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_NOFS);
 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         rc = lfsck_register_namespace(env, ofd->ofd_osd, ofd->ofd_namespace);
 446         /* The LFSCK instance is registered just now, so it must be there when
 447          * register the namespace to such instance. */
 448         LASSERTF(rc == 0, "register namespace failed: rc = %d\n", rc);
 449
 450         lsp.lsp_start = NULL;
 451         lsp.lsp_index_valid = 0;
 452         rc = lfsck_start(env, ofd->ofd_osd, &lsp);
 453         if (rc != 0) {
 454                 CWARN("%s: auto trigger paused LFSCK failed: rc = %d\n",
 455                       obd->obd_name, rc);
 456                 rc = 0;
 457         }
 458
 459         target_recovery_init(&ofd->ofd_lut, tgt_request_handle);
 460         LASSERT(obd->obd_no_conn);
 461         spin_lock(&obd->obd_dev_lock);
 462         obd->obd_no_conn = 0;
 463         spin_unlock(&obd->obd_dev_lock);
 464
 465         if (obd->obd_recovering == 0)
 466                 ofd_postrecov(env, ofd);
 467
 468         RETURN(rc);
 469 }
 470
 471 static int ofd_recovery_complete(const struct lu_env *env,
 472                                  struct lu_device *dev)
 473 {
 474         struct ofd_device       *ofd = ofd_dev(dev);
 475         struct lu_device        *next = &ofd->ofd_osd->dd_lu_dev;
 476         int                      rc = 0, max_precreate;
 477
 478         ENTRY;
 479
 480         /* Grant space for object precreation on the self export.
 481          * This initial reserved space (i.e. 10MB for zfs and 280KB for ldiskfs)
 482          * is enough to create 10k objects. More space is then acquired for
 483          * precreation in ofd_grant_create().
 484          */
 485         max_precreate = OST_MAX_PRECREATE * ofd->ofd_dt_conf.ddp_inodespace / 2;
 486         ofd_grant_connect(env, dev->ld_obd->obd_self_export, max_precreate,
 487                           false);
 488         rc = next->ld_ops->ldo_recovery_complete(env, next);
 489         RETURN(rc);
 490 }
 491
 492 static struct lu_device_operations ofd_lu_ops = {
 493         .ldo_object_alloc       = ofd_object_alloc,
 494         .ldo_process_config     = ofd_process_config,
 495         .ldo_recovery_complete  = ofd_recovery_complete,
 496         .ldo_prepare            = ofd_prepare,
 497 };
 498
 499 static int ofd_procfs_init(struct ofd_device *ofd)
 500 {
 501         struct lprocfs_static_vars       lvars;
 502         struct obd_device               *obd = ofd_obd(ofd);
 503         cfs_proc_dir_entry_t            *entry;
 504         int                              rc = 0;
 505
 506         ENTRY;
 507
 508         /* lprocfs must be setup before the ofd so state can be safely added
 509          * to /proc incrementally as the ofd is setup */
 510         lprocfs_ofd_init_vars(&lvars);
 511         rc = lprocfs_obd_setup(obd, lvars.obd_vars);
 512         if (rc) {
 513                 CERROR("%s: lprocfs_obd_setup failed: %d.\n",
 514                        obd->obd_name, rc);
 515                 RETURN(rc);
 516         }
 517
 518         rc = lprocfs_alloc_obd_stats(obd, LPROC_OFD_STATS_LAST);
 519         if (rc) {
 520                 CERROR("%s: lprocfs_alloc_obd_stats failed: %d.\n",
 521                        obd->obd_name, rc);
 522                 GOTO(obd_cleanup, rc);
 523         }
 524
 525         obd->obd_uses_nid_stats = 1;
 526
 527         entry = lprocfs_register("exports", obd->obd_proc_entry, NULL, NULL);
 528         if (IS_ERR(entry)) {
 529                 rc = PTR_ERR(entry);
 530                 CERROR("%s: error %d setting up lprocfs for %s\n",
 531                        obd->obd_name, rc, "exports");
 532                 GOTO(obd_cleanup, rc);
 533         }
 534         obd->obd_proc_exports_entry = entry;
 535
 536         entry = lprocfs_add_simple(obd->obd_proc_exports_entry, "clear",
 537                                    lprocfs_nid_stats_clear_read,
 538                                    lprocfs_nid_stats_clear_write, obd, NULL);
 539         if (IS_ERR(entry)) {
 540                 rc = PTR_ERR(entry);
