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) 2011, 2012, Whamcloud, Inc.
  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 <lquota.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                 cfs_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                         lut_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 = lut_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                 lut_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         cfs_spin_lock_init(&exp->exp_filter_data.fed_lock);
 370         CFS_INIT_LIST_HEAD(&exp->exp_filter_data.fed_mod_list);
 371         cfs_spin_lock(&exp->exp_lock);
 372         exp->exp_connecting = 1;
 373         cfs_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 = lut_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         lut_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_obd_postrecov(struct obd_device *obd)
 429 {
 430         struct lu_env            env;
 431         struct lu_device        *ldev = obd->obd_lu_dev;
 432         int                      rc;
 433
 434         ENTRY;
 435
 436         rc = lu_env_init(&env, LCT_DT_THREAD);
 437         if (rc)
 438                 RETURN(rc);
 439         ofd_info_init(&env, obd->obd_self_export);
 440
 441         rc = ldev->ld_ops->ldo_recovery_complete(&env, ldev);
 442         lu_env_fini(&env);
 443         RETURN(rc);
 444 }
 445
 446 static int ofd_adapt_sptlrpc_conf(const struct lu_env *env,
 447                                   struct obd_device *obd, int initial)
 448 {
 449         struct filter_obd       *fo = &obd->u.filter;
 450         struct sptlrpc_rule_set  tmp_rset;
 451         int                      rc;
 452
 453         sptlrpc_rule_set_init(&tmp_rset);
 454         rc = sptlrpc_conf_target_get_rules(obd, &tmp_rset, initial);
 455         if (rc) {
 456                 CERROR("%s: failed get sptlrpc rules: rc = %d\n",
 457                        obd->obd_name, rc);
 458                 return rc;
 459         }
 460
 461         sptlrpc_target_update_exp_flavor(obd, &tmp_rset);
 462
 463         cfs_write_lock(&fo->fo_sptlrpc_lock);
 464         sptlrpc_rule_set_free(&fo->fo_sptlrpc_rset);
 465         fo->fo_sptlrpc_rset = tmp_rset;
 466         cfs_write_unlock(&fo->fo_sptlrpc_lock);
 467
 468         return 0;
 469 }
 470
 471 static int ofd_set_mds_conn(struct obd_export *exp, void *val)
 472 {
 473         int rc = 0;
 474
 475         ENTRY;
 476
 477         LCONSOLE_WARN("%s: received MDS connection from %s\n",
 478                       exp->exp_obd->obd_name, obd_export_nid2str(exp));
 479         RETURN(rc);
 480 }
 481
 482 static int ofd_set_info_async(const struct lu_env *env, struct obd_export *exp,
 483                               __u32 keylen, void *key, __u32 vallen, void *val,
 484                               struct ptlrpc_request_set *set)
 485 {
 486         struct ofd_device       *ofd = ofd_exp(exp);
 487         int                      rc = 0;
 488
 489         ENTRY;
 490
 491         if (exp->exp_obd == NULL) {
 492                 CDEBUG(D_IOCTL, "invalid export %p\n", exp);
 493                 RETURN(-EINVAL);
 494         }
 495
 496         if (KEY_IS(KEY_CAPA_KEY)) {
 497                 rc = ofd_update_capa_key(ofd, val);
 498                 if (rc)
 499                         CERROR("%s: update capability key failed: rc = %d\n",
 500                                exp->exp_obd->obd_name, rc);
 501         } else if (KEY_IS(KEY_SPTLRPC_CONF)) {
 502                 ofd_adapt_sptlrpc_conf(env, exp->exp_obd, 0);
 503         } else if (KEY_IS(KEY_MDS_CONN)) {
 504                 rc = ofd_set_mds_conn(exp, val);
 505         } else if (KEY_IS(KEY_GRANT_SHRINK)) {
 506                 struct ost_body *body = val;
 507
 508                 /** handle grant shrink, similar to a read request */
 509                 ofd_grant_prepare_read(env, exp, &body->oa);
 510         } else {
 511                 CERROR("%s: Unsupported key %s\n",
 512                        exp->exp_obd->obd_name, (char*)key);
 513                 rc = -EOPNOTSUPP;
 514         }
 515
 516         RETURN(rc);
 517 }
 518
 519 static int ofd_get_info(const struct lu_env *env, struct obd_export *exp,
 520                         __u32 keylen, void *key, __u32 *vallen, void *val,
 521                         struct lov_stripe_md *lsm)
 522 {
 523         struct ofd_device       *ofd = ofd_exp(exp);
 524         int                      rc = 0;
 525
 526         ENTRY;
 527
 528         if (exp->exp_obd == NULL) {
 529                 CDEBUG(D_IOCTL, "invalid client export %p\n", exp);
 530                 RETURN(-EINVAL);
 531         }
 532
 533         if (KEY_IS(KEY_BLOCKSIZE)) {
 534                 __u32 *blocksize = val;
 535                 if (blocksize) {
 536                         if (*vallen < sizeof(*blocksize))
 537                                 RETURN(-EOVERFLOW);
 538                         *blocksize = 1 << ofd->ofd_dt_conf.ddp_block_shift;
 539                 }
 540                 *vallen = sizeof(*blocksize);
 541         } else if (KEY_IS(KEY_BLOCKSIZE_BITS)) {
 542                 __u32 *blocksize_bits = val;
 543                 if (blocksize_bits) {
 544                         if (*vallen < sizeof(*blocksize_bits))
 545                                 RETURN(-EOVERFLOW);
 546                         *blocksize_bits = ofd->ofd_dt_conf.ddp_block_shift;
 547                 }
 548                 *vallen = sizeof(*blocksize_bits);
 549         } else if (KEY_IS(KEY_LAST_ID)) {
 550                 obd_id *last_id = val;
 551                 if (last_id) {
 552                         if (*vallen < sizeof(*last_id))
 553                                 RETURN(-EOVERFLOW);
 554                         *last_id = ofd_last_id(ofd,
 555                                                exp->exp_filter_data.fed_group);
 556                 }
 557                 *vallen = sizeof(*last_id);
 558         } else if (KEY_IS(KEY_FIEMAP)) {
