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 (OCD_HAS_FLAG(data, PINGLESS)) {
 230                 if (ptlrpc_pinger_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                 rc = ostid_to_fid(&info->fti_fid, &fm_key->oa.o_oi, 0);
 593                 if (rc != 0)
 594                         RETURN(rc);
 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 lu_fid           *fid = val;
 626                 int                     rc;
 627
 628                 if (fid == NULL) {
 629                         *vallen = sizeof(struct lu_fid);
 630                         RETURN(0);
 631                 }
 632
 633                 if (*vallen < sizeof(*fid))
 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                 fid_le_to_cpu(fid, fid);
 642
 643                 oseq = ofd_seq_load(&env, ofd, ostid_seq((struct ost_id *)fid));
 644                 if (IS_ERR(oseq))
 645                         GOTO(out_fini, rc = PTR_ERR(oseq));
 646
 647                 rc = ostid_to_fid(fid, &oseq->os_oi,
 648                              ofd->ofd_lut.lut_lsd.lsd_osd_index);
 649                 if (rc != 0)
 650                         GOTO(out_put, rc);
 651
 652                 CDEBUG(D_HA, "%s: LAST FID is "DFID"\n", ofd_name(ofd),
 653                        PFID(fid));
 654                 *vallen = sizeof(*fid);
 655 out_put:
 656                 ofd_seq_put(&env, oseq);
 657 out_fini:
 658                 lu_env_fini(&env);
 659         } else {
 660                 CERROR("Not supported key %s\n", (char*)key);
 661                 rc = -EOPNOTSUPP;
 662         }
 663
 664         RETURN(rc);
 665 }
 666
 667 /** helper function for statfs, also used by grant code */
 668 int ofd_statfs_internal(const struct lu_env *env, struct ofd_device *ofd,
 669                         struct obd_statfs *osfs, __u64 max_age, int *from_cache)
 670 {
 671         int rc;
 672
 673         spin_lock(&ofd->ofd_osfs_lock);
 674         if (cfs_time_before_64(ofd->ofd_osfs_age, max_age) || max_age == 0) {
 675                 obd_size unstable;
 676
 677                 /* statfs data are too old, get up-to-date one.
 678                  * we must be cautious here since multiple threads might be
 679                  * willing to update statfs data concurrently and we must
 680                  * grant that cached statfs data are always consistent */
 681
 682                 if (ofd->ofd_statfs_inflight == 0)
 683                         /* clear inflight counter if no users, although it would
 684                          * take a while to overflow this 64-bit counter ... */
 685                         ofd->ofd_osfs_inflight = 0;
 686                 /* notify ofd_grant_commit() that we want to track writes
 687                  * completed as of now */
 688                 ofd->ofd_statfs_inflight++;
 689                 /* record value of inflight counter before running statfs to
 690                  * compute the diff once statfs is completed */
 691                 unstable = ofd->ofd_osfs_inflight;
 692                 spin_unlock(&ofd->ofd_osfs_lock);
 693
 694                 /* statfs can sleep ... hopefully not for too long since we can
 695                  * call it fairly often as space fills up */
 696                 rc = dt_statfs(env, ofd->ofd_osd, osfs);
 697                 if (unlikely(rc))
 698                         return rc;
 699
 700                 spin_lock(&ofd->ofd_grant_lock);
 701                 spin_lock(&ofd->ofd_osfs_lock);
 702                 /* calculate how much space was written while we released the
 703                  * ofd_osfs_lock */
 704                 unstable = ofd->ofd_osfs_inflight - unstable;
 705                 ofd->ofd_osfs_unstable = 0;
 706                 if (unstable) {
 707                         /* some writes completed while we were running statfs
 708                          * w/o the ofd_osfs_lock. Those ones got added to
 709                          * the cached statfs data that we are about to crunch.
 710                          * Take them into account in the new statfs data */
 711                         osfs->os_bavail -= min_t(obd_size, osfs->os_bavail,
 712                                                unstable >> ofd->ofd_blockbits);
 713                         /* However, we don't really know if those writes got
 714                          * accounted in the statfs call, so tell
 715                          * ofd_grant_space_left() there is some uncertainty
 716                          * on the accounting of those writes.
 717                          * The purpose is to prevent spurious error messages in
 718                          * ofd_grant_space_left() since those writes might be
 719                          * accounted twice. */
 720                         ofd->ofd_osfs_unstable += unstable;
 721                 }
 722                 /* similarly, there is some uncertainty on write requests
 723                  * between prepare & commit */
 724                 ofd->ofd_osfs_unstable += ofd->ofd_tot_pending;
 725                 spin_unlock(&ofd->ofd_grant_lock);
 726
 727                 /* finally udpate cached statfs data */
 728                 ofd->ofd_osfs = *osfs;
 729                 ofd->ofd_osfs_age = cfs_time_current_64();
 730
 731                 ofd->ofd_statfs_inflight--; /* stop tracking */
 732                 if (ofd->ofd_statfs_inflight == 0)