 541                 CERROR("%s: add proc entry 'clear' failed: %d.\n",
 542                        obd->obd_name, rc);
 543                 GOTO(obd_cleanup, rc);
 544         }
 545
 546         ofd_stats_counter_init(obd->obd_stats);
 547
 548         rc = lprocfs_job_stats_init(obd, LPROC_OFD_STATS_LAST,
 549                                     ofd_stats_counter_init);
 550         if (rc)
 551                 GOTO(remove_entry_clear, rc);
 552         RETURN(0);
 553 remove_entry_clear:
 554         lprocfs_remove_proc_entry("clear", obd->obd_proc_exports_entry);
 555 obd_cleanup:
 556         lprocfs_obd_cleanup(obd);
 557         lprocfs_free_obd_stats(obd);
 558
 559         return rc;
 560 }
 561
 562 static void ofd_procfs_add_brw_stats_symlink(struct ofd_device *ofd)
 563 {
 564         struct obd_device       *obd = ofd_obd(ofd);
 565         struct obd_device       *osd_obd = ofd->ofd_osd_exp->exp_obd;
 566         cfs_proc_dir_entry_t    *osd_root = osd_obd->obd_type->typ_procroot;
 567         cfs_proc_dir_entry_t    *osd_dir;
 568
 569         osd_dir = lprocfs_srch(osd_root, obd->obd_name);
 570         if (osd_dir == NULL)
 571                 return;
 572
 573         if (lprocfs_srch(osd_dir, "brw_stats") != NULL)
 574                 lprocfs_add_symlink("brw_stats", obd->obd_proc_entry,
 575                                     "../../%s/%s/brw_stats",
 576                                     osd_root->name, osd_dir->name);
 577
 578         if (lprocfs_srch(osd_dir, "read_cache_enable") != NULL)
 579                 lprocfs_add_symlink("read_cache_enable", obd->obd_proc_entry,
 580                                     "../../%s/%s/read_cache_enable",
 581                                     osd_root->name, osd_dir->name);
 582
 583         if (lprocfs_srch(osd_dir, "readcache_max_filesize") != NULL)
 584                 lprocfs_add_symlink("readcache_max_filesize",
 585                                     obd->obd_proc_entry,
 586                                     "../../%s/%s/readcache_max_filesize",
 587                                     osd_root->name, osd_dir->name);
 588
 589         if (lprocfs_srch(osd_dir, "writethrough_cache_enable") != NULL)
 590                 lprocfs_add_symlink("writethrough_cache_enable",
 591                                     obd->obd_proc_entry,
 592                                     "../../%s/%s/writethrough_cache_enable",
 593                                     osd_root->name, osd_dir->name);
 594 }
 595
 596 static void ofd_procfs_fini(struct ofd_device *ofd)
 597 {
 598         struct obd_device *obd = ofd_obd(ofd);
 599
 600         lprocfs_remove_proc_entry("writethrough_cache_enable",
 601                                   obd->obd_proc_entry);
 602         lprocfs_remove_proc_entry("readcache_max_filesize",
 603                                   obd->obd_proc_entry);
 604         lprocfs_remove_proc_entry("read_cache_enable", obd->obd_proc_entry);
 605         lprocfs_remove_proc_entry("brw_stats", obd->obd_proc_entry);
 606         lprocfs_remove_proc_entry("clear", obd->obd_proc_exports_entry);
 607         lprocfs_free_per_client_stats(obd);
 608         lprocfs_obd_cleanup(obd);
 609         lprocfs_free_obd_stats(obd);
 610         lprocfs_job_stats_fini(obd);
 611 }
 612
 613 extern int ost_handle(struct ptlrpc_request *req);
 614
 615 int ofd_fid_fini(const struct lu_env *env, struct ofd_device *ofd)
 616 {
 617         return seq_site_fini(env, &ofd->ofd_seq_site);
 618 }
 619
 620 int ofd_fid_init(const struct lu_env *env, struct ofd_device *ofd)
 621 {
 622         struct seq_server_site  *ss = &ofd->ofd_seq_site;
 623         struct lu_device        *lu = &ofd->ofd_dt_dev.dd_lu_dev;
 624         char                    *obd_name = ofd_name(ofd);
 625         char                    *name = NULL;
 626         int                     rc = 0;
 627
 628         ss = &ofd->ofd_seq_site;
 629         lu->ld_site->ld_seq_site = ss;
 630         ss->ss_lu = lu->ld_site;
 631         ss->ss_node_id = ofd->ofd_lut.lut_lsd.lsd_osd_index;
 632
 633         OBD_ALLOC_PTR(ss->ss_server_seq);
 634         if (ss->ss_server_seq == NULL)
