lustre/ofd/ofd_obd.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2009, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2012, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/ofd/ofd_obd.c
  37  *
  38  * Author: Andreas Dilger <adilger@whamcloud.com>
  39  * Author: Alex Zhuravlev <bzzz@whamcloud.com>
  40  * Author: Mike Pershin <tappro@whamcloud.com>
  41  */
  42
  43 #define DEBUG_SUBSYSTEM S_FILTER
  44
  45 #include "ofd_internal.h"
  46 #include <obd_cksum.h>
  47 #include <lustre_quota.h>
  48
  49 static int ofd_export_stats_init(struct ofd_device *ofd,
  50                                  struct obd_export *exp, void *client_nid)
  51 {
  52         struct obd_device       *obd = ofd_obd(ofd);
  53         struct nid_stat         *stats;
  54         int                      num_stats, i;
  55         int                      rc, newnid = 0;
  56
  57         ENTRY;
  58
  59         if (obd_uuid_equals(&exp->exp_client_uuid, &obd->obd_uuid))
  60                 /* Self-export gets no proc entry */
  61                 RETURN(0);
  62
  63         rc = lprocfs_exp_setup(exp, client_nid, &newnid);
  64         if (rc) {
  65                 /* Mask error for already created
  66                  * /proc entries */
  67                 if (rc == -EALREADY)
  68                         rc = 0;
  69                 RETURN(rc);
  70         }
  71
  72         if (newnid == 0)
  73                 RETURN(0);
  74
  75         stats = exp->exp_nid_stats;
  76         LASSERT(stats != NULL);
  77
  78         OBD_ALLOC(stats->nid_brw_stats, sizeof(struct brw_stats));
  79         if (stats->nid_brw_stats == NULL)
  80                 GOTO(clean, rc = -ENOMEM);
  81
  82         for (i = 0; i < BRW_LAST; i++)
  83                 spin_lock_init(&stats->nid_brw_stats->hist[i].oh_lock);
  84
  85         rc = lprocfs_seq_create(stats->nid_proc, "brw_stats", 0644,
  86                                 &ofd_per_nid_stats_fops, stats);
  87         if (rc)
  88                 CWARN("Error adding the brw_stats file\n");
  89
  90         num_stats = (sizeof(*obd->obd_type->typ_dt_ops) / sizeof(void *)) +
  91                      LPROC_OFD_LAST - 1;
  92
  93         stats->nid_stats = lprocfs_alloc_stats(num_stats,
  94                                                LPROCFS_STATS_FLAG_NOPERCPU);
  95         if (stats->nid_stats == NULL)
  96                 return -ENOMEM;
  97
  98         lprocfs_init_ops_stats(LPROC_OFD_LAST, stats->nid_stats);
  99         lprocfs_counter_init(stats->nid_stats, LPROC_OFD_READ_BYTES,
 100                              LPROCFS_CNTR_AVGMINMAX, "read_bytes", "bytes");
 101         lprocfs_counter_init(stats->nid_stats, LPROC_OFD_WRITE_BYTES,
 102                              LPROCFS_CNTR_AVGMINMAX, "write_bytes", "bytes");
 103
 104         rc = lprocfs_register_stats(stats->nid_proc, "stats",
 105                                     stats->nid_stats);
 106         if (rc)
 107                 GOTO(clean, rc);
 108
 109         rc = lprocfs_nid_ldlm_stats_init(stats);
 110         if (rc) {
 111                 lprocfs_free_stats(&stats->nid_stats);
 112                 GOTO(clean, rc);
 113         }
 114
 115         RETURN(0);
 116 clean:
 117         return rc;
 118 }
 119
 120 static int ofd_parse_connect_data(const struct lu_env *env,
 121                                   struct obd_export *exp,
 122                                   struct obd_connect_data *data)
 123 {
 124         struct ofd_device                *ofd = ofd_exp(exp);
 125         struct filter_export_data        *fed = &exp->exp_filter_data;
 126
 127         if (!data)
 128                 RETURN(0);
 129
 130         CDEBUG(D_RPCTRACE, "%s: cli %s/%p ocd_connect_flags: "LPX64
 131                " ocd_version: %x ocd_grant: %d ocd_index: %u\n",
 132                exp->exp_obd->obd_name, exp->exp_client_uuid.uuid, exp,
 133                data->ocd_connect_flags, data->ocd_version,
 134                data->ocd_grant, data->ocd_index);
 135
 136         if (fed->fed_group != 0 && fed->fed_group != data->ocd_group) {
 137                 CWARN("!!! This export (nid %s) used object group %d "
 138                       "earlier; now it's trying to use group %d!  This could "
 139                       "be a bug in the MDS. Please report to "
 140                       "http://bugs.whamcloud.com/\n",
 141                       obd_export_nid2str(exp), fed->fed_group,
 142                       data->ocd_group);
 143                 RETURN(-EPROTO);
 144         }
 145         fed->fed_group = data->ocd_group;
 146
 147         data->ocd_connect_flags &= OST_CONNECT_SUPPORTED;
 148         exp->exp_connect_flags = data->ocd_connect_flags;
 149         data->ocd_version = LUSTRE_VERSION_CODE;
 150
 151         /* Kindly make sure the SKIP_ORPHAN flag is from MDS. */
 152         if (data->ocd_connect_flags & OBD_CONNECT_MDS)
 153                 CDEBUG(D_HA, "%s: Received MDS connection for group %u\n",
 154                        exp->exp_obd->obd_name, data->ocd_group);
 155         else if (data->ocd_connect_flags & OBD_CONNECT_SKIP_ORPHAN)
 156                 RETURN(-EPROTO);
 157
 158         if (ofd_grant_param_supp(exp)) {
 159                 exp->exp_filter_data.fed_pagesize = data->ocd_blocksize;
 160                 /* ocd_{blocksize,inodespace} are log2 values */
 161                 data->ocd_blocksize  = ofd->ofd_blockbits;
 162                 data->ocd_inodespace = ofd->ofd_dt_conf.ddp_inodespace;
 163                 /* ocd_grant_extent is in 1K blocks */
 164                 data->ocd_grant_extent = ofd->ofd_dt_conf.ddp_grant_frag >> 10;
 165         }
 166
 167         if (exp->exp_connect_flags & OBD_CONNECT_GRANT)
 168                 data->ocd_grant = ofd_grant_connect(env, exp, data->ocd_grant);
 169
 170         if (data->ocd_connect_flags & OBD_CONNECT_INDEX) {
 171                 struct lr_server_data *lsd = &ofd->ofd_lut.lut_lsd;
 172                 int                    index = lsd->lsd_ost_index;
 173
 174                 if (!(lsd->lsd_feature_compat & OBD_COMPAT_OST)) {
 175                         /* this will only happen on the first connect */
 176                         lsd->lsd_ost_index = data->ocd_index;
 177                         lsd->lsd_feature_compat |= OBD_COMPAT_OST;
 178                         /* sync is not needed here as lut_client_add will
 179                          * set exp_need_sync flag */
 180                         tgt_server_data_update(env, &ofd->ofd_lut, 0);
 181                 } else if (index != data->ocd_index) {
 182                         LCONSOLE_ERROR_MSG(0x136, "Connection from %s to index"
 183                                            " %u doesn't match actual OST index"
 184                                            " %u in last_rcvd file, bad "
 185                                            "configuration?\n",
 186                                            obd_export_nid2str(exp), index,
 187                                            data->ocd_index);
 188                         RETURN(-EBADF);
 189                 }
 190         }
 191
 192         if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_SIZE)) {
 193                 data->ocd_brw_size = 65536;
 194         } else if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE) {
 195                 data->ocd_brw_size = min(data->ocd_brw_size,
 196                               (__u32)(PTLRPC_MAX_BRW_PAGES << CFS_PAGE_SHIFT));
 197                 if (data->ocd_brw_size == 0) {
 198                         CERROR("%s: cli %s/%p ocd_connect_flags: "LPX64
 199                                " ocd_version: %x ocd_grant: %d ocd_index: %u "
 200                                "ocd_brw_size is unexpectedly zero, "
 201                                "network data corruption?"