 733                         ofd->ofd_osfs_inflight = 0;
 734                 spin_unlock(&ofd->ofd_osfs_lock);
 735
 736                 if (from_cache)
 737                         *from_cache = 0;
 738         } else {
 739                 /* use cached statfs data */
 740                 *osfs = ofd->ofd_osfs;
 741                 spin_unlock(&ofd->ofd_osfs_lock);
 742                 if (from_cache)
 743                         *from_cache = 1;
 744         }
 745         return 0;
 746 }
 747
 748 static int ofd_statfs(const struct lu_env *env,  struct obd_export *exp,
 749                       struct obd_statfs *osfs, __u64 max_age, __u32 flags)
 750 {
 751         struct obd_device       *obd = class_exp2obd(exp);
 752         struct ofd_device       *ofd = ofd_dev(exp->exp_obd->obd_lu_dev);
 753         int                      rc;
 754
 755         ENTRY;
 756
 757         rc = ofd_statfs_internal(env, ofd, osfs, max_age, NULL);
 758         if (unlikely(rc))
 759                 GOTO(out, rc);
 760
 761         /* at least try to account for cached pages.  its still racy and
 762          * might be under-reporting if clients haven't announced their
 763          * caches with brw recently */
 764
 765         CDEBUG(D_SUPER | D_CACHE, "blocks cached "LPU64" granted "LPU64
 766                " pending "LPU64" free "LPU64" avail "LPU64"\n",
 767                ofd->ofd_tot_dirty, ofd->ofd_tot_granted, ofd->ofd_tot_pending,
 768                osfs->os_bfree << ofd->ofd_blockbits,
 769                osfs->os_bavail << ofd->ofd_blockbits);
 770
 771         osfs->os_bavail -= min_t(obd_size, osfs->os_bavail,
 772                                  ((ofd->ofd_tot_dirty + ofd->ofd_tot_pending +
 773                                    osfs->os_bsize - 1) >> ofd->ofd_blockbits));
 774
 775         /* The QoS code on the MDS does not care about space reserved for
 776          * precreate, so take it out. */
 777         if (exp_connect_flags(exp) & OBD_CONNECT_MDS) {
 778                 struct filter_export_data *fed;
 779
 780                 fed = &obd->obd_self_export->exp_filter_data;
 781                 osfs->os_bavail -= min_t(obd_size, osfs->os_bavail,
 782                                          fed->fed_grant >> ofd->ofd_blockbits);
 783         }
 784
 785         ofd_grant_sanity_check(obd, __FUNCTION__);
 786         CDEBUG(D_CACHE, LPU64" blocks: "LPU64" free, "LPU64" avail; "
 787                LPU64" objects: "LPU64" free; state %x\n",
 788                osfs->os_blocks, osfs->os_bfree, osfs->os_bavail,
 789                osfs->os_files, osfs->os_ffree, osfs->os_state);
 790
 791         if (OBD_FAIL_CHECK_VALUE(OBD_FAIL_OST_ENOINO,
 792                                  ofd->ofd_lut.lut_lsd.lsd_osd_index))
 793                 osfs->os_ffree = 0;
 794
 795         /* OS_STATE_READONLY can be set by OSD already */
 796         if (ofd->ofd_raid_degraded)
 797                 osfs->os_state |= OS_STATE_DEGRADED;
 798
 799         if (obd->obd_self_export != exp && ofd_grant_compat(exp, ofd)) {
 800                 /* clients which don't support OBD_CONNECT_GRANT_PARAM
 801                  * should not see a block size > page size, otherwise
 802                  * cl_lost_grant goes mad. Therefore, we emulate a 4KB (=2^12)
 803                  * block size which is the biggest block size known to work
 804                  * with all client's page size. */
 805                 osfs->os_blocks <<= ofd->ofd_blockbits - COMPAT_BSIZE_SHIFT;
 806                 osfs->os_bfree  <<= ofd->ofd_blockbits - COMPAT_BSIZE_SHIFT;
 807                 osfs->os_bavail <<= ofd->ofd_blockbits - COMPAT_BSIZE_SHIFT;
 808                 osfs->os_bsize    = 1 << COMPAT_BSIZE_SHIFT;
 809         }
 810
 811         if (OBD_FAIL_CHECK_VALUE(OBD_FAIL_OST_ENOSPC,
 812                                  ofd->ofd_lut.lut_lsd.lsd_osd_index))
 813                 osfs->os_bfree = osfs->os_bavail = 2;
 814
 815         EXIT;
 816 out:
 817         return rc;
 818 }
 819
 820 int ofd_setattr(const struct lu_env *env, struct obd_export *exp,
 821                 struct obd_info *oinfo, struct obd_trans_info *oti)
 822 {
 823         struct ofd_thread_info  *info;
 824         struct ofd_device       *ofd = ofd_exp(exp);
 825         struct ldlm_namespace   *ns = ofd->ofd_namespace;
 826         struct ldlm_resource    *res;
 827         struct ofd_object       *fo;
 828         struct obdo             *oa = oinfo->oi_oa;
 829         struct filter_fid       *ff = NULL;
 830         int                      rc = 0;
 831
 832         ENTRY;
 833
 834         info = ofd_info_init(env, exp);
 835         ofd_oti2info(info, oti);
 836
 837         rc = ostid_to_fid(&info->fti_fid, &oinfo->oi_oa->o_oi, 0);
 838         if (rc != 0)
 839                 RETURN(rc);
 840
 841         ost_fid_build_resid(&info->fti_fid, &info->fti_resid);
 842         rc = ofd_auth_capa(exp, &info->fti_fid, ostid_seq(&oa->o_oi),
 843                            oinfo_capa(oinfo), CAPA_OPC_META_WRITE);
 844         if (rc)
 845                 GOTO(out, rc);
 846
 847         /* This would be very bad - accidentally truncating a file when
 848          * changing the time or similar - bug 12203. */
 849         if (oinfo->oi_oa->o_valid & OBD_MD_FLSIZE &&
 850             oinfo->oi_policy.l_extent.end != OBD_OBJECT_EOF) {
 851                 static char mdsinum[48];
 852