 559                 struct ofd_thread_info          *info;
 560                 struct ofd_device               *ofd = ofd_exp(exp);
 561                 struct ofd_object               *fo;
 562                 struct ll_fiemap_info_key       *fm_key = key;
 563
 564                 if (val == NULL) {
 565                         *vallen = fiemap_count_to_size(
 566                                                fm_key->fiemap.fm_extent_count);
 567                         RETURN(0);
 568                 }
 569
 570                 info = ofd_info_init(env, exp);
 571
 572                 fid_ostid_unpack(&info->fti_fid, &fm_key->oa.o_oi, 0);
 573
 574                 CDEBUG(D_INODE, "get FIEMAP of object "DFID"\n",
 575                        PFID(&info->fti_fid));
 576
 577                 fo = ofd_object_find(env, ofd, &info->fti_fid);
 578                 if (IS_ERR(fo)) {
 579                         CERROR("%s: error finding object "DFID"\n",
 580                                exp->exp_obd->obd_name, PFID(&info->fti_fid));
 581                         rc = PTR_ERR(fo);
 582                 } else {
 583                         struct ll_user_fiemap *fiemap = val;
 584
 585                         ofd_read_lock(env, fo);
 586                         if (ofd_object_exists(fo)) {
 587                                 *fiemap = fm_key->fiemap;
 588                                 rc = dt_fiemap_get(env,
 589                                                    ofd_object_child(fo),
 590                                                    fiemap);
 591                         } else {
 592                                 rc = -ENOENT;
 593                         }
 594                         ofd_read_unlock(env, fo);
 595                         ofd_object_put(env, fo);
 596                 }
 597         } else if (KEY_IS(KEY_SYNC_LOCK_CANCEL)) {
 598                 *((__u32 *) val) = ofd->ofd_sync_lock_cancel;
 599                 *vallen = sizeof(__u32);
 600         } else {
 601                 CERROR("Not supported key %s\n", (char*)key);
 602                 rc = -EOPNOTSUPP;
 603         }
 604
 605         RETURN(rc);
 606 }
 607
 608 /** helper function for statfs, also used by grant code */
 609 int ofd_statfs_internal(const struct lu_env *env, struct ofd_device *ofd,
 610                         struct obd_statfs *osfs, __u64 max_age, int *from_cache)
 611 {
 612         int rc;
 613
 614         cfs_spin_lock(&ofd->ofd_osfs_lock);
 615         if (cfs_time_before_64(ofd->ofd_osfs_age, max_age) || max_age == 0) {
 616                 obd_size unstable;
 617
 618                 /* statfs data are too old, get up-to-date one.
 619                  * we must be cautious here since multiple threads might be
 620                  * willing to update statfs data concurrently and we must
 621                  * grant that cached statfs data are always consistent */
 622
 623                 if (ofd->ofd_statfs_inflight == 0)
 624                         /* clear inflight counter if no users, although it would
 625                          * take a while to overflow this 64-bit counter ... */
 626                         ofd->ofd_osfs_inflight = 0;
 627                 /* notify ofd_grant_commit() that we want to track writes
 628                  * completed as of now */
 629                 ofd->ofd_statfs_inflight++;
 630                 /* record value of inflight counter before running statfs to
 631                  * compute the diff once statfs is completed */
 632                 unstable = ofd->ofd_osfs_inflight;
 633                 cfs_spin_unlock(&ofd->ofd_osfs_lock);
 634
 635                 /* statfs can sleep ... hopefully not for too long since we can
 636                  * call it fairly often as space fills up */
 637                 rc = dt_statfs(env, ofd->ofd_osd, osfs);
 638                 if (unlikely(rc))
 639                         return rc;
 640
 641                 cfs_spin_lock(&ofd->ofd_grant_lock);
 642                 cfs_spin_lock(&ofd->ofd_osfs_lock);
 643                 /* calculate how much space was written while we released the
 644                  * ofd_osfs_lock */
 645                 unstable = ofd->ofd_osfs_inflight - unstable;
 646                 ofd->ofd_osfs_unstable = 0;
 647                 if (unstable) {
 648                         /* some writes completed while we were running statfs
 649                          * w/o the ofd_osfs_lock. Those ones got added to
 650                          * the cached statfs data that we are about to crunch.
 651                          * Take them into account in the new statfs data */
 652                         osfs->os_bavail -= min_t(obd_size, osfs->os_bavail,
 653                                                unstable >> ofd->ofd_blockbits);
 654                         /* However, we don't really know if those writes got
 655                          * accounted in the statfs call, so tell
 656                          * ofd_grant_space_left() there is some uncertainty
 657                          * on the accounting of those writes.
 658                          * The purpose is to prevent spurious error messages in
 659                          * ofd_grant_space_left() since those writes might be
 660                          * accounted twice. */
 661                         ofd->ofd_osfs_unstable += unstable;
 662                 }
 663                 /* similarly, there is some uncertainty on write requests
 664                  * between prepare & commit */
 665                 ofd->ofd_osfs_unstable += ofd->ofd_tot_pending;
 666                 cfs_spin_unlock(&ofd->ofd_grant_lock);
 667
 668                 /* finally udpate cached statfs data */
 669                 ofd->ofd_osfs = *osfs;
 670                 ofd->ofd_osfs_age = cfs_time_current_64();
 671
 672                 ofd->ofd_statfs_inflight--; /* stop tracking */
 673                 if (ofd->ofd_statfs_inflight == 0)
 674                         ofd->ofd_osfs_inflight = 0;
 675                 cfs_spin_unlock(&ofd->ofd_osfs_lock);