 202                                "Refusing connection of this client\n",
 203                                exp->exp_obd->obd_name,
 204                                exp->exp_client_uuid.uuid,
 205                                exp, data->ocd_connect_flags, data->ocd_version,
 206                                data->ocd_grant, data->ocd_index);
 207                         RETURN(-EPROTO);
 208                 }
 209         }
 210
 211         if (data->ocd_connect_flags & OBD_CONNECT_CKSUM) {
 212                 __u32 cksum_types = data->ocd_cksum_types;
 213
 214                 /* The client set in ocd_cksum_types the checksum types it
 215                  * supports. We have to mask off the algorithms that we don't
 216                  * support */
 217                 data->ocd_cksum_types &= cksum_types_supported_server();
 218
 219                 if (unlikely(data->ocd_cksum_types == 0)) {
 220                         CERROR("%s: Connect with checksum support but no "
 221                                "ocd_cksum_types is set\n",
 222                                exp->exp_obd->obd_name);
 223                         RETURN(-EPROTO);
 224                 }
 225
 226                 CDEBUG(D_RPCTRACE, "%s: cli %s supports cksum type %x, return "
 227                        "%x\n", exp->exp_obd->obd_name, obd_export_nid2str(exp),
 228                        cksum_types, data->ocd_cksum_types);
 229         } else {
 230                 /* This client does not support OBD_CONNECT_CKSUM
 231                  * fall back to CRC32 */
 232                 CDEBUG(D_RPCTRACE, "%s: cli %s does not support "
 233                        "OBD_CONNECT_CKSUM, CRC32 will be used\n",
 234                        exp->exp_obd->obd_name, obd_export_nid2str(exp));
 235         }
 236
 237         if (data->ocd_connect_flags & OBD_CONNECT_MAXBYTES)
 238                 data->ocd_maxbytes = ofd->ofd_dt_conf.ddp_maxbytes;
 239
 240         RETURN(0);
 241 }
 242
 243 static int ofd_obd_reconnect(const struct lu_env *env, struct obd_export *exp,
 244                              struct obd_device *obd, struct obd_uuid *cluuid,
 245                              struct obd_connect_data *data, void *localdata)
 246 {
 247         struct ofd_device       *ofd = ofd_dev(obd->obd_lu_dev);
 248         int                      rc;
 249
 250         ENTRY;
 251
 252         if (exp == NULL || obd == NULL || cluuid == NULL)
 253                 RETURN(-EINVAL);
 254
 255         rc = lu_env_refill((struct lu_env *)env);
 256         if (rc != 0) {
 257                 CERROR("Failure to refill session: '%d'\n", rc);
 258                 RETURN(rc);
 259         }
 260
 261         ofd_info_init(env, exp);
 262         rc = ofd_parse_connect_data(env, exp, data);
 263         if (rc == 0)
 264                 ofd_export_stats_init(ofd, exp, localdata);
 265
 266         RETURN(rc);
 267 }
 268
 269 static int ofd_obd_connect(const struct lu_env *env, struct obd_export **_exp,
 270                            struct obd_device *obd, struct obd_uuid *cluuid,
 271                            struct obd_connect_data *data, void *localdata)
 272 {
 273         struct obd_export       *exp;
 274         struct ofd_device       *ofd;
 275         struct lustre_handle     conn = { 0 };
 276         int                      rc;
 277         ENTRY;
 278
 279         if (_exp == NULL || obd == NULL || cluuid == NULL)
 280                 RETURN(-EINVAL);
 281
 282         ofd = ofd_dev(obd->obd_lu_dev);
 283
 284         rc = class_connect(&conn, obd, cluuid);
 285         if (rc)
 286                 RETURN(rc);
 287
 288         exp = class_conn2export(&conn);
 289         LASSERT(exp != NULL);
 290
 291         rc = lu_env_refill((struct lu_env *)env);
 292         if (rc != 0) {
 293                 CERROR("Failure to refill session: '%d'\n", rc);
 294                 GOTO(out, rc);
 295         }
 296
 297         ofd_info_init(env, exp);
 298
 299         rc = ofd_parse_connect_data(env, exp, data);
 300         if (rc)
 301                 GOTO(out, rc);
 302
 303         if (obd->obd_replayable) {
 304                 struct tg_export_data *ted = &exp->exp_target_data;
 305
 306                 memcpy(ted->ted_lcd->lcd_uuid, cluuid,
 307                        sizeof(ted->ted_lcd->lcd_uuid));
 308                 rc = tgt_client_new(env, exp);
 309                 if (rc != 0)
 310                         GOTO(out, rc);
 311                 ofd_export_stats_init(ofd, exp, localdata);
 312         }
 313
 314         CDEBUG(D_HA, "%s: get connection from MDS %d\n", obd->obd_name,
 315                data->ocd_group);
 316
 317 out:
 318         if (rc != 0) {
 319                 class_disconnect(exp);
 320                 *_exp = NULL;
 321         } else {
 322                 *_exp = exp;
 323         }
 324         RETURN(rc);
 325 }
 326
 327 static int ofd_obd_disconnect(struct obd_export *exp)
 328 {
 329         struct ofd_device       *ofd = ofd_dev(exp->exp_obd->obd_lu_dev);
 330         struct lu_env            env;
 331         int                      rc;
 332
 333         ENTRY;
 334
 335         LASSERT(exp);
 336         class_export_get(exp);
 337
 338         if (!(exp->exp_flags & OBD_OPT_FORCE))
 339                 ofd_grant_sanity_check(ofd_obd(ofd), __FUNCTION__);
 340
 341         rc = server_disconnect_export(exp);
 342
 343         ofd_grant_discard(exp);
 344
 345         rc = lu_env_init(&env, LCT_DT_THREAD);
 346         if (rc)
 347                 RETURN(rc);
 348
 349         /* Do not erase record for recoverable client. */
 350         if (exp->exp_obd->obd_replayable &&
 351             (!exp->exp_obd->obd_fail || exp->exp_failed))
 352                 tgt_client_del(&env, exp);
 353         lu_env_fini(&env);
 354
 355         class_export_put(exp);
 356         RETURN(rc);
 357 }
 358
 359 static int ofd_init_export(struct obd_export *exp)
 360 {
 361         int rc;
 362
 363         spin_lock_init(&exp->exp_filter_data.fed_lock);
 364         CFS_INIT_LIST_HEAD(&exp->exp_filter_data.fed_mod_list);
 365         spin_lock(&exp->exp_lock);
 366         exp->exp_connecting = 1;
 367         spin_unlock(&exp->exp_lock);
 368
 369         /* self-export doesn't need client data and ldlm initialization */
 370         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
 371                                      &exp->exp_client_uuid)))
 372                 return 0;
 373
 374         rc = tgt_client_alloc(exp);
 375         if (rc == 0)
 376                 ldlm_init_export(exp);
 377         if (rc)
 378                 CERROR("%s: Can't initialize export: rc %d\n",
 379                        exp->exp_obd->obd_name, rc);
 380         return rc;
 381 }
 382
 383 static int ofd_destroy_export(struct obd_export *exp)
 384 {
 385         struct ofd_device *ofd = ofd_dev(exp->exp_obd->obd_lu_dev);
 386
 387         if (exp->exp_filter_data.fed_pending)
 388                 CERROR("%s: cli %s/%p has %lu pending on destroyed export"
 389                        "\n", exp->exp_obd->obd_name, exp->exp_client_uuid.uuid,
 390                        exp, exp->exp_filter_data.fed_pending);
 391
 392         target_destroy_export(exp);
 393
 394         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
 395                                      &exp->exp_client_uuid)))
 396                 return 0;
 397
 398         ldlm_destroy_export(exp);
 399         tgt_client_free(exp);
 400
 401         ofd_fmd_cleanup(exp);
 402
 403         /*
 404          * discard grants once we're sure no more
 405          * interaction with the client is possible
 406          */
 407         ofd_grant_discard(exp);
 408         ofd_fmd_cleanup(exp);
 409
 410         if (exp->exp_connect_flags & OBD_CONNECT_GRANT_SHRINK) {
 411                 if (ofd->ofd_tot_granted_clients > 0)
 412                         ofd->ofd_tot_granted_clients --;
 413         }
 414
 415         if (!(exp->exp_flags & OBD_OPT_FORCE))
 416                 ofd_grant_sanity_check(exp->exp_obd, __FUNCTION__);
 417
 418         LASSERT(cfs_list_empty(&exp->exp_filter_data.fed_mod_list));
 419         return 0;
 420 }
 421
 422 int ofd_postrecov(const struct lu_env *env, struct ofd_device *ofd)
 423 {
 424         struct lu_device *ldev = &ofd->ofd_dt_dev.dd_lu_dev;
 425
 426         CDEBUG(D_HA, "%s: recovery is over\n", ofd_obd(ofd)->obd_name);