 853                 if (oinfo->oi_oa->o_valid & OBD_MD_FLFID)
 854                         snprintf(mdsinum, sizeof(mdsinum) - 1,
 855                                  "of parent "DFID, oinfo->oi_oa->o_parent_seq,
 856                                  oinfo->oi_oa->o_parent_oid, 0);
 857                 else
 858                         mdsinum[0] = '\0';
 859
 860                 CERROR("%s: setattr from %s trying to truncate object "DFID
 861                        " %s\n", exp->exp_obd->obd_name,
 862                        obd_export_nid2str(exp), PFID(&info->fti_fid), mdsinum);
 863                 GOTO(out, rc = -EPERM);
 864         }
 865
 866         fo = ofd_object_find(env, ofd, &info->fti_fid);
 867         if (IS_ERR(fo)) {
 868                 CERROR("%s: can't find object "DFID"\n",
 869                        exp->exp_obd->obd_name, PFID(&info->fti_fid));
 870                 GOTO(out, rc = PTR_ERR(fo));
 871         }
 872
 873         la_from_obdo(&info->fti_attr, oinfo->oi_oa, oinfo->oi_oa->o_valid);
 874         info->fti_attr.la_valid &= ~LA_TYPE;
 875
 876         if (oa->o_valid & OBD_MD_FLFID) {
 877                 ff = &info->fti_mds_fid;
 878                 ofd_prepare_fidea(ff, oa);
 879         }
 880
 881         /* setting objects attributes (including owner/group) */
 882         rc = ofd_attr_set(env, fo, &info->fti_attr, ff);
 883         if (rc)
 884                 GOTO(out_unlock, rc);
 885
 886         res = ldlm_resource_get(ns, NULL, &info->fti_resid, LDLM_EXTENT, 0);
 887         if (res != NULL) {
 888                 ldlm_res_lvbo_update(res, NULL, 0);
 889                 ldlm_resource_putref(res);
 890         }
 891
 892         oinfo->oi_oa->o_valid = OBD_MD_FLID;
 893
 894         /* Quota release needs uid/gid info */
 895         rc = ofd_attr_get(env, fo, &info->fti_attr);
 896         obdo_from_la(oinfo->oi_oa, &info->fti_attr,
 897                      OFD_VALID_FLAGS | LA_UID | LA_GID);
 898         ofd_info2oti(info, oti);
 899
 900         ofd_counter_incr(exp, LPROC_OFD_STATS_SETATTR, oti->oti_jobid, 1);
 901         EXIT;
 902 out_unlock:
 903         ofd_object_put(env, fo);
 904 out:
 905         return rc;
 906 }
 907
 908 static int ofd_punch(const struct lu_env *env, struct obd_export *exp,
 909                      struct obd_info *oinfo, struct obd_trans_info *oti,
 910                      struct ptlrpc_request_set *rqset)
 911 {
 912         struct ofd_thread_info  *info;
 913         struct ofd_device       *ofd = ofd_exp(exp);
 914         struct ldlm_namespace   *ns = ofd->ofd_namespace;
 915         struct ldlm_resource    *res;
 916         struct ofd_object       *fo;
 917         struct filter_fid       *ff = NULL;
 918         int                      rc = 0;
 919
 920         ENTRY;
 921
 922         info = ofd_info_init(env, exp);
 923         ofd_oti2info(info, oti);
 924
 925         rc = ostid_to_fid(&info->fti_fid, &oinfo->oi_oa->o_oi, 0);
 926         if (rc != 0)
 927                 RETURN(rc);
 928         ost_fid_build_resid(&info->fti_fid, &info->fti_resid);
 929
 930         CDEBUG(D_INODE, "calling punch for object "DFID", valid = "LPX64
 931                ", start = "LPD64", end = "LPD64"\n", PFID(&info->fti_fid),
 932                oinfo->oi_oa->o_valid, oinfo->oi_policy.l_extent.start,
 933                oinfo->oi_policy.l_extent.end);
 934
 935         rc = ofd_auth_capa(exp, &info->fti_fid, ostid_seq(&oinfo->oi_oa->o_oi),
 936                            oinfo_capa(oinfo), CAPA_OPC_OSS_TRUNC);
 937         if (rc)
 938                 GOTO(out_env, rc);
 939
 940         fo = ofd_object_find(env, ofd, &info->fti_fid);
 941         if (IS_ERR(fo)) {
 942                 CERROR("%s: error finding object "DFID": rc = %ld\n",
 943                        exp->exp_obd->obd_name, PFID(&info->fti_fid),
 944                        PTR_ERR(fo));
 945                 GOTO(out_env, rc = PTR_ERR(fo));
 946         }
 947
 948         LASSERT(oinfo->oi_policy.l_extent.end == OBD_OBJECT_EOF);
 949         if (oinfo->oi_policy.l_extent.end == OBD_OBJECT_EOF) {
 950                 /* Truncate case */
 951                 oinfo->oi_oa->o_size = oinfo->oi_policy.l_extent.start;
 952         } else if (oinfo->oi_policy.l_extent.end >= oinfo->oi_oa->o_size) {
 953                 oinfo->oi_oa->o_size = oinfo->oi_policy.l_extent.end;
 954         }
 955
 956         la_from_obdo(&info->fti_attr, oinfo->oi_oa,
 957                      OBD_MD_FLMTIME | OBD_MD_FLATIME | OBD_MD_FLCTIME);
 958         info->fti_attr.la_valid &= ~LA_TYPE;
 959         info->fti_attr.la_size = oinfo->oi_policy.l_extent.start;
 960         info->fti_attr.la_valid |= LA_SIZE;
 961
 962         if (oinfo->oi_oa->o_valid & OBD_MD_FLFID) {
 963                 ff = &info->fti_mds_fid;
 964                 ofd_prepare_fidea(ff, oinfo->oi_oa);
 965         }
 966
 967         rc = ofd_object_punch(env, fo, oinfo->oi_policy.l_extent.start,
 968                               oinfo->oi_policy.l_extent.end, &info->fti_attr,
 969                               ff);
 970         if (rc)
 971                 GOTO(out, rc);
 972
 973         res = ldlm_resource_get(ns, NULL, &info->fti_resid, LDLM_EXTENT, 0);
 974         if (res != NULL) {
 975                 ldlm_res_lvbo_update(res, NULL, 0);
 976                 ldlm_resource_putref(res);
 977         }
 978
 979         oinfo->oi_oa->o_valid = OBD_MD_FLID;
 980         /* Quota release needs uid/gid info */
