lustre/ofd/ofd_dev.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, 2013, 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.c
  37  *
  38  * Author: Alex Zhuravlev <bzzz@whamcloud.com>
  39  * Author: Mike Pershin <tappro@whamcloud.com>
  40  * Author: Johann Lombardi <johann@whamcloud.com>
  41  */
  42
  43 #define DEBUG_SUBSYSTEM S_FILTER
  44
  45 #include <obd_class.h>
  46 #include <lustre_param.h>
  47 #include <lustre_fid.h>
  48 #include <lustre_lfsck.h>
  49 #include <lustre/lustre_idl.h>
  50 #include <lustre_dlm.h>
  51 #include <lustre_quota.h>
  52
  53 #include "ofd_internal.h"
  54
  55 /* Slab for OFD object allocation */
  56 static struct kmem_cache *ofd_object_kmem;
  57
  58 static struct lu_kmem_descr ofd_caches[] = {
  59         {
  60                 .ckd_cache = &ofd_object_kmem,
  61                 .ckd_name  = "ofd_obj",
  62                 .ckd_size  = sizeof(struct ofd_object)
  63         },
  64         {
  65                 .ckd_cache = NULL
  66         }
  67 };
  68
  69 static int ofd_connect_to_next(const struct lu_env *env, struct ofd_device *m,
  70                                const char *next, struct obd_export **exp)
  71 {
  72         struct obd_connect_data *data = NULL;
  73         struct obd_device       *obd;
  74         int                      rc;
  75         ENTRY;
  76
  77         OBD_ALLOC_PTR(data);
  78         if (data == NULL)
  79                 GOTO(out, rc = -ENOMEM);
  80
  81         obd = class_name2obd(next);
  82         if (obd == NULL) {
  83                 CERROR("%s: can't locate next device: %s\n",
  84                        m->ofd_dt_dev.dd_lu_dev.ld_obd->obd_name, next);
  85                 GOTO(out, rc = -ENOTCONN);
  86         }
  87
  88         data->ocd_connect_flags = OBD_CONNECT_VERSION;
  89         data->ocd_version = LUSTRE_VERSION_CODE;
  90
  91         rc = obd_connect(NULL, exp, obd, &obd->obd_uuid, data, NULL);
  92         if (rc) {
  93                 CERROR("%s: cannot connect to next dev %s: rc = %d\n",
  94                        m->ofd_dt_dev.dd_lu_dev.ld_obd->obd_name, next, rc);
  95                 GOTO(out, rc);
  96         }
  97
  98         m->ofd_dt_dev.dd_lu_dev.ld_site =
  99                 m->ofd_osd_exp->exp_obd->obd_lu_dev->ld_site;
 100         LASSERT(m->ofd_dt_dev.dd_lu_dev.ld_site);
 101         m->ofd_osd = lu2dt_dev(m->ofd_osd_exp->exp_obd->obd_lu_dev);
 102         m->ofd_dt_dev.dd_lu_dev.ld_site->ls_top_dev = &m->ofd_dt_dev.dd_lu_dev;
 103
 104 out:
 105         if (data)
 106                 OBD_FREE_PTR(data);
 107         RETURN(rc);
 108 }
 109
 110 static int ofd_stack_init(const struct lu_env *env,
 111                           struct ofd_device *m, struct lustre_cfg *cfg)
 112 {
 113         const char              *dev = lustre_cfg_string(cfg, 0);
 114         struct lu_device        *d;
 115         struct ofd_thread_info  *info = ofd_info(env);
 116         struct lustre_mount_info *lmi;
 117         int                      rc;
 118         char                    *osdname;
 119
 120         ENTRY;
 121
 122         lmi = server_get_mount(dev);
 123         if (lmi == NULL) {
 124                 CERROR("Cannot get mount info for %s!\n", dev);
 125                 RETURN(-ENODEV);
 126         }
 127
 128         /* find bottom osd */
 129         OBD_ALLOC(osdname, MTI_NAME_MAXLEN);
 130         if (osdname == NULL)
 131                 RETURN(-ENOMEM);
 132
 133         snprintf(osdname, MTI_NAME_MAXLEN, "%s-osd", dev);
 134         rc = ofd_connect_to_next(env, m, osdname, &m->ofd_osd_exp);
 135         OBD_FREE(osdname, MTI_NAME_MAXLEN);
 136         if (rc)
 137                 RETURN(rc);
 138
 139         d = m->ofd_osd_exp->exp_obd->obd_lu_dev;
 140         LASSERT(d);
 141         m->ofd_osd = lu2dt_dev(d);
 142
 143         snprintf(info->fti_u.name, sizeof(info->fti_u.name),
 144                  "%s-osd", lustre_cfg_string(cfg, 0));
 145
 146         RETURN(rc);
 147 }
 148
 149 static void ofd_stack_fini(const struct lu_env *env, struct ofd_device *m,
 150                            struct lu_device *top)
 151 {
 152         struct obd_device       *obd = ofd_obd(m);
 153         struct lustre_cfg_bufs   bufs;
 154         struct lustre_cfg       *lcfg;
 155         char                     flags[3] = "";
 156
 157         ENTRY;
 158
 159         lu_site_purge(env, top->ld_site, ~0);
 160         /* process cleanup, pass mdt obd name to get obd umount flags */
 161         lustre_cfg_bufs_reset(&bufs, obd->obd_name);
 162         if (obd->obd_force)
 163                 strcat(flags, "F");
 164         if (obd->obd_fail)
 165                 strcat(flags, "A");
 166         lustre_cfg_bufs_set_string(&bufs, 1, flags);
 167         lcfg = lustre_cfg_new(LCFG_CLEANUP, &bufs);
 168         if (!lcfg) {
 169                 CERROR("Cannot alloc lcfg!\n");
 170                 RETURN_EXIT;
 171         }
 172
 173         LASSERT(top);
 174         top->ld_ops->ldo_process_config(env, top, lcfg);
 175         lustre_cfg_free(lcfg);
 176
 177         lu_site_purge(env, top->ld_site, ~0);
 178         if (!cfs_hash_is_empty(top->ld_site->ls_obj_hash)) {
 179                 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
 180                 lu_site_print(env, top->ld_site, &msgdata, lu_cdebug_printer);
 181         }
 182
 183         LASSERT(m->ofd_osd_exp);
 184         obd_disconnect(m->ofd_osd_exp);
 185         m->ofd_osd = NULL;
 186
 187         EXIT;
 188 }
 189
 190 /* For interoperability, see mdt_interop_param[]. */
 191 static struct cfg_interop_param ofd_interop_param[] = {
 192         { "ost.quota_type",     NULL },
 193         { NULL }
 194 };
 195
 196 /* Some parameters were moved from ofd to osd and only their
 197  * symlinks were kept in ofd by LU-3106. They are:
 198  * -writehthrough_cache_enable
 199  * -readcache_max_filese
 200  * -read_cache_enable
 201  * -brw_stats
 202  * Since they are not included by the static lprocfs var list,
 203  * a pre-check is added for them to avoid "unknown param" error
 204  * message confuses the customer. If they are matched in this
 205  * check, they will be passed to the osd directly.
 206  */
 207 static bool match_symlink_param(char *param)
 208 {
 209         char *sval;
 210         int paramlen;
 211
 212         if (class_match_param(param, PARAM_OST, &param) == 0) {
 213                 sval = strchr(param, '=');
 214                 if (sval != NULL) {
 215                         paramlen = sval - param;
 216                         if (strncmp(param, "writethrough_cache_enable",
 217                                     paramlen) == 0 ||
 218                             strncmp(param, "readcache_max_filesize",
 219                                     paramlen) == 0 ||
 220                             strncmp(param, "read_cache_enable",
 221                                     paramlen) == 0 ||
 222                             strncmp(param, "brw_stats", paramlen) == 0)
 223                                 return true;
 224                 }
 225         }
 226
 227         return false;
 228 }
 229
 230 /* used by MGS to process specific configurations */
 231 static int ofd_process_config(const struct lu_env *env, struct lu_device *d,
 232                               struct lustre_cfg *cfg)
 233 {
 234         struct ofd_device       *m = ofd_dev(d);
 235         struct dt_device        *dt_next = m->ofd_osd;
 236         struct lu_device        *next = &dt_next->dd_lu_dev;
 237         int                      rc;
 238
 239         ENTRY;
 240
 241         switch (cfg->lcfg_command) {
 242         case LCFG_PARAM: {
 243                 struct lprocfs_static_vars lvars;
 244
 245                 /* For interoperability */
 246                 struct cfg_interop_param   *ptr = NULL;
 247                 struct lustre_cfg          *old_cfg = NULL;
 248                 char                       *param = NULL;
 249
 250                 param = lustre_cfg_string(cfg, 1);
 251                 if (param == NULL) {
 252                         CERROR("param is empty\n");
 253                         rc = -EINVAL;
 254                         break;
 255                 }
 256
 257                 ptr = class_find_old_param(param, ofd_interop_param);
 258                 if (ptr != NULL) {
 259                         if (ptr->new_param == NULL) {
 260                                 rc = 0;
