(env);
1003         struct lustre_handle     lockh;
1004         __u64                    flags = LDLM_AST_DISCARD_DATA, rc = 0;
1005         ldlm_policy_data_t       policy = {
1006                                         .l_extent = { 0, OBD_OBJECT_EOF }
1007                                  };
1008         struct ofd_object       *fo;
1009
1010         ENTRY;
1011
1012         fo = ofd_object_find(env, ofd, fid);
1013         if (IS_ERR(fo))
1014                 RETURN(PTR_ERR(fo));
1015
1016         /* Tell the clients that the object is gone now and that they should
1017          * throw away any cached pages. */
1018         ofd_build_resid(fid, &info->fti_resid);
1019         rc = ldlm_cli_enqueue_local(ofd->ofd_namespace, &info->fti_resid,
1020                                     LDLM_EXTENT, &policy, LCK_PW, &flags,
1021                                     ldlm_blocking_ast, ldlm_completion_ast,
1022                                     NULL, NULL, 0, LVB_T_NONE, NULL, &lockh);
1023
1024         /* We only care about the side-effects, just drop the lock. */
1025         if (rc == ELDLM_OK)
1026                 ldlm_lock_decref(&lockh, LCK_PW);
1027
1028         LASSERT(fo != NULL);
1029
1030         rc = ofd_object_destroy(env, fo, orphan);
1031
1032         ofd_object_put(env, fo);
1033         RETURN(rc);
1034 }
1035
1036 int ofd_destroy(const struct lu_env *env, struct obd_export *exp,
1037                 struct obdo *oa, struct lov_stripe_md *md,
1038                 struct obd_trans_info *oti, struct obd_export *md_exp,
1039                 void *capa)
1040 {
1041         struct ofd_device       *ofd = ofd_exp(exp);
1042         struct ofd_thread_info  *info;
1043         obd_count                count;
1044         int                      rc = 0;
1045
1046         ENTRY;
1047
1048         info = ofd_info_init(env, exp);
1049         ofd_oti2info(info, oti);
1050
1051         if (!(oa->o_valid & OBD_MD_FLGROUP))
1052                 oa->o_seq = 0;
1053
1054         /* check that o_misc makes sense */
1055         if (oa->o_valid & OBD_MD_FLOBJCOUNT)
1056                 count = oa->o_misc;
1057         else
1058                 count = 1; /* default case - single destroy */
1059
1060         /**
1061          * There can be sequence of objects to destroy. Therefore this request
1062          * may have multiple transaction involved in. It is OK, we need only
1063          * the highest used transno to be reported back in reply but not for
1064          * replays, they must report their transno
1065          */
1066         if (info->fti_transno == 0) /* not replay */
1067                 info->fti_mult_trans = 1;
1068         while (count > 0) {
1069                 int lrc;
1070
1071                 fid_ostid_unpack(&info->fti_fid, &oa->o_oi, 0);
1072                 lrc = ofd_destroy_by_fid(env, ofd, &info->fti_fid, 0);
1073                 if (lrc == -ENOENT) {
1074                         CDEBUG(D_INODE,
1075                                "%s: destroying non-existent object "DFID"\n",
1076                                ofd_obd(ofd)->obd_name, PFID(&info->fti_fid));
1077                         /* rewrite rc with -ENOENT only if it is 0 */
1078                         if (rc == 0)
1079                                 rc = lrc;
1080                 } else if (lrc != 0) {
1081                         CERROR("%s: error destroying object "DFID": %d\n",
1082                                ofd_obd(ofd)->obd_name, PFID(&info->fti_fid),
1083                                rc);
1084                         rc = lrc;
1085                 }
1086                 count--;
1087                 oa->o_id++;
1088         }
1089
1090         /* if we have transaction then there were some deletions, we don't
1091          * need to return ENOENT in that case because it will not wait
1092          * for commit of these deletions. The ENOENT must be returned only
1093          * if there were no transations.
1094          */
1095         if (rc == -ENOENT) {
1096                 if (info->fti_transno != 0)
1097                         rc = 0;
1098         } else if (rc != 0) {
1099                 /*
1100                  * If we have at least one transaction then llog record
1101                  * on server will be removed upon commit, so for rc != 0
1102                  * we return no transno and llog record will be reprocessed.