 427         return ldev->ld_ops->ldo_recovery_complete(env, ldev);
 428 }
 429
 430 int ofd_obd_postrecov(struct obd_device *obd)
 431 {
 432         struct lu_env            env;
 433         struct lu_device        *ldev = obd->obd_lu_dev;
 434         int                      rc;
 435
 436         ENTRY;
 437
 438         rc = lu_env_init(&env, LCT_DT_THREAD);
 439         if (rc)
 440                 RETURN(rc);
 441         ofd_info_init(&env, obd->obd_self_export);
 442
 443         rc = ofd_postrecov(&env, ofd_dev(ldev));
 444
 445         lu_env_fini(&env);
 446         RETURN(rc);
 447 }
 448
 449 static int ofd_adapt_sptlrpc_conf(const struct lu_env *env,
 450                                   struct obd_device *obd, int initial)
 451 {
 452         struct filter_obd       *fo = &obd->u.filter;
 453         struct sptlrpc_rule_set  tmp_rset;
 454         int                      rc;
 455
 456         sptlrpc_rule_set_init(&tmp_rset);
 457         rc = sptlrpc_conf_target_get_rules(obd, &tmp_rset, initial);
 458         if (rc) {
 459                 CERROR("%s: failed get sptlrpc rules: rc = %d\n",
 460                        obd->obd_name, rc);
 461                 return rc;
 462         }
 463
 464         sptlrpc_target_update_exp_flavor(obd, &tmp_rset);
 465
 466         write_lock(&fo->fo_sptlrpc_lock);
 467         sptlrpc_rule_set_free(&fo->fo_sptlrpc_rset);
 468         fo->fo_sptlrpc_rset = tmp_rset;
 469         write_unlock(&fo->fo_sptlrpc_lock);
 470
 471         return 0;
 472 }
 473
 474 static int ofd_set_mds_conn(struct obd_export *exp, void *val)
 475 {
 476         int rc = 0;
 477
 478         ENTRY;
 479
 480         LCONSOLE_WARN("%s: received MDS connection from %s\n",
 481                       exp->exp_obd->obd_name, obd_export_nid2str(exp));
 482         RETURN(rc);
 483 }
 484
 485 static int ofd_set_info_async(const struct lu_env *env, struct obd_export *exp,
 486                               __u32 keylen, void *key, __u32 vallen, void *val,
 487                               struct ptlrpc_request_set *set)
 488 {
 489         struct ofd_device       *ofd = ofd_exp(exp);
 490         int                      rc = 0;
 491
 492         ENTRY;
 493
 494         if (exp->exp_obd == NULL) {
 495                 CDEBUG(D_IOCTL, "invalid export %p\n", exp);
 496                 RETURN(-EINVAL);
 497         }
 498
 499         if (KEY_IS(KEY_CAPA_KEY)) {
 500                 rc = ofd_update_capa_key(ofd, val);
 501                 if (rc)
 502                         CERROR("%s: update capability key failed: rc = %d\n",
 503                                exp->exp_obd->obd_name, rc);
 504         } else if (KEY_IS(KEY_SPTLRPC_CONF)) {
 505                 ofd_adapt_sptlrpc_conf(env, exp->exp_obd, 0);
 506         } else if (KEY_IS(KEY_MDS_CONN)) {
 507                 rc = ofd_set_mds_conn(exp, val);
 508         } else if (KEY_IS(KEY_GRANT_SHRINK)) {
 509                 struct ost_body *body = val;
 510
 511                 ofd_info_init(env, exp);
 512                 /** handle grant shrink, similar to a read request */
 513                 ofd_grant_prepare_read(env, exp, &body->oa);
 514         } else {
 515                 CERROR("%s: Unsupported key %s\n",
 516                        exp->exp_obd->obd_name, (char*)key);
 517                 rc = -EOPNOTSUPP;
 518         }
 519
 520         RETURN(rc);
 521 }
 522
 523 static int ofd_get_info(const struct lu_env *env, struct obd_export *exp,
 524                         __u32 keylen, void *key, __u32 *vallen, void *val,
 525                         struct lov_stripe_md *lsm)
 526 {
 527         struct ofd_device       *ofd = ofd_exp(exp);
 528         int                      rc = 0;
 529
 530         ENTRY;
 531
 532         if (exp->exp_obd == NULL) {
 533                 CDEBUG(D_IOCTL, "invalid client export %p\n", exp);
 534                 RETURN(-EINVAL);
 535         }
 536
 537         if (KEY_IS(KEY_BLOCKSIZE)) {
 538                 __u32 *blocksize = val;
 539                 if (blocksize) {
 540                         if (*vallen < sizeof(*blocksize))
 541                                 RETURN(-EOVERFLOW);
 542                         *blocksize = 1 << ofd->ofd_dt_conf.ddp_block_shift;
 543                 }
 544                 *vallen = sizeof(*blocksize);
 545         } else if (KEY_IS(KEY_BLOCKSIZE_BITS)) {
 546                 __u32 *blocksize_bits = val;
 547                 if (blocksize_bits) {
 548                         if (*vallen < sizeof(*blocksize_bits))
 549                                 RETURN(-EOVERFLOW);
 550                         *blocksize_bits = ofd->ofd_dt_conf.ddp_block_shift;
 551                 }
 552                 *vallen = sizeof(*blocksize_bits);
 553         } else if (KEY_IS(KEY_LAST_ID)) {
 554                 obd_id *last_id = val;
 555                 struct ofd_seq *oseq;
 556
 557                 if (val == NULL) {
 558                         *vallen = sizeof(obd_id);
 559                         RETURN(0);
 560                 }
 561                 ofd_info_init(env, exp);
 562                 oseq = ofd_seq_load(env, ofd,
 563                                     (obd_seq)exp->exp_filter_data.fed_group);
 564                 LASSERT(oseq != NULL);
 565                 if (last_id) {
 566                         if (*vallen < sizeof(*last_id)) {
 567                                 ofd_seq_put(env, oseq);
 568                                 RETURN(-EOVERFLOW);
 569                         }
 570                         *last_id = ofd_seq_last_oid(oseq);
 571                 }
 572                 ofd_seq_put(env, oseq);
 573                 *vallen = sizeof(*last_id);
 574         } else if (KEY_IS(KEY_FIEMAP)) {
 575                 struct ofd_thread_info          *info;
 576                 struct ofd_device               *ofd = ofd_exp(exp);
 577                 struct ofd_object               *fo;
 578                 struct ll_fiemap_info_key       *fm_key = key;
 579
 580                 if (val == NULL) {
 581                         *vallen = fiemap_count_to_size(
 582                                                fm_key->fiemap.fm_extent_count);
 583                         RETURN(0);
 584                 }
 585
 586                 info = ofd_info_init(env, exp);
 587
 588                 fid_ostid_unpack(&info->fti_fid, &fm_key->oa.o_oi, 0);
 589
 590                 CDEBUG(D_INODE, "get FIEMAP of object "DFID"\n",
 591                        PFID(&info->fti_fid));
 592
 593                 fo = ofd_object_find(env, ofd, &info->fti_fid);
 594                 if (IS_ERR(fo)) {
 595                         CERROR("%s: error finding object "DFID"\n",
 596                                exp->exp_obd->obd_name, PFID(&info->fti_fid));
 597                         rc = PTR_ERR(fo);
 598                 } else {
 599                         struct ll_user_fiemap *fiemap = val;
 600
 601                         ofd_read_lock(env, fo);
 602                         if (ofd_object_exists(fo)) {
 603                                 *fiemap = fm_key->fiemap;
 604                                 rc = dt_fiemap_get(env,
 605                                                    ofd_object_child(fo),
 606                                                    fiemap);
 607                         } else {
 608                                 rc = -ENOENT;
 609                         }
 610                         ofd_read_unlock(env, fo);
 611                         ofd_object_put(env, fo);
 612                 }
 613         } else if (KEY_IS(KEY_SYNC_LOCK_CANCEL)) {
 614                 *((__u32 *) val) = ofd->ofd_sync_lock_cancel;
 615                 *vallen = sizeof(__u32);
 616         } else {
 617                 CERROR("Not supported key %s\n", (char*)key);
 618                 rc = -EOPNOTSUPP;
 619         }
 620
 621         RETURN(rc);
 622 }
 623
 624 /** helper function for statfs, also used by grant code */
 625 int ofd_statfs_internal(const struct lu_env *env, struct ofd_device *ofd,
 626                         struct obd_statfs *osfs, __u64 max_age, int *from_cache)
 627 {
 628         int rc;
 629
 630         spin_lock(&ofd->ofd_osfs_lock);
 631         if (cfs_time_before_64(ofd->ofd_osfs_age, max_age) || max_age == 0) {
 632                 obd_size unstable;
 633
 634                 /* statfs data are too old, get up-to-date one.