 261                                 CWARN("For interoperability, skip this %s."
 262                                       " It is obsolete.\n", ptr->old_param);
 263                                 break;
 264                         }
 265
 266                         CWARN("Found old param %s, changed it to %s.\n",
 267                               ptr->old_param, ptr->new_param);
 268
 269                         old_cfg = cfg;
 270                         cfg = lustre_cfg_rename(old_cfg, ptr->new_param);
 271                         if (IS_ERR(cfg)) {
 272                                 rc = PTR_ERR(cfg);
 273                                 break;
 274                         }
 275                 }
 276
 277                 if (match_symlink_param(param)) {
 278                         rc = next->ld_ops->ldo_process_config(env, next, cfg);
 279                         break;
 280                 }
 281
 282                 lprocfs_ofd_init_vars(&lvars);
 283                 rc = class_process_proc_param(PARAM_OST, lvars.obd_vars, cfg,
 284                                               d->ld_obd);
 285                 if (rc > 0 || rc == -ENOSYS) {
 286                         CDEBUG(D_CONFIG, "pass param %s down the stack.\n",
 287                                param);
 288                         /* we don't understand; pass it on */
 289                         rc = next->ld_ops->ldo_process_config(env, next, cfg);
 290                 }
 291                 break;
 292         }
 293         case LCFG_SPTLRPC_CONF: {
 294                 rc = -ENOTSUPP;
 295                 break;
 296         }
 297         default:
 298                 /* others are passed further */
 299                 rc = next->ld_ops->ldo_process_config(env, next, cfg);
 300                 break;
 301         }
 302         RETURN(rc);
 303 }
 304
 305 static int ofd_object_init(const struct lu_env *env, struct lu_object *o,
 306                            const struct lu_object_conf *conf)
 307 {
 308         struct ofd_device       *d = ofd_dev(o->lo_dev);
 309         struct lu_device        *under;
 310         struct lu_object        *below;
 311         int                      rc = 0;
 312
 313         ENTRY;
 314
 315         CDEBUG(D_INFO, "object init, fid = "DFID"\n",
 316                PFID(lu_object_fid(o)));
 317
 318         under = &d->ofd_osd->dd_lu_dev;
 319         below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
 320         if (below != NULL)
 321                 lu_object_add(o, below);
 322         else
 323                 rc = -ENOMEM;
 324
 325         RETURN(rc);
 326 }
 327
 328 static void ofd_object_free(const struct lu_env *env, struct lu_object *o)
 329 {
 330         struct ofd_object       *of = ofd_obj(o);
 331         struct lu_object_header *h;
 332
 333         ENTRY;
 334
 335         h = o->lo_header;
 336         CDEBUG(D_INFO, "object free, fid = "DFID"\n",
 337                PFID(lu_object_fid(o)));
 338
 339         lu_object_fini(o);
 340         lu_object_header_fini(h);
 341         OBD_SLAB_FREE_PTR(of, ofd_object_kmem);
 342         EXIT;
 343 }
 344
 345 static int ofd_object_print(const struct lu_env *env, void *cookie,
 346                             lu_printer_t p, const struct lu_object *o)
 347 {
 348         return (*p)(env, cookie, LUSTRE_OST_NAME"-object@%p", o);
 349 }
 350
 351 struct lu_object_operations ofd_obj_ops = {
 352         .loo_object_init        = ofd_object_init,
 353         .loo_object_free        = ofd_object_free,
 354         .loo_object_print       = ofd_object_print
 355 };
 356
 357 static struct lu_object *ofd_object_alloc(const struct lu_env *env,
 358                                           const struct lu_object_header *hdr,
 359                                           struct lu_device *d)
 360 {
 361         struct ofd_object *of;
 362
 363         ENTRY;
 364
 365         OBD_SLAB_ALLOC_PTR_GFP(of, ofd_object_kmem, __GFP_IO);
 366         if (of != NULL) {
 367                 struct lu_object        *o;
 368                 struct lu_object_header *h;
 369
 370                 o = &of->ofo_obj.do_lu;
 371                 h = &of->ofo_header;
 372                 lu_object_header_init(h);
 373                 lu_object_init(o, h, d);
 374                 lu_object_add_top(h, o);
 375                 o->lo_ops = &ofd_obj_ops;
 376                 RETURN(o);
 377         } else {
 378                 RETURN(NULL);
 379         }
 380 }
 381
 382 extern int ost_handle(struct ptlrpc_request *req);
 383
 384 static int ofd_lfsck_out_notify(const struct lu_env *env, void *data,
 385                                 enum lfsck_events event)
 386 {
 387         struct ofd_device *ofd = data;
 388         struct obd_device *obd = ofd_obd(ofd);
 389
 390         switch (event) {
 391         case LE_LASTID_REBUILDING:
 392                 CWARN("%s: Found crashed LAST_ID, deny creating new OST-object "
 393                       "on the device until the LAST_ID rebuilt successfully.\n",
 394                       obd->obd_name);
 395                 down_write(&ofd->ofd_lastid_rwsem);
 396                 ofd->ofd_lastid_rebuilding = 1;
 397                 up_write(&ofd->ofd_lastid_rwsem);
 398                 break;
 399         case LE_LASTID_REBUILT: {
 400                 down_write(&ofd->ofd_lastid_rwsem);
 401                 ofd_seqs_free(env, ofd);
 402                 ofd->ofd_lastid_rebuilding = 0;
 403                 ofd->ofd_lastid_gen++;
 404                 up_write(&ofd->ofd_lastid_rwsem);
 405                 break;
 406         }
 407         default:
 408                 CERROR("%s: unknown lfsck event: rc = %d\n",
 409                        ofd_obd(ofd)->obd_name, event);
 410                 return -EINVAL;
 411         }
 412
 413         return 0;
 414 }
 415
 416 static int ofd_prepare(const struct lu_env *env, struct lu_device *pdev,
 417                        struct lu_device *dev)
 418 {
 419         struct ofd_thread_info          *info;
 420         struct ofd_device               *ofd = ofd_dev(dev);
 421         struct obd_device               *obd = ofd_obd(ofd);
 422         struct lu_device                *next = &ofd->ofd_osd->dd_lu_dev;
 423         struct lfsck_start_param         lsp;
 424         int                              rc;
 425
 426         ENTRY;
 427
 428         info = ofd_info_init(env, NULL);
 429         if (info == NULL)
 430                 RETURN(-EFAULT);
 431
 432         /* initialize lower device */
 433         rc = next->ld_ops->ldo_prepare(env, dev, next);
 434         if (rc != 0)
 435                 RETURN(rc);
 436
 437         rc = lfsck_register(env, ofd->ofd_osd, ofd->ofd_osd, obd,
 438                             ofd_lfsck_out_notify, ofd, false);
 439         if (rc != 0) {
 440                 CERROR("%s: failed to initialize lfsck: rc = %d\n",
 441                        obd->obd_name, rc);
 442                 RETURN(rc);
 443         }
 444
 445         lsp.lsp_namespace = ofd->ofd_namespace;
 446         lsp.lsp_start = NULL;
 447         lsp.lsp_index_valid = 0;
 448         rc = lfsck_start(env, ofd->ofd_osd, &lsp);
 449         if (rc != 0) {
 450                 CWARN("%s: auto trigger paused LFSCK failed: rc = %d\n",
 451                       obd->obd_name, rc);
 452                 rc = 0;
 453         }
 454
 455         target_recovery_init(&ofd->ofd_lut, tgt_request_handle);
 456         LASSERT(obd->obd_no_conn);
 457         spin_lock(&obd->obd_dev_lock);
 458         obd->obd_no_conn = 0;
 459         spin_unlock(&obd->obd_dev_lock);
 460
 461         if (obd->obd_recovering == 0)
 462                 ofd_postrecov(env, ofd);
 463
 464         RETURN(rc);
 465 }
 466
 467 static int ofd_recovery_complete(const struct lu_env *env,
 468                                  struct lu_device *dev)
 469 {
 470         struct ofd_device       *ofd = ofd_dev(dev);
 471         struct lu_device        *next = &ofd->ofd_osd->dd_lu_dev;
 472         int                      rc = 0, max_precreate;
 473
 474         ENTRY;
 475
 476         /* Grant space for object precreation on the self export.
 477          * This initial reserved space (i.e. 10MB for zfs and 280KB for ldiskfs)
 478          * is enough to create 10k objects. More space is then acquired for
 479          * precreation in ofd_grant_create().
 480          */
 481         max_precreate = OST_MAX_PRECREATE * ofd->ofd_dt_conf.ddp_inodespace / 2;
 482         ofd_grant_connect(env, dev->ld_obd->obd_self_export, max_precreate,
 483                           false);
 484         rc = next->ld_ops->ldo_recovery_complete(env, next);
 485         RETURN(rc);
 486 }
 487
 488 static struct lu_device_operations ofd_lu_ops = {
 489         .ldo_object_alloc       = ofd_object_alloc,
 490         .ldo_process_config     = ofd_process_config,
 491         .ldo_recovery_complete  = ofd_recovery_complete,
 492         .ldo_prepare            = ofd_prepare,
 493 };
 494
 495 static int ofd_procfs_init(struct ofd_device *ofd)
 496 {
 497         struct lprocfs_static_vars       lvars;
 498         struct obd_device               *obd = ofd_obd(ofd);
 499         cfs_proc_dir_entry_t            *entry;
 500         int                              rc = 0;
 501
 502         ENTRY;
 503
 504         /* lprocfs must be setup before the ofd so state can be safely added
 505          * to /proc incrementally as the ofd is setup */
 506         lprocfs_ofd_init_vars(&lvars);
 507         rc = lprocfs_obd_setup(obd, lvars.obd_vars);
 508         if (rc) {
 509                 CERROR("%s: lprocfs_obd_setup failed: %d.\n",
 510                        obd->obd_name, rc);
 511                 RETURN(rc);
 512         }
 513
 514         rc = lprocfs_alloc_obd_stats(obd, LPROC_OFD_STATS_LAST);
 515         if (rc) {
 516                 CERROR("%s: lprocfs_alloc_obd_stats failed: %d.\n",
 517                        obd->obd_name, rc);
 518                 GOTO(obd_cleanup, rc);
 519         }
 520
 521         obd->obd_uses_nid_stats = 1;
 522
 523         entry = lprocfs_register("exports", obd->obd_proc_entry, NULL, NULL);
 524         if (IS_ERR(entry)) {
 525                 rc = PTR_ERR(entry);
 526                 CERROR("%s: error %d setting up lprocfs for %s\n",
 527                        obd->obd_name, rc, "exports");
 528                 GOTO(obd_cleanup, rc);
 529         }
 530         obd->obd_proc_exports_entry = entry;
 531
 532         entry = lprocfs_add_simple(obd->obd_proc_exports_entry, "clear",
 533                                    lprocfs_nid_stats_clear_read,
 534                                    lprocfs_nid_stats_clear_write, obd, NULL);
 535         if (IS_ERR(entry)) {
 536                 rc = PTR_ERR(entry);
 537                 CERROR("%s: add proc entry 'clear' failed: %d.\n",
 538                        obd->obd_name, rc);
 539                 GOTO(obd_cleanup, rc);
 540         }
 541
 542         ofd_stats_counter_init(obd->obd_stats);
 543
 544         rc = lprocfs_job_stats_init(obd, LPROC_OFD_STATS_LAST,
 545                                     ofd_stats_counter_init);
 546         if (rc)
 547                 GOTO(remove_entry_clear, rc);
 548         RETURN(0);
 549 remove_entry_clear:
 550         lprocfs_remove_proc_entry("clear", obd->obd_proc_exports_entry);
 551 obd_cleanup:
 552         lprocfs_obd_cleanup(obd);
 553         lprocfs_free_obd_stats(obd);
 554
 555         return rc;
 556 }
 557
 558 static void ofd_procfs_add_brw_stats_symlink(struct ofd_device *ofd)
 559 {
 560         struct obd_device       *obd = ofd_obd(ofd);
 561         struct obd_device       *osd_obd = ofd->ofd_osd_exp->exp_obd;
 562         cfs_proc_dir_entry_t    *osd_root = osd_obd->obd_type->typ_procroot;
 563         cfs_proc_dir_entry_t    *osd_dir;
 564
 565         osd_dir = lprocfs_srch(osd_root, obd->obd_name);
 566         if (osd_dir == NULL)
 567                 return;
 568
 569         if (lprocfs_srch(osd_dir, "brw_stats") != NULL)
 570                 lprocfs_add_symlink("brw_stats", obd->obd_proc_entry,
 571                                     "../../%s/%s/brw_stats",
 572                                     osd_root->name, osd_dir->name);
 573
 574         if (lprocfs_srch(osd_dir, "read_cache_enable") != NULL)
 575                 lprocfs_add_symlink("read_cache_enable", obd->obd_proc_entry,
 576                                     "../../%s/%s/read_cache_enable",
 577                                     osd_root->name, osd_dir->name);
 578
 579         if (lprocfs_srch(osd_dir, "readcache_max_filesize") != NULL)
 580                 lprocfs_add_symlink("readcache_max_filesize",
 581                                     obd->obd_proc_entry,
 582                                     "../../%s/%s/readcache_max_filesize",
 583                                     osd_root->name, osd_dir->name);
 584
 585         if (lprocfs_srch(osd_dir, "writethrough_cache_enable") != NULL)
 586                 lprocfs_add_symlink("writethrough_cache_enable",
 587                                     obd->obd_proc_entry,
 588                                     "../../%s/%s/writethrough_cache_enable",
 589                                     osd_root->name, osd_dir->name);
 590 }
 591
 592 static void ofd_procfs_fini(struct ofd_device *ofd)
 593 {
 594         struct obd_device *obd = ofd_obd(ofd);
 595
 596         lprocfs_remove_proc_entry("writethrough_cache_enable",