1103                  */
1104                 info->fti_transno = 0;
1105         }
1106         ofd_info2oti(info, oti);
1107         RETURN(rc);
1108 }
1109
1110 static int ofd_orphans_destroy(const struct lu_env *env,
1111                                struct obd_export *exp, struct ofd_device *ofd,
1112                                struct obdo *oa)
1113 {
1114         struct ofd_thread_info  *info = ofd_info(env);
1115         obd_id                   last;
1116         int                      skip_orphan;
1117         int                      rc = 0;
1118         struct ost_id            oi = oa->o_oi;
1119         struct ofd_seq          *oseq;
1120
1121         ENTRY;
1122
1123         oseq = ofd_seq_get(ofd, oa->o_seq);
1124         if (oseq == NULL) {
1125                 CERROR("%s: Can not find seq for "LPU64":"LPU64"\n",
1126                        ofd_name(ofd), oa->o_seq, oa->o_id);
1127                 RETURN(-EINVAL);
1128         }
1129
1130         LASSERT(exp != NULL);
1131         skip_orphan = !!(exp_connect_flags(exp) & OBD_CONNECT_SKIP_ORPHAN);
1132
1133         last = ofd_seq_last_oid(oseq);
1134         LCONSOLE_INFO("%s: deleting orphan objects from "LPX64":"LPU64
1135                       " to "LPU64"\n", ofd_name(ofd), oa->o_seq,
1136                       oa->o_id + 1, last);
1137
1138         for (oi.oi_id = last; oi.oi_id > oa->o_id; oi.oi_id--) {
1139                 fid_ostid_unpack(&info->fti_fid, &oi, 0);
1140                 rc = ofd_destroy_by_fid(env, ofd, &info->fti_fid, 1);
1141                 if (rc && rc != -ENOENT) /* this is pretty fatal... */
1142                         CEMERG("error destroying precreated id "LPU64": %d\n",
1143                                oi.oi_id, rc);
1144                 if (!skip_orphan) {
1145                         ofd_seq_last_oid_set(oseq, oi.oi_id - 1);
1146                         /* update last_id on disk periodically so that if we
1147                          * restart * we don't need to re-scan all of the just
1148                          * deleted objects. */
1149                         if ((oi.oi_id & 511) == 0)
1150                                 ofd_seq_last_oid_write(env, ofd, oseq);
1151                 }
1152         }
1153         CDEBUG(D_HA, "%s: after destroy: set last_objids["LPU64"] = "LPU64"\n",
1154                ofd_obd(ofd)->obd_name, oa->o_seq, oa->o_id);
1155         if (!skip_orphan) {
1156                 rc = ofd_seq_last_oid_write(env, ofd, oseq);
1157         } else {
1158                 /* don't reuse orphan object, return last used objid */
1159                 oa->o_id = last;
1160                 rc = 0;
1161         }
1162         ofd_seq_put(env, oseq);
1163         RETURN(rc);
1164 }
1165
1166 int ofd_create(const struct lu_env *env, struct obd_export *exp,
1167                struct obdo *oa, struct lov_stripe_md **ea,
1168                struct obd_trans_info *oti)
1169 {
1170         struct ofd_device       *ofd = ofd_exp(exp);
1171         struct ofd_thread_info  *info;
1172         obd_seq                 seq = oa->o_seq;
1173         struct ofd_seq          *oseq;
1174         int                     rc = 0, diff;
1175         int                     sync_trans = 0;
1176
1177         ENTRY;
1178
1179         info = ofd_info_init(env, exp);
1180         ofd_oti2info(info, oti);
1181
1182         LASSERT(oa->o_seq >= FID_SEQ_OST_MDT0);
1183         LASSERT(oa->o_valid & OBD_MD_FLGROUP);
1184
1185         CDEBUG(D_INFO, "ofd_create(oa->o_seq="LPU64",oa->o_id="LPU64")\n",
1186                seq, oa->o_id);
1187
1188         oseq = ofd_seq_load(env, ofd, seq);
1189         if (IS_ERR(oseq)) {
1190                 CERROR("%s: Can't find FID Sequence "LPX64": rc = %ld\n",
1191                        ofd_name(ofd), seq, PTR_ERR(oseq));
1192                 RETURN(-EINVAL);
1193         }
1194
1195         if ((oa->o_valid & OBD_MD_FLFLAGS) &&
1196             (oa->o_flags & OBD_FL_RECREATE_OBJS)) {
1197                 if (!ofd_obd(ofd)->obd_recovering ||
1198                     oa->o_id > ofd_seq_last_oid(oseq)) {
1199                         CERROR("recreate objid "LPU64" > last id "LPU64"\n",
1200                                         oa->o_id, ofd_seq_last_oid(oseq));
1201                         GOTO(out_nolock, rc = -EINVAL);