 635                  * we must be cautious here since multiple threads might be
 636                  * willing to update statfs data concurrently and we must
 637                  * grant that cached statfs data are always consistent */
 638
 639                 if (ofd->ofd_statfs_inflight == 0)
 640                         /* clear inflight counter if no users, although it would
 641                          * take a while to overflow this 64-bit counter ... */
 642                         ofd->ofd_osfs_inflight = 0;
 643                 /* notify ofd_grant_commit() that we want to track writes
 644                  * completed as of now */
 645                 ofd->ofd_statfs_inflight++;
 646                 /* record value of inflight counter before running statfs to
 647                  * compute the diff once statfs is completed */
 648                 unstable = ofd->ofd_osfs_inflight;
 649                 spin_unlock(&ofd->ofd_osfs_lock);
 650
 651                 /* statfs can sleep ... hopefully not for too long since we can
 652                  * call it fairly often as space fills up */
 653                 rc = dt_statfs(env, ofd->ofd_osd, osfs);
 654                 if (unlikely(rc))
 655                         return rc;
 656
 657                 spin_lock(&ofd->ofd_grant_lock);
 658                 spin_lock(&ofd->ofd_osfs_lock);
 659                 /* calculate how much space was written while we released the
 660                  * ofd_osfs_lock */
 661                 unstable = ofd->ofd_osfs_inflight - unstable;
 662                 ofd->ofd_osfs_unstable = 0;
 663                 if (unstable) {
 664                         /* some writes completed while we were running statfs
 665                          * w/o the ofd_osfs_lock. Those ones got added to
 666                          * the cached statfs data that we are about to crunch.
 667                          * Take them into account in the new statfs data */
 668                         osfs->os_bavail -= min_t(obd_size, osfs->os_bavail,
 669                                                unstable >> ofd->ofd_blockbits);
 670                         /* However, we don't really know if those writes got
 671                          * accounted in the statfs call, so tell
 672                          * ofd_grant_space_left() there is some uncertainty
 673                          * on the accounting of those writes.
 674                          * The purpose is to prevent spurious error messages in
 675                          * ofd_grant_space_left() since those writes might be
 676                          * accounted twice. */
 677                         ofd->ofd_osfs_unstable += unstable;
 678                 }
 679                 /* similarly, there is some uncertainty on write requests
 680                  * between prepare & commit */
 681                 ofd->ofd_osfs_unstable += ofd->ofd_tot_pending;
 682                 spin_unlock(&ofd->ofd_grant_lock);
 683
 684                 /* finally udpate cached statfs data */
 685                 ofd->ofd_osfs = *osfs;
 686                 ofd->ofd_osfs_age = cfs_time_current_64();
 687
 688                 ofd->ofd_statfs_inflight--; /* stop tracking */
 689                 if (ofd->ofd_statfs_inflight == 0)
 690                         ofd->ofd_osfs_inflight = 0;
 691                 spin_unlock(&ofd->ofd_osfs_lock);
 692
 693                 if (from_cache)
 694                         *from_cache = 0;
 695         } else {
 696                 /* use cached statfs data */
 697                 *osfs = ofd->ofd_osfs;
 698                 spin_unlock(&ofd->ofd_osfs_lock);
 699                 if (from_cache)
 700                         *from_cache = 1;
 701         }
 702         return 0;
 703 }
 704
 705 static int ofd_statfs(const struct lu_env *env,  struct obd_export *exp,
 706                       struct obd_statfs *osfs, __u64 max_age, __u32 flags)
 707 {
 708         struct obd_device       *obd = class_exp2obd(exp);
 709         struct ofd_device       *ofd = ofd_dev(exp->exp_obd->obd_lu_dev);
 710         int                      rc;
 711
 712         ENTRY;
 713
 714         rc = ofd_statfs_internal(env, ofd, osfs, max_age, NULL);
 715         if (unlikely(rc))
 716                 GOTO(out, rc);
 717
 718         /* at least try to account for cached pages.  its still racy and
 719          * might be under-reporting if clients haven't announced their
 720          * caches with brw recently */
 721
 722         CDEBUG(D_SUPER | D_CACHE, "blocks cached "LPU64" granted "LPU64
 723                " pending "LPU64" free "LPU64" avail "LPU64"\n",
 724                ofd->ofd_tot_dirty, ofd->ofd_tot_granted, ofd->ofd_tot_pending,
 725                osfs->os_bfree << ofd->ofd_blockbits,
 726                osfs->os_bavail << ofd->ofd_blockbits);
 727
 728         osfs->os_bavail -= min_t(obd_size, osfs->os_bavail,
 729                                  ((ofd->ofd_tot_dirty + ofd->ofd_tot_pending +
 730                                    osfs->os_bsize - 1) >> ofd->ofd_blockbits));
 731
 732         /* The QoS code on the MDS does not care about space reserved for
 733          * precreate, so take it out. */
 734         if (exp->exp_connect_flags & OBD_CONNECT_MDS) {
 735                 struct filter_export_data *fed;
 736
 737                 fed = &obd->obd_self_export->exp_filter_data;
 738                 osfs->os_bavail -= min_t(obd_size, osfs->os_bavail,
 739                                          fed->fed_grant >> ofd->ofd_blockbits);
 740         }
 741
 742         ofd_grant_sanity_check(obd, __FUNCTION__);
 743         CDEBUG(D_CACHE, LPU64" blocks: "LPU64" free, "LPU64" avail; "
 744                LPU64" objects: "LPU64" free; state %x\n",
 745                osfs->os_blocks, osfs->os_bfree, osfs->os_bavail,
 746                osfs->os_files, osfs->os_ffree, osfs->os_state);
 747
 748         if (OBD_FAIL_CHECK_VALUE(OBD_FAIL_OST_ENOINO,
 749                                  ofd->ofd_lut.lut_lsd.lsd_ost_index))
 750                 osfs->os_ffree = 0;
 751
 752         /* OS_STATE_READONLY can be set by OSD already */
 753         if (ofd->ofd_raid_degraded)
 754                 osfs->os_state |= OS_STATE_DEGRADED;
 755
 756         if (obd->obd_self_export != exp && ofd_grant_compat(exp, ofd)) {
 757                 /* clients which don't support OBD_CONNECT_GRANT_PARAM
 758                  * should not see a block size > page size, otherwise
 759                  * cl_lost_grant goes mad. Therefore, we emulate a 4KB (=2^12)
 760                  * block size which is the biggest block size known to work
 761                  * with all client's page size. */
 762                 osfs->os_blocks <<= ofd->ofd_blockbits - COMPAT_BSIZE_SHIFT;
 763                 osfs->os_bfree  <<= ofd->ofd_blockbits - COMPAT_BSIZE_SHIFT;
 764                 osfs->os_bavail <<= ofd->ofd_blockbits - COMPAT_BSIZE_SHIFT;
 765                 osfs->os_bsize    = 1 << COMPAT_BSIZE_SHIFT;
 766         }
 767
 768         if (OBD_FAIL_CHECK_VALUE(OBD_FAIL_OST_ENOSPC,
 769                                  ofd->ofd_lut.lut_lsd.lsd_ost_index))
 770                 osfs->os_bfree = osfs->os_bavail = 2;
 771
 772         EXIT;
 773 out:
 774         return rc;
 775 }
 776
 777 int ofd_setattr(const struct lu_env *env, struct obd_export *exp,
 778                 struct obd_info *oinfo, struct obd_trans_info *oti)
 779 {
 780         struct ofd_thread_info  *info;
 781         struct ofd_device       *ofd = ofd_exp(exp);
 782         struct ldlm_namespace   *ns = ofd->ofd_namespace;
 783         struct ldlm_resource    *res;
 784         struct ofd_object       *fo;
 785         struct obdo             *oa = oinfo->oi_oa;
 786         struct filter_fid       *ff = NULL;
 787         int                      rc = 0;
 788
 789         ENTRY;
 790