 635                 GOTO(out_free, rc = -ENOMEM);
 636
 637         OBD_ALLOC(name, strlen(obd_name) + 10);
 638         if (!name) {
 639                 OBD_FREE_PTR(ss->ss_server_seq);
 640                 ss->ss_server_seq = NULL;
 641                 GOTO(out_free, rc = -ENOMEM);
 642         }
 643
 644         rc = seq_server_init(env, ss->ss_server_seq, ofd->ofd_osd, obd_name,
 645                              LUSTRE_SEQ_SERVER, ss);
 646         if (rc) {
 647                 CERROR("%s : seq server init error %d\n", obd_name, rc);
 648                 GOTO(out_free, rc);
 649         }
 650         ss->ss_server_seq->lss_space.lsr_index = ss->ss_node_id;
 651
 652         OBD_ALLOC_PTR(ss->ss_client_seq);
 653         if (ss->ss_client_seq == NULL)
 654                 GOTO(out_free, rc = -ENOMEM);
 655
 656         snprintf(name, strlen(obd_name) + 6, "%p-super", obd_name);
 657         rc = seq_client_init(ss->ss_client_seq, NULL, LUSTRE_SEQ_DATA,
 658                              name, NULL);
 659         if (rc) {
 660                 CERROR("%s : seq client init error %d\n", obd_name, rc);
 661                 GOTO(out_free, rc);
 662         }
 663         OBD_FREE(name, strlen(obd_name) + 10);
 664         name = NULL;
 665
 666         rc = seq_server_set_cli(env, ss->ss_server_seq, ss->ss_client_seq);
 667
 668 out_free:
 669         if (rc) {
 670                 if (ss->ss_server_seq) {
 671                         seq_server_fini(ss->ss_server_seq, env);
 672                         OBD_FREE_PTR(ss->ss_server_seq);
 673                         ss->ss_server_seq = NULL;
 674                 }
 675
 676                 if (ss->ss_client_seq) {
 677                         seq_client_fini(ss->ss_client_seq);
 678                         OBD_FREE_PTR(ss->ss_client_seq);
 679                         ss->ss_client_seq = NULL;
 680                 }
 681
 682                 if (name) {
 683                         OBD_FREE(name, strlen(obd_name) + 10);
 684                         name = NULL;
 685                 }
 686         }
 687
 688         return rc;
 689 }
 690
 691 int ofd_set_info_hdl(struct tgt_session_info *tsi)
 692 {
 693         struct ptlrpc_request   *req = tgt_ses_req(tsi);
 694         struct ost_body         *body = NULL, *repbody;
 695         void                    *key, *val = NULL;
 696         int                      keylen, vallen, rc = 0;
 697         bool                     is_grant_shrink;
 698         struct ofd_device       *ofd = ofd_exp(tsi->tsi_exp);
 699
 700         ENTRY;
 701
 702         key = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_KEY);
 703         if (key == NULL) {
 704                 DEBUG_REQ(D_HA, req, "no set_info key");
 705                 RETURN(err_serious(-EFAULT));
 706         }
 707         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_KEY,
 708                                       RCL_CLIENT);
 709
 710         val = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_VAL);
 711         if (val == NULL) {
 712                 DEBUG_REQ(D_HA, req, "no set_info val");
 713                 RETURN(err_serious(-EFAULT));
 714         }
 715         vallen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_VAL,
 716                                       RCL_CLIENT);
 717
 718         is_grant_shrink = KEY_IS(KEY_GRANT_SHRINK);
 719         if (is_grant_shrink)
 720                 /* In this case the value is actually an RMF_OST_BODY, so we
 721                  * transmutate the type of this PTLRPC */
 722                 req_capsule_extend(tsi->tsi_pill, &RQF_OST_SET_GRANT_INFO);
 723
 724         rc = req_capsule_server_pack(tsi->tsi_pill);
 725         if (rc < 0)
 726                 RETURN(rc);
 727
 728         if (is_grant_shrink) {
 729                 body = req_capsule_client_get(tsi->tsi_pill, &RMF_OST_BODY);
 730
 731                 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
 732                 *repbody = *body;
 733
 734                 /** handle grant shrink, similar to a read request */
 735                 ofd_grant_prepare_read(tsi->tsi_env, tsi->tsi_exp,
