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