 597                                   obd->obd_proc_entry);
 598         lprocfs_remove_proc_entry("readcache_max_filesize",
 599                                   obd->obd_proc_entry);
 600         lprocfs_remove_proc_entry("read_cache_enable", obd->obd_proc_entry);
 601         lprocfs_remove_proc_entry("brw_stats", obd->obd_proc_entry);
 602         lprocfs_remove_proc_entry("clear", obd->obd_proc_exports_entry);
 603         lprocfs_free_per_client_stats(obd);
 604         lprocfs_obd_cleanup(obd);
 605         lprocfs_free_obd_stats(obd);
 606         lprocfs_job_stats_fini(obd);
 607 }
 608
 609 extern int ost_handle(struct ptlrpc_request *req);
 610
 611 int ofd_fid_fini(const struct lu_env *env, struct ofd_device *ofd)
 612 {
 613         return seq_site_fini(env, &ofd->ofd_seq_site);
 614 }
 615
 616 int ofd_fid_init(const struct lu_env *env, struct ofd_device *ofd)
 617 {
 618         struct seq_server_site  *ss = &ofd->ofd_seq_site;
 619         struct lu_device        *lu = &ofd->ofd_dt_dev.dd_lu_dev;
 620         char                    *obd_name = ofd_name(ofd);
 621         char                    *name = NULL;
 622         int                     rc = 0;
 623
 624         ss = &ofd->ofd_seq_site;
 625         lu->ld_site->ld_seq_site = ss;
 626         ss->ss_lu = lu->ld_site;
 627         ss->ss_node_id = ofd->ofd_lut.lut_lsd.lsd_osd_index;
 628
 629         OBD_ALLOC_PTR(ss->ss_server_seq);
 630         if (ss->ss_server_seq == NULL)
 631                 GOTO(out_free, rc = -ENOMEM);
 632
 633         OBD_ALLOC(name, strlen(obd_name) + 10);
 634         if (!name) {
 635                 OBD_FREE_PTR(ss->ss_server_seq);
 636                 ss->ss_server_seq = NULL;
 637                 GOTO(out_free, rc = -ENOMEM);
 638         }
 639
 640         rc = seq_server_init(env, ss->ss_server_seq, ofd->ofd_osd, obd_name,
 641                              LUSTRE_SEQ_SERVER, ss);
 642         if (rc) {
 643                 CERROR("%s : seq server init error %d\n", obd_name, rc);
 644                 GOTO(out_free, rc);
 645         }
 646         ss->ss_server_seq->lss_space.lsr_index = ss->ss_node_id;
 647
 648         OBD_ALLOC_PTR(ss->ss_client_seq);
 649         if (ss->ss_client_seq == NULL)
 650                 GOTO(out_free, rc = -ENOMEM);
 651
 652         snprintf(name, strlen(obd_name) + 6, "%p-super", obd_name);
 653         rc = seq_client_init(ss->ss_client_seq, NULL, LUSTRE_SEQ_DATA,
 654                              name, NULL);
 655         if (rc) {
 656                 CERROR("%s : seq client init error %d\n", obd_name, rc);
 657                 GOTO(out_free, rc);
 658         }
 659         OBD_FREE(name, strlen(obd_name) + 10);
 660         name = NULL;
 661
 662         rc = seq_server_set_cli(env, ss->ss_server_seq, ss->ss_client_seq);
 663
 664 out_free:
 665         if (rc) {
 666                 if (ss->ss_server_seq) {
 667                         seq_server_fini(ss->ss_server_seq, env);
 668                         OBD_FREE_PTR(ss->ss_server_seq);
 669                         ss->ss_server_seq = NULL;
 670                 }
 671
 672                 if (ss->ss_client_seq) {
 673                         seq_client_fini(ss->ss_client_seq);
 674                         OBD_FREE_PTR(ss->ss_client_seq);
 675                         ss->ss_client_seq = NULL;
 676                 }
 677
 678                 if (name) {
 679                         OBD_FREE(name, strlen(obd_name) + 10);
 680                         name = NULL;
 681                 }
 682         }
 683
 684         return rc;
 685 }
 686
 687 int ofd_set_info_hdl(struct tgt_session_info *tsi)
 688 {
 689         struct ptlrpc_request   *req = tgt_ses_req(tsi);
 690         struct ost_body         *body = NULL, *repbody;
 691         void                    *key, *val = NULL;
 692         int                      keylen, vallen, rc = 0;
 693         bool                     is_grant_shrink;
 694         struct ofd_device       *ofd = ofd_exp(tsi->tsi_exp);
 695
 696         ENTRY;
 697
 698         key = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_KEY);
 699         if (key == NULL) {
 700                 DEBUG_REQ(D_HA, req, "no set_info key");
 701                 RETURN(err_serious(-EFAULT));
 702         }
 703         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_KEY,
 704                                       RCL_CLIENT);
 705
 706         val = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_VAL);
 707         if (val == NULL) {
 708                 DEBUG_REQ(D_HA, req, "no set_info val");
 709                 RETURN(err_serious(-EFAULT));
 710         }
 711         vallen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_VAL,
 712                                       RCL_CLIENT);
 713
 714         is_grant_shrink = KEY_IS(KEY_GRANT_SHRINK);
 715         if (is_grant_shrink)
 716                 /* In this case the value is actually an RMF_OST_BODY, so we
 717                  * transmutate the type of this PTLRPC */
 718                 req_capsule_extend(tsi->tsi_pill, &RQF_OST_SET_GRANT_INFO);
 719
 720         rc = req_capsule_server_pack(tsi->tsi_pill);
 721         if (rc < 0)
 722                 RETURN(rc);
 723
 724         if (is_grant_shrink) {
 725                 body = req_capsule_client_get(tsi->tsi_pill, &RMF_OST_BODY);
 726
 727                 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
 728                 *repbody = *body;
 729
 730                 /** handle grant shrink, similar to a read request */
 731                 ofd_grant_prepare_read(tsi->tsi_env, tsi->tsi_exp,
 732                                        &repbody->oa);
 733         } else if (KEY_IS(KEY_EVICT_BY_NID)) {
 734                 if (vallen > 0)
 735                         obd_export_evict_by_nid(tsi->tsi_exp->exp_obd, val);
 736                 rc = 0;
 737         } else if (KEY_IS(KEY_CAPA_KEY)) {
 738                 rc = ofd_update_capa_key(ofd, val);
 739         } else if (KEY_IS(KEY_SPTLRPC_CONF)) {
 740                 rc = tgt_adapt_sptlrpc_conf(tsi->tsi_tgt, 0);
 741         } else {
 742                 CERROR("%s: Unsupported key %s\n",
 743                        tgt_name(tsi->tsi_tgt), (char *)key);
 744                 rc = -EOPNOTSUPP;
 745         }
 746         ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_SET_INFO,
 747                          tsi->tsi_jobid, 1);
 748
 749         RETURN(rc);
 750 }
 751
 752 int ofd_fiemap_get(const struct lu_env *env, struct ofd_device *ofd,
 753                    struct lu_fid *fid, struct ll_user_fiemap *fiemap)
 754 {
 755         struct ofd_object       *fo;
 756         int                      rc;
 757
 758         fo = ofd_object_find(env, ofd, fid);
 759         if (IS_ERR(fo)) {
 760                 CERROR("%s: error finding object "DFID"\n",
 761                        ofd_name(ofd), PFID(fid));
 762                 return PTR_ERR(fo);
 763         }
 764
 765         ofd_read_lock(env, fo);
 766         if (ofd_object_exists(fo))
 767                 rc = dt_fiemap_get(env, ofd_object_child(fo), fiemap);
 768         else
 769                 rc = -ENOENT;
 770         ofd_read_unlock(env, fo);
 771         ofd_object_put(env, fo);
 772         return rc;
 773 }
 774
 775 struct locked_region {
 776         cfs_list_t              list;
 777         struct lustre_handle    lh;
 778 };
 779
 780 static int lock_region(struct ldlm_namespace *ns, struct ldlm_res_id *res_id,
 781                        unsigned long long begin, unsigned long long end,
 782                        cfs_list_t *locked)
 783 {
 784         struct locked_region    *region = NULL;
 785         __u64                    flags = 0;
 786         int                      rc;
 787
 788         LASSERT(begin <= end);
 789         OBD_ALLOC_PTR(region);
 790         if (region == NULL)
 791                 return -ENOMEM;
 792
 793         rc = tgt_extent_lock(ns, res_id, begin, end, &region->lh,
 794                              LCK_PR, &flags);
 795         if (rc != 0)
 796                 return rc;
 797
 798         CDEBUG(D_OTHER, "ost lock [%llu,%llu], lh=%p\n", begin, end,
 799                &region->lh);
 800         cfs_list_add(&region->list, locked);
 801
 802         return 0;
 803 }
 804
 805 static int lock_zero_regions(struct ldlm_namespace *ns,
 806                              struct ldlm_res_id *res_id,
 807                              struct ll_user_fiemap *fiemap,
 808                              cfs_list_t *locked)
 809 {
 810         __u64 begin = fiemap->fm_start;
 811         unsigned int i;
 812         int rc = 0;
 813         struct ll_fiemap_extent *fiemap_start = fiemap->fm_extents;
 814
 815         ENTRY;
 816
 817         CDEBUG(D_OTHER, "extents count %u\n", fiemap->fm_mapped_extents);
 818         for (i = 0; i < fiemap->fm_mapped_extents; i++) {
 819                 if (fiemap_start[i].fe_logical > begin) {
 820                         CDEBUG(D_OTHER, "ost lock [%llu,%llu]\n",
 821                                begin, fiemap_start[i].fe_logical);
 822                         rc = lock_region(ns, res_id, begin,
 823                                          fiemap_start[i].fe_logical, locked);
 824                         if (rc)
 825                                 RETURN(rc);
 826                 }
 827
 828                 begin = fiemap_start[i].fe_logical + fiemap_start[i].fe_length;
 829         }
 830
 831         if (begin < (fiemap->fm_start + fiemap->fm_length)) {
 832                 CDEBUG(D_OTHER, "ost lock [%llu,%llu]\n",
 833                        begin, fiemap->fm_start + fiemap->fm_length);
 834                 rc = lock_region(ns, res_id, begin,
 835                                  fiemap->fm_start + fiemap->fm_length, locked);
 836         }
 837
 838         RETURN(rc);
 839 }
 840
 841 static void unlock_zero_regions(struct ldlm_namespace *ns, cfs_list_t *locked)
 842 {
 843         struct locked_region *entry, *temp;
 844
 845         cfs_list_for_each_entry_safe(entry, temp, locked, list) {
 846                 CDEBUG(D_OTHER, "ost unlock lh=%p\n", &entry->lh);
 847                 tgt_extent_unlock(&entry->lh, LCK_PR);
 848                 cfs_list_del(&entry->list);
 849                 OBD_FREE_PTR(entry);
 850         }
 851 }
 852
 853 int ofd_get_info_hdl(struct tgt_session_info *tsi)
 854 {
 855         struct obd_export               *exp = tsi->tsi_exp;
 856         struct ofd_device               *ofd = ofd_exp(exp);
 857         struct ofd_thread_info          *fti = tsi2ofd_info(tsi);
 858         void                            *key;
 859         int                              keylen;
 860         int                              replylen, rc = 0;
 861
 862         ENTRY;
 863
 864         /* this common part for get_info rpc */
 865         key = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_KEY);
 866         if (key == NULL) {
 867                 DEBUG_REQ(D_HA, tgt_ses_req(tsi), "no get_info key");
 868                 RETURN(err_serious(-EPROTO));
 869         }
 870         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_GETINFO_KEY,
 871                                       RCL_CLIENT);
 872
 873         if (KEY_IS(KEY_LAST_ID)) {
 874                 obd_id          *last_id;
 875                 struct ofd_seq  *oseq;
 876
 877                 req_capsule_extend(tsi->tsi_pill, &RQF_OST_GET_INFO_LAST_ID);
 878                 rc = req_capsule_server_pack(tsi->tsi_pill);
 879                 if (rc)
 880                         RETURN(err_serious(rc));
 881
 882                 last_id = req_capsule_server_get(tsi->tsi_pill, &RMF_OBD_ID);
 883
 884                 oseq = ofd_seq_load(tsi->tsi_env, ofd,
 885                                     (obd_seq)exp->exp_filter_data.fed_group);
 886                 if (IS_ERR(oseq))
 887                         rc = -EFAULT;
 888                 else
 889                         *last_id = ofd_seq_last_oid(oseq);
 890                 ofd_seq_put(tsi->tsi_env, oseq);
 891         } else if (KEY_IS(KEY_FIEMAP)) {
 892                 struct ll_fiemap_info_key       *fm_key;
 893                 struct ll_user_fiemap           *fiemap;
 894                 struct lu_fid                   *fid;
 895
 896                 req_capsule_extend(tsi->tsi_pill, &RQF_OST_GET_INFO_FIEMAP);
 897
 898                 fm_key = req_capsule_client_get(tsi->tsi_pill, &RMF_FIEMAP_KEY);
 899                 rc = tgt_validate_obdo(tsi, &fm_key->oa);
 900                 if (rc)
 901                         RETURN(err_serious(rc));
 902
 903                 fid = &fm_key->oa.o_oi.oi_fid;
 904
 905                 CDEBUG(D_INODE, "get FIEMAP of object "DFID"\n", PFID(fid));
 906
 907                 replylen = fiemap_count_to_size(fm_key->fiemap.fm_extent_count);
 908                 req_capsule_set_size(tsi->tsi_pill, &RMF_FIEMAP_VAL,
 909                                      RCL_SERVER, replylen);
 910
 911                 rc = req_capsule_server_pack(tsi->tsi_pill);
 912                 if (rc)
 913                         RETURN(err_serious(rc));
 914
 915                 fiemap = req_capsule_server_get(tsi->tsi_pill, &RMF_FIEMAP_VAL);
 916                 if (fiemap == NULL)
 917                         RETURN(-ENOMEM);
 918
 919                 *fiemap = fm_key->fiemap;
 920                 rc = ofd_fiemap_get(tsi->tsi_env, ofd, fid, fiemap);
 921
