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 <lustre/lustre_idl.h>
  46 #include "ofd_internal.h"
  47 #include <obd_cksum.h>
  48 #include <lustre_quota.h>
  49
  50 static int ofd_export_stats_init(struct ofd_device *ofd,
  51                                  struct obd_export *exp, void *client_nid)
  52 {
  53         struct obd_device       *obd = ofd_obd(ofd);
  54         struct nid_stat         *stats;
  55         int                      num_stats;
  56         int                      rc, newnid = 0;
  57
  58         ENTRY;
  59
  60         if (obd_uuid_equals(&exp->exp_client_uuid, &obd->obd_uuid))
  61                 /* Self-export gets no proc entry */
  62                 RETURN(0);
  63
  64         rc = lprocfs_exp_setup(exp, client_nid, &newnid);
  65         if (rc) {
  66                 /* Mask error for already created
  67                  * /proc entries */
  68                 if (rc == -EALREADY)
  69                         rc = 0;
  70                 RETURN(rc);
  71         }
  72
  73         if (newnid == 0)
  74                 RETURN(0);
  75
  76         stats = exp->exp_nid_stats;
  77         LASSERT(stats != NULL);
  78
  79         num_stats = (sizeof(*obd->obd_type->typ_dt_ops) / sizeof(void *)) +
  80                      LPROC_OFD_LAST - 1;
  81
  82         stats->nid_stats = lprocfs_alloc_stats(num_stats,
  83                                                LPROCFS_STATS_FLAG_NOPERCPU);
  84         if (stats->nid_stats == NULL)
  85                 return -ENOMEM;
  86
  87         lprocfs_init_ops_stats(LPROC_OFD_LAST, stats->nid_stats);
  88         lprocfs_counter_init(stats->nid_stats, LPROC_OFD_READ_BYTES,
  89                              LPROCFS_CNTR_AVGMINMAX, "read_bytes", "bytes");
  90         lprocfs_counter_init(stats->nid_stats, LPROC_OFD_WRITE_BYTES,
  91                              LPROCFS_CNTR_AVGMINMAX, "write_bytes", "bytes");
  92
  93         rc = lprocfs_register_stats(stats->nid_proc, "stats",
  94                                     stats->nid_stats);
  95         if (rc)
  96                 GOTO(clean, rc);
  97
  98         rc = lprocfs_nid_ldlm_stats_init(stats);
  99         if (rc) {
 100                 lprocfs_free_stats(&stats->nid_stats);
 101                 GOTO(clean, rc);
 102         }
 103
 104         RETURN(0);
 105 clean:
 106         return rc;
 107 }
 108
 109 static int ofd_parse_connect_data(const struct lu_env *env,
 110                                   struct obd_export *exp,
 111                                   struct obd_connect_data *data)
 112 {
 113         struct ofd_device                *ofd = ofd_exp(exp);
 114         struct filter_export_data        *fed = &exp->exp_filter_data;
 115
 116         if (!data)
 117                 RETURN(0);
 118
 119         CDEBUG(D_RPCTRACE, "%s: cli %s/%p ocd_connect_flags: "LPX64
 120                " ocd_version: %x ocd_grant: %d ocd_index: %u"
 121                " ocd_group %u\n",
 122                exp->exp_obd->obd_name, exp->exp_client_uuid.uuid, exp,
 123                data->ocd_connect_flags, data->ocd_version,
 124                data->ocd_grant, data->ocd_index, data->ocd_group);
 125
 126         if (fed->fed_group != 0 && fed->fed_group != data->ocd_group) {
 127                 CWARN("!!! This export (nid %s) used object group %d "
 128                       "earlier; now it's trying to use group %d!  This could "
 129                       "be a bug in the MDS. Please report to "
 130                       "http://bugs.whamcloud.com/\n",
 131                       obd_export_nid2str(exp), fed->fed_group,
 132                       data->ocd_group);
 133                 RETURN(-EPROTO);
 134         }
 135         fed->fed_group = data->ocd_group;
 136
 137         data->ocd_connect_flags &= OST_CONNECT_SUPPORTED;
 138         exp->exp_connect_data = *data;
 139         data->ocd_version = LUSTRE_VERSION_CODE;
 140
 141         /* Kindly make sure the SKIP_ORPHAN flag is from MDS. */
 142         if (data->ocd_connect_flags & OBD_CONNECT_MDS)
 143                 CDEBUG(D_HA, "%s: Received MDS connection for group %u\n",
 144                        exp->exp_obd->obd_name, data->ocd_group);
 145         else if (data->ocd_connect_flags & OBD_CONNECT_SKIP_ORPHAN)
 146                 RETURN(-EPROTO);
 147
 148         if (ofd_grant_param_supp(exp)) {
 149                 exp->exp_filter_data.fed_pagesize = data->ocd_blocksize;
 150                 /* ocd_{blocksize,inodespace} are log2 values */
 151                 data->ocd_blocksize  = ofd->ofd_blockbits;
 152                 data->ocd_inodespace = ofd->ofd_dt_conf.ddp_inodespace;
 153                 /* ocd_grant_extent is in 1K blocks */
 154                 data->ocd_grant_extent = ofd->ofd_dt_conf.ddp_grant_frag >> 10;
 155         }
 156
 157         if (exp_connect_flags(exp) & OBD_CONNECT_GRANT)
 158                 data->ocd_grant = ofd_grant_connect(env, exp, data->ocd_grant);
 159
 160         if (data->ocd_connect_flags & OBD_CONNECT_INDEX) {
 161                 struct lr_server_data *lsd = &ofd->ofd_lut.lut_lsd;
 162                 int                    index = lsd->lsd_osd_index;
 163
 164                 if (index != data->ocd_index) {
 165                         LCONSOLE_ERROR_MSG(0x136, "Connection from %s to index"
 166                                            " %u doesn't match actual OST index"
 167                                            " %u in last_rcvd file, bad "
 168                                            "configuration?\n",
 169                                            obd_export_nid2str(exp), index,
 170                                            data->ocd_index);
 171                         RETURN(-EBADF);
 172                 }
 173                 if (!(lsd->lsd_feature_compat & OBD_COMPAT_OST)) {
 174                         /* this will only happen on the first connect */
 175                         lsd->lsd_feature_compat |= OBD_COMPAT_OST;
 176                         /* sync is not needed here as lut_client_add will
 177                          * set exp_need_sync flag */
 178                         tgt_server_data_update(env, &ofd->ofd_lut, 0);
 179                 }
 180         }
 181         if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_SIZE)) {
 182                 data->ocd_brw_size = 65536;
 183         } else if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE) {
 184                 data->ocd_brw_size = min(data->ocd_brw_size,
 185                                          (__u32)DT_MAX_BRW_SIZE);
 186                 if (data->ocd_brw_size == 0) {
 187                         CERROR("%s: cli %s/%p ocd_connect_flags: "LPX64
 188                                " ocd_version: %x ocd_grant: %d ocd_index: %u "
 189                                "ocd_brw_size is unexpectedly zero, "
 190                                "network data corruption?"