 981         rc = ofd_attr_get(env, fo, &info->fti_attr);
 982         obdo_from_la(oinfo->oi_oa, &info->fti_attr,
 983                      OFD_VALID_FLAGS | LA_UID | LA_GID);
 984         ofd_info2oti(info, oti);
 985
 986         ofd_counter_incr(exp, LPROC_OFD_STATS_PUNCH, oti->oti_jobid, 1);
 987         EXIT;
 988 out:
 989         ofd_object_put(env, fo);
 990 out_env:
 991         return rc;
 992 }
 993
 994 static int ofd_destroy_by_fid(const struct lu_env *env,
 995                               struct ofd_device *ofd,
 996                               const struct lu_fid *fid, int orphan)
 997 {
 998         struct ofd_thread_info  *info = ofd_info(env);
 999         struct lustre_handle     lockh;
1000         __u64                    flags = LDLM_AST_DISCARD_DATA, rc = 0;
1001         ldlm_policy_data_t       policy = {
1002                                         .l_extent = { 0, OBD_OBJECT_EOF }
1003                                  };
1004         struct ofd_object       *fo;
1005
1006         ENTRY;
1007
1008         fo = ofd_object_find(env, ofd, fid);
1009         if (IS_ERR(fo))
1010                 RETURN(PTR_ERR(fo));
1011
1012         /* Tell the clients that the object is gone now and that they should
1013          * throw away any cached pages. */
1014         ost_fid_build_resid(fid, &info->fti_resid);
1015         rc = ldlm_cli_enqueue_local(ofd->ofd_namespace, &info->fti_resid,
1016                                     LDLM_EXTENT, &policy, LCK_PW, &flags,
1017                                     ldlm_blocking_ast, ldlm_completion_ast,
1018                                     NULL, NULL, 0, LVB_T_NONE, NULL, &lockh);
1019
1020         /* We only care about the side-effects, just drop the lock. */
1021         if (rc == ELDLM_OK)
1022                 ldlm_lock_decref(&lockh, LCK_PW);
1023
1024         LASSERT(fo != NULL);
1025
1026         rc = ofd_object_destroy(env, fo, orphan);
1027
1028         ofd_object_put(env, fo);
1029         RETURN(rc);
1030 }
1031
1032 int ofd_destroy(const struct lu_env *env, struct obd_export *exp,
1033                 struct obdo *oa, struct lov_stripe_md *md,
1034                 struct obd_trans_info *oti, struct obd_export *md_exp,
1035                 void *capa)
1036 {
1037         struct ofd_device       *ofd = ofd_exp(exp);
1038         struct ofd_thread_info  *info;
1039         obd_count                count;
1040         int                      rc = 0;
1041
1042         ENTRY;
1043
1044         info = ofd_info_init(env, exp);
1045         ofd_oti2info(info, oti);
1046
1047         if (!(oa->o_valid & OBD_MD_FLGROUP))
1048                 ostid_set_seq_mdt0(&oa->o_oi);
1049
1050         /* check that o_misc makes sense */
1051         if (oa->o_valid & OBD_MD_FLOBJCOUNT)
1052                 count = oa->o_misc;
1053         else
1054                 count = 1; /* default case - single destroy */
1055
1056         /**
1057          * There can be sequence of objects to destroy. Therefore this request
1058          * may have multiple transaction involved in. It is OK, we need only
1059          * the highest used transno to be reported back in reply but not for
1060          * replays, they must report their transno
1061          */
1062         if (info->fti_transno == 0) /* not replay */
1063                 info->fti_mult_trans = 1;
1064
1065         CDEBUG(D_HA, "%s: Destroy object "DOSTID" count %d\n", ofd_name(ofd),
1066                POSTID(&oa->o_oi), count);
1067         while (count > 0) {
1068                 int lrc;
1069
1070                 lrc = ostid_to_fid(&info->fti_fid, &oa->o_oi, 0);
1071                 if (lrc != 0) {
1072                         if (rc == 0)
1073                                 rc = lrc;
1074                         GOTO(out, rc);
1075                 }
1076                 lrc = ofd_destroy_by_fid(env, ofd, &info->fti_fid, 0);
1077                 if (lrc == -ENOENT) {
1078                         CDEBUG(D_INODE,
1079                                "%s: destroying non-existent object "DFID"\n",
1080                                ofd_obd(ofd)->obd_name, PFID(&info->fti_fid));
1081                         /* rewrite rc with -ENOENT only if it is 0 */
1082                         if (rc == 0)
1083                                 rc = lrc;
1084                 } else if (lrc != 0) {
1085                         CERROR("%s: error destroying object "DFID": %d\n",
1086                                ofd_obd(ofd)->obd_name, PFID(&info->fti_fid),
1087                                rc);
1088                         rc = lrc;
1089                 }
1090                 count--;
1091                 ostid_inc_id(&oa->o_oi);
1092         }
1093
1094         /* if we have transaction then there were some deletions, we don't
1095          * need to return ENOENT in that case because it will not wait
1096          * for commit of these deletions. The ENOENT must be returned only
1097          * if there were no transations.
1098          */
1099         if (rc == -ENOENT) {
1100                 if (info->fti_transno != 0)
1101                         rc = 0;
1102         } else if (rc != 0) {
1103                 /*
1104                  * If we have at least one transaction then llog record
1105                  * on server will be removed upon commit, so for rc != 0
1106                  * we return no transno and llog record will be reprocessed.