1202                 }
1203                 /* do nothing because we create objects during first write */
1204                 GOTO(out_nolock, rc = 0);
1205         }
1206         /* former ofd_handle_precreate */
1207         if ((oa->o_valid & OBD_MD_FLFLAGS) &&
1208             (oa->o_flags & OBD_FL_DELORPHAN)) {
1209                 /* destroy orphans */
1210                 if (oti->oti_conn_cnt < exp->exp_conn_cnt) {
1211                         CERROR("%s: dropping old orphan cleanup request\n",
1212                                ofd_name(ofd));
1213                         GOTO(out_nolock, rc = 0);
1214                 }
1215                 /* This causes inflight precreates to abort and drop lock */
1216                 oseq->os_destroys_in_progress = 1;
1217                 mutex_lock(&oseq->os_create_lock);
1218                 if (!oseq->os_destroys_in_progress) {
1219                         CERROR("%s:["LPU64"] destroys_in_progress already cleared\n",
1220                                exp->exp_obd->obd_name, oa->o_seq);
1221                         GOTO(out, rc = 0);
1222                 }
1223                 diff = oa->o_id - ofd_seq_last_oid(oseq);
1224                 CDEBUG(D_HA, "ofd_last_id() = "LPU64" -> diff = %d\n",
1225                         ofd_seq_last_oid(oseq), diff);
1226                 if (-diff > OST_MAX_PRECREATE) {
1227                         /* FIXME: should reset precreate_next_id on MDS */
1228                         rc = 0;
1229                 } else if (diff < 0) {
1230                         rc = ofd_orphans_destroy(env, exp, ofd, oa);
1231                         oseq->os_destroys_in_progress = 0;
1232                 } else {
1233                         /* XXX: Used by MDS for the first time! */
1234                         oseq->os_destroys_in_progress = 0;
1235                 }
1236         } else {
1237                 mutex_lock(&oseq->os_create_lock);
1238                 if (oti->oti_conn_cnt < exp->exp_conn_cnt) {
1239                         CERROR("%s: dropping old precreate request\n",
1240                                 ofd_obd(ofd)->obd_name);
1241                         GOTO(out, rc = 0);
1242                 }
1243                 /* only precreate if seq is 0, IDIF or normal and also o_id
1244                  * must be specfied */
1245                 if ((!fid_seq_is_mdt(oa->o_seq) &&
1246                      !fid_seq_is_norm(oa->o_seq) &&
1247                      !fid_seq_is_idif(oa->o_seq)) || oa->o_id == 0) {
1248                         diff = 1; /* shouldn't we create this right now? */
1249                 } else {
1250                         diff = oa->o_id - ofd_seq_last_oid(oseq);
1251                         /* Do sync create if the seq is about to used up */
1252                         if (fid_seq_is_idif(oa->o_seq) ||
1253                             fid_seq_is_mdt0(oa->o_seq)) {
1254                                 if (unlikely(oa->o_id >= IDIF_MAX_OID - 1))
1255                                         sync_trans = 1;
1256                         } else if (fid_seq_is_norm(oa->o_seq)) {
1257                                 if (unlikely(oa->o_id >=
1258                                              LUSTRE_DATA_SEQ_MAX_WIDTH - 1))
1259                                         sync_trans = 1;
1260                         } else {
1261                                 CERROR("%s : invalid o_seq "LPX64": rc = %d\n",
1262                                        ofd_name(ofd), oa->o_seq, -EINVAL);
1263                                 GOTO(out, rc = -EINVAL);
1264                         }
1265                 }
1266         }
1267         if (diff > 0) {
1268                 cfs_time_t       enough_time = cfs_time_shift(DISK_TIMEOUT);
1269                 obd_id           next_id;
1270                 int              created = 0;
1271                 int              count;
1272
1273                 if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
1274                     !(oa->o_flags & OBD_FL_DELORPHAN)) {
1275                         /* don't enforce grant during orphan recovery */
1276                         rc = ofd_grant_create(env,
1277                                               ofd_obd(ofd)->obd_self_export,
1278                                               &diff);