 191                                "Refusing connection of this client\n",
 192                                exp->exp_obd->obd_name,
 193                                exp->exp_client_uuid.uuid,
 194                                exp, data->ocd_connect_flags, data->ocd_version,
 195                                data->ocd_grant, data->ocd_index);
 196                         RETURN(-EPROTO);
 197                 }
 198         }
 199
 200         if (data->ocd_connect_flags & OBD_CONNECT_CKSUM) {
 201                 __u32 cksum_types = data->ocd_cksum_types;
 202
 203                 /* The client set in ocd_cksum_types the checksum types it
 204                  * supports. We have to mask off the algorithms that we don't
 205                  * support */
 206                 data->ocd_cksum_types &= cksum_types_supported_server();
 207
 208                 if (unlikely(data->ocd_cksum_types == 0)) {
 209                         CERROR("%s: Connect with checksum support but no "
 210                                "ocd_cksum_types is set\n",
 211                                exp->exp_obd->obd_name);
 212                         RETURN(-EPROTO);
 213                 }
 214
 215                 CDEBUG(D_RPCTRACE, "%s: cli %s supports cksum type %x, return "
 216                        "%x\n", exp->exp_obd->obd_name, obd_export_nid2str(exp),
 217                        cksum_types, data->ocd_cksum_types);
 218         } else {
 219                 /* This client does not support OBD_CONNECT_CKSUM
 220                  * fall back to CRC32 */
 221                 CDEBUG(D_RPCTRACE, "%s: cli %s does not support "
 222                        "OBD_CONNECT_CKSUM, CRC32 will be used\n",
 223                        exp->exp_obd->obd_name, obd_export_nid2str(exp));
 224         }
 225
 226         if (data->ocd_connect_flags & OBD_CONNECT_MAXBYTES)
 227                 data->ocd_maxbytes = ofd->ofd_dt_conf.ddp_maxbytes;
 228
 229         if (data->ocd_connect_flags & OBD_CONNECT_PINGLESS) {
 230                 if (suppress_pings) {
 231                         spin_lock(&exp->exp_obd->obd_dev_lock);
 232                         list_del_init(&exp->exp_obd_chain_timed);
 233                         spin_unlock(&exp->exp_obd->obd_dev_lock);
 234                 } else {
 235                         data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
 236                 }
 237         }
 238
 239         RETURN(0);
 240 }
 241
 242 static int ofd_obd_reconnect(const struct lu_env *env, struct obd_export *exp,
 243                              struct obd_device *obd, struct obd_uuid *cluuid,
 244                              struct obd_connect_data *data, void *localdata)
 245 {
 246         struct ofd_device       *ofd;
 247         int                      rc;
 248
 249         ENTRY;
 250
 251         if (exp == NULL || obd == NULL || cluuid == NULL)
 252                 RETURN(-EINVAL);
 253
 254         ofd = ofd_dev(obd->obd_lu_dev);
 255
 256         rc = lu_env_refill((struct lu_env *)env);
 257         if (rc != 0) {
 258                 CERROR("Failure to refill session: '%d'\n", rc);
 259                 RETURN(rc);
 260         }
 261
 262         ofd_info_init(env, exp);
 263         rc = ofd_parse_connect_data(env, exp, data);
 264         if (rc == 0)
 265                 ofd_export_stats_init(ofd, exp, localdata);
 266
 267         RETURN(rc);
 268 }
 269
 270 static int ofd_obd_connect(const struct lu_env *env, struct obd_export **_exp,
 271                            struct obd_device *obd, struct obd_uuid *cluuid,
 272                            struct obd_connect_data *data, void *localdata)
 273 {
 274         struct obd_export       *exp;
 275         struct ofd_device       *ofd;
 276         struct lustre_handle     conn = { 0 };
 277         int                      rc;
 278         ENTRY;
 279
 280         if (_exp == NULL || obd == NULL || cluuid == NULL)
 281                 RETURN(-EINVAL);
 282
 283         ofd = ofd_dev(obd->obd_lu_dev);
 284
 285         rc = class_connect(&conn, obd, cluuid);
 286         if (rc)
 287                 RETURN(rc);
 288
 289         exp = class_conn2export(&conn);
 290         LASSERT(exp != NULL);
 291
 292         rc = lu_env_refill((struct lu_env *)env);
 293         if (rc != 0) {
 294                 CERROR("Failure to refill session: '%d'\n", rc);
 295                 GOTO(out, rc);
 296         }
 297
 298         ofd_info_init(env, exp);
 299
 300         rc = ofd_parse_connect_data(env, exp, data);
 301         if (rc)
 302                 GOTO(out, rc);
 303
 304         if (obd->obd_replayable) {
 305                 struct tg_export_data *ted = &exp->exp_target_data;
 306
 307                 memcpy(ted->ted_lcd->lcd_uuid, cluuid,
 308                        sizeof(ted->ted_lcd->lcd_uuid));
 309                 rc = tgt_client_new(env, exp);
 310                 if (rc != 0)
 311                         GOTO(out, rc);
 312                 ofd_export_stats_init(ofd, exp, localdata);
 313         }
 314
 315         CDEBUG(D_HA, "%s: get connection from MDS %d\n", obd->obd_name,
 316                data->ocd_group);
 317
 318 out:
 319         if (rc != 0) {
 320                 class_disconnect(exp);
 321                 *_exp = NULL;
 322         } else {
 323                 *_exp = exp;
 324         }
 325         RETURN(rc);
 326 }
 327
 328 static int ofd_obd_disconnect(struct obd_export *exp)
 329 {
 330         struct ofd_device       *ofd = ofd_dev(exp->exp_obd->obd_lu_dev);
 331         struct lu_env            env;
 332         int                      rc;
 333
 334         ENTRY;
 335
 336         LASSERT(exp);
 337         class_export_get(exp);
 338
 339         if (!(exp->exp_flags & OBD_OPT_FORCE))
 340                 ofd_grant_sanity_check(ofd_obd(ofd), __FUNCTION__);
 341
 342         rc = server_disconnect_export(exp);
 343
 344         ofd_grant_discard(exp);
 345
 346         rc = lu_env_init(&env, LCT_DT_THREAD);
 347         if (rc)
 348                 RETURN(rc);
 349
 350         /* Do not erase record for recoverable client. */
 351         if (exp->exp_obd->obd_replayable &&
 352             (!exp->exp_obd->obd_fail || exp->exp_failed))
 353                 tgt_client_del(&env, exp);
 354         lu_env_fini(&env);
 355
 356         class_export_put(exp);
 357         RETURN(rc);
 358 }
 359
 360 static int ofd_init_export(struct obd_export *exp)
 361 {
 362         int rc;
 363
 364         spin_lock_init(&exp->exp_filter_data.fed_lock);
 365         CFS_INIT_LIST_HEAD(&exp->exp_filter_data.fed_mod_list);
 366         spin_lock(&exp->exp_lock);
 367         exp->exp_connecting = 1;
 368         spin_unlock(&exp->exp_lock);
 369
 370         /* self-export doesn't need client data and ldlm initialization */
 371         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
 372                                      &exp->exp_client_uuid)))
 373                 return 0;
 374
 375         rc = tgt_client_alloc(exp);
 376         if (rc == 0)
 377                 ldlm_init_export(exp);
 378         if (rc)
 379                 CERROR("%s: Can't initialize export: rc %d\n",
 380                        exp->exp_obd->obd_name, rc);
 381         return rc;
 382 }
 383
 384 static int ofd_destroy_export(struct obd_export *exp)
 385 {
 386         struct ofd_device *ofd = ofd_dev(exp->exp_obd->obd_lu_dev);
 387
 388         if (exp->exp_filter_data.fed_pending)
 389                 CERROR("%s: cli %s/%p has %lu pending on destroyed export"
 390                        "\n", exp->exp_obd->obd_name, exp->exp_client_uuid.uuid,
 391                        exp, exp->exp_filter_data.fed_pending);
 392
 393         target_destroy_export(exp);
 394
 395         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
 396                                      &exp->exp_client_uuid)))
 397                 return 0;
 398
 399         ldlm_destroy_export(exp);
 400         tgt_client_free(exp);
 401
 402         ofd_fmd_cleanup(exp);
 403
 404         /*
 405          * discard grants once we're sure no more
 406          * interaction with the client is possible
 407          */
 408         ofd_grant_discard(exp);
 409         ofd_fmd_cleanup(exp);
 410
 411         if (exp_connect_flags(exp) & OBD_CONNECT_GRANT_SHRINK) {
 412                 if (ofd->ofd_tot_granted_clients > 0)
 413                         ofd->ofd_tot_granted_clients --;
 414         }
 415
 416         if (!(exp->exp_flags & OBD_OPT_FORCE))
 417                 ofd_grant_sanity_check(exp->exp_obd, __FUNCTION__);
 418
 419         LASSERT(cfs_list_empty(&exp->exp_filter_data.fed_mod_list));
 420         return 0;
 421 }
 422
 423 int ofd_postrecov(const struct lu_env *env, struct ofd_device *ofd)
 424 {
 425         struct lu_device *ldev = &ofd->ofd_dt_dev.dd_lu_dev;
 426
 427         CDEBUG(D_HA, "%s: recovery is over\n", ofd_obd(ofd)->obd_name);
 428         return ldev->ld_ops->ldo_recovery_complete(env, ldev);
 429 }
 430
 431 int ofd_obd_postrecov(struct obd_device *obd)
 432 {
 433         struct lu_env            env;
 434         struct lu_device        *ldev = obd->obd_lu_dev;
 435         int                      rc;
 436
 437         ENTRY;
 438
 439         rc = lu_env_init(&env, LCT_DT_THREAD);
 440         if (rc)
 441                 RETURN(rc);