1107                  */
1108                 info->fti_transno = 0;
1109         }
1110         ofd_info2oti(info, oti);
1111 out:
1112         RETURN(rc);
1113 }
1114
1115 static int ofd_orphans_destroy(const struct lu_env *env,
1116                                struct obd_export *exp, struct ofd_device *ofd,
1117                                struct obdo *oa)
1118 {
1119         struct ofd_thread_info  *info = ofd_info(env);
1120         obd_id                   last;
1121         int                      skip_orphan;
1122         int                      rc = 0;
1123         struct ost_id            oi = oa->o_oi;
1124         __u64                    end_id = ostid_id(&oa->o_oi);
1125         struct ofd_seq          *oseq;
1126
1127         ENTRY;
1128
1129         oseq = ofd_seq_get(ofd, ostid_seq(&oa->o_oi));
1130         if (oseq == NULL) {
1131                 CERROR("%s: Can not find seq for "DOSTID"\n",
1132                        ofd_name(ofd), POSTID(&oa->o_oi));
1133                 RETURN(-EINVAL);
1134         }
1135
1136         LASSERT(exp != NULL);
1137         skip_orphan = !!(exp_connect_flags(exp) & OBD_CONNECT_SKIP_ORPHAN);
1138
1139         last = ofd_seq_last_oid(oseq);
1140         LCONSOLE_INFO("%s: deleting orphan objects from "LPX64":"LPU64
1141                       " to "LPU64"\n", ofd_name(ofd), ostid_seq(&oa->o_oi),
1142                       end_id + 1, last);
1143
1144         for (ostid_set_id(&oi, last); ostid_id(&oi) > end_id;
1145                           ostid_dec_id(&oi)) {
1146                 rc = ostid_to_fid(&info->fti_fid, &oi, 0);
1147                 if (rc != 0)
1148                         GOTO(out_put, rc);
1149                 rc = ofd_destroy_by_fid(env, ofd, &info->fti_fid, 1);
1150                 if (rc && rc != -ENOENT) /* this is pretty fatal... */
1151                         CEMERG("error destroying precreated id "DOSTID": %d\n",
1152                                POSTID(&oi), rc);
1153                 if (!skip_orphan) {
1154                         ofd_seq_last_oid_set(oseq, ostid_id(&oi) - 1);
1155                         /* update last_id on disk periodically so that if we
1156                          * restart * we don't need to re-scan all of the just
1157                          * deleted objects. */
1158                         if ((ostid_id(&oi) & 511) == 0)
1159                                 ofd_seq_last_oid_write(env, ofd, oseq);
1160                 }
1161         }
1162         CDEBUG(D_HA, "%s: after destroy: set last_objids"DOSTID"\n",
1163                ofd_obd(ofd)->obd_name, POSTID(&oa->o_oi));
1164         if (!skip_orphan) {
1165                 rc = ofd_seq_last_oid_write(env, ofd, oseq);
1166         } else {
1167                 /* don't reuse orphan object, return last used objid */
1168                 ostid_set_id(&oa->o_oi, last);
1169                 rc = 0;
1170         }
1171 out_put:
1172         ofd_seq_put(env, oseq);
1173         RETURN(rc);
1174 }
1175
1176 int ofd_create(const struct lu_env *env, struct obd_export *exp,
1177                struct obdo *oa, struct lov_stripe_md **ea,
1178                struct obd_trans_info *oti)
1179 {
1180         struct ofd_device       *ofd = ofd_exp(exp);
1181         struct ofd_thread_info  *info;
1182         obd_seq                 seq = ostid_seq(&oa->o_oi);
1183         struct ofd_seq          *oseq;
1184         int                     rc = 0, diff;
1185         int                     sync_trans = 0;
1186
1187         ENTRY;
1188
1189         info = ofd_info_init(env, exp);
1190         ofd_oti2info(info, oti);
1191
1192         LASSERT(ostid_seq(&oa->o_oi) >= FID_SEQ_OST_MDT0);
1193         LASSERT(oa->o_valid & OBD_MD_FLGROUP);
1194
1195         CDEBUG(D_INFO, "ofd_create("DOSTID")\n", POSTID(&oa->o_oi));
1196
1197         oseq = ofd_seq_load(env, ofd, seq);
1198         if (IS_ERR(oseq)) {
1199                 CERROR("%s: Can't find FID Sequence "LPX64": rc = %ld\n",
1200                        ofd_name(ofd), seq, PTR_ERR(oseq));
1201                 RETURN(-EINVAL);
1202         }
1203
1204         if ((oa->o_valid & OBD_MD_FLFLAGS) &&
1205             (oa->o_flags & OBD_FL_RECREATE_OBJS)) {
1206                 if (!ofd_obd(ofd)->obd_recovering ||
1207                     ostid_id(&oa->o_oi) > ofd_seq_last_oid(oseq)) {
1208                         CERROR("recreate objid "DOSTID" > last id "LPU64"\n",
1209                                POSTID(&oa->o_oi), ofd_seq_last_oid(oseq));
1210                         GOTO(out_nolock, rc = -EINVAL);
1211                 }
1212                 /* do nothing because we create objects during first write */
1213                 GOTO(out_nolock, rc = 0);
1214         }
1215         /* former ofd_handle_precreate */
1216         if ((oa->o_valid & OBD_MD_FLFLAGS) &&
1217             (oa->o_flags & OBD_FL_DELORPHAN)) {
1218                 /* destroy orphans */
1219                 if (oti->oti_conn_cnt < exp->exp_conn_cnt) {
1220                         CERROR("%s: dropping old orphan cleanup request\n",
1221                                ofd_name(ofd));
1222                         GOTO(out_nolock, rc = 0);
1223                 }
1224                 /* This causes inflight precreates to abort and drop lock */
1225                 oseq->os_destroys_in_progress = 1;
1226                 mutex_lock(&oseq->os_create_lock);
1227                 if (!oseq->os_destroys_in_progress) {
1228                         CERROR("%s:["LPU64"] destroys_in_progress already"
1229                                " cleared\n", exp->exp_obd->obd_name,
1230                                ostid_seq(&oa->o_oi));
1231                         GOTO(out, rc = 0);
1232                 }
1233                 diff = ostid_id(&oa->o_oi) - ofd_seq_last_oid(oseq);
1234                 CDEBUG(D_HA, "ofd_last_id() = "LPU64" -> diff = %d\n",
1235                         ofd_seq_last_oid(oseq), diff);
1236                 if (-diff > OST_MAX_PRECREATE) {
1237                         /* FIXME: should reset precreate_next_id on MDS */
1238                         rc = 0;
1239                 } else if (diff < 0) {