1279                         if (rc) {
1280                                 CDEBUG(D_HA, "%s: failed to acquire grant space"
1281                                        "for precreate (%d)\n",
1282                                        ofd_name(ofd), diff);
1283                                 diff = 0;
1284                         }
1285                 }
1286
1287                 while (diff > 0) {
1288                         next_id = ofd_seq_last_oid(oseq) + 1;
1289                         count = ofd_precreate_batch(ofd, diff);
1290
1291                         CDEBUG(D_HA, "%s: reserve %d objects in group "LPU64
1292                                " at "LPU64"\n", ofd_obd(ofd)->obd_name,
1293                                count, oa->o_seq, next_id);
1294
1295                         if (cfs_time_after(jiffies, enough_time)) {
1296                                 LCONSOLE_WARN("%s: Slow creates, %d/%d objects"
1297                                               " created at a rate of %d/s\n",
1298                                               ofd_obd(ofd)->obd_name,
1299                                               created, diff + created,
1300                                               created / DISK_TIMEOUT);
1301                                 break;
1302                         }
1303
1304                         rc = ofd_precreate_objects(env, ofd, next_id,
1305                                                    oseq, count, sync_trans);
1306                         if (rc > 0) {
1307                                 created += rc;
1308                                 diff -= rc;
1309                         } else if (rc < 0) {
1310                                 break;
1311                         }
1312                 }
1313                 if (created > 0)
1314                         /* some objects got created, we can return
1315                          * them, even if last creation failed */
1316                         rc = 0;
1317                 else
1318                         CERROR("unable to precreate: %d\n", rc);
1319
1320                 oa->o_id = ofd_seq_last_oid(oseq);
1321                 oa->o_valid |= OBD_MD_FLID | OBD_MD_FLGROUP;
1322
1323                 if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
1324                     !(oa->o_flags & OBD_FL_DELORPHAN))
1325                         ofd_grant_commit(env, ofd_obd(ofd)->obd_self_export,
1326                                          rc);
1327         }
1328
1329         ofd_info2oti(info, oti);
1330 out:
1331         mutex_unlock(&oseq->os_create_lock);
1332 out_nolock:
1333         if (rc == 0 && ea != NULL) {
1334                 struct lov_stripe_md *lsm = *ea;
1335
1336                 lsm->lsm_object_id = oa->o_id;
1337         }
1338         ofd_seq_put(env, oseq);
1339         RETURN(rc);
1340 }
1341
1342 int ofd_getattr(const struct lu_env *env, struct obd_export *exp,
1343                 struct obd_info *oinfo)
1344 {
1345         struct ofd_device       *ofd = ofd_exp(exp);
1346         struct ofd_thread_info  *info;
1347         struct ofd_object       *fo;
1348         __u64                    curr_version;
1349         int                      rc = 0;
1350
1351         ENTRY;
1352
1353         info = ofd_info_init(env, exp);
1354
1355         fid_ostid_unpack(&info->fti_fid, &oinfo->oi_oa->o_oi, 0);
1356         rc = ofd_auth_capa(exp, &info->fti_fid, oinfo->oi_oa->o_seq,
1357                            oinfo_capa(oinfo), CAPA_OPC_META_READ);
1358         if (rc)
1359                 GOTO(out, rc);
1360
1361         fo = ofd_object_find(env, ofd, &info->fti_fid);
1362         if (IS_ERR(fo))
1363                 GOTO(out, rc = PTR_ERR(fo));
1364         LASSERT(fo != NULL);
1365         rc = ofd_attr_get(env, fo, &info->fti_attr);
1366         oinfo->oi_oa->o_valid = OBD_MD_FLID;
1367         if (rc == 0)
1368                 obdo_from_la(oinfo->oi_oa, &info->fti_attr,
1369                              OFD_VALID_FLAGS | LA_UID | LA_GID);
1370
1371         /* Store object version in reply */
1372         curr_version = dt_version_get(env, ofd_object_child(fo));
1373         if ((__s64)curr_version != -EOPNOTSUPP) {
1374                 oinfo->oi_oa->o_valid |= OBD_MD_FLDATAVERSION;
1375                 oinfo->oi_oa->o_data_version = curr_version;