 442         ofd_info_init(&env, obd->obd_self_export);
 443
 444         rc = ofd_postrecov(&env, ofd_dev(ldev));
 445
 446         lu_env_fini(&env);
 447         RETURN(rc);
 448 }
 449
 450 static int ofd_adapt_sptlrpc_conf(const struct lu_env *env,
 451                                   struct obd_device *obd, int initial)
 452 {
 453         struct filter_obd       *fo = &obd->u.filter;
 454         struct sptlrpc_rule_set  tmp_rset;
 455         int                      rc;
 456
 457         sptlrpc_rule_set_init(&tmp_rset);
 458         rc = sptlrpc_conf_target_get_rules(obd, &tmp_rset, initial);
 459         if (rc) {
 460                 CERROR("%s: failed get sptlrpc rules: rc = %d\n",
 461                        obd->obd_name, rc);
 462                 return rc;
 463         }
 464
 465         sptlrpc_target_update_exp_flavor(obd, &tmp_rset);
 466
 467         write_lock(&fo->fo_sptlrpc_lock);
 468         sptlrpc_rule_set_free(&fo->fo_sptlrpc_rset);
 469         fo->fo_sptlrpc_rset = tmp_rset;
 470         write_unlock(&fo->fo_sptlrpc_lock);
 471
 472         return 0;
 473 }
 474
 475 static int ofd_set_mds_conn(struct obd_export *exp, void *val)
 476 {
 477         int rc = 0;
 478
 479         ENTRY;
 480
 481         LCONSOLE_WARN("%s: received MDS connection from %s\n",
 482                       exp->exp_obd->obd_name, obd_export_nid2str(exp));
 483         RETURN(rc);
 484 }
 485
 486 static int ofd_set_info_async(const struct lu_env *env, struct obd_export *exp,
 487                               __u32 keylen, void *key, __u32 vallen, void *val,
 488                               struct ptlrpc_request_set *set)
 489 {
 490         struct ofd_device       *ofd;
 491         int                      rc = 0;
 492
 493         ENTRY;
 494
 495         if (exp->exp_obd == NULL) {
 496                 CDEBUG(D_IOCTL, "invalid export %p\n", exp);
 497                 RETURN(-EINVAL);
 498         }
 499
 500         ofd = ofd_exp(exp);
 501
 502         if (KEY_IS(KEY_CAPA_KEY)) {
 503                 rc = ofd_update_capa_key(ofd, val);
 504                 if (rc)
 505                         CERROR("%s: update capability key failed: rc = %d\n",
 506                                exp->exp_obd->obd_name, rc);
 507         } else if (KEY_IS(KEY_SPTLRPC_CONF)) {
 508                 ofd_adapt_sptlrpc_conf(env, exp->exp_obd, 0);
 509         } else if (KEY_IS(KEY_MDS_CONN)) {
 510                 rc = ofd_set_mds_conn(exp, val);
 511         } else if (KEY_IS(KEY_GRANT_SHRINK)) {
 512                 struct ost_body *body = val;
 513
 514                 ofd_info_init(env, exp);
 515                 /** handle grant shrink, similar to a read request */
 516                 ofd_grant_prepare_read(env, exp, &body->oa);
 517         } else {
 518                 CERROR("%s: Unsupported key %s\n",
 519                        exp->exp_obd->obd_name, (char*)key);
 520                 rc = -EOPNOTSUPP;
 521         }
 522
 523         RETURN(rc);
 524 }
 525
 526 static int ofd_get_info(const struct lu_env *env, struct obd_export *exp,
 527                         __u32 keylen, void *key, __u32 *vallen, void *val,
 528                         struct lov_stripe_md *lsm)
 529 {
 530         struct ofd_device       *ofd;
 531         int                      rc = 0;
 532
 533         ENTRY;
 534
 535         if (exp->exp_obd == NULL) {
 536                 CDEBUG(D_IOCTL, "invalid client export %p\n", exp);
 537                 RETURN(-EINVAL);
 538         }
 539
 540         ofd = ofd_exp(exp);
 541
 542         if (KEY_IS(KEY_BLOCKSIZE)) {
 543                 __u32 *blocksize = val;
 544                 if (blocksize) {
 545                         if (*vallen < sizeof(*blocksize))
 546                                 RETURN(-EOVERFLOW);
 547                         *blocksize = 1 << ofd->ofd_dt_conf.ddp_block_shift;
 548                 }
 549                 *vallen = sizeof(*blocksize);
 550         } else if (KEY_IS(KEY_BLOCKSIZE_BITS)) {
 551                 __u32 *blocksize_bits = val;
 552                 if (blocksize_bits) {
 553                         if (*vallen < sizeof(*blocksize_bits))
 554                                 RETURN(-EOVERFLOW);
 555                         *blocksize_bits = ofd->ofd_dt_conf.ddp_block_shift;
 556                 }
 557                 *vallen = sizeof(*blocksize_bits);
 558         } else if (KEY_IS(KEY_LAST_ID)) {
 559                 obd_id *last_id = val;
 560                 struct ofd_seq *oseq;
 561
 562                 if (val == NULL) {
 563                         *vallen = sizeof(obd_id);
 564                         RETURN(0);
 565                 }
 566                 ofd_info_init(env, exp);
 567                 oseq = ofd_seq_load(env, ofd,
 568                                     (obd_seq)exp->exp_filter_data.fed_group);
 569                 LASSERT(!IS_ERR(oseq));
 570                 if (last_id) {
 571                         if (*vallen < sizeof(*last_id)) {
 572                                 ofd_seq_put(env, oseq);
 573                                 RETURN(-EOVERFLOW);
 574                         }
 575                         *last_id = ofd_seq_last_oid(oseq);
 576                 }
 577                 ofd_seq_put(env, oseq);
 578                 *vallen = sizeof(*last_id);
 579         } else if (KEY_IS(KEY_FIEMAP)) {
 580                 struct ofd_thread_info          *info;
 581                 struct ofd_device               *ofd = ofd_exp(exp);
 582                 struct ofd_object               *fo;
 583                 struct ll_fiemap_info_key       *fm_key = key;
 584
 585                 if (val == NULL) {
 586                         *vallen = fiemap_count_to_size(
 587                                                fm_key->fiemap.fm_extent_count);
 588                         RETURN(0);
 589                 }
 590
 591                 info = ofd_info_init(env, exp);
 592
 593                 fid_ostid_unpack(&info->fti_fid, &fm_key->oa.o_oi, 0);
 594
 595                 CDEBUG(D_INODE, "get FIEMAP of object "DFID"\n",
 596                        PFID(&info->fti_fid));
 597
 598                 fo = ofd_object_find(env, ofd, &info->fti_fid);
 599                 if (IS_ERR(fo)) {
 600                         CERROR("%s: error finding object "DFID"\n",
 601                                exp->exp_obd->obd_name, PFID(&info->fti_fid));
 602                         rc = PTR_ERR(fo);
 603                 } else {
 604                         struct ll_user_fiemap *fiemap = val;
 605
 606                         ofd_read_lock(env, fo);
 607                         if (ofd_object_exists(fo)) {
 608                                 *fiemap = fm_key->fiemap;
 609                                 rc = dt_fiemap_get(env,
 610                                                    ofd_object_child(fo),
 611                                                    fiemap);
 612                         } else {
 613                                 rc = -ENOENT;
 614                         }
 615                         ofd_read_unlock(env, fo);
 616                         ofd_object_put(env, fo);
 617                 }
 618         } else if (KEY_IS(KEY_SYNC_LOCK_CANCEL)) {
 619                 *((__u32 *) val) = ofd->ofd_sync_lock_cancel;
 620                 *vallen = sizeof(__u32);
 621         } else if (KEY_IS(KEY_LAST_FID)) {
 622                 struct lu_env           env;
 623                 struct ofd_device       *ofd = ofd_exp(exp);
 624                 struct ofd_seq          *oseq;
 625                 struct ost_id           *oid = val;
 626                 int                     rc;
 627
 628                 if (oid == NULL) {
 629                         *vallen = sizeof(struct ost_id);
 630                         RETURN(0);
 631                 }
 632
 633                 if (*vallen < sizeof(*oid))
 634                         RETURN(-EOVERFLOW);
 635
 636                 rc = lu_env_init(&env, LCT_DT_THREAD);
 637                 if (rc != 0)
 638                         RETURN(rc);
 639                 ofd_info_init(&env, exp);
 640
 641                 ostid_le_to_cpu(oid, oid);
 642
 643                 oseq = ofd_seq_load(&env, ofd, oid->oi_seq);
 644                 if (IS_ERR(oseq))
 645                         GOTO(out_fini, rc = PTR_ERR(oseq));
 646
 647                 CDEBUG(D_HA, "LAST FID is "DOSTID"\n", POSTID(&oseq->os_oi));
 648
 649                 *oid = oseq->os_oi;
 650                 *vallen = sizeof(*oid);
 651                 ofd_seq_put(&env, oseq);
 652 out_fini:
 653                 lu_env_fini(&env);
 654         } else {
 655                 CERROR("Not supported key %s\n", (char*)key);
 656                 rc = -EOPNOTSUPP;
 657         }
 658
 659         RETURN(rc);
 660 }
 661
 662 /** helper function for statfs, also used by grant code */
 663 int ofd_statfs_internal(const struct lu_env *env, struct ofd_device *ofd,
 664                         struct obd_statfs *osfs, __u64 max_age, int *from_cache)
 665 {
 666         int rc;
 667
 668         spin_lock(&ofd->ofd_osfs_lock);
 669         if (cfs_time_before_64(ofd->ofd_osfs_age, max_age) || max_age == 0) {
 670                 obd_size unstable;
 671
 672                 /* statfs data are too old, get up-to-date one.