 791         info = ofd_info_init(env, exp);
 792         ofd_oti2info(info, oti);
 793
 794         fid_ostid_unpack(&info->fti_fid, &oinfo->oi_oa->o_oi, 0);
 795         ofd_build_resid(&info->fti_fid, &info->fti_resid);
 796
 797         rc = ofd_auth_capa(exp, &info->fti_fid, oa->o_seq,
 798                            oinfo_capa(oinfo), CAPA_OPC_META_WRITE);
 799         if (rc)
 800                 GOTO(out, rc);
 801
 802         /* This would be very bad - accidentally truncating a file when
 803          * changing the time or similar - bug 12203. */
 804         if (oinfo->oi_oa->o_valid & OBD_MD_FLSIZE &&
 805             oinfo->oi_policy.l_extent.end != OBD_OBJECT_EOF) {
 806                 static char mdsinum[48];
 807
 808                 if (oinfo->oi_oa->o_valid & OBD_MD_FLFID)
 809                         snprintf(mdsinum, sizeof(mdsinum) - 1,
 810                                  "of parent "DFID, oinfo->oi_oa->o_parent_seq,
 811                                  oinfo->oi_oa->o_parent_oid, 0);
 812                 else
 813                         mdsinum[0] = '\0';
 814
 815                 CERROR("%s: setattr from %s trying to truncate object "DFID
 816                        " %s\n", exp->exp_obd->obd_name,
 817                        obd_export_nid2str(exp), PFID(&info->fti_fid), mdsinum);
 818                 GOTO(out, rc = -EPERM);
 819         }
 820
 821         fo = ofd_object_find(env, ofd, &info->fti_fid);
 822         if (IS_ERR(fo)) {
 823                 CERROR("%s: can't find object "DFID"\n",
 824                        exp->exp_obd->obd_name, PFID(&info->fti_fid));
 825                 GOTO(out, rc = PTR_ERR(fo));
 826         }
 827
 828         la_from_obdo(&info->fti_attr, oinfo->oi_oa, oinfo->oi_oa->o_valid);
 829         info->fti_attr.la_valid &= ~LA_TYPE;
 830
 831         if (oa->o_valid & OBD_MD_FLFID) {
 832                 ff = &info->fti_mds_fid;
 833                 ofd_prepare_fidea(ff, oa);
 834         }
 835
 836         /* setting objects attributes (including owner/group) */
 837         rc = ofd_attr_set(env, fo, &info->fti_attr, ff);
 838         if (rc)
 839                 GOTO(out_unlock, rc);
 840
 841         res = ldlm_resource_get(ns, NULL, &info->fti_resid, LDLM_EXTENT, 0);
 842         if (res != NULL) {
 843                 ldlm_res_lvbo_update(res, NULL, 0);
 844                 ldlm_resource_putref(res);
 845         }
 846
 847         oinfo->oi_oa->o_valid = OBD_MD_FLID;
 848
 849         /* Quota release needs uid/gid info */
 850         rc = ofd_attr_get(env, fo, &info->fti_attr);
 851         obdo_from_la(oinfo->oi_oa, &info->fti_attr,
 852                      OFD_VALID_FLAGS | LA_UID | LA_GID);
 853         ofd_info2oti(info, oti);
 854
 855         ofd_counter_incr(exp, LPROC_OFD_STATS_SETATTR, oti->oti_jobid, 1);
 856         EXIT;
 857 out_unlock:
 858         ofd_object_put(env, fo);
 859 out:
 860         return rc;
 861 }
 862
 863 static int ofd_punch(const struct lu_env *env, struct obd_export *exp,
 864                      struct obd_info *oinfo, struct obd_trans_info *oti,
 865                      struct ptlrpc_request_set *rqset)
 866 {
 867         struct ofd_thread_info  *info;
 868         struct ofd_device       *ofd = ofd_exp(exp);
 869         struct ldlm_namespace   *ns = ofd->ofd_namespace;
 870         struct ldlm_resource    *res;
 871         struct ofd_object       *fo;
 872         struct filter_fid       *ff = NULL;
 873         int                      rc = 0;
 874
 875         ENTRY;
 876
 877         info = ofd_info_init(env, exp);
 878         ofd_oti2info(info, oti);
 879
 880         fid_ostid_unpack(&info->fti_fid, &oinfo->oi_oa->o_oi, 0);
 881         ofd_build_resid(&info->fti_fid, &info->fti_resid);
 882
 883         CDEBUG(D_INODE, "calling punch for object "DFID", valid = "LPX64
 884                ", start = "LPD64", end = "LPD64"\n", PFID(&info->fti_fid),
 885                oinfo->oi_oa->o_valid, oinfo->oi_policy.l_extent.start,
 886                oinfo->oi_policy.l_extent.end);
 887
 888         rc = ofd_auth_capa(exp, &info->fti_fid, oinfo->oi_oa->o_seq,
 889                            oinfo_capa(oinfo), CAPA_OPC_OSS_TRUNC);
 890         if (rc)
 891                 GOTO(out_env, rc);
 892
 893         fo = ofd_object_find(env, ofd, &info->fti_fid);
 894         if (IS_ERR(fo)) {
 895                 CERROR("%s: error finding object "DFID": rc = %ld\n",
 896                        exp->exp_obd->obd_name, PFID(&info->fti_fid),
 897                        PTR_ERR(fo));
 898                 GOTO(out_env, rc = PTR_ERR(fo));
 899         }
 900
 901         LASSERT(oinfo->oi_policy.l_extent.end == OBD_OBJECT_EOF);
 902         if (oinfo->oi_policy.l_extent.end == OBD_OBJECT_EOF) {
 903                 /* Truncate case */
 904                 oinfo->oi_oa->o_size = oinfo->oi_policy.l_extent.start;
 905         } else if (oinfo->oi_policy.l_extent.end >= oinfo->oi_oa->o_size) {
 906                 oinfo->oi_oa->o_size = oinfo->oi_policy.l_extent.end;
 907         }
 908
 909         la_from_obdo(&info->fti_attr, oinfo->oi_oa,
 910                      OBD_MD_FLMTIME | OBD_MD_FLATIME | OBD_MD_FLCTIME);
 911         info->fti_attr.la_valid &= ~LA_TYPE;
 912         info->fti_attr.la_size = oinfo->oi_policy.l_extent.start;
 913         info->fti_attr.la_valid |= LA_SIZE;
 914
 915         if (oinfo->oi_oa->o_valid & OBD_MD_FLFID) {
 916                 ff = &info->fti_mds_fid;
 917                 ofd_prepare_fidea(ff, oinfo->oi_oa);
 918         }
 919
 920         rc = ofd_object_punch(env, fo, oinfo->oi_policy.l_extent.start,
 921                               oinfo->oi_policy.l_extent.end, &info->fti_attr,
 922                               ff);
 923         if (rc)
 924                 GOTO(out, rc);
 925
 926         res = ldlm_resource_get(ns, NULL, &info->fti_resid, LDLM_EXTENT, 0);
 927         if (res != NULL) {
 928                 ldlm_res_lvbo_update(res, NULL, 0);
 929                 ldlm_resource_putref(res);
 930         }
 931
 932         oinfo->oi_oa->o_valid = OBD_MD_FLID;
 933         /* Quota release needs uid/gid info */
 934         rc = ofd_attr_get(env, fo, &info->fti_attr);
 935         obdo_from_la(oinfo->oi_oa, &info->fti_attr,
 936                      OFD_VALID_FLAGS | LA_UID | LA_GID);
 937         ofd_info2oti(info, oti);
 938
 939         ofd_counter_incr(exp, LPROC_OFD_STATS_PUNCH, oti->oti_jobid, 1);
 940         EXIT;
 941 out:
 942         ofd_object_put(env, fo);
 943 out_env:
 944         return rc;
 945 }
 946
 947 static int ofd_destroy_by_fid(const struct lu_env *env,
 948                               struct ofd_device *ofd,
 949                               const struct lu_fid *fid, int orphan)
 950 {
 951         struct ofd_thread_info  *info = ofd_info(env);
 952         struct lustre_handle     lockh;
 953         __u64                    flags = LDLM_AST_DISCARD_DATA, rc = 0;
 954         ldlm_policy_data_t       policy = {
 955                                         .l_extent = { 0, OBD_OBJECT_EOF }
 956                                  };
 957         struct ofd_object       *fo;
 958
 959         ENTRY;
 960
 961         fo = ofd_object_find(env, ofd, fid);
 962         if (IS_ERR(fo))
 963                 RETURN(PTR_ERR(fo));
 964
 965         /* Tell the clients that the object is gone now and that they should
 966          * throw away any cached pages. */
 967         ofd_build_resid(fid, &info->fti_resid);
 968         rc = ldlm_cli_enqueue_local(ofd->ofd_namespace, &info->fti_resid,
 969                                     LDLM_EXTENT, &policy, LCK_PW, &flags,
 970                                     ldlm_blocking_ast, ldlm_completion_ast,
 971                                     NULL, NULL, 0, LVB_T_NONE, NULL, &lockh);
 972
 973         /* We only care about the side-effects, just drop the lock. */
 974         if (rc == ELDLM_OK)