1240                         rc = ofd_orphans_destroy(env, exp, ofd, oa);
1241                         oseq->os_destroys_in_progress = 0;
1242                 } else {
1243                         /* XXX: Used by MDS for the first time! */
1244                         oseq->os_destroys_in_progress = 0;
1245                 }
1246         } else {
1247                 mutex_lock(&oseq->os_create_lock);
1248                 if (oti->oti_conn_cnt < exp->exp_conn_cnt) {
1249                         CERROR("%s: dropping old precreate request\n",
1250                                 ofd_obd(ofd)->obd_name);
1251                         GOTO(out, rc = 0);
1252                 }
1253                 /* only precreate if seq is 0, IDIF or normal and also o_id
1254                  * must be specfied */
1255                 if ((!fid_seq_is_mdt(ostid_seq(&oa->o_oi)) &&
1256                      !fid_seq_is_norm(ostid_seq(&oa->o_oi)) &&
1257                      !fid_seq_is_idif(ostid_seq(&oa->o_oi))) ||
1258                                 ostid_id(&oa->o_oi) == 0) {
1259                         diff = 1; /* shouldn't we create this right now? */
1260                 } else {
1261                         diff = ostid_id(&oa->o_oi) - ofd_seq_last_oid(oseq);
1262                         /* Do sync create if the seq is about to used up */
1263                         if (fid_seq_is_idif(ostid_seq(&oa->o_oi)) ||
1264                             fid_seq_is_mdt0(ostid_seq(&oa->o_oi))) {
1265                                 if (unlikely(ostid_id(&oa->o_oi) >= IDIF_MAX_OID - 1))
1266                                         sync_trans = 1;
1267                         } else if (fid_seq_is_norm(ostid_seq(&oa->o_oi))) {
1268                                 if (unlikely(ostid_id(&oa->o_oi) >=
1269                                              LUSTRE_DATA_SEQ_MAX_WIDTH - 1))
1270                                         sync_trans = 1;
1271                         } else {
1272                                 CERROR("%s : invalid o_seq "DOSTID": rc = %d\n",
1273                                        ofd_name(ofd), POSTID(&oa->o_oi), -EINVAL);
1274                                 GOTO(out, rc = -EINVAL);
1275                         }
1276                 }
1277         }
1278         if (diff > 0) {
1279                 cfs_time_t       enough_time = cfs_time_shift(DISK_TIMEOUT);
1280                 obd_id           next_id;
1281                 int              created = 0;
1282                 int              count;
1283
1284                 if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
1285                     !(oa->o_flags & OBD_FL_DELORPHAN)) {
1286                         /* don't enforce grant during orphan recovery */
1287                         rc = ofd_grant_create(env,
1288                                               ofd_obd(ofd)->obd_self_export,
1289                                               &diff);
1290                         if (rc) {
1291                                 CDEBUG(D_HA, "%s: failed to acquire grant "
1292                                        "space for precreate (%d): rc = %d\n",
1293                                        ofd_name(ofd), diff, rc);
1294                                 diff = 0;
1295                         }
1296                 }
1297
1298                 while (diff > 0) {
1299                         next_id = ofd_seq_last_oid(oseq) + 1;
1300                         count = ofd_precreate_batch(ofd, diff);
1301
1302                         CDEBUG(D_HA, "%s: reserve %d objects in group "LPX64
1303                                " at "LPU64"\n", ofd_obd(ofd)->obd_name,
1304                                count, ostid_seq(&oa->o_oi), next_id);
1305
1306                         if (cfs_time_after(jiffies, enough_time)) {
1307                                 LCONSOLE_WARN("%s: Slow creates, %d/%d objects"
1308                                               " created at a rate of %d/s\n",
1309                                               ofd_obd(ofd)->obd_name,
1310                                               created, diff + created,
1311                                               created / DISK_TIMEOUT);
1312                                 break;
1313                         }
1314
1315                         rc = ofd_precreate_objects(env, ofd, next_id,
1316                                                    oseq, count, sync_trans);
1317                         if (rc > 0) {
1318                                 created += rc;
1319                                 diff -= rc;
1320                         } else if (rc < 0) {
1321                                 break;
1322                         }
1323                 }
1324                 if (created > 0)
1325                         /* some objects got created, we can return
1326                          * them, even if last creation failed */
1327                         rc = 0;
1328                 else
1329                         CERROR("%s: unable to precreate: rc = %d\n",
1330                                ofd_name(ofd), rc);
1331
1332                 ostid_set_id(&oa->o_oi, ofd_seq_last_oid(oseq));
1333                 oa->o_valid |= OBD_MD_FLID | OBD_MD_FLGROUP;
1334
1335                 if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
1336                     !(oa->o_flags & OBD_FL_DELORPHAN))
1337                         ofd_grant_commit(env, ofd_obd(ofd)->obd_self_export,
1338                                          rc);
1339         }
1340
1341         ofd_info2oti(info, oti);
1342 out:
1343         mutex_unlock(&oseq->os_create_lock);
1344 out_nolock:
1345         if (rc == 0 && ea != NULL) {
1346                 struct lov_stripe_md *lsm = *ea;
1347
1348                 lsm->lsm_oi = oa->o_oi;
1349         }
1350         ofd_seq_put(env, oseq);
1351         RETURN(rc);
1352 }
1353
1354 int ofd_getattr(const struct lu_env *env, struct obd_export *exp,
1355                 struct obd_info *oinfo)
1356 {
1357         struct ofd_device       *ofd = ofd_exp(exp);
1358         struct ofd_thread_info  *info;