 673                  * we must be cautious here since multiple threads might be
 674                  * willing to update statfs data concurrently and we must
 675                  * grant that cached statfs data are always consistent */
 676
 677                 if (ofd->ofd_statfs_inflight == 0)
 678                         /* clear inflight counter if no users, although it would
 679                          * take a while to overflow this 64-bit counter ... */
 680                         ofd->ofd_osfs_inflight = 0;
 681                 /* notify ofd_grant_commit() that we want to track writes
 682                  * completed as of now */
 683                 ofd->ofd_statfs_inflight++;
 684                 /* record value of inflight counter before running statfs to
 685                  * compute the diff once statfs is completed */
 686                 unstable = ofd->ofd_osfs_inflight;
 687                 spin_unlock(&ofd->ofd_osfs_lock);
 688
 689                 /* statfs can sleep ... hopefully not for too long since we can
 690                  * call it fairly often as space fills up */
 691                 rc = dt_statfs(env, ofd->ofd_osd, osfs);
 692                 if (unlikely(rc))
 693                         return rc;
 694
 695                 spin_lock(&ofd->ofd_grant_lock);
 696                 spin_lock(&ofd->ofd_osfs_lock);
 697                 /* calculate how much space was written while we released the
 698                  * ofd_osfs_lock */
 699                 unstable = ofd->ofd_osfs_inflight - unstable;
 700                 ofd->ofd_osfs_unstable = 0;
 701                 if (unstable) {
 702                         /* some writes completed while we were running statfs
 703                          * w/o the ofd_osfs_lock. Those ones got added to
 704                          * the cached statfs data that we are about to crunch.
 705                          * Take them into account in the new statfs data */
 706                         osfs->os_bavail -= min_t(obd_size, osfs->os_bavail,
 707                                                unstable >> ofd->ofd_blockbits);
 708                         /* However, we don't really know if those writes got
 709                          * accounted in the statfs call, so tell
 710                          * ofd_grant_space_left() there is some uncertainty
 711                          * on the accounting of those writes.
 712                          * The purpose is to prevent spurious error messages in
 713                          * ofd_grant_space_left() since those writes might be
 714                          * accounted twice. */
 715                         ofd->ofd_osfs_unstable += unstable;
 716                 }
 717                 /* similarly, there is some uncertainty on write requests
 718                  * between prepare & commit */
 719                 ofd->ofd_osfs_unstable += ofd->ofd_tot_pending;
 720                 spin_unlock(&ofd->ofd_grant_lock);
 721
 722                 /* finally udpate cached statfs data */
 723                 ofd->ofd_osfs = *osfs;
 724                 ofd->ofd_osfs_age = cfs_time_current_64();
 725
 726                 ofd->ofd_statfs_inflight--; /* stop tracking */
 727                 if (ofd->ofd_statfs_inflight == 0)
 728                         ofd->ofd_osfs_inflight = 0;
 729                 spin_unlock(&ofd->ofd_osfs_lock);
 730
 731                 if (from_cache)
 732                         *from_cache = 0;
 733         } else {
 734                 /* use cached statfs data */
 735                 *osfs = ofd->ofd_osfs;
 736                 spin_unlock(&ofd->ofd_osfs_lock);
 737                 if (from_cache)
 738                         *from_cache = 1;
 739         }
 740         return 0;
 741 }
 742
 743 static int ofd_statfs(const struct lu_env *env,  struct obd_export *exp,
 744                       struct obd_statfs *osfs, __u64 max_age, __u32 flags)
 745 {
 746         struct obd_device       *obd = class_exp2obd(exp);
 747         struct ofd_device       *ofd = ofd_dev(exp->exp_obd->obd_lu_dev);
 748         int                      rc;
 749
 750         ENTRY;
 751
 752         rc = ofd_statfs_internal(env, ofd, osfs, max_age, NULL);
 753         if (unlikely(rc))
 754                 GOTO(out, rc);
 755
 756         /* at least try to account for cached pages.  its still racy and
 757          * might be under-reporting if clients haven't announced their
 758          * caches with brw recently */
 759
 760         CDEBUG(D_SUPER | D_CACHE, "blocks cached "LPU64" granted "LPU64
 761                " pending "LPU64" free "LPU64" avail "LPU64"\n",
 762                ofd->ofd_tot_dirty, ofd->ofd_tot_granted, ofd->ofd_tot_pending,
 763                osfs->os_bfree << ofd->ofd_blockbits,
 764                osfs->os_bavail << ofd->ofd_blockbits);
 765
 766         osfs->os_bavail -= min_t(obd_size, osfs->os_bavail,
 767                                  ((ofd->ofd_tot_dirty + ofd->ofd_tot_pending +
 768                                    osfs->os_bsize - 1) >> ofd->ofd_blockbits));
 769
 770         /* The QoS code on the MDS does not care about space reserved for
 771          * precreate, so take it out. */
 772         if (exp_connect_flags(exp) & OBD_CONNECT_MDS) {
 773                 struct filter_export_data *fed;
 774
 775                 fed = &obd->obd_self_export->exp_filter_data;
 776                 osfs->os_bavail -= min_t(obd_size, osfs->os_bavail,
 777                                          fed->fed_grant >> ofd->ofd_blockbits);
 778         }
 779
 780         ofd_grant_sanity_check(obd, __FUNCTION__);
 781         CDEBUG(D_CACHE, LPU64" blocks: "LPU64" free, "LPU64" avail; "
 782                LPU64" objects: "LPU64" free; state %x\n",
 783                osfs->os_blocks, osfs->os_bfree, osfs->os_bavail,
 784                osfs->os_files, osfs->os_ffree, osfs->os_state);
 785
 786         if (OBD_FAIL_CHECK_VALUE(OBD_FAIL_OST_ENOINO,
 787                                  ofd->ofd_lut.lut_lsd.lsd_osd_index))
 788                 osfs->os_ffree = 0;
 789
 790         /* OS_STATE_READONLY can be set by OSD already */
 791         if (ofd->ofd_raid_degraded)
 792                 osfs->os_state |= OS_STATE_DEGRADED;
 793
 794         if (obd->obd_self_export != exp && ofd_grant_compat(exp, ofd)) {
 795                 /* clients which don't support OBD_CONNECT_GRANT_PARAM
 796                  * should not see a block size > page size, otherwise
 797                  * cl_lost_grant goes mad. Therefore, we emulate a 4KB (=2^12)
 798                  * block size which is the biggest block size known to work
 799                  * with all client's page size. */
 800                 osfs->os_blocks <<= ofd->ofd_blockbits - COMPAT_BSIZE_SHIFT;
 801                 osfs->os_bfree  <<= ofd->ofd_blockbits - COMPAT_BSIZE_SHIFT;
 802                 osfs->os_bavail <<= ofd->ofd_blockbits - COMPAT_BSIZE_SHIFT;
 803                 osfs->os_bsize    = 1 << COMPAT_BSIZE_SHIFT;
 804         }
 805
 806         if (OBD_FAIL_CHECK_VALUE(OBD_FAIL_OST_ENOSPC,
 807                                  ofd->ofd_lut.lut_lsd.lsd_osd_index))
 808                 osfs->os_bfree = osfs->os_bavail = 2;
 809
 810         EXIT;
 811 out:
 812         return rc;
 813 }
 814
 815 int ofd_setattr(const struct lu_env *env, struct obd_export *exp,
 816                 struct obd_info *oinfo, struct obd_trans_info *oti)
 817 {
 818         struct ofd_thread_info  *info;
 819         struct ofd_device       *ofd = ofd_exp(exp);
 820         struct ldlm_namespace   *ns = ofd->ofd_namespace;
 821         struct ldlm_resource    *res;
 822         struct ofd_object       *fo;
 823         struct obdo             *oa = oinfo->oi_oa;
 824         struct filter_fid       *ff = NULL;
 825         int                      rc = 0;
 826
 827         ENTRY;
 828
 829         info = ofd_info_init(env, exp);
 830         ofd_oti2info(info, oti);
 831
 832         fid_ostid_unpack(&info->fti_fid, &oinfo->oi_oa->o_oi, 0);
 833         ofd_build_resid(&info->fti_fid, &info->fti_resid);
 834
 835         rc = ofd_auth_capa(exp, &info->fti_fid, oa->o_seq,
 836                            oinfo_capa(oinfo), CAPA_OPC_META_WRITE);
 837         if (rc)
 838                 GOTO(out, rc);
 839
 840         /* This would be very bad - accidentally truncating a file when
 841          * changing the time or similar - bug 12203. */
 842         if (oinfo->oi_oa->o_valid & OBD_MD_FLSIZE &&
 843             oinfo->oi_policy.l_extent.end != OBD_OBJECT_EOF) {
 844                 static char mdsinum[48];
 845
 846                 if (oinfo->oi_oa->o_valid & OBD_MD_FLFID)
 847                         snprintf(mdsinum, sizeof(mdsinum) - 1,
 848                                  "of parent "DFID, oinfo->oi_oa->o_parent_seq,
 849                                  oinfo->oi_oa->o_parent_oid, 0);
 850                 else
 851                         mdsinum[0] = '\0';
 852
 853                 CERROR("%s: setattr from %s trying to truncate object "DFID
 854                        " %s\n", exp->exp_obd->obd_name,
 855                        obd_export_nid2str(exp), PFID(&info->fti_fid), mdsinum);
 856                 GOTO(out, rc = -EPERM);
 857         }
 858
 859         fo = ofd_object_find(env, ofd, &info->fti_fid);