 676
 677                 if (from_cache)
 678                         *from_cache = 0;
 679         } else {
 680                 /* use cached statfs data */
 681                 *osfs = ofd->ofd_osfs;
 682                 cfs_spin_unlock(&ofd->ofd_osfs_lock);
 683                 if (from_cache)
 684                         *from_cache = 1;
 685         }
 686         return 0;
 687 }
 688
 689 static int ofd_statfs(const struct lu_env *env,  struct obd_export *exp,
 690                       struct obd_statfs *osfs, __u64 max_age, __u32 flags)
 691 {
 692         struct obd_device       *obd = class_exp2obd(exp);
 693         struct ofd_device       *ofd = ofd_dev(exp->exp_obd->obd_lu_dev);
 694         int                      rc;
 695
 696         ENTRY;
 697
 698         rc = ofd_statfs_internal(env, ofd, osfs, max_age, NULL);
 699         if (unlikely(rc))
 700                 GOTO(out, rc);
 701
 702         /* at least try to account for cached pages.  its still racy and
 703          * might be under-reporting if clients haven't announced their
 704          * caches with brw recently */
 705
 706         CDEBUG(D_SUPER | D_CACHE, "blocks cached "LPU64" granted "LPU64
 707                " pending "LPU64" free "LPU64" avail "LPU64"\n",
 708                ofd->ofd_tot_dirty, ofd->ofd_tot_granted, ofd->ofd_tot_pending,
 709                osfs->os_bfree << ofd->ofd_blockbits,
 710                osfs->os_bavail << ofd->ofd_blockbits);
 711
 712         osfs->os_bavail -= min_t(obd_size, osfs->os_bavail,
 713                                  ((ofd->ofd_tot_dirty + ofd->ofd_tot_pending +
 714                                    osfs->os_bsize - 1) >> ofd->ofd_blockbits));
 715
 716         /* The QoS code on the MDS does not care about space reserved for
 717          * precreate, so take it out. */
 718         if (exp->exp_connect_flags & OBD_CONNECT_MDS) {
 719                 struct filter_export_data *fed;
 720
 721                 fed = &obd->obd_self_export->exp_filter_data;
 722                 osfs->os_bavail -= min_t(obd_size, osfs->os_bavail,
 723                                          fed->fed_grant >> ofd->ofd_blockbits);
 724         }
 725
 726         ofd_grant_sanity_check(obd, __FUNCTION__);
 727         CDEBUG(D_CACHE, LPU64" blocks: "LPU64" free, "LPU64" avail; "
 728                LPU64" objects: "LPU64" free; state %x\n",
 729                osfs->os_blocks, osfs->os_bfree, osfs->os_bavail,
 730                osfs->os_files, osfs->os_ffree, osfs->os_state);
 731
 732         if (OBD_FAIL_CHECK_VALUE(OBD_FAIL_OST_ENOSPC,
 733                                  ofd->ofd_lut.lut_lsd.lsd_ost_index))
 734                 osfs->os_bfree = osfs->os_bavail = 2;
 735
 736         if (OBD_FAIL_CHECK_VALUE(OBD_FAIL_OST_ENOINO,
 737                                  ofd->ofd_lut.lut_lsd.lsd_ost_index))
 738                 osfs->os_ffree = 0;
 739
 740         /* OS_STATE_READONLY can be set by OSD already */
 741         if (ofd->ofd_raid_degraded)
 742                 osfs->os_state |= OS_STATE_DEGRADED;
 743
 744         if (obd->obd_self_export != exp && ofd_grant_compat(exp, ofd)) {
 745                 /* clients which don't support OBD_CONNECT_GRANT_PARAM
 746                  * should not see a block size > page size, otherwise
 747                  * cl_lost_grant goes mad. Therefore, we emulate a 4KB (=2^12)
 748                  * block size which is the biggest block size known to work
 749                  * with all client's page size. */
 750                 osfs->os_blocks <<= ofd->ofd_blockbits - COMPAT_BSIZE_SHIFT;
 751                 osfs->os_bfree  <<= ofd->ofd_blockbits - COMPAT_BSIZE_SHIFT;
 752                 osfs->os_bavail <<= ofd->ofd_blockbits - COMPAT_BSIZE_SHIFT;
 753                 osfs->os_bsize    = 1 << COMPAT_BSIZE_SHIFT;
 754         }
 755
 756         EXIT;
 757 out:
 758         return rc;
 759 }
 760
 761 int ofd_setattr(const struct lu_env *env, struct obd_export *exp,
 762                 struct obd_info *oinfo, struct obd_trans_info *oti)
 763 {
 764         struct ofd_thread_info  *info;
 765         struct ofd_device       *ofd = ofd_exp(exp);
 766         struct ldlm_namespace   *ns = ofd->ofd_namespace;
 767         struct ldlm_resource    *res;
 768         struct ofd_object       *fo;
 769         struct obdo             *oa = oinfo->oi_oa;
 770         struct filter_fid       *ff = NULL;
 771         int                      rc = 0;
 772
 773         ENTRY;
 774
 775         info = ofd_info_init(env, exp);
 776         ofd_oti2info(info, oti);
 777
 778         fid_ostid_unpack(&info->fti_fid, &oinfo->oi_oa->o_oi, 0);
 779         ofd_build_resid(&info->fti_fid, &info->fti_resid);
 780
 781         rc = ofd_auth_capa(exp, &info->fti_fid, oa->o_seq,
 782                            oinfo_capa(oinfo), CAPA_OPC_META_WRITE);
 783         if (rc)
 784                 GOTO(out, rc);
 785
 786         /* This would be very bad - accidentally truncating a file when
 787          * changing the time or similar - bug 12203. */
 788         if (oinfo->oi_oa->o_valid & OBD_MD_FLSIZE &&
 789             oinfo->oi_policy.l_extent.end != OBD_OBJECT_EOF) {
 790                 static char mdsinum[48];
 791
 792                 if (oinfo->oi_oa->o_valid & OBD_MD_FLFID)
 793                         snprintf(mdsinum, sizeof(mdsinum) - 1,
 794                                  "of parent "DFID, oinfo->oi_oa->o_parent_seq,
 795                                  oinfo->oi_oa->o_parent_oid, 0);
 796                 else
 797                         mdsinum[0] = '\0';
 798
 799                 CERROR("%s: setattr from %s trying to truncate object "DFID
 800                        " %s\n", exp->exp_obd->obd_name,
 801                        obd_export_nid2str(exp), PFID(&info->fti_fid), mdsinum);
 802                 GOTO(out, rc = -EPERM);
 803         }
 804
 805         fo = ofd_object_find(env, ofd, &info->fti_fid);
 806         if (IS_ERR(fo)) {
 807                 CERROR("%s: can't find object "DFID"\n",