1359         struct ofd_object       *fo;
1360         __u64                    curr_version;
1361         int                      rc = 0;
1362
1363         ENTRY;
1364
1365         info = ofd_info_init(env, exp);
1366
1367         rc = ostid_to_fid(&info->fti_fid, &oinfo->oi_oa->o_oi, 0);
1368         if (rc != 0)
1369                 GOTO(out, rc);
1370         rc = ofd_auth_capa(exp, &info->fti_fid, ostid_seq(&oinfo->oi_oa->o_oi),
1371                            oinfo_capa(oinfo), CAPA_OPC_META_READ);
1372         if (rc)
1373                 GOTO(out, rc);
1374
1375         fo = ofd_object_find(env, ofd, &info->fti_fid);
1376         if (IS_ERR(fo))
1377                 GOTO(out, rc = PTR_ERR(fo));
1378         LASSERT(fo != NULL);
1379         rc = ofd_attr_get(env, fo, &info->fti_attr);
1380         oinfo->oi_oa->o_valid = OBD_MD_FLID;
1381         if (rc == 0)
1382                 obdo_from_la(oinfo->oi_oa, &info->fti_attr,
1383                              OFD_VALID_FLAGS | LA_UID | LA_GID);
1384
1385         /* Store object version in reply */
1386         curr_version = dt_version_get(env, ofd_object_child(fo));
1387         if ((__s64)curr_version != -EOPNOTSUPP) {
1388                 oinfo->oi_oa->o_valid |= OBD_MD_FLDATAVERSION;
1389                 oinfo->oi_oa->o_data_version = curr_version;
1390         }
1391         ofd_object_put(env, fo);
1392 out:
1393         RETURN(rc);
1394 }
1395
1396 static int ofd_sync(const struct lu_env *env, struct obd_export *exp,
1397                     struct obd_info *oinfo, obd_size start, obd_size end,
1398                     struct ptlrpc_request_set *set)
1399 {
1400         struct ofd_device       *ofd = ofd_exp(exp);
1401         struct ofd_thread_info  *info;
1402         struct ofd_object       *fo;
1403         int                      rc = 0;
1404
1405         ENTRY;
1406
1407         /* if no objid is specified, it means "sync whole filesystem" */
1408         if (oinfo->oi_oa == NULL || !(oinfo->oi_oa->o_valid & OBD_MD_FLID)) {
1409                 rc = dt_sync(env, ofd->ofd_osd);
1410                 GOTO(out, rc);
1411         }
1412
1413         info = ofd_info_init(env, exp);
1414         rc = ostid_to_fid(&info->fti_fid, &oinfo->oi_oa->o_oi, 0);
1415         if (rc != 0)
1416                 GOTO(out, rc);
1417
1418         rc = ofd_auth_capa(exp, &info->fti_fid, ostid_seq(&oinfo->oi_oa->o_oi),
1419                            oinfo_capa(oinfo), CAPA_OPC_OSS_TRUNC);
1420         if (rc)
1421                 GOTO(out, rc);
1422
1423         fo = ofd_object_find(env, ofd, &info->fti_fid);
1424         if (IS_ERR(fo)) {
1425                 CERROR("%s: error finding object "DFID": rc = %ld\n",
1426                        exp->exp_obd->obd_name, PFID(&info->fti_fid),
1427                        PTR_ERR(fo));
1428                 GOTO(out, rc = PTR_ERR(fo));
1429         }
1430
1431         if (!ofd_object_exists(fo))
1432                 GOTO(put, rc = -ENOENT);
1433
1434         if (dt_version_get(env, ofd_object_child(fo)) >
1435             ofd_obd(ofd)->obd_last_committed) {
1436                 rc = dt_object_sync(env, ofd_object_child(fo));
1437                 if (rc)
1438                         GOTO(put, rc);
1439         }
1440
1441         oinfo->oi_oa->o_valid = OBD_MD_FLID;
1442         rc = ofd_attr_get(env, fo, &info->fti_attr);
1443         obdo_from_la(oinfo->oi_oa, &info->fti_attr, OFD_VALID_FLAGS);
1444
1445         ofd_counter_incr(exp, LPROC_OFD_STATS_SYNC, oinfo->oi_jobid, 1);
1446         EXIT;
1447 put:
1448         ofd_object_put(env, fo);
1449 out:
1450         return rc;
1451 }
1452
1453 static int ofd_ioc_get_obj_version(const struct lu_env *env,
1454                                    struct ofd_device *ofd, void *karg)
1455 {
1456         struct obd_ioctl_data *data = karg;
1457         struct lu_fid          fid;
1458         struct ofd_object     *fo;
1459         dt_obj_version_t       version;
1460         int                    rc = 0;
1461
1462         ENTRY;
1463
1464         if (data->ioc_inlbuf2 == NULL || data->ioc_inllen2 != sizeof(version))
1465                 GOTO(out, rc = -EINVAL);
1466
1467         if (data->ioc_inlbuf1 != NULL && data->ioc_inllen1 == sizeof(fid)) {
1468                 fid = *(struct lu_fid *)data->ioc_inlbuf1;
1469         } else if (data->ioc_inlbuf3 != NULL &&
1470                    data->ioc_inllen3 == sizeof(__u64) &&
1471                    data->ioc_inlbuf4 != NULL &&
1472                    data->ioc_inllen4 == sizeof(__u64)) {
1473                 struct ost_id ostid;
1474
1475                 ostid_set_seq(&ostid, *(__u64 *)data->ioc_inlbuf4);
1476                 ostid_set_id(&ostid, *(__u64 *)data->ioc_inlbuf3);
1477                 rc = ostid_to_fid(&fid, &ostid, 0);
1478                 if (rc != 0)
1479                         GOTO(out, rc);
1480         } else {
1481                 GOTO(out, rc = -EINVAL);
1482         }
1483
1484         if (!fid_is_sane(&fid))
1485                 GOTO(out, rc = -EINVAL);
1486
1487         fo = ofd_object_find(env, ofd, &fid);
1488         if (IS_ERR(fo))
1489                 GOTO(out, rc = PTR_ERR(fo));
1490
1491         if (!ofd_object_exists(fo))
1492                 GOTO(out_fo, rc = -ENOENT);
1493
1494         if (lu_object_remote(&fo->ofo_obj.do_lu))
1495                 GOTO(out_fo, rc = -EREMOTE);
1496
1497         version = dt_version_get(env, ofd_object_child(fo));
1498         if (version == 0)
1499                 GOTO(out_fo, rc = -EIO);
1500
1501         *(dt_obj_version_t *)data->ioc_inlbuf2 = version;
1502
1503         EXIT;
1504 out_fo:
1505         ofd_object_put(env, fo);
1506 out:
1507         return rc;
1508 }
1509
1510 int ofd_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1511                   void *karg, void *uarg)
1512 {
1513         struct lu_env            env;
1514         struct ofd_device       *ofd = ofd_exp(exp);
1515         struct obd_device       *obd = ofd_obd(ofd);
1516         int                      rc;
1517
1518         ENTRY;
1519
1520         CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