 860         if (IS_ERR(fo)) {
 861                 CERROR("%s: can't find object "DFID"\n",
 862                        exp->exp_obd->obd_name, PFID(&info->fti_fid));
 863                 GOTO(out, rc = PTR_ERR(fo));
 864         }
 865
 866         la_from_obdo(&info->fti_attr, oinfo->oi_oa, oinfo->oi_oa->o_valid);
 867         info->fti_attr.la_valid &= ~LA_TYPE;
 868
 869         if (oa->o_valid & OBD_MD_FLFID) {
 870                 ff = &info->fti_mds_fid;
 871                 ofd_prepare_fidea(ff, oa);
 872         }
 873
 874         /* setting objects attributes (including owner/group) */
 875         rc = ofd_attr_set(env, fo, &info->fti_attr, ff);
 876         if (rc)
 877                 GOTO(out_unlock, rc);
 878
 879         res = ldlm_resource_get(ns, NULL, &info->fti_resid, LDLM_EXTENT, 0);
 880         if (res != NULL) {
 881                 ldlm_res_lvbo_update(res, NULL, 0);
 882                 ldlm_resource_putref(res);
 883         }
 884
 885         oinfo->oi_oa->o_valid = OBD_MD_FLID;
 886
 887         /* Quota release needs uid/gid info */
 888         rc = ofd_attr_get(env, fo, &info->fti_attr);
 889         obdo_from_la(oinfo->oi_oa, &info->fti_attr,
 890                      OFD_VALID_FLAGS | LA_UID | LA_GID);
 891         ofd_info2oti(info, oti);
 892
 893         ofd_counter_incr(exp, LPROC_OFD_STATS_SETATTR, oti->oti_jobid, 1);
 894         EXIT;
 895 out_unlock:
 896         ofd_object_put(env, fo);
 897 out:
 898         return rc;
 899 }
 900
 901 static int ofd_punch(const struct lu_env *env, struct obd_export *exp,
 902                      struct obd_info *oinfo, struct obd_trans_info *oti,
 903                      struct ptlrpc_request_set *rqset)
 904 {
 905         struct ofd_thread_info  *info;
 906         struct ofd_device       *ofd = ofd_exp(exp);
 907         struct ldlm_namespace   *ns = ofd->ofd_namespace;
 908         struct ldlm_resource    *res;
 909         struct ofd_object       *fo;
 910         struct filter_fid       *ff = NULL;
 911         int                      rc = 0;
 912
 913         ENTRY;
 914
 915         info = ofd_info_init(env, exp);
 916         ofd_oti2info(info, oti);
 917
 918         fid_ostid_unpack(&info->fti_fid, &oinfo->oi_oa->o_oi, 0);
 919         ofd_build_resid(&info->fti_fid, &info->fti_resid);
 920
 921         CDEBUG(D_INODE, "calling punch for object "DFID", valid = "LPX64
 922                ", start = "LPD64", end = "LPD64"\n", PFID(&info->fti_fid),
 923                oinfo->oi_oa->o_valid, oinfo->oi_policy.l_extent.start,
 924                oinfo->oi_policy.l_extent.end);
 925
 926         rc = ofd_auth_capa(exp, &info->fti_fid, oinfo->oi_oa->o_seq,
 927                            oinfo_capa(oinfo), CAPA_OPC_OSS_TRUNC);
 928         if (rc)
 929                 GOTO(out_env, rc);
 930
 931         fo = ofd_object_find(env, ofd, &info->fti_fid);
 932         if (IS_ERR(fo)) {
 933                 CERROR("%s: error finding object "DFID": rc = %ld\n",
 934                        exp->exp_obd->obd_name, PFID(&info->fti_fid),
 935                        PTR_ERR(fo));
 936                 GOTO(out_env, rc = PTR_ERR(fo));
 937         }
 938
 939         LASSERT(oinfo->oi_policy.l_extent.end == OBD_OBJECT_EOF);
 940         if (oinfo->oi_policy.l_extent.end == OBD_OBJECT_EOF) {
 941                 /* Truncate case */
 942                 oinfo->oi_oa->o_size = oinfo->oi_policy.l_extent.start;
 943         } else if (oinfo->oi_policy.l_extent.end >= oinfo->oi_oa->o_size) {
 944                 oinfo->oi_oa->o_size = oinfo->oi_policy.l_extent.end;
 945         }
 946
 947         la_from_obdo(&info->fti_attr, oinfo->oi_oa,
 948                      OBD_MD_FLMTIME | OBD_MD_FLATIME | OBD_MD_FLCTIME);
 949         info->fti_attr.la_valid &= ~LA_TYPE;
 950         info->fti_attr.la_size = oinfo->oi_policy.l_extent.start;
 951         info->fti_attr.la_valid |= LA_SIZE;
 952
 953         if (oinfo->oi_oa->o_valid & OBD_MD_FLFID) {
 954                 ff = &info->fti_mds_fid;
 955                 ofd_prepare_fidea(ff, oinfo->oi_oa);
 956         }
 957
 958         rc = ofd_object_punch(env, fo, oinfo->oi_policy.l_extent.start,
 959                               oinfo->oi_policy.l_extent.end, &info->fti_attr,
 960                               ff);
 961         if (rc)
 962                 GOTO(out, rc);
 963
 964         res = ldlm_resource_get(ns, NULL, &info->fti_resid, LDLM_EXTENT, 0);
 965         if (res != NULL) {
 966                 ldlm_res_lvbo_update(res, NULL, 0);
 967                 ldlm_resource_putref(res);
 968         }
 969
 970         oinfo->oi_oa->o_valid = OBD_MD_FLID;
 971         /* Quota release needs uid/gid info */
 972         rc = ofd_attr_get(env, fo, &info->fti_attr);
 973         obdo_from_la(oinfo->oi_oa, &info->fti_attr,
 974                      OFD_VALID_FLAGS | LA_UID | LA_GID);
 975         ofd_info2oti(info, oti);
 976
 977         ofd_counter_incr(exp, LPROC_OFD_STATS_PUNCH, oti->oti_jobid, 1);
 978         EXIT;
 979 out:
 980         ofd_object_put(env, fo);
 981 out_env:
 982         return rc;
 983 }
 984
 985 static int ofd_destroy_by_fid(const struct lu_env *env,
 986                               struct ofd_device *ofd,
 987                               const struct lu_fid *fid, int orphan)
 988 {
 989         struct ofd_thread_info  *info = ofd_info(env);
 990         struct lustre_handle     lockh;
 991         __u64                    flags = LDLM_AST_DISCARD_DATA, rc = 0;
 992         ldlm_policy_data_t       policy = {
 993                                         .l_extent = { 0, OBD_OBJECT_EOF }
 994                                  };
 995         struct ofd_object       *fo;
 996
 997         ENTRY;
 998
 999         fo = ofd_object_find(env, ofd, fid);
1000         if (IS_ERR(fo))
1001                 RETURN(PTR_ERR(fo));
1002
1003         /* Tell the clients that the object is gone now and that they should
1004          * throw away any cached pages. */
1005         ofd_build_resid(fid, &info->fti_resid);
1006         rc = ldlm_cli_enqueue_local(ofd->ofd_namespace, &info->fti_resid,
1007                                     LDLM_EXTENT, &policy, LCK_PW, &flags,
1008                                     ldlm_blocking_ast, ldlm_completion_ast,
1009                                     NULL, NULL, 0, LVB_T_NONE, NULL, &lockh);
1010
1011         /* We only care about the side-effects, just drop the lock. */
1012         if (rc == ELDLM_OK)
1013                 ldlm_lock_decref(&lockh, LCK_PW);
1014
1015         LASSERT(fo != NULL);
1016
1017         rc = ofd_object_destroy(env, fo, orphan);
1018
1019         ofd_object_put(env, fo);
1020         RETURN(rc);
1021 }
1022
1023 int ofd_destroy(const struct lu_env *env, struct obd_export *exp,
1024                 struct obdo *oa, struct lov_stripe_md *md,
1025                 struct obd_trans_info *oti, struct obd_export *md_exp,
1026                 void *capa)
1027 {
1028         struct ofd_device       *ofd = ofd_exp(exp);
1029         struct ofd_thread_info  *info;
1030         obd_count                count;
1031         int                      rc = 0;
1032
1033         ENTRY;
1034
1035         info = ofd_info_init(env, exp);
1036         ofd_oti2info(info, oti);
1037
1038         if (!(oa->o_valid & OBD_MD_FLGROUP))
1039                 oa->o_seq = 0;
1040
1041         /* check that o_misc makes sense */
1042         if (oa->o_valid & OBD_MD_FLOBJCOUNT)
1043                 count = oa->o_misc;
1044         else
1045                 count = 1; /* default case - single destroy */
1046
1047         /**
1048          * There can be sequence of objects to destroy. Therefore this request
1049          * may have multiple transaction involved in. It is OK, we need only
1050          * the highest used transno to be reported back in reply but not for
1051          * replays, they must report their transno
1052          */
1053         if (info->fti_transno == 0) /* not replay */
1054                 info->fti_mult_trans = 1;
1055         while (count > 0) {
1056                 int lrc;
1057
1058                 fid_ostid_unpack(&info->fti_fid, &oa->o_oi, 0);
1059                 lrc = ofd_destroy_by_fid(env, ofd, &info->fti_fid, 0);
1060                 if (lrc == -ENOENT) {
1061                         CDEBUG(D_INODE,
1062                                "%s: destroying non-existent object "DFID"\n",
1063                                ofd_obd(ofd)->obd_name, PFID(&info->fti_fid));
1064                         /* rewrite rc with -ENOENT only if it is 0 */
1065                         if (rc == 0)
1066                                 rc = lrc;
1067                 } else if (lrc != 0) {
1068                         CERROR("%s: error destroying object "DFID": %d\n",
1069                                ofd_obd(ofd)->obd_name, PFID(&info->fti_fid),
1070                                rc);
1071                         rc = lrc;
1072                 }
1073                 count--;
1074                 oa->o_id++;
1075         }
1076
1077         /* if we have transaction then there were some deletions, we don't
1078          * need to return ENOENT in that case because it will not wait
1079          * for commit of these deletions. The ENOENT must be returned only
1080          * if there were no transations.