1376         }
1377         ofd_object_put(env, fo);
1378 out:
1379         RETURN(rc);
1380 }
1381
1382 static int ofd_sync(const struct lu_env *env, struct obd_export *exp,
1383                     struct obd_info *oinfo, obd_size start, obd_size end,
1384                     struct ptlrpc_request_set *set)
1385 {
1386         struct ofd_device       *ofd = ofd_exp(exp);
1387         struct ofd_thread_info  *info;
1388         struct ofd_object       *fo;
1389         int                      rc = 0;
1390
1391         ENTRY;
1392
1393         /* if no objid is specified, it means "sync whole filesystem" */
1394         if (oinfo->oi_oa == NULL || !(oinfo->oi_oa->o_valid & OBD_MD_FLID)) {
1395                 rc = dt_sync(env, ofd->ofd_osd);
1396                 GOTO(out, rc);
1397         }
1398
1399         info = ofd_info_init(env, exp);
1400         fid_ostid_unpack(&info->fti_fid, &oinfo->oi_oa->o_oi, 0);
1401
1402         rc = ofd_auth_capa(exp, &info->fti_fid, oinfo->oi_oa->o_seq,
1403                            oinfo_capa(oinfo), CAPA_OPC_OSS_TRUNC);
1404         if (rc)
1405                 GOTO(out, rc);
1406
1407         fo = ofd_object_find(env, ofd, &info->fti_fid);
1408         if (IS_ERR(fo)) {
1409                 CERROR("%s: error finding object "DFID": rc = %ld\n",
1410                        exp->exp_obd->obd_name, PFID(&info->fti_fid),
1411                        PTR_ERR(fo));
1412                 GOTO(out, rc = PTR_ERR(fo));
1413         }
1414
1415         if (!ofd_object_exists(fo))
1416                 GOTO(put, rc = -ENOENT);
1417
1418         if (dt_version_get(env, ofd_object_child(fo)) >
1419             ofd_obd(ofd)->obd_last_committed) {
1420                 rc = dt_object_sync(env, ofd_object_child(fo));
1421                 if (rc)
1422                         GOTO(put, rc);
1423         }
1424
1425         oinfo->oi_oa->o_valid = OBD_MD_FLID;
1426         rc = ofd_attr_get(env, fo, &info->fti_attr);
1427         obdo_from_la(oinfo->oi_oa, &info->fti_attr, OFD_VALID_FLAGS);
1428
1429         ofd_counter_incr(exp, LPROC_OFD_STATS_SYNC, oinfo->oi_jobid, 1);
1430         EXIT;
1431 put:
1432         ofd_object_put(env, fo);
1433 out:
1434         return rc;
1435 }
1436
1437 int ofd_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1438                   void *karg, void *uarg)
1439 {
1440         struct lu_env            env;
1441         struct ofd_device       *ofd = ofd_exp(exp);
1442         struct obd_device       *obd = ofd_obd(ofd);
1443         int                      rc;
1444
1445         ENTRY;
1446
1447         CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
1448         rc = lu_env_init(&env, LCT_LOCAL);
1449         if (rc)
1450                 RETURN(rc);
1451
1452         switch (cmd) {
1453         case OBD_IOC_ABORT_RECOVERY:
1454                 CERROR("%s: aborting recovery\n", obd->obd_name);
1455                 target_stop_recovery_thread(obd);
1456                 break;
1457         case OBD_IOC_SYNC:
1458                 CDEBUG(D_RPCTRACE, "syncing ost %s\n", obd->obd_name);
1459                 rc = dt_sync(&env, ofd->ofd_osd);
1460                 break;
1461         case OBD_IOC_SET_READONLY:
1462                 rc = dt_sync(&env, ofd->ofd_osd);
1463                 if (rc == 0)
1464                         rc = dt_ro(&env, ofd->ofd_osd);
1465                 break;
1466         default:
1467                 CERROR("%s: not supported cmd = %d\n", obd->obd_name, cmd);
1468                 rc = -ENOTTY;
1469         }
1470
1471         lu_env_fini(&env);
1472         RETURN(rc);
1473 }
1474
1475 static int ofd_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
1476 {
1477         int rc = 0;
1478
1479         ENTRY;
1480
1481         switch(stage) {
1482         case OBD_CLEANUP_EARLY:
1483                 break;
1484         case OBD_CLEANUP_EXPORTS:
1485                 target_cleanup_recovery(obd);
1486                 break;
1487         }
1488         RETURN(rc);
1489 }
1490
1491 static int ofd_ping(const struct lu_env *env, struct obd_export *exp)
1492 {
1493         ofd_fmd_expire(exp);
1494         return 0;
1495 }
1496
1497 static int ofd_health_check(const struct lu_env *nul, struct obd_device *obd)
1498 {
1499         struct ofd_device       *ofd = ofd_dev(obd->obd_lu_dev);