 975                 ldlm_lock_decref(&lockh, LCK_PW);
 976
 977         LASSERT(fo != NULL);
 978
 979         rc = ofd_object_destroy(env, fo, orphan);
 980
 981         ofd_object_put(env, fo);
 982         RETURN(rc);
 983 }
 984
 985 int ofd_destroy(const struct lu_env *env, struct obd_export *exp,
 986                 struct obdo *oa, struct lov_stripe_md *md,
 987                 struct obd_trans_info *oti, struct obd_export *md_exp,
 988                 void *capa)
 989 {
 990         struct ofd_device       *ofd = ofd_exp(exp);
 991         struct ofd_thread_info  *info;
 992         obd_count                count;
 993         int                      rc = 0;
 994
 995         ENTRY;
 996
 997         info = ofd_info_init(env, exp);
 998         ofd_oti2info(info, oti);
 999
1000         if (!(oa->o_valid & OBD_MD_FLGROUP))
1001                 oa->o_seq = 0;
1002
1003         /* check that o_misc makes sense */
1004         if (oa->o_valid & OBD_MD_FLOBJCOUNT)
1005                 count = oa->o_misc;
1006         else
1007                 count = 1; /* default case - single destroy */
1008
1009         /**
1010          * There can be sequence of objects to destroy. Therefore this request
1011          * may have multiple transaction involved in. It is OK, we need only
1012          * the highest used transno to be reported back in reply but not for
1013          * replays, they must report their transno
1014          */
1015         if (info->fti_transno == 0) /* not replay */
1016                 info->fti_mult_trans = 1;
1017         while (count > 0) {
1018                 int lrc;
1019
1020                 fid_ostid_unpack(&info->fti_fid, &oa->o_oi, 0);
1021                 lrc = ofd_destroy_by_fid(env, ofd, &info->fti_fid, 0);
1022                 if (lrc == -ENOENT) {
1023                         CDEBUG(D_INODE,
1024                                "%s: destroying non-existent object "DFID"\n",
1025                                ofd_obd(ofd)->obd_name, PFID(&info->fti_fid));
1026                         /* rewrite rc with -ENOENT only if it is 0 */
1027                         if (rc == 0)
1028                                 rc = lrc;
1029                 } else if (lrc != 0) {
1030                         CERROR("%s: error destroying object "DFID": %d\n",
1031                                ofd_obd(ofd)->obd_name, PFID(&info->fti_fid),
1032                                rc);
1033                         rc = lrc;
1034                 }
1035                 count--;
1036                 oa->o_id++;
1037         }
1038
1039         /* if we have transaction then there were some deletions, we don't
1040          * need to return ENOENT in that case because it will not wait
1041          * for commit of these deletions. The ENOENT must be returned only
1042          * if there were no transations.
1043          */
1044         if (rc == -ENOENT) {
1045                 if (info->fti_transno != 0)
1046                         rc = 0;
1047         } else if (rc != 0) {
1048                 /*
1049                  * If we have at least one transaction then llog record
1050                  * on server will be removed upon commit, so for rc != 0
1051                  * we return no transno and llog record will be reprocessed.
1052                  */
1053                 info->fti_transno = 0;
1054         }
1055         ofd_info2oti(info, oti);
1056         RETURN(rc);
1057 }
1058
1059 static int ofd_orphans_destroy(const struct lu_env *env,
1060                                struct obd_export *exp, struct ofd_device *ofd,
1061                                struct obdo *oa)
1062 {
1063         struct ofd_thread_info  *info = ofd_info(env);
1064         obd_id                   last;
1065         int                      skip_orphan;
1066         int                      rc = 0;
1067         struct ost_id            oi = oa->o_oi;
1068         struct ofd_seq          *oseq;
1069
1070         ENTRY;
1071
1072         oseq = ofd_seq_get(ofd, oa->o_seq);
1073         if (oseq == NULL) {
1074                 CERROR("%s: Can not find seq for "LPU64":"LPU64"\n",
1075                        ofd_name(ofd), oa->o_seq, oa->o_id);
1076                 RETURN(-EINVAL);
1077         }
1078
1079         LASSERT(exp != NULL);
1080         skip_orphan = !!(exp->exp_connect_flags & OBD_CONNECT_SKIP_ORPHAN);
1081
1082         last = ofd_seq_last_oid(oseq);
1083         LCONSOLE_INFO("%s: deleting orphan objects from "LPX64":"LPU64
1084                       " to "LPU64"\n", ofd_name(ofd), oa->o_seq,
1085                       oa->o_id + 1, last);
1086
1087         for (oi.oi_id = last; oi.oi_id > oa->o_id; oi.oi_id--) {
1088                 fid_ostid_unpack(&info->fti_fid, &oi, 0);
1089                 rc = ofd_destroy_by_fid(env, ofd, &info->fti_fid, 1);
1090                 if (rc && rc != -ENOENT) /* this is pretty fatal... */
1091                         CEMERG("error destroying precreated id "LPU64": %d\n",
1092                                oi.oi_id, rc);
1093                 if (!skip_orphan) {
1094                         ofd_seq_last_oid_set(oseq, oi.oi_id - 1);
1095                         /* update last_id on disk periodically so that if we
1096                          * restart * we don't need to re-scan all of the just
1097                          * deleted objects. */
1098                         if ((oi.oi_id & 511) == 0)
1099                                 ofd_seq_last_oid_write(env, ofd, oseq);
1100                 }
1101         }
1102         CDEBUG(D_HA, "%s: after destroy: set last_objids["LPU64"] = "LPU64"\n",
1103                ofd_obd(ofd)->obd_name, oa->o_seq, oa->o_id);
1104         if (!skip_orphan) {
1105                 rc = ofd_seq_last_oid_write(env, ofd, oseq);
1106         } else {
1107                 /* don't reuse orphan object, return last used objid */
1108                 oa->o_id = last;
1109                 rc = 0;
1110         }
1111         ofd_seq_put(env, oseq);
1112         RETURN(rc);
1113 }
1114
1115 int ofd_create(const struct lu_env *env, struct obd_export *exp,
1116                struct obdo *oa, struct lov_stripe_md **ea,
1117                struct obd_trans_info *oti)
1118 {
1119         struct ofd_device       *ofd = ofd_exp(exp);
1120         struct ofd_thread_info  *info;
1121         obd_seq                 seq = oa->o_seq;
1122         struct ofd_seq          *oseq;
1123         int                      rc = 0, diff;
1124
1125         ENTRY;
1126
1127         info = ofd_info_init(env, exp);
1128         ofd_oti2info(info, oti);
1129
1130         LASSERT(oa->o_seq >= FID_SEQ_OST_MDT0);
1131         LASSERT(oa->o_valid & OBD_MD_FLGROUP);
1132
1133         CDEBUG(D_INFO, "ofd_create(oa->o_seq="LPU64",oa->o_id="LPU64")\n",
1134                seq, oa->o_id);
1135
1136         oseq = ofd_seq_load(env, ofd, seq);
1137         if (oseq == NULL) {
1138                 CERROR("%s: Can't find oseq "LPX64"\n", ofd_name(ofd), seq);
1139                 RETURN(-EINVAL);
1140         }
1141
1142         if ((oa->o_valid & OBD_MD_FLFLAGS) &&
1143             (oa->o_flags & OBD_FL_RECREATE_OBJS)) {
1144                 if (!ofd_obd(ofd)->obd_recovering ||
1145                     oa->o_id > ofd_seq_last_oid(oseq)) {
1146                         CERROR("recreate objid "LPU64" > last id "LPU64"\n",
1147                                         oa->o_id, ofd_seq_last_oid(oseq));
1148                         GOTO(out_nolock, rc = -EINVAL);
1149                 }
1150                 /* do nothing because we create objects during first write */
1151                 GOTO(out_nolock, rc = 0);
1152         }
1153         /* former ofd_handle_precreate */
1154         if ((oa->o_valid & OBD_MD_FLFLAGS) &&
1155             (oa->o_flags & OBD_FL_DELORPHAN)) {
1156                 /* destroy orphans */
1157                 if (oti->oti_conn_cnt < exp->exp_conn_cnt) {