1081          */
1082         if (rc == -ENOENT) {
1083                 if (info->fti_transno != 0)
1084                         rc = 0;
1085         } else if (rc != 0) {
1086                 /*
1087                  * If we have at least one transaction then llog record
1088                  * on server will be removed upon commit, so for rc != 0
1089                  * we return no transno and llog record will be reprocessed.
1090                  */
1091                 info->fti_transno = 0;
1092         }
1093         ofd_info2oti(info, oti);
1094         RETURN(rc);
1095 }
1096
1097 static int ofd_orphans_destroy(const struct lu_env *env,
1098                                struct obd_export *exp, struct ofd_device *ofd,
1099                                struct obdo *oa)
1100 {
1101         struct ofd_thread_info  *info = ofd_info(env);
1102         obd_id                   last;
1103         int                      skip_orphan;
1104         int                      rc = 0;
1105         struct ost_id            oi = oa->o_oi;
1106         struct ofd_seq          *oseq;
1107
1108         ENTRY;
1109
1110         oseq = ofd_seq_get(ofd, oa->o_seq);
1111         if (oseq == NULL) {
1112                 CERROR("%s: Can not find seq for "LPU64":"LPU64"\n",
1113                        ofd_name(ofd), oa->o_seq, oa->o_id);
1114                 RETURN(-EINVAL);
1115         }
1116
1117         LASSERT(exp != NULL);
1118         skip_orphan = !!(exp_connect_flags(exp) & OBD_CONNECT_SKIP_ORPHAN);
1119
1120         last = ofd_seq_last_oid(oseq);
1121         LCONSOLE_INFO("%s: deleting orphan objects from "LPX64":"LPU64
1122                       " to "LPU64"\n", ofd_name(ofd), oa->o_seq,
1123                       oa->o_id + 1, last);
1124
1125         for (oi.oi_id = last; oi.oi_id > oa->o_id; oi.oi_id--) {
1126                 fid_ostid_unpack(&info->fti_fid, &oi, 0);
1127                 rc = ofd_destroy_by_fid(env, ofd, &info->fti_fid, 1);
1128                 if (rc && rc != -ENOENT) /* this is pretty fatal... */
1129                         CEMERG("error destroying precreated id "LPU64": %d\n",
1130                                oi.oi_id, rc);
1131                 if (!skip_orphan) {
1132                         ofd_seq_last_oid_set(oseq, oi.oi_id - 1);
1133                         /* update last_id on disk periodically so that if we
1134                          * restart * we don't need to re-scan all of the just
1135                          * deleted objects. */
1136                         if ((oi.oi_id & 511) == 0)
1137                                 ofd_seq_last_oid_write(env, ofd, oseq);
1138                 }
1139         }
1140         CDEBUG(D_HA, "%s: after destroy: set last_objids["LPU64"] = "LPU64"\n",
1141                ofd_obd(ofd)->obd_name, oa->o_seq, oa->o_id);
1142         if (!skip_orphan) {
1143                 rc = ofd_seq_last_oid_write(env, ofd, oseq);
1144         } else {
1145                 /* don't reuse orphan object, return last used objid */
1146                 oa->o_id = last;
1147                 rc = 0;
1148         }
1149         ofd_seq_put(env, oseq);
1150         RETURN(rc);
1151 }
1152
1153 int ofd_create(const struct lu_env *env, struct obd_export *exp,
1154                struct obdo *oa, struct lov_stripe_md **ea,
1155                struct obd_trans_info *oti)
1156 {
1157         struct ofd_device       *ofd = ofd_exp(exp);
1158         struct ofd_thread_info  *info;
1159         obd_seq                 seq = oa->o_seq;
1160         struct ofd_seq          *oseq;
1161         int                     rc = 0, diff;
1162         int                     sync_trans = 0;
1163
1164         ENTRY;
1165
1166         info = ofd_info_init(env, exp);
1167         ofd_oti2info(info, oti);
1168
1169         LASSERT(oa->o_seq >= FID_SEQ_OST_MDT0);
1170         LASSERT(oa->o_valid & OBD_MD_FLGROUP);
1171
1172         CDEBUG(D_INFO, "ofd_create(oa->o_seq="LPU64",oa->o_id="LPU64")\n",
1173                seq, oa->o_id);
1174
1175         oseq = ofd_seq_load(env, ofd, seq);
1176         if (IS_ERR(oseq)) {
1177                 CERROR("%s: Can't find FID Sequence "LPX64": rc = %ld\n",
1178                        ofd_name(ofd), seq, PTR_ERR(oseq));
1179                 RETURN(-EINVAL);
1180         }
1181
1182         if ((oa->o_valid & OBD_MD_FLFLAGS) &&
1183             (oa->o_flags & OBD_FL_RECREATE_OBJS)) {
1184                 if (!ofd_obd(ofd)->obd_recovering ||
1185                     oa->o_id > ofd_seq_last_oid(oseq)) {
1186                         CERROR("recreate objid "LPU64" > last id "LPU64"\n",
1187                                         oa->o_id, ofd_seq_last_oid(oseq));
1188                         GOTO(out_nolock, rc = -EINVAL);
1189                 }
1190                 /* do nothing because we create objects during first write */
1191                 GOTO(out_nolock, rc = 0);
1192         }
1193         /* former ofd_handle_precreate */
1194         if ((oa->o_valid & OBD_MD_FLFLAGS) &&
1195             (oa->o_flags & OBD_FL_DELORPHAN)) {
1196                 /* destroy orphans */
1197                 if (oti->oti_conn_cnt < exp->exp_conn_cnt) {
1198                         CERROR("%s: dropping old orphan cleanup request\n",
1199                                ofd_name(ofd));
1200                         GOTO(out_nolock, rc = 0);
1201                 }
1202                 /* This causes inflight precreates to abort and drop lock */
1203                 oseq->os_destroys_in_progress = 1;
1204                 mutex_lock(&oseq->os_create_lock);
1205                 if (!oseq->os_destroys_in_progress) {
1206                         CERROR("%s:["LPU64"] destroys_in_progress already cleared\n",
1207                                exp->exp_obd->obd_name, oa->o_seq);
1208                         GOTO(out, rc = 0);
1209                 }
1210                 diff = oa->o_id - ofd_seq_last_oid(oseq);
1211                 CDEBUG(D_HA, "ofd_last_id() = "LPU64" -> diff = %d\n",
1212                         ofd_seq_last_oid(oseq), diff);
1213                 if (-diff > OST_MAX_PRECREATE) {
1214                         /* FIXME: should reset precreate_next_id on MDS */
1215                         rc = 0;
1216                 } else if (diff < 0) {
1217                         rc = ofd_orphans_destroy(env, exp, ofd, oa);
1218                         oseq->os_destroys_in_progress = 0;
1219                 } else {
1220                         /* XXX: Used by MDS for the first time! */
1221                         oseq->os_destroys_in_progress = 0;
1222                 }
1223         } else {
1224                 mutex_lock(&oseq->os_create_lock);
1225                 if (oti->oti_conn_cnt < exp->exp_conn_cnt) {
1226                         CERROR("%s: dropping old precreate request\n",
1227                                 ofd_obd(ofd)->obd_name);
1228                         GOTO(out, rc = 0);
1229                 }
1230                 /* only precreate if seq is 0, IDIF or normal and also o_id
1231                  * must be specfied */
1232                 if ((!fid_seq_is_mdt(oa->o_seq) &&
1233                      !fid_seq_is_norm(oa->o_seq) &&
1234                      !fid_seq_is_idif(oa->o_seq)) || oa->o_id == 0) {
1235                         diff = 1; /* shouldn't we create this right now? */
1236                 } else {
1237                         diff = oa->o_id - ofd_seq_last_oid(oseq);
1238                         /* Do sync create if the seq is about to used up */
1239                         if (fid_seq_is_idif(oa->o_seq) ||
1240                             fid_seq_is_mdt0(oa->o_seq)) {
1241                                 if (unlikely(oa->o_id >= IDIF_MAX_OID - 1))
1242                                         sync_trans = 1;
1243                         } else if (fid_seq_is_norm(oa->o_seq)) {
1244                                 if (unlikely(oa->o_id >=
1245                                              LUSTRE_DATA_SEQ_MAX_WIDTH - 1))
1246                                         sync_trans = 1;
1247                         } else {