 922                 /* LU-3219: Lock the sparse areas to make sure dirty
 923                  * flushed back from client, then call fiemap again. */
 924                 if (fm_key->oa.o_valid & OBD_MD_FLFLAGS &&
 925                     fm_key->oa.o_flags & OBD_FL_SRVLOCK) {
 926                         cfs_list_t locked = CFS_LIST_HEAD_INIT(locked);
 927
 928                         ost_fid_build_resid(fid, &fti->fti_resid);
 929                         rc = lock_zero_regions(ofd->ofd_namespace,
 930                                                &fti->fti_resid, fiemap,
 931                                                &locked);
 932                         if (rc == 0 && !cfs_list_empty(&locked)) {
 933                                 rc = ofd_fiemap_get(tsi->tsi_env, ofd, fid,
 934                                                     fiemap);
 935                                 unlock_zero_regions(ofd->ofd_namespace,
 936                                                     &locked);
 937                         }
 938                 }
 939         } else if (KEY_IS(KEY_LAST_FID)) {
 940                 struct ofd_device       *ofd = ofd_exp(exp);
 941                 struct ofd_seq          *oseq;
 942                 struct lu_fid           *fid;
 943                 int                      rc;
 944
 945                 req_capsule_extend(tsi->tsi_pill, &RQF_OST_GET_INFO_LAST_FID);
 946                 rc = req_capsule_server_pack(tsi->tsi_pill);
 947                 if (rc)
 948                         RETURN(err_serious(rc));
 949
 950                 fid = req_capsule_client_get(tsi->tsi_pill, &RMF_FID);
 951                 if (fid == NULL)
 952                         RETURN(err_serious(-EPROTO));
 953
 954                 fid_le_to_cpu(&fti->fti_ostid.oi_fid, fid);
 955
 956                 fid = req_capsule_server_get(tsi->tsi_pill, &RMF_FID);
 957                 if (fid == NULL)
 958                         RETURN(-ENOMEM);
 959
 960                 oseq = ofd_seq_load(tsi->tsi_env, ofd,
 961                                     ostid_seq(&fti->fti_ostid));
 962                 if (IS_ERR(oseq))
 963                         RETURN(PTR_ERR(oseq));
 964
 965                 rc = ostid_to_fid(fid, &oseq->os_oi,
 966                                   ofd->ofd_lut.lut_lsd.lsd_osd_index);
 967                 if (rc != 0)
 968                         GOTO(out_put, rc);
 969
 970                 CDEBUG(D_HA, "%s: LAST FID is "DFID"\n", ofd_name(ofd),
 971                        PFID(fid));
 972 out_put:
 973                 ofd_seq_put(tsi->tsi_env, oseq);
 974         } else {
 975                 CERROR("%s: not supported key %s\n", tgt_name(tsi->tsi_tgt),
 976                        (char *)key);
 977                 rc = -EOPNOTSUPP;
 978         }
 979         ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_GET_INFO,
 980                          tsi->tsi_jobid, 1);
 981
 982         RETURN(rc);
 983 }
 984
 985 static int ofd_getattr_hdl(struct tgt_session_info *tsi)
 986 {
 987         struct ofd_thread_info  *fti = tsi2ofd_info(tsi);
 988         struct ofd_device       *ofd = ofd_exp(tsi->tsi_exp);
 989         struct ost_body         *repbody;
 990         struct lustre_handle     lh = { 0 };
 991         struct ofd_object       *fo;
 992         __u64                    flags = 0;
 993         ldlm_mode_t              lock_mode = LCK_PR;
 994         bool                     srvlock;
 995         int                      rc;
 996         ENTRY;
 997
 998         LASSERT(tsi->tsi_ost_body != NULL);
 999
1000         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1001         if (repbody == NULL)
1002                 RETURN(-ENOMEM);
1003
1004         repbody->oa.o_oi = tsi->tsi_ost_body->oa.o_oi;
1005         repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1006
1007         srvlock = tsi->tsi_ost_body->oa.o_valid & OBD_MD_FLFLAGS &&
1008                   tsi->tsi_ost_body->oa.o_flags & OBD_FL_SRVLOCK;
1009
1010         if (srvlock) {
1011                 if (unlikely(tsi->tsi_ost_body->oa.o_flags & OBD_FL_FLUSH))
1012                         lock_mode = LCK_PW;
1013
1014                 rc = tgt_extent_lock(tsi->tsi_tgt->lut_obd->obd_namespace,
1015                                      &tsi->tsi_resid, 0, OBD_OBJECT_EOF, &lh,
1016                                      lock_mode, &flags);
1017                 if (rc != 0)
1018                         RETURN(rc);
1019         }
1020
1021         fo = ofd_object_find_exists(tsi->tsi_env, ofd, &tsi->tsi_fid);
1022         if (IS_ERR(fo))
1023                 GOTO(out, rc = PTR_ERR(fo));
1024
1025         rc = ofd_attr_get(tsi->tsi_env, fo, &fti->fti_attr);
1026         if (rc == 0) {
1027                 __u64    curr_version;
1028
1029                 obdo_from_la(&repbody->oa, &fti->fti_attr,
1030                              OFD_VALID_FLAGS | LA_UID | LA_GID);
1031                 tgt_drop_id(tsi->tsi_exp, &repbody->oa);
1032
1033                 /* Store object version in reply */
1034                 curr_version = dt_version_get(tsi->tsi_env,
1035                                               ofd_object_child(fo));
1036                 if ((__s64)curr_version != -EOPNOTSUPP) {
1037                         repbody->oa.o_valid |= OBD_MD_FLDATAVERSION;
1038                         repbody->oa.o_data_version = curr_version;
1039                 }
1040         }
1041
1042         ofd_object_put(tsi->tsi_env, fo);
1043 out:
1044         if (srvlock)
1045                 tgt_extent_unlock(&lh, lock_mode);
1046
1047         ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_GETATTR,
1048                          tsi->tsi_jobid, 1);
1049
1050         repbody->oa.o_valid |= OBD_MD_FLFLAGS;
1051         repbody->oa.o_flags = OBD_FL_FLUSH;
1052
1053         RETURN(rc);
1054 }
1055
1056 static int ofd_setattr_hdl(struct tgt_session_info *tsi)
1057 {
1058         struct ofd_thread_info  *fti = tsi2ofd_info(tsi);
1059         struct ofd_device       *ofd = ofd_exp(tsi->tsi_exp);
1060         struct ost_body         *body = tsi->tsi_ost_body;
1061         struct ost_body         *repbody;
1062         struct ldlm_resource    *res;
1063         struct ofd_object       *fo;
1064         struct filter_fid       *ff = NULL;
1065         int                      rc = 0;
1066
1067         ENTRY;
1068
1069         LASSERT(body != NULL);
1070
1071         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1072         if (repbody == NULL)
1073                 RETURN(-ENOMEM);
1074
1075         repbody->oa.o_oi = body->oa.o_oi;
1076         repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1077
1078         /* This would be very bad - accidentally truncating a file when
1079          * changing the time or similar - bug 12203. */
1080         if (body->oa.o_valid & OBD_MD_FLSIZE &&
1081             body->oa.o_size != OBD_OBJECT_EOF) {
1082                 static char mdsinum[48];
1083
1084                 if (body->oa.o_valid & OBD_MD_FLFID)
1085                         snprintf(mdsinum, sizeof(mdsinum) - 1,
1086                                  "of parent "DFID, body->oa.o_parent_seq,
1087                                  body->oa.o_parent_oid, 0);
1088                 else
1089                         mdsinum[0] = '\0';
1090
1091                 CERROR("%s: setattr from %s is trying to truncate object "DFID
1092                        " %s\n", ofd_name(ofd), obd_export_nid2str(tsi->tsi_exp),
1093                        PFID(&tsi->tsi_fid), mdsinum);
1094                 RETURN(-EPERM);
1095         }
1096
1097         fo = ofd_object_find_exists(tsi->tsi_env, ofd, &tsi->tsi_fid);
1098         if (IS_ERR(fo))
1099                 GOTO(out, rc = PTR_ERR(fo));
1100
1101         la_from_obdo(&fti->fti_attr, &body->oa, body->oa.o_valid);
1102         fti->fti_attr.la_valid &= ~LA_TYPE;
1103
1104         if (body->oa.o_valid & OBD_MD_FLFID) {
1105                 ff = &fti->fti_mds_fid;
1106                 ofd_prepare_fidea(ff, &body->oa);
1107         }
1108
1109         /* setting objects attributes (including owner/group) */
1110         rc = ofd_attr_set(tsi->tsi_env, fo, &fti->fti_attr, ff);
1111         if (rc != 0)
1112                 GOTO(out_put, rc);
1113
1114         obdo_from_la(&repbody->oa, &fti->fti_attr,
1115                      OFD_VALID_FLAGS | LA_UID | LA_GID);
1116         tgt_drop_id(tsi->tsi_exp, &repbody->oa);
1117
1118         ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_SETATTR,
1119                          tsi->tsi_jobid, 1);
1120         EXIT;
1121 out_put:
1122         ofd_object_put(tsi->tsi_env, fo);
1123 out:
1124         if (rc == 0) {
1125                 /* we do not call this before to avoid lu_object_find() in
1126                  *  ->lvbo_update() holding another reference on the object.
1127                  * otherwise concurrent destroy can make the object unavailable
1128                  * for 2nd lu_object_find() waiting for the first reference
1129                  * to go... deadlock! */
1130                 res = ldlm_resource_get(ofd->ofd_namespace, NULL,
1131                                         &tsi->tsi_resid, LDLM_EXTENT, 0);
1132                 if (res != NULL) {
1133                         ldlm_res_lvbo_update(res, NULL, 0);
1134                         ldlm_resource_putref(res);
1135                 }
1136         }
1137         return rc;
1138 }
1139
1140 static int ofd_orphans_destroy(const struct lu_env *env,
1141                                struct obd_export *exp,
1142                                struct ofd_device *ofd, struct obdo *oa)
1143 {
1144         struct ofd_thread_info  *info   = ofd_info(env);
1145         struct lu_fid           *fid    = &info->fti_fid;
1146         struct ost_id           *oi     = &oa->o_oi;
1147         struct ofd_seq          *oseq;
1148         obd_seq                  seq    = ostid_seq(oi);
1149         obd_id                   end_id = ostid_id(oi);
1150         obd_id                   last;
1151         obd_id                   oid;
1152         int                      skip_orphan;
1153         int                      rc     = 0;
1154
1155         ENTRY;
1156
1157         oseq = ofd_seq_get(ofd, seq);
1158         if (oseq == NULL) {
1159                 CERROR("%s: Can not find seq for "DOSTID"\n",
1160                        ofd_name(ofd), POSTID(oi));
1161                 RETURN(-EINVAL);
1162         }
1163
1164         *fid = oi->oi_fid;
1165         last = ofd_seq_last_oid(oseq);
1166         oid = last;
1167
1168         LASSERT(exp != NULL);
1169         skip_orphan = !!(exp_connect_flags(exp) & OBD_CONNECT_SKIP_ORPHAN);
1170
1171         LCONSOLE(D_INFO, "%s: deleting orphan objects from "DOSTID
1172                  " to "DOSTID"\n", ofd_name(ofd), seq, end_id + 1, seq, last);
1173
1174         while (oid > end_id) {
1175                 rc = fid_set_id(fid, oid);
1176                 if (unlikely(rc != 0))
1177                         GOTO(out_put, rc);
1178
1179                 rc = ofd_destroy_by_fid(env, ofd, fid, 1);
1180                 if (rc != 0 && rc != -ENOENT) /* this is pretty fatal... */
1181                         CEMERG("%s: error destroying precreated id "DFID
1182                                ": rc = %d\n", ofd_name(ofd), PFID(fid), rc);
1183
1184                 oid--;
1185                 if (!skip_orphan) {
1186                         ofd_seq_last_oid_set(oseq, oid);
1187                         /* update last_id on disk periodically so that if we
1188                          * restart * we don't need to re-scan all of the just
1189                          * deleted objects. */
1190                         if ((oid & 511) == 0)
1191                                 ofd_seq_last_oid_write(env, ofd, oseq);
1192                 }
1193         }
1194
1195         CDEBUG(D_HA, "%s: after destroy: set last_id to "DOSTID"\n",
1196                ofd_name(ofd), seq, oid);
1197
1198         if (!skip_orphan) {
1199                 rc = ofd_seq_last_oid_write(env, ofd, oseq);
1200         } else {
1201                 /* don't reuse orphan object, return last used objid */
1202                 ostid_set_id(oi, last);
1203                 rc = 0;
1204         }
1205
1206         GOTO(out_put, rc);
1207
1208 out_put:
1209         ofd_seq_put(env, oseq);
1210         return rc;
1211 }
1212
1213 static int ofd_create_hdl(struct tgt_session_info *tsi)
1214 {
1215         struct ost_body         *repbody;
1216         const struct obdo       *oa = &tsi->tsi_ost_body->oa;
1217         struct obdo             *rep_oa;
1218         struct obd_export       *exp = tsi->tsi_exp;
1219         struct ofd_device       *ofd = ofd_exp(exp);
1220         obd_seq                  seq = ostid_seq(&oa->o_oi);
1221         obd_id                   oid = ostid_id(&oa->o_oi);
1222         struct ofd_seq          *oseq;
1223         int                      rc = 0, diff;
1224         int                      sync_trans = 0;
1225
1226         ENTRY;
1227
1228         if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
1229                 RETURN(-EROFS);
1230
1231         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1232         if (repbody == NULL)
1233                 RETURN(-ENOMEM);
1234
1235         down_read(&ofd->ofd_lastid_rwsem);
1236         /* Currently, for safe, we do not distinguish which LAST_ID is broken,
1237          * we may do that in the future.