 808                        exp->exp_obd->obd_name, PFID(&info->fti_fid));
 809                 GOTO(out, rc = PTR_ERR(fo));
 810         }
 811
 812         la_from_obdo(&info->fti_attr, oinfo->oi_oa, oinfo->oi_oa->o_valid);
 813         info->fti_attr.la_valid &= ~LA_TYPE;
 814
 815         if (oa->o_valid & OBD_MD_FLFID) {
 816                 ff = &info->fti_mds_fid;
 817                 ofd_prepare_fidea(ff, oa);
 818         }
 819
 820         /* setting objects attributes (including owner/group) */
 821         rc = ofd_attr_set(env, fo, &info->fti_attr, ff);
 822         if (rc)
 823                 GOTO(out_unlock, rc);
 824
 825         res = ldlm_resource_get(ns, NULL, &info->fti_resid, LDLM_EXTENT, 0);
 826         if (res != NULL) {
 827                 ldlm_res_lvbo_update(res, NULL, 0);
 828                 ldlm_resource_putref(res);
 829         }
 830
 831         oinfo->oi_oa->o_valid = OBD_MD_FLID;
 832
 833         /* Quota release needs uid/gid info */
 834         rc = ofd_attr_get(env, fo, &info->fti_attr);
 835         obdo_from_la(oinfo->oi_oa, &info->fti_attr,
 836                      OFD_VALID_FLAGS | LA_UID | LA_GID);
 837         ofd_info2oti(info, oti);
 838
 839         ofd_counter_incr(exp, LPROC_OFD_STATS_SETATTR, oti->oti_jobid, 1);
 840         EXIT;
 841 out_unlock:
 842         ofd_object_put(env, fo);
 843 out:
 844         return rc;
 845 }
 846
 847 static int ofd_punch(const struct lu_env *env, struct obd_export *exp,
 848                      struct obd_info *oinfo, struct obd_trans_info *oti,
 849                      struct ptlrpc_request_set *rqset)
 850 {
 851         struct ofd_thread_info  *info;
 852         struct ofd_device       *ofd = ofd_exp(exp);
 853         struct ldlm_namespace   *ns = ofd->ofd_namespace;
 854         struct ldlm_resource    *res;
 855         struct ofd_object       *fo;
 856         struct filter_fid       *ff = NULL;
 857         int                      rc = 0;
 858
 859         ENTRY;
 860
 861         info = ofd_info_init(env, exp);
 862         ofd_oti2info(info, oti);
 863
 864         fid_ostid_unpack(&info->fti_fid, &oinfo->oi_oa->o_oi, 0);
 865         ofd_build_resid(&info->fti_fid, &info->fti_resid);
 866
 867         CDEBUG(D_INODE, "calling punch for object "DFID", valid = "LPX64
 868                ", start = "LPD64", end = "LPD64"\n", PFID(&info->fti_fid),
 869                oinfo->oi_oa->o_valid, oinfo->oi_policy.l_extent.start,
 870                oinfo->oi_policy.l_extent.end);
 871
 872         rc = ofd_auth_capa(exp, &info->fti_fid, oinfo->oi_oa->o_seq,
 873                            oinfo_capa(oinfo), CAPA_OPC_OSS_TRUNC);
 874         if (rc)
 875                 GOTO(out_env, rc);
 876
 877         fo = ofd_object_find(env, ofd, &info->fti_fid);
 878         if (IS_ERR(fo)) {
 879                 CERROR("%s: error finding object "DFID": rc = %ld\n",
 880                        exp->exp_obd->obd_name, PFID(&info->fti_fid),
 881                        PTR_ERR(fo));
 882                 GOTO(out_env, rc = PTR_ERR(fo));
 883         }
 884
 885         LASSERT(oinfo->oi_policy.l_extent.end == OBD_OBJECT_EOF);
 886         if (oinfo->oi_policy.l_extent.end == OBD_OBJECT_EOF) {
 887                 /* Truncate case */
 888                 oinfo->oi_oa->o_size = oinfo->oi_policy.l_extent.start;
 889         } else if (oinfo->oi_policy.l_extent.end >= oinfo->oi_oa->o_size) {
 890                 oinfo->oi_oa->o_size = oinfo->oi_policy.l_extent.end;
 891         }
 892
 893         la_from_obdo(&info->fti_attr, oinfo->oi_oa,
 894                      OBD_MD_FLMTIME | OBD_MD_FLATIME | OBD_MD_FLCTIME);
 895         info->fti_attr.la_valid &= ~LA_TYPE;
 896         info->fti_attr.la_size = oinfo->oi_policy.l_extent.start;
 897         info->fti_attr.la_valid |= LA_SIZE;
 898
 899         if (oinfo->oi_oa->o_valid & OBD_MD_FLFID) {
 900                 ff = &info->fti_mds_fid;
 901                 ofd_prepare_fidea(ff, oinfo->oi_oa);
 902         }
 903
 904         rc = ofd_object_punch(env, fo, oinfo->oi_policy.l_extent.start,
 905                               oinfo->oi_policy.l_extent.end, &info->fti_attr,
 906                               ff);
 907         if (rc)
 908                 GOTO(out, rc);
 909
 910         res = ldlm_resource_get(ns, NULL, &info->fti_resid, LDLM_EXTENT, 0);
 911         if (res != NULL) {
 912                 ldlm_res_lvbo_update(res, NULL, 0);
 913                 ldlm_resource_putref(res);
 914         }
 915
 916         oinfo->oi_oa->o_valid = OBD_MD_FLID;
 917         /* Quota release needs uid/gid info */
 918         rc = ofd_attr_get(env, fo, &info->fti_attr);
 919         obdo_from_la(oinfo->oi_oa, &info->fti_attr,
 920                      OFD_VALID_FLAGS | LA_UID | LA_GID);
 921         ofd_info2oti(info, oti);
 922
 923         ofd_counter_incr(exp, LPROC_OFD_STATS_PUNCH, oti->oti_jobid, 1);
 924         EXIT;
 925 out:
 926         ofd_object_put(env, fo);
 927 out_env:
 928         return rc;
 929 }
 930
 931 static int ofd_destroy_by_fid(const struct lu_env *env,
 932                               struct ofd_device *ofd,
 933                               const struct lu_fid *fid, int orphan)
 934 {
 935         struct ofd_thread_info  *info = ofd_info(env);
 936         struct lustre_handle     lockh;
 937         int                      flags = LDLM_AST_DISCARD_DATA, rc = 0;
 938         ldlm_policy_data_t       policy = {
 939                                         .l_extent = { 0, OBD_OBJECT_EOF }
 940                                  };
 941         struct ofd_object       *fo;
 942
 943         ENTRY;
 944
 945         /* Tell the clients that the object is gone now and that they should
 946          * throw away any cached pages. */
 947         ofd_build_resid(fid, &info->fti_resid);
 948         rc = ldlm_cli_enqueue_local(ofd->ofd_namespace, &info->fti_resid,
 949                                     LDLM_EXTENT, &policy, LCK_PW, &flags,
 950                                     ldlm_blocking_ast, ldlm_completion_ast,