1158                         CERROR("%s: dropping old orphan cleanup request\n",
1159                                ofd_name(ofd));
1160                         GOTO(out_nolock, rc = 0);
1161                 }
1162                 /* This causes inflight precreates to abort and drop lock */
1163                 oseq->os_destroys_in_progress = 1;
1164                 mutex_lock(&oseq->os_create_lock);
1165                 if (!oseq->os_destroys_in_progress) {
1166                         CERROR("%s:["LPU64"] destroys_in_progress already cleared\n",
1167                                exp->exp_obd->obd_name, oa->o_seq);
1168                         GOTO(out, rc = 0);
1169                 }
1170                 diff = oa->o_id - ofd_seq_last_oid(oseq);
1171                 CDEBUG(D_HA, "ofd_last_id() = "LPU64" -> diff = %d\n",
1172                         ofd_seq_last_oid(oseq), diff);
1173                 if (-diff > OST_MAX_PRECREATE) {
1174                         /* FIXME: should reset precreate_next_id on MDS */
1175                         rc = 0;
1176                 } else if (diff < 0) {
1177                         rc = ofd_orphans_destroy(env, exp, ofd, oa);
1178                         oseq->os_destroys_in_progress = 0;
1179                 } else {
1180                         /* XXX: Used by MDS for the first time! */
1181                         oseq->os_destroys_in_progress = 0;
1182                 }
1183         } else {
1184                 mutex_lock(&oseq->os_create_lock);
1185                 if (oti->oti_conn_cnt < exp->exp_conn_cnt) {
1186                         CERROR("%s: dropping old precreate request\n",
1187                                 ofd_obd(ofd)->obd_name);
1188                         GOTO(out, rc = 0);
1189                 }
1190                 /* only precreate if seq == 0 and o_id is specfied */
1191                 if (!fid_seq_is_mdt(oa->o_seq) || oa->o_id == 0) {
1192                         diff = 1; /* shouldn't we create this right now? */
1193                 } else {
1194                         diff = oa->o_id - ofd_seq_last_oid(oseq);
1195                 }
1196         }
1197         if (diff > 0) {
1198                 cfs_time_t       enough_time = cfs_time_shift(DISK_TIMEOUT);
1199                 obd_id           next_id;
1200                 int              created = 0;
1201                 int              count;
1202
1203                 if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
1204                     !(oa->o_flags & OBD_FL_DELORPHAN)) {
1205                         /* don't enforce grant during orphan recovery */
1206                         rc = ofd_grant_create(env,
1207                                               ofd_obd(ofd)->obd_self_export,
1208                                               &diff);
1209                         if (rc) {
1210                                 CDEBUG(D_HA, "%s: failed to acquire grant space"
1211                                        "for precreate (%d)\n",
1212                                        ofd_name(ofd), diff);
1213                                 diff = 0;
1214                         }
1215                 }
1216
1217                 while (diff > 0) {
1218                         next_id = ofd_seq_last_oid(oseq) + 1;
1219                         count = ofd_precreate_batch(ofd, diff);
1220
1221                         CDEBUG(D_HA, "%s: reserve %d objects in group "LPU64
1222                                " at "LPU64"\n", ofd_obd(ofd)->obd_name,
1223                                count, oa->o_seq, next_id);
1224
1225                         if (cfs_time_after(jiffies, enough_time)) {
1226                                 LCONSOLE_WARN("%s: Slow creates, %d/%d objects"
1227                                               " created at a rate of %d/s\n",
1228                                               ofd_obd(ofd)->obd_name,
1229                                               created, diff + created,
1230                                               created / DISK_TIMEOUT);
1231                                 break;
1232                         }
1233
1234                         rc = ofd_precreate_objects(env, ofd, next_id,
1235                                                    oseq, count);
1236                         if (rc > 0) {
1237                                 created += rc;
1238                                 diff -= rc;
1239                         } else if (rc < 0) {
1240                                 break;
1241                         }
1242                 }
1243                 if (created > 0)
1244                         /* some objects got created, we can return
1245                          * them, even if last creation failed */
1246                         rc = 0;
1247                 else
1248                         CERROR("unable to precreate: %d\n", rc);
1249
1250                 oa->o_id = ofd_seq_last_oid(oseq);
1251                 oa->o_valid |= OBD_MD_FLID | OBD_MD_FLGROUP;
1252
1253                 if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
1254                     !(oa->o_flags & OBD_FL_DELORPHAN))
1255                         ofd_grant_commit(env, ofd_obd(ofd)->obd_self_export,
1256                                          rc);
1257         }
1258
1259         ofd_info2oti(info, oti);
1260 out:
1261         mutex_unlock(&oseq->os_create_lock);
1262 out_nolock:
1263         if (rc == 0 && ea != NULL) {
1264                 struct lov_stripe_md *lsm = *ea;
1265
1266                 lsm->lsm_object_id = oa->o_id;
1267         }
1268         ofd_seq_put(env, oseq);
1269         RETURN(rc);
1270 }
1271
1272 int ofd_getattr(const struct lu_env *env, struct obd_export *exp,
1273                 struct obd_info *oinfo)
1274 {
1275         struct ofd_device       *ofd = ofd_exp(exp);
1276         struct ofd_thread_info  *info;
1277         struct ofd_object       *fo;
1278         __u64                    curr_version;
1279         int                      rc = 0;
1280
1281         ENTRY;
1282
1283         info = ofd_info_init(env, exp);
1284
1285         fid_ostid_unpack(&info->fti_fid, &oinfo->oi_oa->o_oi, 0);
1286         rc = ofd_auth_capa(exp, &info->fti_fid, oinfo->oi_oa->o_seq,
1287                            oinfo_capa(oinfo), CAPA_OPC_META_READ);
1288         if (rc)
1289                 GOTO(out, rc);
1290
1291         fo = ofd_object_find(env, ofd, &info->fti_fid);
1292         if (IS_ERR(fo))
1293                 GOTO(out, rc = PTR_ERR(fo));
1294         LASSERT(fo != NULL);
1295         rc = ofd_attr_get(env, fo, &info->fti_attr);
1296         oinfo->oi_oa->o_valid = OBD_MD_FLID;
1297         if (rc == 0)
1298                 obdo_from_la(oinfo->oi_oa, &info->fti_attr,
1299                              OFD_VALID_FLAGS | LA_UID | LA_GID);
1300
1301         /* Store object version in reply */
1302         curr_version = dt_version_get(env, ofd_object_child(fo));
1303         if ((__s64)curr_version != -EOPNOTSUPP) {
1304                 oinfo->oi_oa->o_valid |= OBD_MD_FLDATAVERSION;
1305                 oinfo->oi_oa->o_data_version = curr_version;
1306         }
1307         ofd_object_put(env, fo);
1308 out:
1309         RETURN(rc);
1310 }
1311
1312 static int ofd_sync(const struct lu_env *env, struct obd_export *exp,
1313                     struct obd_info *oinfo, obd_size start, obd_size end,
1314                     struct ptlrpc_request_set *set)
1315 {
1316         struct ofd_device       *ofd = ofd_exp(exp);
1317         struct ofd_thread_info  *info;
1318         struct ofd_object       *fo;
1319         int                      rc = 0;
1320
1321         ENTRY;
1322
1323         /* if no objid is specified, it means "sync whole filesystem" */
1324         if (oinfo->oi_oa == NULL || !(oinfo->oi_oa->o_valid & OBD_MD_FLID)) {
1325                 rc = dt_sync(env, ofd->ofd_osd);