1248                                 CERROR("%s : invalid o_seq "LPX64": rc = %d\n",
1249                                        ofd_name(ofd), oa->o_seq, -EINVAL);
1250                                 GOTO(out, rc = -EINVAL);
1251                         }
1252                 }
1253         }
1254         if (diff > 0) {
1255                 cfs_time_t       enough_time = cfs_time_shift(DISK_TIMEOUT);
1256                 obd_id           next_id;
1257                 int              created = 0;
1258                 int              count;
1259
1260                 if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
1261                     !(oa->o_flags & OBD_FL_DELORPHAN)) {
1262                         /* don't enforce grant during orphan recovery */
1263                         rc = ofd_grant_create(env,
1264                                               ofd_obd(ofd)->obd_self_export,
1265                                               &diff);
1266                         if (rc) {
1267                                 CDEBUG(D_HA, "%s: failed to acquire grant "
1268                                        "space for precreate (%d): rc = %d\n",
1269                                        ofd_name(ofd), diff, rc);
1270                                 diff = 0;
1271                         }
1272                 }
1273
1274                 while (diff > 0) {
1275                         next_id = ofd_seq_last_oid(oseq) + 1;
1276                         count = ofd_precreate_batch(ofd, diff);
1277
1278                         CDEBUG(D_HA, "%s: reserve %d objects in group "LPU64
1279                                " at "LPU64"\n", ofd_obd(ofd)->obd_name,
1280                                count, oa->o_seq, next_id);
1281
1282                         if (cfs_time_after(jiffies, enough_time)) {
1283                                 LCONSOLE_WARN("%s: Slow creates, %d/%d objects"
1284                                               " created at a rate of %d/s\n",
1285                                               ofd_obd(ofd)->obd_name,
1286                                               created, diff + created,
1287                                               created / DISK_TIMEOUT);
1288                                 break;
1289                         }
1290
1291                         rc = ofd_precreate_objects(env, ofd, next_id,
1292                                                    oseq, count, sync_trans);
1293                         if (rc > 0) {
1294                                 created += rc;
1295                                 diff -= rc;
1296                         } else if (rc < 0) {
1297                                 break;
1298                         }
1299                 }
1300                 if (created > 0)
1301                         /* some objects got created, we can return
1302                          * them, even if last creation failed */
1303                         rc = 0;
1304                 else
1305                         CERROR("%s: unable to precreate: rc = %d\n",
1306                                ofd_name(ofd), rc);
1307
1308                 oa->o_id = ofd_seq_last_oid(oseq);
1309                 oa->o_valid |= OBD_MD_FLID | OBD_MD_FLGROUP;
1310
1311                 if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
1312                     !(oa->o_flags & OBD_FL_DELORPHAN))
1313                         ofd_grant_commit(env, ofd_obd(ofd)->obd_self_export,
1314                                          rc);
1315         }
1316
1317         ofd_info2oti(info, oti);
1318 out:
1319         mutex_unlock(&oseq->os_create_lock);
1320 out_nolock:
1321         if (rc == 0 && ea != NULL) {
1322                 struct lov_stripe_md *lsm = *ea;
1323
1324                 lsm->lsm_oi = oa->o_oi;
1325         }
1326         ofd_seq_put(env, oseq);
1327         RETURN(rc);
1328 }
1329
1330 int ofd_getattr(const struct lu_env *env, struct obd_export *exp,
1331                 struct obd_info *oinfo)
1332 {
1333         struct ofd_device       *ofd = ofd_exp(exp);
1334         struct ofd_thread_info  *info;
1335         struct ofd_object       *fo;
1336         __u64                    curr_version;
1337         int                      rc = 0;
1338
1339         ENTRY;
1340
1341         info = ofd_info_init(env, exp);
1342
1343         fid_ostid_unpack(&info->fti_fid, &oinfo->oi_oa->o_oi, 0);
1344         rc = ofd_auth_capa(exp, &info->fti_fid, oinfo->oi_oa->o_seq,
1345                            oinfo_capa(oinfo), CAPA_OPC_META_READ);
1346         if (rc)
1347                 GOTO(out, rc);
1348
1349         fo = ofd_object_find(env, ofd, &info->fti_fid);
1350         if (IS_ERR(fo))
1351                 GOTO(out, rc = PTR_ERR(fo));
1352         LASSERT(fo != NULL);
1353         rc = ofd_attr_get(env, fo, &info->fti_attr);
1354         oinfo->oi_oa->o_valid = OBD_MD_FLID;
1355         if (rc == 0)
1356                 obdo_from_la(oinfo->oi_oa, &info->fti_attr,
1357                              OFD_VALID_FLAGS | LA_UID | LA_GID);
1358
1359         /* Store object version in reply */
1360         curr_version = dt_version_get(env, ofd_object_child(fo));
1361         if ((__s64)curr_version != -EOPNOTSUPP) {
1362                 oinfo->oi_oa->o_valid |= OBD_MD_FLDATAVERSION;
1363                 oinfo->oi_oa->o_data_version = curr_version;
1364         }
1365         ofd_object_put(env, fo);
1366 out:
1367         RETURN(rc);
1368 }
1369
1370 static int ofd_sync(const struct lu_env *env, struct obd_export *exp,
1371                     struct obd_info *oinfo, obd_size start, obd_size end,
1372                     struct ptlrpc_request_set *set)
1373 {
1374         struct ofd_device       *ofd = ofd_exp(exp);
1375         struct ofd_thread_info  *info;
1376         struct ofd_object       *fo;
1377         int                      rc = 0;
1378
1379         ENTRY;
1380
1381         /* if no objid is specified, it means "sync whole filesystem" */
1382         if (oinfo->oi_oa == NULL || !(oinfo->oi_oa->o_valid & OBD_MD_FLID)) {
1383                 rc = dt_sync(env, ofd->ofd_osd);
1384                 GOTO(out, rc);
1385         }
1386
1387         info = ofd_info_init(env, exp);
1388         fid_ostid_unpack(&info->fti_fid, &oinfo->oi_oa->o_oi, 0);
1389
1390         rc = ofd_auth_capa(exp, &info->fti_fid, oinfo->oi_oa->o_seq,
1391                            oinfo_capa(oinfo), CAPA_OPC_OSS_TRUNC);
1392         if (rc)
1393                 GOTO(out, rc);
1394
1395         fo = ofd_object_find(env, ofd, &info->fti_fid);
1396         if (IS_ERR(fo)) {
1397                 CERROR("%s: error finding object "DFID": rc = %ld\n",
1398                        exp->exp_obd->obd_name, PFID(&info->fti_fid),
1399                        PTR_ERR(fo));
1400                 GOTO(out, rc = PTR_ERR(fo));
1401         }
1402
1403         if (!ofd_object_exists(fo))
1404                 GOTO(put, rc = -ENOENT);
1405
1406         if (dt_version_get(env, ofd_object_child(fo)) >
1407             ofd_obd(ofd)->obd_last_committed) {
1408                 rc = dt_object_sync(env, ofd_object_child(fo));
1409                 if (rc)
1410                         GOTO(put, rc);
1411         }
1412
1413         oinfo->oi_oa->o_valid = OBD_MD_FLID;
1414         rc = ofd_attr_get(env, fo, &info->fti_attr);
1415         obdo_from_la(oinfo->oi_oa, &info->fti_attr, OFD_VALID_FLAGS);
1416
1417         ofd_counter_incr(exp, LPROC_OFD_STATS_SYNC, oinfo->oi_jobid, 1);
1418         EXIT;
1419 put:
1420         ofd_object_put(env, fo);
1421 out:
1422         return rc;
1423 }
1424
1425 static int ofd_ioc_get_obj_version(const struct lu_env *env,
1426                                    struct ofd_device *ofd, void *karg)
1427 {
1428         struct obd_ioctl_data *data = karg;
1429         struct lu_fid          fid;
1430         struct ofd_object     *fo;
1431         dt_obj_version_t       version;
1432         int                    rc = 0;
1433
1434         ENTRY;
1435
1436         if (data->ioc_inlbuf2 == NULL || data->ioc_inllen2 != sizeof(version))
1437                 GOTO(out, rc = -EINVAL);