1521         rc = lu_env_init(&env, LCT_DT_THREAD);
1522         if (rc)
1523                 RETURN(rc);
1524
1525         switch (cmd) {
1526         case OBD_IOC_ABORT_RECOVERY:
1527                 CERROR("%s: aborting recovery\n", obd->obd_name);
1528                 target_stop_recovery_thread(obd);
1529                 break;
1530         case OBD_IOC_SYNC:
1531                 CDEBUG(D_RPCTRACE, "syncing ost %s\n", obd->obd_name);
1532                 rc = dt_sync(&env, ofd->ofd_osd);
1533                 break;
1534         case OBD_IOC_SET_READONLY:
1535                 rc = dt_sync(&env, ofd->ofd_osd);
1536                 if (rc == 0)
1537                         rc = dt_ro(&env, ofd->ofd_osd);
1538                 break;
1539         case OBD_IOC_GET_OBJ_VERSION:
1540                 rc = ofd_ioc_get_obj_version(&env, ofd, karg);
1541                 break;
1542         default:
1543                 CERROR("%s: not supported cmd = %d\n", obd->obd_name, cmd);
1544                 rc = -ENOTTY;
1545         }
1546
1547         lu_env_fini(&env);
1548         RETURN(rc);
1549 }
1550
1551 static int ofd_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
1552 {
1553         int rc = 0;
1554
1555         ENTRY;
1556
1557         switch(stage) {
1558         case OBD_CLEANUP_EARLY:
1559                 break;
1560         case OBD_CLEANUP_EXPORTS:
1561                 target_cleanup_recovery(obd);
1562                 break;
1563         }
1564         RETURN(rc);
1565 }
1566
1567 static int ofd_ping(const struct lu_env *env, struct obd_export *exp)
1568 {
1569         ofd_fmd_expire(exp);
1570         return 0;
1571 }
1572
1573 static int ofd_health_check(const struct lu_env *nul, struct obd_device *obd)
1574 {
1575         struct ofd_device       *ofd = ofd_dev(obd->obd_lu_dev);
1576         struct ofd_thread_info  *info;
1577         struct lu_env            env;
1578 #ifdef USE_HEALTH_CHECK_WRITE
1579         struct thandle          *th;
1580 #endif
1581         int                      rc = 0;
1582
1583         /* obd_proc_read_health pass NULL env, we need real one */
1584         rc = lu_env_init(&env, LCT_DT_THREAD);
1585         if (rc)
1586                 RETURN(rc);
1587
1588         info = ofd_info_init(&env, NULL);
1589         rc = dt_statfs(&env, ofd->ofd_osd, &info->fti_u.osfs);
1590         if (unlikely(rc))
1591                 GOTO(out, rc);
1592
1593         if (info->fti_u.osfs.os_state == OS_STATE_READONLY)
1594                 GOTO(out, rc = -EROFS);
1595
1596 #ifdef USE_HEALTH_CHECK_WRITE
1597         OBD_ALLOC(info->fti_buf.lb_buf, CFS_PAGE_SIZE);
1598         if (info->fti_buf.lb_buf == NULL)
1599                 GOTO(out, rc = -ENOMEM);
1600
1601         info->fti_buf.lb_len = CFS_PAGE_SIZE;
1602         info->fti_off = 0;
1603
1604         th = dt_trans_create(&env, ofd->ofd_osd);
1605         if (IS_ERR(th))
1606                 GOTO(out, rc = PTR_ERR(th));
1607
1608         rc = dt_declare_record_write(&env, ofd->ofd_health_check_file,
1609                                      info->fti_buf.lb_len, info->fti_off, th);
1610         if (rc == 0) {
1611                 th->th_sync = 1; /* sync IO is needed */
1612                 rc = dt_trans_start_local(&env, ofd->ofd_osd, th);
1613                 if (rc == 0)
1614                         rc = dt_record_write(&env, ofd->ofd_health_check_file,
1615                                              &info->fti_buf, &info->fti_off,
1616                                              th);
1617         }
1618         dt_trans_stop(&env, ofd->ofd_osd, th);
1619
1620         OBD_FREE(info->fti_buf.lb_buf, CFS_PAGE_SIZE);
1621
1622         CDEBUG(D_INFO, "write 1 page synchronously for checking io rc %d\n",rc);
1623 #endif
1624 out:
1625         lu_env_fini(&env);
1626         return !!rc;
1627 }
1628
1629 /*
1630  * Handle quota control requests to consult current usage/limit.
1631  *
1632  * \param obd - is the obd device associated with the ofd
1633  * \param exp - is the client's export
1634  * \param oqctl - is the obd_quotactl request to be processed
1635  */
1636 static int ofd_quotactl(struct obd_device *obd, struct obd_export *exp,
1637                         struct obd_quotactl *oqctl)
1638 {
1639         struct ofd_device  *ofd = ofd_dev(obd->obd_lu_dev);
1640         struct lu_env       env;
1641         int                 rc;
1642         ENTRY;
1643
1644         /* report success for quota on/off for interoperability with current MDT
1645          * stack */
1646         if (oqctl->qc_cmd == Q_QUOTAON || oqctl->qc_cmd == Q_QUOTAOFF)
1647                 RETURN(0);
1648
1649         rc = lu_env_init(&env, LCT_DT_THREAD);
1650         if (rc)
1651                 RETURN(rc);
1652
1653         rc = lquotactl_slv(&env, ofd->ofd_osd, oqctl);
1654         lu_env_fini(&env);
1655
1656         RETURN(rc);
1657 }
1658
1659 struct obd_ops ofd_obd_ops = {
1660         .o_owner                = THIS_MODULE,
1661         .o_connect              = ofd_obd_connect,
1662         .o_reconnect            = ofd_obd_reconnect,
1663         .o_disconnect           = ofd_obd_disconnect,
1664         .o_set_info_async       = ofd_set_info_async,
1665         .o_get_info             = ofd_get_info,
1666         .o_create               = ofd_create,
1667         .o_statfs               = ofd_statfs,
1668         .o_setattr              = ofd_setattr,
1669         .o_preprw               = ofd_preprw,
1670         .o_commitrw             = ofd_commitrw,
1671         .o_destroy              = ofd_destroy,
1672         .o_init_export          = ofd_init_export,
1673         .o_destroy_export       = ofd_destroy_export,
1674         .o_postrecov            = ofd_obd_postrecov,
1675         .o_punch                = ofd_punch,
1676         .o_getattr              = ofd_getattr,
1677         .o_sync                 = ofd_sync,
1678         .o_iocontrol            = ofd_iocontrol,
1679         .o_precleanup           = ofd_precleanup,
1680         .o_ping                 = ofd_ping,
1681         .o_health_check         = ofd_health_check,
1682         .o_quotactl             = ofd_quotactl,
1683 };