1238          * Return -ENOSPC until the LAST_ID rebuilt. */
1239         if (unlikely(ofd->ofd_lastid_rebuilding))
1240                 GOTO(out_sem, rc = -ENOSPC);
1241
1242         rep_oa = &repbody->oa;
1243         rep_oa->o_oi = oa->o_oi;
1244
1245         LASSERT(seq >= FID_SEQ_OST_MDT0);
1246         LASSERT(oa->o_valid & OBD_MD_FLGROUP);
1247
1248         CDEBUG(D_INFO, "ofd_create("DOSTID")\n", POSTID(&oa->o_oi));
1249
1250         oseq = ofd_seq_load(tsi->tsi_env, ofd, seq);
1251         if (IS_ERR(oseq)) {
1252                 CERROR("%s: Can't find FID Sequence "LPX64": rc = %ld\n",
1253                        ofd_name(ofd), seq, PTR_ERR(oseq));
1254                 GOTO(out_sem, rc = -EINVAL);
1255         }
1256
1257         if ((oa->o_valid & OBD_MD_FLFLAGS) &&
1258             (oa->o_flags & OBD_FL_RECREATE_OBJS)) {
1259                 if (!ofd_obd(ofd)->obd_recovering ||
1260                     oid > ofd_seq_last_oid(oseq)) {
1261                         CERROR("%s: recreate objid "DOSTID" > last id "LPU64
1262                                "\n", ofd_name(ofd), POSTID(&oa->o_oi),
1263                                ofd_seq_last_oid(oseq));
1264                         GOTO(out_nolock, rc = -EINVAL);
1265                 }
1266                 /* Do nothing here, we re-create objects during recovery
1267                  * upon write replay, see ofd_preprw_write() */
1268                 GOTO(out_nolock, rc = 0);
1269         }
1270         /* former ofd_handle_precreate */
1271         if ((oa->o_valid & OBD_MD_FLFLAGS) &&
1272             (oa->o_flags & OBD_FL_DELORPHAN)) {
1273                 exp->exp_filter_data.fed_lastid_gen = ofd->ofd_lastid_gen;
1274
1275                 /* destroy orphans */
1276                 if (lustre_msg_get_conn_cnt(tgt_ses_req(tsi)->rq_reqmsg) <
1277                     exp->exp_conn_cnt) {
1278                         CERROR("%s: dropping old orphan cleanup request\n",
1279                                ofd_name(ofd));
1280                         GOTO(out_nolock, rc = 0);
1281                 }
1282                 /* This causes inflight precreates to abort and drop lock */
1283                 oseq->os_destroys_in_progress = 1;
1284                 mutex_lock(&oseq->os_create_lock);
1285                 if (!oseq->os_destroys_in_progress) {
1286                         CERROR("%s:["LPU64"] destroys_in_progress already"
1287                                " cleared\n", ofd_name(ofd), seq);
1288                         ostid_set_id(&rep_oa->o_oi, ofd_seq_last_oid(oseq));
1289                         GOTO(out, rc = 0);
1290                 }
1291                 diff = oid - ofd_seq_last_oid(oseq);
1292                 CDEBUG(D_HA, "ofd_last_id() = "LPU64" -> diff = %d\n",
1293                         ofd_seq_last_oid(oseq), diff);
1294                 if (-diff > OST_MAX_PRECREATE) {
1295                         /* FIXME: should reset precreate_next_id on MDS */
1296                         rc = 0;
1297                 } else if (diff < 0) {
1298                         rc = ofd_orphans_destroy(tsi->tsi_env, exp,
1299                                                  ofd, rep_oa);
1300                         oseq->os_destroys_in_progress = 0;
1301                 } else {
1302                         /* XXX: Used by MDS for the first time! */
1303                         oseq->os_destroys_in_progress = 0;
1304                 }
1305         } else {
1306                 if (unlikely(exp->exp_filter_data.fed_lastid_gen !=
1307                              ofd->ofd_lastid_gen)) {
1308                         ofd_obd_disconnect(exp);
1309                         GOTO(out_nolock, rc = -ENOTCONN);
1310                 }
1311
1312                 mutex_lock(&oseq->os_create_lock);
1313                 if (lustre_msg_get_conn_cnt(tgt_ses_req(tsi)->rq_reqmsg) <
1314                     exp->exp_conn_cnt) {
1315                         CERROR("%s: dropping old precreate request\n",
1316                                ofd_name(ofd));
1317                         GOTO(out, rc = 0);
1318                 }
1319                 /* only precreate if seq is 0, IDIF or normal and also o_id
1320                  * must be specfied */
1321                 if ((!fid_seq_is_mdt(seq) && !fid_seq_is_norm(seq) &&
1322                      !fid_seq_is_idif(seq)) || oid == 0) {
1323                         diff = 1; /* shouldn't we create this right now? */
1324                 } else {
1325                         diff = oid - ofd_seq_last_oid(oseq);
1326                         /* Do sync create if the seq is about to used up */
1327                         if (fid_seq_is_idif(seq) || fid_seq_is_mdt0(seq)) {
1328                                 if (unlikely(oid >= IDIF_MAX_OID - 1))
1329                                         sync_trans = 1;
1330                         } else if (fid_seq_is_norm(seq)) {
1331                                 if (unlikely(oid >=
1332                                              LUSTRE_DATA_SEQ_MAX_WIDTH - 1))
1333                                         sync_trans = 1;
1334                         } else {
1335                                 CERROR("%s : invalid o_seq "DOSTID"\n",
1336                                        ofd_name(ofd), POSTID(&oa->o_oi));
1337                                 GOTO(out, rc = -EINVAL);
1338                         }
1339                 }
1340         }
1341         if (diff > 0) {
1342                 cfs_time_t       enough_time = cfs_time_shift(DISK_TIMEOUT);
1343                 obd_id           next_id;
1344                 int              created = 0;
1345                 int              count;
1346
1347                 if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
1348                     !(oa->o_flags & OBD_FL_DELORPHAN)) {
1349                         /* don't enforce grant during orphan recovery */
1350                         rc = ofd_grant_create(tsi->tsi_env,
1351                                               ofd_obd(ofd)->obd_self_export,
1352                                               &diff);
1353                         if (rc) {
1354                                 CDEBUG(D_HA, "%s: failed to acquire grant "
1355                                        "space for precreate (%d): rc = %d\n",
1356                                        ofd_name(ofd), diff, rc);
1357                                 diff = 0;
1358                         }
1359                 }
1360
1361                 /* This can happen if a new OST is formatted and installed
1362                  * in place of an old one at the same index.  Instead of
1363                  * precreating potentially millions of deleted old objects
1364                  * (possibly filling the OST), only precreate the last batch.
1365                  * LFSCK will eventually clean up any orphans. LU-14 */
1366                 if (diff > 5 * OST_MAX_PRECREATE) {
1367                         diff = OST_MAX_PRECREATE / 2;
1368                         LCONSOLE_WARN("%s: precreate FID "DOSTID" is over %u "
1369                                       "larger than the LAST_ID "DOSTID", only "
1370                                       "precreating the last %u objects.\n",
1371                                       ofd_name(ofd), POSTID(&oa->o_oi),
1372                                       5 * OST_MAX_PRECREATE,
1373                                       POSTID(&oseq->os_oi), diff);
1374                         ofd_seq_last_oid_set(oseq, ostid_id(&oa->o_oi) - diff);
1375                 }
1376
1377                 while (diff > 0) {
1378                         next_id = ofd_seq_last_oid(oseq) + 1;
1379                         count = ofd_precreate_batch(ofd, diff);
1380
1381                         CDEBUG(D_HA, "%s: reserve %d objects in group "LPX64
1382                                " at "LPU64"\n", ofd_name(ofd),
1383                                count, seq, next_id);
1384
1385                         if (cfs_time_after(jiffies, enough_time)) {
1386                                 LCONSOLE_WARN("%s: Slow creates, %d/%d objects"
1387                                               " created at a rate of %d/s\n",
1388                                               ofd_name(ofd), created,
1389                                               diff + created,
1390                                               created / DISK_TIMEOUT);
1391                                 break;
1392                         }
1393
1394                         rc = ofd_precreate_objects(tsi->tsi_env, ofd, next_id,
1395                                                    oseq, count, sync_trans);
1396                         if (rc > 0) {
1397                                 created += rc;
1398                                 diff -= rc;
1399                         } else if (rc < 0) {
1400                                 break;
1401                         }
1402                 }
1403                 if (created > 0)
1404                         /* some objects got created, we can return
1405                          * them, even if last creation failed */
1406                         rc = 0;
1407                 else
1408                         CERROR("%s: unable to precreate: rc = %d\n",
1409                                ofd_name(ofd), rc);
1410
1411                 if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
1412                     !(oa->o_flags & OBD_FL_DELORPHAN))
1413                         ofd_grant_commit(tsi->tsi_env,
1414                                          ofd_obd(ofd)->obd_self_export, rc);
1415
1416                 ostid_set_id(&rep_oa->o_oi, ofd_seq_last_oid(oseq));
1417         }
1418         EXIT;
1419         ofd_counter_incr(exp, LPROC_OFD_STATS_CREATE,
1420                          tsi->tsi_jobid, 1);
1421 out:
1422         mutex_unlock(&oseq->os_create_lock);
1423 out_nolock:
1424         if (rc == 0)
1425                 rep_oa->o_valid |= OBD_MD_FLID | OBD_MD_FLGROUP;
1426
1427         ofd_seq_put(tsi->tsi_env, oseq);
1428
1429 out_sem:
1430         up_read(&ofd->ofd_lastid_rwsem);
1431         return rc;
1432 }
1433
1434 static int ofd_destroy_hdl(struct tgt_session_info *tsi)
1435 {
1436         const struct ost_body   *body = tsi->tsi_ost_body;
1437         struct ost_body         *repbody;
1438         struct ofd_device       *ofd = ofd_exp(tsi->tsi_exp);
1439         struct ofd_thread_info  *fti = tsi2ofd_info(tsi);
1440         struct lu_fid           *fid = &fti->fti_fid;
1441         obd_id                   oid;
1442         obd_count                count;
1443         int                      rc = 0;
1444
1445         ENTRY;
1446
1447         if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
1448                 RETURN(-EROFS);
1449
1450         /* This is old case for clients before Lustre 2.4 */
1451         /* If there's a DLM request, cancel the locks mentioned in it */
1452         if (req_capsule_field_present(tsi->tsi_pill, &RMF_DLM_REQ,
1453                                       RCL_CLIENT)) {
1454                 struct ldlm_request *dlm;
1455
1456                 dlm = req_capsule_client_get(tsi->tsi_pill, &RMF_DLM_REQ);
1457                 if (dlm == NULL)
1458                         RETURN(-EFAULT);
1459                 ldlm_request_cancel(tgt_ses_req(tsi), dlm, 0);
1460         }
1461
1462         *fid = body->oa.o_oi.oi_fid;
1463         oid = ostid_id(&body->oa.o_oi);
1464         LASSERT(oid != 0);
1465
1466         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1467
1468         /* check that o_misc makes sense */
1469         if (body->oa.o_valid & OBD_MD_FLOBJCOUNT)
1470                 count = body->oa.o_misc;
1471         else
1472                 count = 1; /* default case - single destroy */
1473
1474         CDEBUG(D_HA, "%s: Destroy object "DOSTID" count %d\n", ofd_name(ofd),
1475                POSTID(&body->oa.o_oi), count);
1476
1477         while (count > 0) {
1478                 int lrc;
1479
1480                 lrc = ofd_destroy_by_fid(tsi->tsi_env, ofd, fid, 0);
1481                 if (lrc == -ENOENT) {
1482                         CDEBUG(D_INODE,
1483                                "%s: destroying non-existent object "DFID"\n",
1484                                ofd_name(ofd), PFID(fid));
1485                         /* rewrite rc with -ENOENT only if it is 0 */
1486                         if (rc == 0)
1487                                 rc = lrc;
1488                 } else if (lrc != 0) {
1489                         CERROR("%s: error destroying object "DFID": %d\n",
1490                                ofd_name(ofd), PFID(fid), lrc);
1491                         rc = lrc;
1492                 }
1493
1494                 count--;
1495                 oid++;
1496                 lrc = fid_set_id(fid, oid);
1497                 if (unlikely(lrc != 0 && count > 0))
1498                         GOTO(out, rc = lrc);
1499         }
1500
1501         ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_DESTROY,
1502                          tsi->tsi_jobid, 1);
1503
1504         GOTO(out, rc);
1505
1506 out:
1507         fid_to_ostid(fid, &repbody->oa.o_oi);
1508         return rc;
1509 }
1510
1511 static int ofd_statfs_hdl(struct tgt_session_info *tsi)
1512 {
1513         struct obd_statfs       *osfs;
1514         int                      rc;
1515