1438
1439         if (data->ioc_inlbuf1 != NULL && data->ioc_inllen1 == sizeof(fid)) {
1440                 fid = *(struct lu_fid *)data->ioc_inlbuf1;
1441         } else if (data->ioc_inlbuf3 != NULL &&
1442                    data->ioc_inllen3 == sizeof(__u64) &&
1443                    data->ioc_inlbuf4 != NULL &&
1444                    data->ioc_inllen4 == sizeof(__u64)) {
1445                 struct ost_id ostid;
1446
1447                 ostid.oi_id = *(__u64 *)data->ioc_inlbuf3;
1448                 ostid.oi_seq = *(__u64 *)data->ioc_inlbuf4;
1449                 rc = fid_ostid_unpack(&fid, &ostid, 0);
1450                 if (rc != 0)
1451                         GOTO(out, rc);
1452         } else {
1453                 GOTO(out, rc = -EINVAL);
1454         }
1455
1456         if (!fid_is_sane(&fid))
1457                 GOTO(out, rc = -EINVAL);
1458
1459         fo = ofd_object_find(env, ofd, &fid);
1460         if (IS_ERR(fo))
1461                 GOTO(out, rc = PTR_ERR(fo));
1462
1463         if (!ofd_object_exists(fo))
1464                 GOTO(out_fo, rc = -ENOENT);
1465
1466         if (lu_object_remote(&fo->ofo_obj.do_lu))
1467                 GOTO(out_fo, rc = -EREMOTE);
1468
1469         version = dt_version_get(env, ofd_object_child(fo));
1470         if (version == 0)
1471                 GOTO(out_fo, rc = -EIO);
1472
1473         *(dt_obj_version_t *)data->ioc_inlbuf2 = version;
1474
1475         EXIT;
1476 out_fo:
1477         ofd_object_put(env, fo);
1478 out:
1479         return rc;
1480 }
1481
1482 int ofd_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1483                   void *karg, void *uarg)
1484 {
1485         struct lu_env            env;
1486         struct ofd_device       *ofd = ofd_exp(exp);
1487         struct obd_device       *obd = ofd_obd(ofd);
1488         int                      rc;
1489
1490         ENTRY;
1491
1492         CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
1493         rc = lu_env_init(&env, LCT_DT_THREAD);
1494         if (rc)
1495                 RETURN(rc);
1496
1497         switch (cmd) {
1498         case OBD_IOC_ABORT_RECOVERY:
1499                 CERROR("%s: aborting recovery\n", obd->obd_name);
1500                 target_stop_recovery_thread(obd);
1501                 break;
1502         case OBD_IOC_SYNC:
1503                 CDEBUG(D_RPCTRACE, "syncing ost %s\n", obd->obd_name);
1504                 rc = dt_sync(&env, ofd->ofd_osd);
1505                 break;
1506         case OBD_IOC_SET_READONLY:
1507                 rc = dt_sync(&env, ofd->ofd_osd);
1508                 if (rc == 0)
1509                         rc = dt_ro(&env, ofd->ofd_osd);
1510                 break;
1511         case OBD_IOC_GET_OBJ_VERSION:
1512                 rc = ofd_ioc_get_obj_version(&env, ofd, karg);
1513                 break;
1514         default:
1515                 CERROR("%s: not supported cmd = %d\n", obd->obd_name, cmd);
1516                 rc = -ENOTTY;
1517         }
1518
1519         lu_env_fini(&env);
1520         RETURN(rc);
1521 }
1522
1523 static int ofd_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
1524 {
1525         int rc = 0;
1526
1527         ENTRY;
1528
1529         switch(stage) {
1530         case OBD_CLEANUP_EARLY:
1531                 break;
1532         case OBD_CLEANUP_EXPORTS:
1533                 target_cleanup_recovery(obd);
1534                 break;
1535         }
1536         RETURN(rc);
1537 }
1538
1539 static int ofd_ping(const struct lu_env *env, struct obd_export *exp)
1540 {
1541         ofd_fmd_expire(exp);
1542         return 0;
1543 }
1544
1545 static int ofd_health_check(const struct lu_env *nul, struct obd_device *obd)
1546 {
1547         struct ofd_device       *ofd = ofd_dev(obd->obd_lu_dev);
1548         struct ofd_thread_info  *info;
1549         struct lu_env            env;
1550 #ifdef USE_HEALTH_CHECK_WRITE
1551         struct thandle          *th;
1552 #endif
1553         int                      rc = 0;
1554
1555         /* obd_proc_read_health pass NULL env, we need real one */
1556         rc = lu_env_init(&env, LCT_DT_THREAD);
1557         if (rc)
1558                 RETURN(rc);
1559
1560         info = ofd_info_init(&env, NULL);
1561         rc = dt_statfs(&env, ofd->ofd_osd, &info->fti_u.osfs);
1562         if (unlikely(rc))
1563                 GOTO(out, rc);
1564
1565         if (info->fti_u.osfs.os_state == OS_STATE_READONLY)
1566                 GOTO(out, rc = -EROFS);
1567
1568 #ifdef USE_HEALTH_CHECK_WRITE
1569         OBD_ALLOC(info->fti_buf.lb_buf, CFS_PAGE_SIZE);
1570         if (info->fti_buf.lb_buf == NULL)
1571                 GOTO(out, rc = -ENOMEM);
1572
1573         info->fti_buf.lb_len = CFS_PAGE_SIZE;
1574         info->fti_off = 0;
1575
1576         th = dt_trans_create(&env, ofd->ofd_osd);
1577         if (IS_ERR(th))
1578                 GOTO(out, rc = PTR_ERR(th));
1579
1580         rc = dt_declare_record_write(&env, ofd->ofd_health_check_file,
1581                                      info->fti_buf.lb_len, info->fti_off, th);
1582         if (rc == 0) {
1583                 th->th_sync = 1; /* sync IO is needed */
1584                 rc = dt_trans_start_local(&env, ofd->ofd_osd, th);
1585                 if (rc == 0)
1586                         rc = dt_record_write(&env, ofd->ofd_health_check_file,
1587                                              &info->fti_buf, &info->fti_off,
1588                                              th);
1589         }
1590         dt_trans_stop(&env, ofd->ofd_osd, th);
1591
1592         OBD_FREE(info->fti_buf.lb_buf, CFS_PAGE_SIZE);
1593
1594         CDEBUG(D_INFO, "write 1 page synchronously for checking io rc %d\n",rc);
1595 #endif
1596 out:
1597         lu_env_fini(&env);
1598         return !!rc;
1599 }
1600
1601 /*
1602  * Handle quota control requests to consult current usage/limit.
1603  *
1604  * \param obd - is the obd device associated with the ofd
1605  * \param exp - is the client's export
1606  * \param oqctl - is the obd_quotactl request to be processed
1607  */
1608 static int ofd_quotactl(struct obd_device *obd, struct obd_export *exp,
1609                         struct obd_quotactl *oqctl)
1610 {
1611         struct ofd_device  *ofd = ofd_dev(obd->obd_lu_dev);
1612         struct lu_env       env;
1613         int                 rc;
1614         ENTRY;
1615
1616         /* report success for quota on/off for interoperability with current MDT
1617          * stack */
1618         if (oqctl->qc_cmd == Q_QUOTAON || oqctl->qc_cmd == Q_QUOTAOFF)
1619                 RETURN(0);
1620
1621         rc = lu_env_init(&env, LCT_DT_THREAD);
1622         if (rc)
1623                 RETURN(rc);
1624
1625         rc = lquotactl_slv(&env, ofd->ofd_osd, oqctl);
1626         lu_env_fini(&env);
1627
1628         RETURN(rc);
1629 }
1630
1631 struct obd_ops ofd_obd_ops = {
1632         .o_owner                = THIS_MODULE,
1633         .o_connect              = ofd_obd_connect,
1634         .o_reconnect            = ofd_obd_reconnect,
1635         .o_disconnect           = ofd_obd_disconnect,
1636         .o_set_info_async       = ofd_set_info_async,
1637         .o_get_info             = ofd_get_info,
1638         .o_create               = ofd_create,
1639         .o_statfs               = ofd_statfs,
1640         .o_setattr              = ofd_setattr,
1641         .o_preprw               = ofd_preprw,
1642         .o_commitrw             = ofd_commitrw,
1643         .o_destroy              = ofd_destroy,
1644         .o_init_export          = ofd_init_export,
1645         .o_destroy_export       = ofd_destroy_export,
1646         .o_postrecov            = ofd_obd_postrecov,
1647         .o_punch                = ofd_punch,
1648         .o_getattr              = ofd_getattr,
1649         .o_sync                 = ofd_sync,
1650         .o_iocontrol            = ofd_iocontrol,
1651         .o_precleanup           = ofd_precleanup,
1652         .o_ping                 = ofd_ping,
1653         .o_health_check         = ofd_health_check,
1654         .o_quotactl             = ofd_quotactl,
1655 };