 951                                     NULL, NULL, 0, NULL, &lockh);
 952
 953         /* We only care about the side-effects, just drop the lock. */
 954         if (rc == ELDLM_OK)
 955                 ldlm_lock_decref(&lockh, LCK_PW);
 956
 957         fo = ofd_object_find(env, ofd, fid);
 958         if (IS_ERR(fo))
 959                 RETURN(PTR_ERR(fo));
 960         LASSERT(fo != NULL);
 961
 962         rc = ofd_object_destroy(env, fo, orphan);
 963
 964         ofd_object_put(env, fo);
 965         RETURN(rc);
 966 }
 967
 968 int ofd_destroy(const struct lu_env *env, struct obd_export *exp,
 969                 struct obdo *oa, struct lov_stripe_md *md,
 970                 struct obd_trans_info *oti, struct obd_export *md_exp,
 971                 void *capa)
 972 {
 973         struct ofd_device       *ofd = ofd_exp(exp);
 974         struct ofd_thread_info  *info;
 975         obd_count                count;
 976         int                      rc = 0;
 977
 978         ENTRY;
 979
 980         info = ofd_info_init(env, exp);
 981         ofd_oti2info(info, oti);
 982
 983         if (!(oa->o_valid & OBD_MD_FLGROUP))
 984                 oa->o_seq = 0;
 985
 986         /* check that o_misc makes sense */
 987         if (oa->o_valid & OBD_MD_FLOBJCOUNT)
 988                 count = oa->o_misc;
 989         else
 990                 count = 1; /* default case - single destroy */
 991
 992         /**
 993          * There can be sequence of objects to destroy. Therefore this request
 994          * may have multiple transaction involved in. It is OK, we need only
 995          * the highest used transno to be reported back in reply but not for
 996          * replays, they must report their transno
 997          */
 998         if (info->fti_transno == 0) /* not replay */
 999                 info->fti_mult_trans = 1;
1000         while (count > 0) {
1001                 int lrc;
1002
1003                 fid_ostid_unpack(&info->fti_fid, &oa->o_oi, 0);
1004                 lrc = ofd_destroy_by_fid(env, ofd, &info->fti_fid, 0);
1005                 if (lrc == -ENOENT) {
1006                         CDEBUG(D_INODE,
1007                                "destroying non-existent object "LPU64"\n",
1008                                oa->o_id);
1009                         /* rewrite rc with -ENOENT only if it is 0 */
1010                         if (rc == 0)
1011                                 rc = lrc;
1012                 } else if (lrc != 0) {
1013                         CEMERG("error destroying object "LPU64": %d\n",
1014                                oa->o_id, rc);
1015                         rc = lrc;
1016                 }
1017                 count--;
1018                 oa->o_id++;
1019         }
1020
1021         /* if we have transaction then there were some deletions, we don't
1022          * need to return ENOENT in that case because it will not wait
1023          * for commit of these deletions. The ENOENT must be returned only
1024          * if there were no transations.
1025          */
1026         if (rc == -ENOENT) {
1027                 if (info->fti_transno != 0)
1028                         rc = 0;
1029         } else if (rc != 0) {
1030                 /*
1031                  * If we have at least one transaction then llog record
1032                  * on server will be removed upon commit, so for rc != 0
1033                  * we return no transno and llog record will be reprocessed.
1034                  */
1035                 info->fti_transno = 0;
1036         }
1037         ofd_info2oti(info, oti);
1038         RETURN(rc);
1039 }
1040
1041 static int ofd_orphans_destroy(const struct lu_env *env,
1042                                struct obd_export *exp, struct ofd_device *ofd,
1043                                struct obdo *oa)
1044 {
1045         struct ofd_thread_info  *info = ofd_info(env);
1046         obd_id                   last;
1047         int                      skip_orphan;
1048         int                      rc = 0;
1049         struct ost_id            oi = oa->o_oi;
1050
1051         ENTRY;
1052
1053         LASSERT(exp != NULL);
1054         skip_orphan = !!(exp->exp_connect_flags & OBD_CONNECT_SKIP_ORPHAN);
1055
1056         last = ofd_last_id(ofd, oa->o_seq);
1057         CWARN("%s: deleting orphan objects from "LPU64" to "LPU64"\n",
1058               ofd_obd(ofd)->obd_name, oa->o_id + 1, last);
1059
1060         for (oi.oi_id = last; oi.oi_id > oa->o_id; oi.oi_id--) {
1061                 fid_ostid_unpack(&info->fti_fid, &oi, 0);
1062                 rc = ofd_destroy_by_fid(env, ofd, &info->fti_fid, 1);
1063                 if (rc && rc != -ENOENT) /* this is pretty fatal... */
1064                         CEMERG("error destroying precreated id "LPU64": %d\n",
1065                                oi.oi_id, rc);
1066                 if (!skip_orphan) {
1067                         ofd_last_id_set(ofd, oi.oi_id - 1, oa->o_seq);
1068                         /* update last_id on disk periodically so that if we
1069                          * restart * we don't need to re-scan all of the just
1070                          * deleted objects. */
1071                         if ((oi.oi_id & 511) == 0)
1072                                 ofd_last_id_write(env, ofd, oa->o_seq);
1073                 }
1074         }
1075         CDEBUG(D_HA, "%s: after destroy: set last_objids["LPU64"] = "LPU64"\n",
1076                ofd_obd(ofd)->obd_name, oa->o_seq, oa->o_id);
1077         if (!skip_orphan) {
1078                 rc = ofd_last_id_write(env, ofd, oa->o_seq);
1079         } else {
1080                 /* don't reuse orphan object, return last used objid */
1081                 oa->o_id = last;
1082                 rc = 0;
1083         }
1084         RETURN(rc);
1085 }
1086
1087 int ofd_create(const struct lu_env *env, struct obd_export *exp,
1088                struct obdo *oa, struct lov_stripe_md **ea,
1089                struct obd_trans_info *oti)
1090 {
1091         struct ofd_device       *ofd = ofd_exp(exp);