1516         ENTRY;
1517
1518         osfs = req_capsule_server_get(tsi->tsi_pill, &RMF_OBD_STATFS);
1519
1520         rc = ofd_statfs(tsi->tsi_env, tsi->tsi_exp, osfs,
1521                         cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS), 0);
1522         if (rc != 0)
1523                 CERROR("%s: statfs failed: rc = %d\n",
1524                        tgt_name(tsi->tsi_tgt), rc);
1525
1526         if (OBD_FAIL_CHECK(OBD_FAIL_OST_STATFS_EINPROGRESS))
1527                 rc = -EINPROGRESS;
1528
1529         ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_STATFS,
1530                          tsi->tsi_jobid, 1);
1531
1532         RETURN(rc);
1533 }
1534
1535 static int ofd_sync_hdl(struct tgt_session_info *tsi)
1536 {
1537         struct ost_body         *body = tsi->tsi_ost_body;
1538         struct ost_body         *repbody;
1539         struct ofd_thread_info  *fti = tsi2ofd_info(tsi);
1540         struct ofd_device       *ofd = ofd_exp(tsi->tsi_exp);
1541         struct ofd_object       *fo = NULL;
1542         int                      rc = 0;
1543
1544         ENTRY;
1545
1546         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1547
1548         /* if no objid is specified, it means "sync whole filesystem" */
1549         if (!fid_is_zero(&tsi->tsi_fid)) {
1550                 fo = ofd_object_find_exists(tsi->tsi_env, ofd, &tsi->tsi_fid);
1551                 if (IS_ERR(fo))
1552                         RETURN(PTR_ERR(fo));
1553         }
1554
1555         rc = tgt_sync(tsi->tsi_env, tsi->tsi_tgt,
1556                       fo != NULL ? ofd_object_child(fo) : NULL);
1557         if (rc)
1558                 GOTO(put, rc);
1559
1560         ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_SYNC,
1561                          tsi->tsi_jobid, 1);
1562         if (fo == NULL)
1563                 RETURN(0);
1564
1565         repbody->oa.o_oi = body->oa.o_oi;
1566         repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1567
1568         rc = ofd_attr_get(tsi->tsi_env, fo, &fti->fti_attr);
1569         if (rc == 0)
1570                 obdo_from_la(&repbody->oa, &fti->fti_attr,
1571                              OFD_VALID_FLAGS);
1572         else
1573                 /* don't return rc from getattr */
1574                 rc = 0;
1575         EXIT;
1576 put:
1577         if (fo != NULL)
1578                 ofd_object_put(tsi->tsi_env, fo);
1579         return rc;
1580 }
1581
1582 static int ofd_punch_hdl(struct tgt_session_info *tsi)
1583 {
1584         const struct obdo       *oa = &tsi->tsi_ost_body->oa;
1585         struct ost_body         *repbody;
1586         struct ofd_thread_info  *info = tsi2ofd_info(tsi);
1587         struct ldlm_namespace   *ns = tsi->tsi_tgt->lut_obd->obd_namespace;
1588         struct ldlm_resource    *res;
1589         struct ofd_object       *fo;
1590         struct filter_fid       *ff = NULL;
1591         __u64                    flags = 0;
1592         struct lustre_handle     lh = { 0, };
1593         int                      rc;
1594         __u64                    start, end;
1595         bool                     srvlock;
1596
1597         ENTRY;
1598
1599         /* check that we do support OBD_CONNECT_TRUNCLOCK. */
1600         CLASSERT(OST_CONNECT_SUPPORTED & OBD_CONNECT_TRUNCLOCK);
1601
1602         if ((oa->o_valid & (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS)) !=
1603             (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS))
1604                 RETURN(err_serious(-EPROTO));
1605
1606         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1607         if (repbody == NULL)
1608                 RETURN(err_serious(-ENOMEM));
1609
1610         /* punch start,end are passed in o_size,o_blocks throught wire */
1611         start = oa->o_size;
1612         end = oa->o_blocks;
1613
1614         if (end != OBD_OBJECT_EOF) /* Only truncate is supported */
1615                 RETURN(-EPROTO);
1616
1617         /* standard truncate optimization: if file body is completely
1618          * destroyed, don't send data back to the server. */
1619         if (start == 0)
1620                 flags |= LDLM_FL_AST_DISCARD_DATA;
1621
1622         repbody->oa.o_oi = oa->o_oi;
1623         repbody->oa.o_valid = OBD_MD_FLID;
1624
1625         srvlock = oa->o_valid & OBD_MD_FLFLAGS &&
1626                   oa->o_flags & OBD_FL_SRVLOCK;
1627
1628         if (srvlock) {
1629                 rc = tgt_extent_lock(ns, &tsi->tsi_resid, start, end, &lh,
1630                                      LCK_PW, &flags);
1631                 if (rc != 0)
1632                         RETURN(rc);
1633         }
1634
1635         CDEBUG(D_INODE, "calling punch for object "DFID", valid = "LPX64
1636                ", start = "LPD64", end = "LPD64"\n", PFID(&tsi->tsi_fid),
1637                oa->o_valid, start, end);
1638
1639         fo = ofd_object_find_exists(tsi->tsi_env, ofd_exp(tsi->tsi_exp),
1640                                     &tsi->tsi_fid);
1641         if (IS_ERR(fo))
1642                 GOTO(out, rc = PTR_ERR(fo));
1643
1644         la_from_obdo(&info->fti_attr, oa,
1645                      OBD_MD_FLMTIME | OBD_MD_FLATIME | OBD_MD_FLCTIME);
1646         info->fti_attr.la_size = start;
1647         info->fti_attr.la_valid |= LA_SIZE;
1648
1649         if (oa->o_valid & OBD_MD_FLFID) {
1650                 ff = &info->fti_mds_fid;
1651                 ofd_prepare_fidea(ff, oa);
1652         }
1653
1654         rc = ofd_object_punch(tsi->tsi_env, fo, start, end, &info->fti_attr,
1655                               ff);
1656         if (rc)
1657                 GOTO(out_put, rc);
1658
1659         ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_PUNCH,
1660                          tsi->tsi_jobid, 1);
1661         EXIT;
1662 out_put:
1663         ofd_object_put(tsi->tsi_env, fo);
1664 out:
1665         if (srvlock)
1666                 tgt_extent_unlock(&lh, LCK_PW);
1667         if (rc == 0) {
1668                 /* we do not call this before to avoid lu_object_find() in
1669                  *  ->lvbo_update() holding another reference on the object.
1670                  * otherwise concurrent destroy can make the object unavailable
1671                  * for 2nd lu_object_find() waiting for the first reference
1672                  * to go... deadlock! */
1673                 res = ldlm_resource_get(ns, NULL, &tsi->tsi_resid,
1674                                         LDLM_EXTENT, 0);
1675                 if (res != NULL) {
1676                         ldlm_res_lvbo_update(res, NULL, 0);
1677                         ldlm_resource_putref(res);
1678                 }
1679         }
1680         return rc;
1681 }
1682
1683 static int ofd_quotactl(struct tgt_session_info *tsi)
1684 {
1685         struct obd_quotactl     *oqctl, *repoqc;
1686         int                      rc;
1687
1688         ENTRY;
1689
1690         oqctl = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_QUOTACTL);
1691         if (oqctl == NULL)
1692                 RETURN(err_serious(-EPROTO));
1693
1694         repoqc = req_capsule_server_get(tsi->tsi_pill, &RMF_OBD_QUOTACTL);
1695         if (repoqc == NULL)
1696                 RETURN(err_serious(-ENOMEM));
1697
1698         /* report success for quota on/off for interoperability with current MDT
1699          * stack */
1700         if (oqctl->qc_cmd == Q_QUOTAON || oqctl->qc_cmd == Q_QUOTAOFF)
1701                 RETURN(0);
1702
1703         *repoqc = *oqctl;
1704         rc = lquotactl_slv(tsi->tsi_env, tsi->tsi_tgt->lut_bottom, repoqc);
1705
1706         ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_QUOTACTL,
1707                          tsi->tsi_jobid, 1);
1708
1709         RETURN(rc);
1710 }
1711
1712 /* High priority request handlers for OFD */
1713
1714 /* prolong locks for the current service time of the corresponding
1715  * portal (= OST_IO_PORTAL)
1716  */
1717 static inline int prolong_timeout(struct ptlrpc_request *req)
1718 {
1719         struct ptlrpc_service_part *svcpt = req->rq_rqbd->rqbd_svcpt;
1720
1721         if (AT_OFF)
1722                 return obd_timeout / 2;
1723
1724         return max(at_est2timeout(at_get(&svcpt->scp_at_estimate)),
1725                    ldlm_timeout);
1726 }
1727
1728 static int ofd_prolong_one_lock(struct tgt_session_info *tsi,
1729                                 struct ldlm_lock *lock,
1730                                 struct ldlm_extent *extent, int timeout)
1731 {
1732
1733         if (lock->l_flags & LDLM_FL_DESTROYED) /* lock already cancelled */
1734                 return 0;
1735
1736         /* XXX: never try to grab resource lock here because we're inside
1737          * exp_bl_list_lock; in ldlm_lockd.c to handle waiting list we take
1738          * res lock and then exp_bl_list_lock. */
1739
1740         if (!(lock->l_flags & LDLM_FL_AST_SENT))
1741                 /* ignore locks not being cancelled */
1742                 return 0;
1743
1744         LDLM_DEBUG(lock, "refreshed for req x"LPU64" ext("LPU64"->"LPU64") "
1745                          "to %ds.\n", tgt_ses_req(tsi)->rq_xid, extent->start,
1746                          extent->end, timeout);
1747
1748         /* OK. this is a possible lock the user holds doing I/O
1749          * let's refresh eviction timer for it */
1750         ldlm_refresh_waiting_lock(lock, timeout);
1751         return 1;
1752 }
1753
1754 static int ofd_prolong_extent_locks(struct tgt_session_info *tsi,
1755                                     __u64 start, __u64 end)
1756 {
1757         struct obd_export       *exp = tsi->tsi_exp;
1758         struct obdo             *oa  = &tsi->tsi_ost_body->oa;
1759         struct ldlm_extent       extent = {
1760                 .start = start,
1761                 .end = end
1762         };
1763         struct ldlm_lock        *lock;
1764         int                      timeout = prolong_timeout(tgt_ses_req(tsi));
1765         int                      lock_count = 0;
1766
1767         ENTRY;
1768
1769         if (oa->o_valid & OBD_MD_FLHANDLE) {
1770                 /* mostly a request should be covered by only one lock, try
1771                  * fast path. */
1772                 lock = ldlm_handle2lock(&oa->o_handle);
1773                 if (lock != NULL) {
1774                         /* Fast path to check if the lock covers the whole IO
1775                          * region exclusively. */
1776                         if (lock->l_granted_mode == LCK_PW &&
1777                             ldlm_extent_contain(&lock->l_policy_data.l_extent,
1778                                                 &extent)) {
1779                                 /* bingo */
1780                                 LASSERT(lock->l_export == exp);
1781                                 lock_count = ofd_prolong_one_lock(tsi, lock,
1782                                                              &extent, timeout);
1783                                 LDLM_LOCK_PUT(lock);
1784                                 RETURN(lock_count);
1785                         }
1786                         LDLM_LOCK_PUT(lock);
1787                 }
1788         }
1789
1790         spin_lock_bh(&exp->exp_bl_list_lock);
1791         list_for_each_entry(lock, &exp->exp_bl_list, l_exp_list) {
1792                 LASSERT(lock->l_flags & LDLM_FL_AST_SENT);
1793                 LASSERT(lock->l_resource->lr_type == LDLM_EXTENT);
1794
1795                 if (!ldlm_res_eq(&tsi->tsi_resid, &lock->l_resource->lr_name))
1796                         continue;
1797
1798                 if (!ldlm_extent_overlap(&lock->l_policy_data.l_extent,
1799                                          &extent))
1800                         continue;
1801
1802                 lock_count += ofd_prolong_one_lock(tsi, lock, &extent, timeout);
1803         }
1804         spin_unlock_bh(&exp->exp_bl_list_lock);
1805
1806         RETURN(lock_count);
1807 }
1808
1809 /**
1810  * Returns 1 if the given PTLRPC matches the given LDLM lock, or 0 if it does
1811  * not.
1812  */
1813 static int ofd_rw_hpreq_lock_match(struct ptlrpc_request *req,
1814                                    struct ldlm_lock *lock)
1815 {
1816         struct niobuf_remote    *rnb;
1817         struct obd_ioobj        *ioo;
1818         ldlm_mode_t              mode;
1819         struct ldlm_extent       ext;
1820         __u32                    opc = lustre_msg_get_opc(req->rq_reqmsg);
1821
1822         ENTRY;
1823
1824         ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
1825         LASSERT(ioo != NULL);
1826
1827         rnb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
1828         LASSERT(rnb != NULL);
1829
1830         ext.start = rnb->offset;
1831         rnb += ioo->ioo_bufcnt - 1;
1832         ext.end = rnb->offset + rnb->len - 1;
1833
1834         LASSERT(lock->l_resource != NULL);
1835         if (!ostid_res_name_eq(&ioo->ioo_oid, &lock->l_resource->lr_name))
1836                 RETURN(0);
1837
1838         mode = LCK_PW;
1839         if (opc == OST_READ)
1840                 mode |= LCK_PR;
1841
1842         if (!(lock->l_granted_mode & mode))
1843                 RETURN(0);
1844
1845         RETURN(ldlm_extent_overlap(&lock->l_policy_data.l_extent, &ext));
1846 }
1847
1848 /**
1849  * High-priority queue request check for whether the given PTLRPC request
1850  * (\a req) is blocking an LDLM lock cancel.
