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