1092         struct ofd_thread_info  *info;
1093         int                      rc = 0, diff;
1094
1095         ENTRY;
1096
1097         info = ofd_info_init(env, exp);
1098         ofd_oti2info(info, oti);
1099
1100         LASSERT(oa->o_seq >= FID_SEQ_OST_MDT0);
1101         LASSERT(oa->o_valid & OBD_MD_FLGROUP);
1102
1103         CDEBUG(D_INFO, "ofd_create(oa->o_seq="LPU64",oa->o_id="LPU64")\n",
1104                oa->o_seq, oa->o_id);
1105
1106         if ((oa->o_valid & OBD_MD_FLFLAGS) &&
1107             (oa->o_flags & OBD_FL_RECREATE_OBJS)) {
1108                 if (!ofd_obd(ofd)->obd_recovering ||
1109                     oa->o_id > ofd_last_id(ofd, oa->o_seq)) {
1110                         CERROR("recreate objid "LPU64" > last id "LPU64"\n",
1111                                         oa->o_id, ofd_last_id(ofd, oa->o_seq));
1112                         GOTO(out_nolock, rc = -EINVAL);
1113                 }
1114                 /* do nothing because we create objects during first write */
1115                 GOTO(out_nolock, rc = 0);
1116         }
1117         /* former ofd_handle_precreate */
1118         if ((oa->o_valid & OBD_MD_FLFLAGS) &&
1119             (oa->o_flags & OBD_FL_DELORPHAN)) {
1120                 /* destroy orphans */
1121                 if (oti->oti_conn_cnt < exp->exp_conn_cnt) {
1122                         CERROR("%s: dropping old orphan cleanup request\n",
1123                                ofd_obd(ofd)->obd_name);
1124                         GOTO(out_nolock, rc = 0);
1125                 }
1126                 /* This causes inflight precreates to abort and drop lock */
1127                 cfs_set_bit(oa->o_seq, &ofd->ofd_destroys_in_progress);
1128                 cfs_mutex_lock(&ofd->ofd_create_locks[oa->o_seq]);
1129                 if (!cfs_test_bit(oa->o_seq, &ofd->ofd_destroys_in_progress)) {
1130                         CERROR("%s:["LPU64"] destroys_in_progress already cleared\n",
1131                                exp->exp_obd->obd_name, oa->o_seq);
1132                         GOTO(out, rc = 0);
1133                 }
1134                 diff = oa->o_id - ofd_last_id(ofd, oa->o_seq);
1135                 CDEBUG(D_HA, "ofd_last_id() = "LPU64" -> diff = %d\n",
1136                        ofd_last_id(ofd, oa->o_seq), diff);
1137                 if (-diff > OST_MAX_PRECREATE) {
1138                         /* FIXME: should reset precreate_next_id on MDS */
1139                         rc = 0;
1140                 } else if (diff < 0) {
1141                         rc = ofd_orphans_destroy(env, exp, ofd, oa);
1142                         cfs_clear_bit(oa->o_seq, &ofd->ofd_destroys_in_progress);
1143                 } else {
1144                         /* XXX: Used by MDS for the first time! */
1145                         cfs_clear_bit(oa->o_seq, &ofd->ofd_destroys_in_progress);
1146                 }
1147         } else {
1148                 cfs_mutex_lock(&ofd->ofd_create_locks[oa->o_seq]);
1149                 if (oti->oti_conn_cnt < exp->exp_conn_cnt) {
1150                         CERROR("%s: dropping old precreate request\n",
1151                                ofd_obd(ofd)->obd_name);
1152                         GOTO(out, rc = 0);
1153                 }
1154                 /* only precreate if group == 0 and o_id is specfied */
1155                 if (!fid_seq_is_mdt(oa->o_seq) || oa->o_id == 0) {
1156                         diff = 1; /* shouldn't we create this right now? */
1157                 } else {
1158                         diff = oa->o_id - ofd_last_id(ofd, oa->o_seq);
1159                 }
1160         }
1161         if (diff > 0) {
1162                 cfs_time_t       enough_time = cfs_time_shift(DISK_TIMEOUT);
1163                 obd_id           next_id;
1164                 int              created = 0;
1165                 int              count;
1166
1167                 if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
1168                     !(oa->o_flags & OBD_FL_DELORPHAN)) {
1169                         /* don't enforce grant during orphan recovery */
1170                         rc = ofd_grant_create(env,
1171                                               ofd_obd(ofd)->obd_self_export,
1172                                               &diff);
1173                         if (rc) {
1174                                 CDEBUG(D_HA, "%s: failed to acquire grant space"
1175                                        "for precreate (%d)\n",
1176                                        ofd_obd(ofd)->obd_name, diff);
1177                                 diff = 0;
1178                         }
1179                 }
1180
1181                 while (diff > 0) {
1182                         next_id = ofd_last_id(ofd, oa->o_seq) + 1;
1183                         count = ofd_precreate_batch(ofd, diff);
1184
1185                         CDEBUG(D_HA, "%s: reserve %d objects in group "LPU64
1186                                " at "LPU64"\n", ofd_obd(ofd)->obd_name,
1187                                count, oa->o_seq, next_id);
1188
1189                         if (cfs_time_after(jiffies, enough_time)) {
1190                                 LCONSOLE_WARN("%s: Slow creates, %d/%d objects"
1191                                               " created at a rate of %d/s\n",
1192                                               ofd_obd(ofd)->obd_name,
1193                                               created, diff + created,
1194                                               created / DISK_TIMEOUT);
1195                                 break;
1196                 }
1197
1198                         rc = ofd_precreate_objects(env, ofd, next_id,
1199                                                    oa->o_seq, count);
1200                         if (rc > 0) {
1201                                 created += rc;
1202                                 diff -= rc;
1203                         } else if (rc < 0) {
1204                                 break;
1205                         }