1851  *
1852  * Returns 1 if the given given PTLRPC request (\a req) is blocking an LDLM lock
1853  * cancel, 0 if it is not, and -EFAULT if the request is malformed.
1854  *
1855  * Only OST_READs, OST_WRITEs and OST_PUNCHes go on the h-p RPC queue.  This
1856  * function looks only at OST_READs and OST_WRITEs.
1857  */
1858 static int ofd_rw_hpreq_check(struct ptlrpc_request *req)
1859 {
1860         struct tgt_session_info *tsi;
1861         struct obd_ioobj        *ioo;
1862         struct niobuf_remote    *rnb;
1863         __u64                    start, end;
1864         int                      lock_count;
1865
1866         ENTRY;
1867
1868         /* Don't use tgt_ses_info() to get session info, because lock_match()
1869          * can be called while request has no processing thread yet. */
1870         tsi = lu_context_key_get(&req->rq_session, &tgt_session_key);
1871         LASSERT(tsi != NULL);
1872
1873         /*
1874          * Use LASSERT below because malformed RPCs should have
1875          * been filtered out in tgt_hpreq_handler().
1876          */
1877         ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
1878         LASSERT(ioo != NULL);
1879
1880         rnb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
1881         LASSERT(rnb != NULL);
1882         LASSERT(!(rnb->flags & OBD_BRW_SRVLOCK));
1883
1884         start = rnb->offset;
1885         rnb += ioo->ioo_bufcnt - 1;
1886         end = rnb->offset + rnb->len - 1;
1887
1888         DEBUG_REQ(D_RPCTRACE, req, "%s %s: refresh rw locks: "DFID
1889                                    " ("LPU64"->"LPU64")\n",
1890                   tgt_name(tsi->tsi_tgt), current->comm,
1891                   PFID(&tsi->tsi_fid), start, end);
1892
1893         lock_count = ofd_prolong_extent_locks(tsi, start, end);
1894
1895         CDEBUG(D_DLMTRACE, "%s: refreshed %u locks timeout for req %p.\n",
1896                tgt_name(tsi->tsi_tgt), lock_count, req);
1897
1898         RETURN(lock_count > 0);
1899 }
1900
1901 static void ofd_rw_hpreq_fini(struct ptlrpc_request *req)
1902 {
1903         ofd_rw_hpreq_check(req);
1904 }
1905
1906 /**
1907  * Like tgt_rw_hpreq_lock_match(), but for OST_PUNCH RPCs.
1908  */
1909 static int ofd_punch_hpreq_lock_match(struct ptlrpc_request *req,
1910                                       struct ldlm_lock *lock)
1911 {
1912         struct tgt_session_info *tsi;
1913
1914         /* Don't use tgt_ses_info() to get session info, because lock_match()
1915          * can be called while request has no processing thread yet. */
1916         tsi = lu_context_key_get(&req->rq_session, &tgt_session_key);
1917         LASSERT(tsi != NULL);
1918
1919         LASSERT(tsi->tsi_ost_body != NULL);
1920         if (tsi->tsi_ost_body->oa.o_valid & OBD_MD_FLHANDLE &&
1921             tsi->tsi_ost_body->oa.o_handle.cookie == lock->l_handle.h_cookie)
1922                 return 1;
1923
1924         return 0;
1925 }
1926
1927 /**
1928  * Like ost_rw_hpreq_check(), but for OST_PUNCH RPCs.
1929  */
1930 static int ofd_punch_hpreq_check(struct ptlrpc_request *req)
1931 {
1932         struct tgt_session_info *tsi;
1933         struct obdo             *oa;
1934         int                      lock_count;
1935
1936         ENTRY;
1937
1938         /* Don't use tgt_ses_info() to get session info, because lock_match()
1939          * can be called while request has no processing thread yet. */
1940         tsi = lu_context_key_get(&req->rq_session, &tgt_session_key);
1941         LASSERT(tsi != NULL);
1942         oa = &tsi->tsi_ost_body->oa;
1943
1944         LASSERT(!(oa->o_valid & OBD_MD_FLFLAGS &&
1945                   oa->o_flags & OBD_FL_SRVLOCK));
1946
1947         CDEBUG(D_DLMTRACE,
1948                "%s: refresh locks: "LPU64"/"LPU64" ("LPU64"->"LPU64")\n",
1949                tgt_name(tsi->tsi_tgt), tsi->tsi_resid.name[0],
1950                tsi->tsi_resid.name[1], oa->o_size, oa->o_blocks);
1951
1952         lock_count = ofd_prolong_extent_locks(tsi, oa->o_size, oa->o_blocks);
1953
1954         CDEBUG(D_DLMTRACE, "%s: refreshed %u locks timeout for req %p.\n",
1955                tgt_name(tsi->tsi_tgt), lock_count, req);
1956
1957         RETURN(lock_count > 0);
1958 }
1959
1960 static void ofd_punch_hpreq_fini(struct ptlrpc_request *req)
1961 {
1962         ofd_punch_hpreq_check(req);
1963 }
1964
1965 struct ptlrpc_hpreq_ops ofd_hpreq_rw = {
1966         .hpreq_lock_match       = ofd_rw_hpreq_lock_match,
1967         .hpreq_check            = ofd_rw_hpreq_check,
1968         .hpreq_fini             = ofd_rw_hpreq_fini
1969 };
1970
1971 struct ptlrpc_hpreq_ops ofd_hpreq_punch = {
1972         .hpreq_lock_match       = ofd_punch_hpreq_lock_match,
1973         .hpreq_check            = ofd_punch_hpreq_check,
1974         .hpreq_fini             = ofd_punch_hpreq_fini
1975 };
1976
1977 /** Assign high priority operations to the IO requests */
1978 static void ofd_hp_brw(struct tgt_session_info *tsi)
1979 {
1980         struct niobuf_remote    *rnb;
1981         struct obd_ioobj        *ioo;
1982
1983         ENTRY;
1984
1985         ioo = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_IOOBJ);
1986         LASSERT(ioo != NULL); /* must exist after request preprocessing */
1987         if (ioo->ioo_bufcnt > 0) {
1988                 rnb = req_capsule_client_get(tsi->tsi_pill, &RMF_NIOBUF_REMOTE);
1989                 LASSERT(rnb != NULL); /* must exist after request preprocessing */
1990
1991                 /* no high priority if server lock is needed */
1992                 if (rnb->flags & OBD_BRW_SRVLOCK)
1993                         return;
1994         }
1995         tgt_ses_req(tsi)->rq_ops = &ofd_hpreq_rw;
1996 }
1997
1998 static void ofd_hp_punch(struct tgt_session_info *tsi)
1999 {
2000         LASSERT(tsi->tsi_ost_body != NULL); /* must exists if we are here */
2001         /* no high-priority if server lock is needed */
2002         if (tsi->tsi_ost_body->oa.o_valid & OBD_MD_FLFLAGS &&
2003             tsi->tsi_ost_body->oa.o_flags & OBD_FL_SRVLOCK)
2004                 return;
2005         tgt_ses_req(tsi)->rq_ops = &ofd_hpreq_punch;
2006 }
2007
2008 #define OBD_FAIL_OST_READ_NET   OBD_FAIL_OST_BRW_NET
2009 #define OBD_FAIL_OST_WRITE_NET  OBD_FAIL_OST_BRW_NET
2010 #define OST_BRW_READ    OST_READ
2011 #define OST_BRW_WRITE   OST_WRITE
2012
2013 static struct tgt_handler ofd_tgt_handlers[] = {
2014 TGT_RPC_HANDLER(OST_FIRST_OPC,
2015                 0,                      OST_CONNECT,    tgt_connect,
2016                 &RQF_CONNECT, LUSTRE_OBD_VERSION),
2017 TGT_RPC_HANDLER(OST_FIRST_OPC,
2018                 0,                      OST_DISCONNECT, tgt_disconnect,
2019                 &RQF_OST_DISCONNECT, LUSTRE_OBD_VERSION),
2020 TGT_RPC_HANDLER(OST_FIRST_OPC,
2021                 0,                      OST_SET_INFO,   ofd_set_info_hdl,
2022                 &RQF_OBD_SET_INFO, LUSTRE_OST_VERSION),
2023 TGT_OST_HDL(0,                          OST_GET_INFO,   ofd_get_info_hdl),
2024 TGT_OST_HDL(HABEO_CORPUS| HABEO_REFERO, OST_GETATTR,    ofd_getattr_hdl),
2025 TGT_OST_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR,
2026                                         OST_SETATTR,    ofd_setattr_hdl),
2027 TGT_OST_HDL(0           | HABEO_REFERO | MUTABOR,
2028                                         OST_CREATE,     ofd_create_hdl),
2029 TGT_OST_HDL(0           | HABEO_REFERO | MUTABOR,
2030                                         OST_DESTROY,    ofd_destroy_hdl),
2031 TGT_OST_HDL(0           | HABEO_REFERO, OST_STATFS,     ofd_statfs_hdl),
2032 TGT_OST_HDL_HP(HABEO_CORPUS| HABEO_REFERO,
2033                                         OST_BRW_READ,   tgt_brw_read,
2034                                                         ofd_hp_brw),
2035 /* don't set CORPUS flag for brw_write because -ENOENT may be valid case */
2036 TGT_OST_HDL_HP(HABEO_CORPUS| MUTABOR,   OST_BRW_WRITE,  tgt_brw_write,
2037                                                         ofd_hp_brw),
2038 TGT_OST_HDL_HP(HABEO_CORPUS| HABEO_REFERO | MUTABOR,
2039                                         OST_PUNCH,      ofd_punch_hdl,
2040                                                         ofd_hp_punch),
2041 TGT_OST_HDL(HABEO_CORPUS| HABEO_REFERO, OST_SYNC,       ofd_sync_hdl),
2042 TGT_OST_HDL(0           | HABEO_REFERO, OST_QUOTACTL,   ofd_quotactl),
2043 };
2044
2045 static struct tgt_opc_slice ofd_common_slice[] = {
2046         {
2047                 .tos_opc_start  = OST_FIRST_OPC,
2048                 .tos_opc_end    = OST_LAST_OPC,
2049                 .tos_hs         = ofd_tgt_handlers
2050         },
2051         {
2052                 .tos_opc_start  = OBD_FIRST_OPC,
2053                 .tos_opc_end    = OBD_LAST_OPC,
2054                 .tos_hs         = tgt_obd_handlers
2055         },
2056         {
2057                 .tos_opc_start  = LDLM_FIRST_OPC,
2058                 .tos_opc_end    = LDLM_LAST_OPC,
2059                 .tos_hs         = tgt_dlm_handlers
2060         },
2061         {
2062                 .tos_opc_start  = UPDATE_OBJ,
2063                 .tos_opc_end    = UPDATE_LAST_OPC,
2064                 .tos_hs         = tgt_out_handlers
2065         },
2066         {
2067                 .tos_opc_start  = SEQ_FIRST_OPC,
2068                 .tos_opc_end    = SEQ_LAST_OPC,
2069                 .tos_hs         = seq_handlers
2070         },
2071         {
2072                 .tos_opc_start  = LFSCK_FIRST_OPC,
2073                 .tos_opc_end    = LFSCK_LAST_OPC,
2074                 .tos_hs         = tgt_lfsck_handlers
2075         },
2076         {
2077                 .tos_hs         = NULL
2078         }
2079 };
2080
2081 static int ofd_init0(const struct lu_env *env, struct ofd_device *m,
2082                      struct lu_device_type *ldt, struct lustre_cfg *cfg)
2083 {
2084         const char              *dev = lustre_cfg_string(cfg, 0);
2085         struct ofd_thread_info  *info = NULL;
2086         struct obd_device       *obd;
2087         struct obd_statfs       *osfs;
2088         int                      rc;
2089
2090         ENTRY;
2091
2092         obd = class_name2obd(dev);
2093         if (obd == NULL) {
2094                 CERROR("Cannot find obd with name %s\n", dev);
2095                 RETURN(-ENODEV);
2096         }
2097
2098         rc = lu_env_refill((struct lu_env *)env);
2099         if (rc != 0)
2100                 RETURN(rc);
2101
2102         obd->u.obt.obt_magic = OBT_MAGIC;
2103
2104         m->ofd_fmd_max_num = OFD_FMD_MAX_NUM_DEFAULT;
2105         m->ofd_fmd_max_age = OFD_FMD_MAX_AGE_DEFAULT;
2106
2107         spin_lock_init(&m->ofd_flags_lock);
2108         m->ofd_raid_degraded = 0;
2109         m->ofd_syncjournal = 0;
2110         ofd_slc_set(m);
2111         m->ofd_grant_compat_disable = 0;
2112         m->ofd_soft_sync_limit = OFD_SOFT_SYNC_LIMIT_DEFAULT;