 736                                        &repbody->oa);
 737         } else if (KEY_IS(KEY_EVICT_BY_NID)) {
 738                 if (vallen > 0)
 739                         obd_export_evict_by_nid(tsi->tsi_exp->exp_obd, val);
 740                 rc = 0;
 741         } else if (KEY_IS(KEY_CAPA_KEY)) {
 742                 rc = ofd_update_capa_key(ofd, val);
 743         } else if (KEY_IS(KEY_SPTLRPC_CONF)) {
 744                 rc = tgt_adapt_sptlrpc_conf(tsi->tsi_tgt, 0);
 745         } else {
 746                 CERROR("%s: Unsupported key %s\n",
 747                        tgt_name(tsi->tsi_tgt), (char *)key);
 748                 rc = -EOPNOTSUPP;
 749         }
 750         ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_SET_INFO,
 751                          tsi->tsi_jobid, 1);
 752
 753         RETURN(rc);
 754 }
 755
 756 int ofd_fiemap_get(const struct lu_env *env, struct ofd_device *ofd,
 757                    struct lu_fid *fid, struct ll_user_fiemap *fiemap)
 758 {
 759         struct ofd_object       *fo;
 760         int                      rc;
 761
 762         fo = ofd_object_find(env, ofd, fid);
 763         if (IS_ERR(fo)) {
 764                 CERROR("%s: error finding object "DFID"\n",
 765                        ofd_name(ofd), PFID(fid));
 766                 return PTR_ERR(fo);
 767         }
 768
 769         ofd_read_lock(env, fo);
 770         if (ofd_object_exists(fo))
 771                 rc = dt_fiemap_get(env, ofd_object_child(fo), fiemap);
 772         else
 773                 rc = -ENOENT;
 774         ofd_read_unlock(env, fo);
 775         ofd_object_put(env, fo);
 776         return rc;
 777 }
 778
 779 struct locked_region {
 780         cfs_list_t              list;
 781         struct lustre_handle    lh;
 782 };
 783
 784 static int lock_region(struct ldlm_namespace *ns, struct ldlm_res_id *res_id,
 785                        unsigned long long begin, unsigned long long end,
 786                        cfs_list_t *locked)
 787 {
 788         struct locked_region    *region = NULL;
 789         __u64                    flags = 0;
 790         int                      rc;
 791
 792         LASSERT(begin <= end);
 793         OBD_ALLOC_PTR(region);
 794         if (region == NULL)
 795                 return -ENOMEM;
 796
 797         rc = tgt_extent_lock(ns, res_id, begin, end, &region->lh,
 798                              LCK_PR, &flags);
 799         if (rc != 0)
 800                 return rc;
 801
 802         CDEBUG(D_OTHER, "ost lock [%llu,%llu], lh=%p\n", begin, end,
 803                &region->lh);
 804         cfs_list_add(&region->list, locked);
 805
 806         return 0;
 807 }
 808
 809 static int lock_zero_regions(struct ldlm_namespace *ns,
 810                              struct ldlm_res_id *res_id,
 811                              struct ll_user_fiemap *fiemap,
 812                              cfs_list_t *locked)
 813 {
 814         __u64 begin = fiemap->fm_start;
 815         unsigned int i;
 816         int rc = 0;
 817         struct ll_fiemap_extent *fiemap_start = fiemap->fm_extents;
 818
 819         ENTRY;
 820
 821         CDEBUG(D_OTHER, "extents count %u\n", fiemap->fm_mapped_extents);
 822         for (i = 0; i < fiemap->fm_mapped_extents; i++) {
 823                 if (fiemap_start[i].fe_logical > begin) {
 824                         CDEBUG(D_OTHER, "ost lock [%llu,%llu]\n",
 825                                begin, fiemap_start[i].fe_logical);
 826                         rc = lock_region(ns, res_id, begin,
 827                                          fiemap_start[i].fe_logical, locked);
 828                         if (rc)
 829                                 RETURN(rc);
 830                 }
 831
 832                 begin = fiemap_start[i].fe_logical + fiemap_start[i].fe_length;
 833         }
 834
 835         if (begin < (fiemap->fm_start + fiemap->fm_length)) {
 836                 CDEBUG(D_OTHER, "ost lock [%llu,%llu]\n",
 837                        begin, fiemap->fm_start + fiemap->fm_length);
 838                 rc = lock_region(ns, res_id, begin,
 839                                  fiemap->fm_start + fiemap->fm_length, locked);