1206                 }
1207                 if (created > 0) {
1208                         /* some objects got created, we can return
1209                          * them, even if last creation failed */
1210                         oa->o_id = ofd_last_id(ofd, oa->o_seq);
1211                         rc = 0;
1212                 } else {
1213                         CERROR("unable to precreate: %d\n", rc);
1214                         oa->o_id = ofd_last_id(ofd, oa->o_seq);
1215                 }
1216
1217                 oa->o_valid |= OBD_MD_FLID | OBD_MD_FLGROUP;
1218
1219                 if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
1220                     !(oa->o_flags & OBD_FL_DELORPHAN))
1221                         ofd_grant_commit(env, ofd_obd(ofd)->obd_self_export,
1222                                          rc);
1223         }
1224
1225         ofd_info2oti(info, oti);
1226 out:
1227         cfs_mutex_unlock(&ofd->ofd_create_locks[oa->o_seq]);
1228 out_nolock:
1229         if (rc == 0 && ea != NULL) {
1230                 struct lov_stripe_md *lsm = *ea;
1231
1232                 lsm->lsm_object_id = oa->o_id;
1233         }
1234         return rc;
1235 }
1236
1237 int ofd_getattr(const struct lu_env *env, struct obd_export *exp,
1238                 struct obd_info *oinfo)
1239 {
1240         struct ofd_device       *ofd = ofd_exp(exp);
1241         struct ofd_thread_info  *info;
1242         struct ofd_object       *fo;
1243         __u64                    curr_version;
1244         int                      rc = 0;
1245
1246         ENTRY;
1247
1248         info = ofd_info_init(env, exp);
1249
1250         fid_ostid_unpack(&info->fti_fid, &oinfo->oi_oa->o_oi, 0);
1251         rc = ofd_auth_capa(exp, &info->fti_fid, oinfo->oi_oa->o_seq,
1252                            oinfo_capa(oinfo), CAPA_OPC_META_READ);
1253         if (rc)
1254                 GOTO(out, rc);
1255
1256         fo = ofd_object_find(env, ofd, &info->fti_fid);
1257         if (IS_ERR(fo))
1258                 GOTO(out, rc = PTR_ERR(fo));
1259         LASSERT(fo != NULL);
1260         rc = ofd_attr_get(env, fo, &info->fti_attr);
1261         oinfo->oi_oa->o_valid = OBD_MD_FLID;
1262         if (rc == 0)
1263                 obdo_from_la(oinfo->oi_oa, &info->fti_attr,
1264                              OFD_VALID_FLAGS | LA_UID | LA_GID);
1265
1266         /* Store object version in reply */
1267         curr_version = dt_version_get(env, ofd_object_child(fo));
1268         if ((__s64)curr_version != -EOPNOTSUPP) {
1269                 oinfo->oi_oa->o_valid |= OBD_MD_FLDATAVERSION;
1270                 oinfo->oi_oa->o_data_version = curr_version;
1271         }
1272         ofd_object_put(env, fo);
1273 out:
1274         RETURN(rc);
1275 }
1276
1277 static int ofd_sync(const struct lu_env *env, struct obd_export *exp,
1278                     struct obd_info *oinfo, obd_size start, obd_size end,
1279                     struct ptlrpc_request_set *set)
1280 {
1281         struct ofd_device       *ofd = ofd_exp(exp);
1282         struct ofd_thread_info  *info;
1283         struct ofd_object       *fo;
1284         int                      rc = 0;
1285
1286         ENTRY;
1287
1288         /* if no objid is specified, it means "sync whole filesystem" */
1289         if (oinfo->oi_oa == NULL || !(oinfo->oi_oa->o_valid & OBD_MD_FLID)) {
1290                 rc = dt_sync(env, ofd->ofd_osd);
1291                 GOTO(out, rc);
1292         }
1293
1294         info = ofd_info_init(env, exp);
1295         fid_ostid_unpack(&info->fti_fid, &oinfo->oi_oa->o_oi, 0);
1296
1297         rc = ofd_auth_capa(exp, &info->fti_fid, oinfo->oi_oa->o_seq,
1298                            oinfo_capa(oinfo), CAPA_OPC_OSS_TRUNC);
1299         if (rc)
1300                 GOTO(out, rc);
1301
1302         fo = ofd_object_find(env, ofd, &info->fti_fid);
1303         if (IS_ERR(fo)) {
1304                 CERROR("%s: error finding object "DFID": rc = %ld\n",
1305                        exp->exp_obd->obd_name, PFID(&info->fti_fid),
1306                        PTR_ERR(fo));
1307                 GOTO(out, rc = PTR_ERR(fo));
1308         }
1309
1310         ofd_write_lock(env, fo);
1311         if (!ofd_object_exists(fo))
1312                 GOTO(unlock, rc = -ENOENT);
1313
1314         if (dt_version_get(env, ofd_object_child(fo)) >
1315             ofd_obd(ofd)->obd_last_committed) {
1316                 rc = dt_object_sync(env, ofd_object_child(fo));
1317                 if (rc)
1318                         GOTO(unlock, rc);
1319         }
1320
1321         oinfo->oi_oa->o_valid = OBD_MD_FLID;
1322         rc = ofd_attr_get(env, fo, &info->fti_attr);
1323         obdo_from_la(oinfo->oi_oa, &info->fti_attr, OFD_VALID_FLAGS);
1324
1325         ofd_counter_incr(exp, LPROC_OFD_STATS_SYNC, oinfo->oi_jobid, 1);
1326         EXIT;
1327 unlock:
1328         ofd_write_unlock(env, fo);
1329         ofd_object_put(env, fo);
1330 out:
1331         return rc;
1332 }
1333
1334 int ofd_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1335                   void *karg, void *uarg)
1336 {
1337         struct lu_env            env;
1338         struct ofd_device       *ofd = ofd_exp(exp);
1339         struct obd_device       *obd = ofd_obd(ofd);
1340         int                      rc;
1341
1342         ENTRY;
1343
1344         CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
1345         rc = lu_env_init(&env, LCT_LOCAL);
1346         if (rc)
1347                 RETURN(rc);
1348
1349         switch (cmd) {
1350         case OBD_IOC_ABORT_RECOVERY:
1351                 CERROR("%s: aborting recovery\n", obd->obd_name);
1352                 target_stop_recovery_thread(obd);
1353                 break;
1354         case OBD_IOC_SYNC:
1355                 CDEBUG(D_RPCTRACE, "syncing ost %s\n", obd->obd_name);
1356                 rc = dt_sync(&env, ofd->ofd_osd);
1357                 break;
1358         case OBD_IOC_SET_READONLY:
1359                 rc = dt_sync(&env, ofd->ofd_osd);
1360                 if (rc == 0)