1500         struct ofd_thread_info  *info;
1501         struct lu_env            env;
1502 #ifdef USE_HEALTH_CHECK_WRITE
1503         struct thandle          *th;
1504 #endif
1505         int                      rc = 0;
1506
1507         /* obd_proc_read_health pass NULL env, we need real one */
1508         rc = lu_env_init(&env, LCT_DT_THREAD);
1509         if (rc)
1510                 RETURN(rc);
1511
1512         info = ofd_info_init(&env, NULL);
1513         rc = dt_statfs(&env, ofd->ofd_osd, &info->fti_u.osfs);
1514         if (unlikely(rc))
1515                 GOTO(out, rc);
1516
1517         if (info->fti_u.osfs.os_state == OS_STATE_READONLY)
1518                 GOTO(out, rc = -EROFS);
1519
1520 #ifdef USE_HEALTH_CHECK_WRITE
1521         OBD_ALLOC(info->fti_buf.lb_buf, CFS_PAGE_SIZE);
1522         if (info->fti_buf.lb_buf == NULL)
1523                 GOTO(out, rc = -ENOMEM);
1524
1525         info->fti_buf.lb_len = CFS_PAGE_SIZE;
1526         info->fti_off = 0;
1527
1528         th = dt_trans_create(&env, ofd->ofd_osd);
1529         if (IS_ERR(th))
1530                 GOTO(out, rc = PTR_ERR(th));
1531
1532         rc = dt_declare_record_write(&env, ofd->ofd_health_check_file,
1533                                      info->fti_buf.lb_len, info->fti_off, th);
1534         if (rc == 0) {
1535                 th->th_sync = 1; /* sync IO is needed */
1536                 rc = dt_trans_start_local(&env, ofd->ofd_osd, th);
1537                 if (rc == 0)
1538                         rc = dt_record_write(&env, ofd->ofd_health_check_file,
1539                                              &info->fti_buf, &info->fti_off,
1540                                              th);
1541         }
1542         dt_trans_stop(&env, ofd->ofd_osd, th);
1543
1544         OBD_FREE(info->fti_buf.lb_buf, CFS_PAGE_SIZE);
1545
1546         CDEBUG(D_INFO, "write 1 page synchronously for checking io rc %d\n",rc);
1547 #endif
1548 out:
1549         lu_env_fini(&env);
1550         return !!rc;
1551 }
1552
1553 /*
1554  * Handle quota control requests to consult current usage/limit.
1555  *
1556  * \param obd - is the obd device associated with the ofd
1557  * \param exp - is the client's export
1558  * \param oqctl - is the obd_quotactl request to be processed
1559  */
1560 static int ofd_quotactl(struct obd_device *obd, struct obd_export *exp,
1561                         struct obd_quotactl *oqctl)
1562 {
1563         struct ofd_device  *ofd = ofd_dev(obd->obd_lu_dev);
1564         struct lu_env       env;
1565         int                 rc;
1566         ENTRY;
1567
1568         /* report success for quota on/off for interoperability with current MDT
1569          * stack */
1570         if (oqctl->qc_cmd == Q_QUOTAON || oqctl->qc_cmd == Q_QUOTAOFF)
1571                 RETURN(0);
1572
1573         rc = lu_env_init(&env, LCT_DT_THREAD);
1574         if (rc)
1575                 RETURN(rc);
1576
1577         rc = lquotactl_slv(&env, ofd->ofd_osd, oqctl);
1578         lu_env_fini(&env);
1579
1580         RETURN(rc);
1581 }
1582
1583 struct obd_ops ofd_obd_ops = {
1584         .o_owner                = THIS_MODULE,
1585         .o_connect              = ofd_obd_connect,
1586         .o_reconnect            = ofd_obd_reconnect,
1587         .o_disconnect           = ofd_obd_disconnect,
1588         .o_set_info_async       = ofd_set_info_async,
1589         .o_get_info             = ofd_get_info,
1590         .o_create               = ofd_create,
1591         .o_statfs               = ofd_statfs,
1592         .o_setattr              = ofd_setattr,
1593         .o_preprw               = ofd_preprw,
1594         .o_commitrw             = ofd_commitrw,
1595         .o_destroy              = ofd_destroy,
1596         .o_init_export          = ofd_init_export,
1597         .o_destroy_export       = ofd_destroy_export,
1598         .o_postrecov            = ofd_obd_postrecov,
1599         .o_punch                = ofd_punch,
1600         .o_getattr              = ofd_getattr,
1601         .o_sync                 = ofd_sync,
1602         .o_iocontrol            = ofd_iocontrol,
1603         .o_precleanup           = ofd_precleanup,
1604         .o_ping                 = ofd_ping,
1605         .o_health_check         = ofd_health_check,
1606         .o_quotactl             = ofd_quotactl,
1607 };