1326                 GOTO(out, rc);
1327         }
1328
1329         info = ofd_info_init(env, exp);
1330         fid_ostid_unpack(&info->fti_fid, &oinfo->oi_oa->o_oi, 0);
1331
1332         rc = ofd_auth_capa(exp, &info->fti_fid, oinfo->oi_oa->o_seq,
1333                            oinfo_capa(oinfo), CAPA_OPC_OSS_TRUNC);
1334         if (rc)
1335                 GOTO(out, rc);
1336
1337         fo = ofd_object_find(env, ofd, &info->fti_fid);
1338         if (IS_ERR(fo)) {
1339                 CERROR("%s: error finding object "DFID": rc = %ld\n",
1340                        exp->exp_obd->obd_name, PFID(&info->fti_fid),
1341                        PTR_ERR(fo));
1342                 GOTO(out, rc = PTR_ERR(fo));
1343         }
1344
1345         if (!ofd_object_exists(fo))
1346                 GOTO(put, rc = -ENOENT);
1347
1348         if (dt_version_get(env, ofd_object_child(fo)) >
1349             ofd_obd(ofd)->obd_last_committed) {
1350                 rc = dt_object_sync(env, ofd_object_child(fo));
1351                 if (rc)
1352                         GOTO(put, rc);
1353         }
1354
1355         oinfo->oi_oa->o_valid = OBD_MD_FLID;
1356         rc = ofd_attr_get(env, fo, &info->fti_attr);
1357         obdo_from_la(oinfo->oi_oa, &info->fti_attr, OFD_VALID_FLAGS);
1358
1359         ofd_counter_incr(exp, LPROC_OFD_STATS_SYNC, oinfo->oi_jobid, 1);
1360         EXIT;
1361 put:
1362         ofd_object_put(env, fo);
1363 out:
1364         return rc;
1365 }
1366
1367 int ofd_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1368                   void *karg, void *uarg)
1369 {
1370         struct lu_env            env;
1371         struct ofd_device       *ofd = ofd_exp(exp);
1372         struct obd_device       *obd = ofd_obd(ofd);
1373         int                      rc;
1374
1375         ENTRY;
1376
1377         CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
1378         rc = lu_env_init(&env, LCT_LOCAL);
1379         if (rc)
1380                 RETURN(rc);
1381
1382         switch (cmd) {
1383         case OBD_IOC_ABORT_RECOVERY:
1384                 CERROR("%s: aborting recovery\n", obd->obd_name);
1385                 target_stop_recovery_thread(obd);
1386                 break;
1387         case OBD_IOC_SYNC:
1388                 CDEBUG(D_RPCTRACE, "syncing ost %s\n", obd->obd_name);
1389                 rc = dt_sync(&env, ofd->ofd_osd);
1390                 break;
1391         case OBD_IOC_SET_READONLY:
1392                 rc = dt_sync(&env, ofd->ofd_osd);
1393                 if (rc == 0)
1394                         rc = dt_ro(&env, ofd->ofd_osd);
1395                 break;
1396         default:
1397                 CERROR("%s: not supported cmd = %d\n", obd->obd_name, cmd);
1398                 rc = -ENOTTY;
1399         }
1400
1401         lu_env_fini(&env);
1402         RETURN(rc);
1403 }
1404
1405 static int ofd_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
1406 {
1407         int rc = 0;
1408
1409         ENTRY;
1410
1411         switch(stage) {
1412         case OBD_CLEANUP_EARLY:
1413                 break;
1414         case OBD_CLEANUP_EXPORTS:
1415                 target_cleanup_recovery(obd);
1416                 break;
1417         }
1418         RETURN(rc);
1419 }
1420
1421 static int ofd_ping(const struct lu_env *env, struct obd_export *exp)
1422 {
1423         ofd_fmd_expire(exp);
1424         return 0;
1425 }
1426
1427 static int ofd_health_check(const struct lu_env *nul, struct obd_device *obd)
1428 {
1429         struct ofd_device       *ofd = ofd_dev(obd->obd_lu_dev);
1430         struct ofd_thread_info  *info;
1431         struct lu_env            env;
1432 #ifdef USE_HEALTH_CHECK_WRITE
1433         struct thandle          *th;
1434 #endif
1435         int                      rc = 0;
1436
1437         /* obd_proc_read_health pass NULL env, we need real one */
1438         rc = lu_env_init(&env, LCT_DT_THREAD);
1439         if (rc)
1440                 RETURN(rc);
1441
1442         info = ofd_info_init(&env, NULL);
1443         rc = dt_statfs(&env, ofd->ofd_osd, &info->fti_u.osfs);
1444         if (unlikely(rc))
1445                 GOTO(out, rc);
1446
1447         if (info->fti_u.osfs.os_state == OS_STATE_READONLY)
1448                 GOTO(out, rc = -EROFS);
1449
1450 #ifdef USE_HEALTH_CHECK_WRITE
1451         OBD_ALLOC(info->fti_buf.lb_buf, CFS_PAGE_SIZE);
1452         if (info->fti_buf.lb_buf == NULL)
1453                 GOTO(out, rc = -ENOMEM);
1454
1455         info->fti_buf.lb_len = CFS_PAGE_SIZE;
1456         info->fti_off = 0;
1457
1458         th = dt_trans_create(&env, ofd->ofd_osd);
1459         if (IS_ERR(th))
1460                 GOTO(out, rc = PTR_ERR(th));
1461
1462         rc = dt_declare_record_write(&env, ofd->ofd_health_check_file,
1463                                      info->fti_buf.lb_len, info->fti_off, th);
1464         if (rc == 0) {
1465                 th->th_sync = 1; /* sync IO is needed */
1466                 rc = dt_trans_start_local(&env, ofd->ofd_osd, th);
1467                 if (rc == 0)
1468                         rc = dt_record_write(&env, ofd->ofd_health_check_file,
1469                                              &info->fti_buf, &info->fti_off,
1470                                              th);
1471         }
1472         dt_trans_stop(&env, ofd->ofd_osd, th);
1473
1474         OBD_FREE(info->fti_buf.lb_buf, CFS_PAGE_SIZE);
1475
1476         CDEBUG(D_INFO, "write 1 page synchronously for checking io rc %d\n",rc);
1477 #endif
1478 out:
1479         lu_env_fini(&env);
1480         return !!rc;
1481 }
1482
1483 /*
1484  * Handle quota control requests to consult current usage/limit.
1485  *
1486  * \param obd - is the obd device associated with the ofd
1487  * \param exp - is the client's export
1488  * \param oqctl - is the obd_quotactl request to be processed
1489  */
1490 static int ofd_quotactl(struct obd_device *obd, struct obd_export *exp,
1491                         struct obd_quotactl *oqctl)
1492 {
1493         struct ofd_device  *ofd = ofd_dev(obd->obd_lu_dev);
1494         struct lu_env       env;
1495         int                 rc;
1496         ENTRY;
1497
1498         /* report success for quota on/off for interoperability with current MDT
1499          * stack */
1500         if (oqctl->qc_cmd == Q_QUOTAON || oqctl->qc_cmd == Q_QUOTAOFF)
1501                 RETURN(0);
1502
1503         rc = lu_env_init(&env, LCT_DT_THREAD);
1504         if (rc)
1505                 RETURN(rc);
1506
1507         rc = lquotactl_slv(&env, ofd->ofd_osd, oqctl);
1508         lu_env_fini(&env);
1509
1510         RETURN(rc);
1511 }
1512
1513 struct obd_ops ofd_obd_ops = {
1514         .o_owner                = THIS_MODULE,
1515         .o_connect              = ofd_obd_connect,
1516         .o_reconnect            = ofd_obd_reconnect,
1517         .o_disconnect           = ofd_obd_disconnect,
1518         .o_set_info_async       = ofd_set_info_async,
1519         .o_get_info             = ofd_get_info,
1520         .o_create               = ofd_create,
1521         .o_statfs               = ofd_statfs,
1522         .o_setattr              = ofd_setattr,
1523         .o_preprw               = ofd_preprw,
1524         .o_commitrw             = ofd_commitrw,
1525         .o_destroy              = ofd_destroy,
1526         .o_init_export          = ofd_init_export,
1527         .o_destroy_export       = ofd_destroy_export,
1528         .o_postrecov            = ofd_obd_postrecov,
1529         .o_punch                = ofd_punch,
1530         .o_getattr              = ofd_getattr,
1531         .o_sync                 = ofd_sync,
1532         .o_iocontrol            = ofd_iocontrol,
1533         .o_precleanup           = ofd_precleanup,
1534         .o_ping                 = ofd_ping,
1535         .o_health_check         = ofd_health_check,
1536         .o_quotactl             = ofd_quotactl,
1537 };