2113
2114         /* statfs data */
2115         spin_lock_init(&m->ofd_osfs_lock);
2116         m->ofd_osfs_age = cfs_time_shift_64(-1000);
2117         m->ofd_osfs_unstable = 0;
2118         m->ofd_statfs_inflight = 0;
2119         m->ofd_osfs_inflight = 0;
2120
2121         /* grant data */
2122         spin_lock_init(&m->ofd_grant_lock);
2123         m->ofd_tot_dirty = 0;
2124         m->ofd_tot_granted = 0;
2125         m->ofd_tot_pending = 0;
2126         m->ofd_seq_count = 0;
2127
2128         spin_lock_init(&m->ofd_batch_lock);
2129         rwlock_init(&obd->u.filter.fo_sptlrpc_lock);
2130         sptlrpc_rule_set_init(&obd->u.filter.fo_sptlrpc_rset);
2131         init_rwsem(&m->ofd_lastid_rwsem);
2132
2133         obd->u.filter.fo_fl_oss_capa = 0;
2134         CFS_INIT_LIST_HEAD(&obd->u.filter.fo_capa_keys);
2135         obd->u.filter.fo_capa_hash = init_capa_hash();
2136         if (obd->u.filter.fo_capa_hash == NULL)
2137                 RETURN(-ENOMEM);
2138
2139         m->ofd_dt_dev.dd_lu_dev.ld_ops = &ofd_lu_ops;
2140         m->ofd_dt_dev.dd_lu_dev.ld_obd = obd;
2141         /* set this lu_device to obd, because error handling need it */
2142         obd->obd_lu_dev = &m->ofd_dt_dev.dd_lu_dev;
2143
2144         rc = ofd_procfs_init(m);
2145         if (rc) {
2146                 CERROR("Can't init ofd lprocfs, rc %d\n", rc);
2147                 RETURN(rc);
2148         }
2149
2150         /* No connection accepted until configurations will finish */
2151         spin_lock(&obd->obd_dev_lock);
2152         obd->obd_no_conn = 1;
2153         spin_unlock(&obd->obd_dev_lock);
2154         obd->obd_replayable = 1;
2155         if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
2156                 char *str = lustre_cfg_string(cfg, 4);
2157
2158                 if (strchr(str, 'n')) {
2159                         CWARN("%s: recovery disabled\n", obd->obd_name);
2160                         obd->obd_replayable = 0;
2161                 }
2162         }
2163
2164         info = ofd_info_init(env, NULL);
2165         if (info == NULL)
2166                 RETURN(-EFAULT);
2167
2168         rc = ofd_stack_init(env, m, cfg);
2169         if (rc) {
2170                 CERROR("Can't init device stack, rc %d\n", rc);
2171                 GOTO(err_fini_proc, rc);
2172         }
2173
2174         ofd_procfs_add_brw_stats_symlink(m);
2175
2176         /* populate cached statfs data */
2177         osfs = &ofd_info(env)->fti_u.osfs;
2178         rc = ofd_statfs_internal(env, m, osfs, 0, NULL);
2179         if (rc != 0) {
2180                 CERROR("%s: can't get statfs data, rc %d\n", obd->obd_name, rc);
2181                 GOTO(err_fini_stack, rc);
2182         }
2183         if (!IS_PO2(osfs->os_bsize)) {
2184                 CERROR("%s: blocksize (%d) is not a power of 2\n",
2185                                 obd->obd_name, osfs->os_bsize);
2186                 GOTO(err_fini_stack, rc = -EPROTO);
2187         }
2188         m->ofd_blockbits = fls(osfs->os_bsize) - 1;
2189
2190         m->ofd_precreate_batch = OFD_PRECREATE_BATCH_DEFAULT;
2191         if (osfs->os_bsize * osfs->os_blocks < OFD_PRECREATE_SMALL_FS)
2192                 m->ofd_precreate_batch = OFD_PRECREATE_BATCH_SMALL;
2193
2194         snprintf(info->fti_u.name, sizeof(info->fti_u.name), "%s-%s",
2195                  "filter"/*LUSTRE_OST_NAME*/, obd->obd_uuid.uuid);
2196         m->ofd_namespace = ldlm_namespace_new(obd, info->fti_u.name,
2197                                               LDLM_NAMESPACE_SERVER,
2198                                               LDLM_NAMESPACE_GREEDY,
2199                                               LDLM_NS_TYPE_OST);
2200         if (m->ofd_namespace == NULL)
2201                 GOTO(err_fini_stack, rc = -ENOMEM);
2202         /* set obd_namespace for compatibility with old code */
2203         obd->obd_namespace = m->ofd_namespace;
2204         ldlm_register_intent(m->ofd_namespace, ofd_intent_policy);
2205         m->ofd_namespace->ns_lvbo = &ofd_lvbo;
2206         m->ofd_namespace->ns_lvbp = m;
2207
2208         ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
2209                            "filter_ldlm_cb_client", &obd->obd_ldlm_client);
2210
2211         dt_conf_get(env, m->ofd_osd, &m->ofd_dt_conf);
2212
2213         /* Allow at most ddp_grant_reserved% of the available filesystem space
2214          * to be granted to clients, so that any errors in the grant overhead
2215          * calculations do not allow granting more space to clients than can be
2216          * written. Assumes that in aggregate the grant overhead calculations do
2217          * not have more than ddp_grant_reserved% estimation error in them. */
2218         m->ofd_grant_ratio =
2219                 ofd_grant_ratio_conv(m->ofd_dt_conf.ddp_grant_reserved);
2220
2221         rc = tgt_init(env, &m->ofd_lut, obd, m->ofd_osd, ofd_common_slice,
2222                       OBD_FAIL_OST_ALL_REQUEST_NET,
2223                       OBD_FAIL_OST_ALL_REPLY_NET);
2224         if (rc)
2225                 GOTO(err_free_ns, rc);
2226
2227         rc = ofd_fs_setup(env, m, obd);
2228         if (rc)
2229                 GOTO(err_fini_lut, rc);
2230
2231         RETURN(0);
2232 err_fini_lut:
2233         tgt_fini(env, &m->ofd_lut);
2234 err_free_ns:
2235         ldlm_namespace_free(m->ofd_namespace, 0, obd->obd_force);
2236         obd->obd_namespace = m->ofd_namespace = NULL;
2237 err_fini_stack:
2238         ofd_stack_fini(env, m, &m->ofd_osd->dd_lu_dev);
2239 err_fini_proc:
2240         ofd_procfs_fini(m);
2241         return rc;
2242 }
2243
2244 static void ofd_fini(const struct lu_env *env, struct ofd_device *m)
2245 {
2246         struct obd_device       *obd = ofd_obd(m);
2247         struct lu_device        *d   = &m->ofd_dt_dev.dd_lu_dev;
2248         struct lfsck_stop        stop;
2249
2250         stop.ls_status = LS_PAUSED;
2251         lfsck_stop(env, m->ofd_osd, &stop);
2252         lfsck_degister(env, m->ofd_osd);
2253         target_recovery_fini(obd);
2254         obd_exports_barrier(obd);
2255         obd_zombie_barrier();
2256
2257         tgt_fini(env, &m->ofd_lut);
2258         ofd_fs_cleanup(env, m);
2259
2260         ofd_free_capa_keys(m);
2261         cleanup_capa_hash(obd->u.filter.fo_capa_hash);
2262
2263         if (m->ofd_namespace != NULL) {
2264                 ldlm_namespace_free(m->ofd_namespace, NULL,
2265                                     d->ld_obd->obd_force);
2266                 d->ld_obd->obd_namespace = m->ofd_namespace = NULL;
2267         }
2268
2269         ofd_stack_fini(env, m, &m->ofd_dt_dev.dd_lu_dev);
2270         ofd_procfs_fini(m);
2271         LASSERT(atomic_read(&d->ld_ref) == 0);
2272         server_put_mount(obd->obd_name);
2273         EXIT;
2274 }
2275
2276 static struct lu_device *ofd_device_fini(const struct lu_env *env,
2277                                          struct lu_device *d)
2278 {
2279         ENTRY;
2280         ofd_fini(env, ofd_dev(d));
2281         RETURN(NULL);
2282 }
2283
2284 static struct lu_device *ofd_device_free(const struct lu_env *env,
2285                                          struct lu_device *d)
2286 {
2287         struct ofd_device *m = ofd_dev(d);
2288
2289         dt_device_fini(&m->ofd_dt_dev);
2290         OBD_FREE_PTR(m);
2291         RETURN(NULL);
2292 }
2293
2294 static struct lu_device *ofd_device_alloc(const struct lu_env *env,
2295                                           struct lu_device_type *t,
2296                                           struct lustre_cfg *cfg)
2297 {
2298         struct ofd_device *m;
2299         struct lu_device  *l;
2300         int                rc;
2301
2302         OBD_ALLOC_PTR(m);
2303         if (m == NULL)
2304                 return ERR_PTR(-ENOMEM);
2305
2306         l = &m->ofd_dt_dev.dd_lu_dev;
2307         dt_device_init(&m->ofd_dt_dev, t);
2308         rc = ofd_init0(env, m, t, cfg);
2309         if (rc != 0) {
2310                 ofd_device_free(env, l);
2311                 l = ERR_PTR(rc);
2312         }
2313
2314         return l;
2315 }
2316
2317 /* thread context key constructor/destructor */
2318 LU_KEY_INIT_FINI(ofd, struct ofd_thread_info);
2319
2320 static void ofd_key_exit(const struct lu_context *ctx,
2321                          struct lu_context_key *key, void *data)
2322 {
2323         struct ofd_thread_info *info = data;
2324
2325         info->fti_env = NULL;
2326         info->fti_exp = NULL;
2327
2328         info->fti_xid = 0;
2329         info->fti_pre_version = 0;
2330         info->fti_used = 0;
2331
2332         memset(&info->fti_attr, 0, sizeof info->fti_attr);
2333 }
2334
2335 struct lu_context_key ofd_thread_key = {
2336         .lct_tags = LCT_DT_THREAD,
2337         .lct_init = ofd_key_init,
2338         .lct_fini = ofd_key_fini,
2339         .lct_exit = ofd_key_exit
2340 };
2341
2342 /* type constructor/destructor: mdt_type_init, mdt_type_fini */
2343 LU_TYPE_INIT_FINI(ofd, &ofd_thread_key);
2344
2345 static struct lu_device_type_operations ofd_device_type_ops = {
2346         .ldto_init              = ofd_type_init,
2347         .ldto_fini              = ofd_type_fini,
2348
2349         .ldto_start             = ofd_type_start,
2350         .ldto_stop              = ofd_type_stop,
2351
2352         .ldto_device_alloc      = ofd_device_alloc,
2353         .ldto_device_free       = ofd_device_free,
2354         .ldto_device_fini       = ofd_device_fini
2355 };
2356
2357 static struct lu_device_type ofd_device_type = {
2358         .ldt_tags       = LU_DEVICE_DT,
2359         .ldt_name       = LUSTRE_OST_NAME,
2360         .ldt_ops        = &ofd_device_type_ops,
2361         .ldt_ctx_tags   = LCT_DT_THREAD
2362 };
2363
2364 int __init ofd_init(void)
2365 {
2366         struct lprocfs_static_vars      lvars;
2367         int                             rc;
2368
2369         rc = lu_kmem_init(ofd_caches);
2370         if (rc)
2371                 return rc;
2372
2373         rc = ofd_fmd_init();
2374         if (rc) {
2375                 lu_kmem_fini(ofd_caches);
2376                 return(rc);
2377         }
2378
2379         lprocfs_ofd_init_vars(&lvars);
2380
2381         rc = class_register_type(&ofd_obd_ops, NULL, NULL,
2382 #ifndef HAVE_ONLY_PROCFS_SEQ
2383                                 lvars.module_vars,
2384 #endif
2385                                 LUSTRE_OST_NAME, &ofd_device_type);
2386         return rc;
2387 }
2388
2389 void __exit ofd_exit(void)
2390 {
2391         ofd_fmd_exit();
2392         lu_kmem_fini(ofd_caches);
2393         class_unregister_type(LUSTRE_OST_NAME);
2394 }
2395
2396 MODULE_AUTHOR("Whamcloud, Inc. <http://www.whamcloud.com/>");
2397 MODULE_DESCRIPTION("Lustre Object Filtering Device");
2398 MODULE_LICENSE("GPL");
2399
2400 module_init(ofd_init);
2401 module_exit(ofd_exit);