1361                         rc = dt_ro(&env, ofd->ofd_osd);
1362                 break;
1363         default:
1364                 CERROR("%s: not supported cmd = %d\n", obd->obd_name, cmd);
1365                 rc = -ENOTTY;
1366         }
1367
1368         lu_env_fini(&env);
1369         RETURN(rc);
1370 }
1371
1372 static int ofd_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
1373 {
1374         int rc = 0;
1375
1376         ENTRY;
1377
1378         switch(stage) {
1379         case OBD_CLEANUP_EARLY:
1380                 break;
1381         case OBD_CLEANUP_EXPORTS:
1382                 target_cleanup_recovery(obd);
1383                 break;
1384         }
1385         RETURN(rc);
1386 }
1387
1388 static int ofd_ping(const struct lu_env *env, struct obd_export *exp)
1389 {
1390         ofd_fmd_expire(exp);
1391         return 0;
1392 }
1393
1394 static int ofd_health_check(const struct lu_env *nul, struct obd_device *obd)
1395 {
1396         struct ofd_device       *ofd = ofd_dev(obd->obd_lu_dev);
1397         struct ofd_thread_info  *info;
1398         struct lu_env            env;
1399 #ifdef USE_HEALTH_CHECK_WRITE
1400         struct thandle          *th;
1401 #endif
1402         int                      rc = 0;
1403
1404         /* obd_proc_read_health pass NULL env, we need real one */
1405         rc = lu_env_init(&env, LCT_DT_THREAD);
1406         if (rc)
1407                 RETURN(rc);
1408
1409         info = ofd_info_init(&env, NULL);
1410         rc = dt_statfs(&env, ofd->ofd_osd, &info->fti_u.osfs);
1411         if (unlikely(rc))
1412                 GOTO(out, rc);
1413
1414         if (info->fti_u.osfs.os_state == OS_STATE_READONLY)
1415                 GOTO(out, rc = -EROFS);
1416
1417 #ifdef USE_HEALTH_CHECK_WRITE
1418         OBD_ALLOC(info->fti_buf.lb_buf, CFS_PAGE_SIZE);
1419         if (info->fti_buf.lb_buf == NULL)
1420                 GOTO(out, rc = -ENOMEM);
1421
1422         info->fti_buf.lb_len = CFS_PAGE_SIZE;
1423         info->fti_off = 0;
1424
1425         th = dt_trans_create(&env, ofd->ofd_osd);
1426         if (IS_ERR(th))
1427                 GOTO(out, rc = PTR_ERR(th));
1428
1429         rc = dt_declare_record_write(&env, ofd->ofd_health_check_file,
1430                                      info->fti_buf.lb_len, info->fti_off, th);
1431         if (rc == 0) {
1432                 th->th_sync = 1; /* sync IO is needed */
1433                 rc = dt_trans_start_local(&env, ofd->ofd_osd, th);
1434                 if (rc == 0)
1435                         rc = dt_record_write(&env, ofd->ofd_health_check_file,
1436                                              &info->fti_buf, &info->fti_off,
1437                                              th);
1438         }
1439         dt_trans_stop(&env, ofd->ofd_osd, th);
1440
1441         OBD_FREE(info->fti_buf.lb_buf, CFS_PAGE_SIZE);
1442
1443         CDEBUG(D_INFO, "write 1 page synchronously for checking io rc %d\n",rc);
1444 #endif
1445 out:
1446         lu_env_fini(&env);
1447         return !!rc;
1448 }
1449
1450 static int ofd_obd_notify(struct obd_device *obd, struct obd_device *unused,
1451                           enum obd_notify_event ev, void *data)
1452 {
1453         switch (ev) {
1454         case OBD_NOTIFY_CONFIG:
1455                 LASSERT(obd->obd_no_conn);
1456                 cfs_spin_lock(&obd->obd_dev_lock);
1457                 obd->obd_no_conn = 0;
1458                 cfs_spin_unlock(&obd->obd_dev_lock);
1459                 break;
1460         default:
1461                 CDEBUG(D_INFO, "%s: Unhandled notification %#x\n",
1462                        obd->obd_name, ev);
1463         }
1464         return 0;
1465 }
1466
1467 /*
1468  * Handle quota control requests to consult current usage/limit.
1469  *
1470  * \param obd - is the obd device associated with the ofd
1471  * \param exp - is the client's export
1472  * \param oqctl - is the obd_quotactl request to be processed
1473  */
1474 static int ofd_quotactl(struct obd_device *obd, struct obd_export *exp,
1475                         struct obd_quotactl *oqctl)
1476 {
1477         struct ofd_device  *ofd = ofd_dev(obd->obd_lu_dev);
1478         struct lu_env       env;
1479         int                 rc;
1480         ENTRY;
1481
1482         /* report success for quota on/off for interoperability with current MDT
1483          * stack */
1484         if (oqctl->qc_cmd == Q_QUOTAON || oqctl->qc_cmd == Q_QUOTAOFF)
1485                 RETURN(0);
1486
1487         rc = lu_env_init(&env, LCT_DT_THREAD);
1488         if (rc)
1489                 RETURN(rc);
1490
1491         rc = lquotactl_slv(&env, ofd->ofd_osd, oqctl);
1492         lu_env_fini(&env);
1493
1494         RETURN(rc);
1495 }
1496
1497 struct obd_ops ofd_obd_ops = {
1498         .o_owner                = THIS_MODULE,
1499         .o_connect              = ofd_obd_connect,
1500         .o_reconnect            = ofd_obd_reconnect,
1501         .o_disconnect           = ofd_obd_disconnect,
1502         .o_set_info_async       = ofd_set_info_async,
1503         .o_get_info             = ofd_get_info,
1504         .o_create               = ofd_create,
1505         .o_statfs               = ofd_statfs,
1506         .o_setattr              = ofd_setattr,
1507         .o_preprw               = ofd_preprw,
1508         .o_commitrw             = ofd_commitrw,
1509         .o_destroy              = ofd_destroy,
1510         .o_init_export          = ofd_init_export,
1511         .o_destroy_export       = ofd_destroy_export,
1512         .o_postrecov            = ofd_obd_postrecov,
1513         .o_punch                = ofd_punch,
1514         .o_getattr              = ofd_getattr,
1515         .o_sync                 = ofd_sync,
1516         .o_iocontrol            = ofd_iocontrol,
1517         .o_precleanup           = ofd_precleanup,
1518         .o_ping                 = ofd_ping,
1519         .o_health_check         = ofd_health_check,
1520         .o_notify               = ofd_obd_notify,
1521         .o_quotactl             = ofd_quotactl,
1522 };