 840         }
 841
 842         RETURN(rc);
 843 }
 844
 845 static void unlock_zero_regions(struct ldlm_namespace *ns, cfs_list_t *locked)
 846 {
 847         struct locked_region *entry, *temp;
 848
 849         cfs_list_for_each_entry_safe(entry, temp, locked, list) {
 850                 CDEBUG(D_OTHER, "ost unlock lh=%p\n", &entry->lh);
 851                 tgt_extent_unlock(&entry->lh, LCK_PR);
 852                 cfs_list_del(&entry->list);
 853                 OBD_FREE_PTR(entry);
 854         }
 855 }
 856
 857 int ofd_get_info_hdl(struct tgt_session_info *tsi)
 858 {
 859         struct obd_export               *exp = tsi->tsi_exp;
 860         struct ofd_device               *ofd = ofd_exp(exp);
 861         struct ofd_thread_info          *fti = tsi2ofd_info(tsi);
 862         void                            *key;
 863         int                              keylen;
 864         int                              replylen, rc = 0;
 865
 866         ENTRY;
 867
 868         /* this common part for get_info rpc */
 869         key = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_KEY);
 870         if (key == NULL) {
 871                 DEBUG_REQ(D_HA, tgt_ses_req(tsi), "no get_info key");
 872                 RETURN(err_serious(-EPROTO));
 873         }
 874         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_GETINFO_KEY,
 875                                       RCL_CLIENT);
 876
 877         if (KEY_IS(KEY_LAST_ID)) {
 878                 obd_id          *last_id;
 879                 struct ofd_seq  *oseq;
 880
 881                 req_capsule_extend(tsi->tsi_pill, &RQF_OST_GET_INFO_LAST_ID);
 882                 rc = req_capsule_server_pack(tsi->tsi_pill);
 883                 if (rc)
 884                         RETURN(err_serious(rc));
 885
 886                 last_id = req_capsule_server_get(tsi->tsi_pill, &RMF_OBD_ID);
 887
 888                 oseq = ofd_seq_load(tsi->tsi_env, ofd,
 889                                     (obd_seq)exp->exp_filter_data.fed_group);
 890                 if (IS_ERR(oseq))
 891                         rc = -EFAULT;
 892                 else
 893                         *last_id = ofd_seq_last_oid(oseq);
 894                 ofd_seq_put(tsi->tsi_env, oseq);
 895         } else if (KEY_IS(KEY_FIEMAP)) {
 896                 struct ll_fiemap_info_key       *fm_key;
 897                 struct ll_user_fiemap           *fiemap;
 898                 struct lu_fid                   *fid;
 899
 900                 req_capsule_extend(tsi->tsi_pill, &RQF_OST_GET_INFO_FIEMAP);
 901
 902                 fm_key = req_capsule_client_get(tsi->tsi_pill, &RMF_FIEMAP_KEY);
 903                 rc = tgt_validate_obdo(tsi, &fm_key->oa);
 904                 if (rc)
 905                         RETURN(err_serious(rc));
 906
 907                 fid = &fm_key->oa.o_oi.oi_fid;
 908
 909                 CDEBUG(D_INODE, "get FIEMAP of object "DFID"\n", PFID(fid));
 910
 911                 replylen = fiemap_count_to_size(fm_key->fiemap.fm_extent_count);
 912                 req_capsule_set_size(tsi->tsi_pill, &RMF_FIEMAP_VAL,
 913                                      RCL_SERVER, replylen);
 914
 915                 rc = req_capsule_server_pack(tsi->tsi_pill);
 916                 if (rc)
 917                         RETURN(err_serious(rc));
 918
 919                 fiemap = req_capsule_server_get(tsi->tsi_pill, &RMF_FIEMAP_VAL);
 920                 if (fiemap == NULL)
 921                         RETURN(-ENOMEM);
 922
 923                 *fiemap = fm_key->fiemap;
 924                 rc = ofd_fiemap_get(tsi->tsi_env, ofd, fid, fiemap);
 925
 926                 /* LU-3219: Lock the sparse areas to make sure dirty
 927                  * flushed back from client, then call fiemap again. */
 928                 if (fm_key->oa.o_valid & OBD_MD_FLFLAGS &&
 929                     fm_key->oa.o_flags & OBD_FL_SRVLOCK) {
 930                         cfs_list_t locked = CFS_LIST_HEAD_INIT(locked);
 931
 932                         ost_fid_build_resid(fid, &fti->fti_resid);
 933                         rc = lock_zero_regions(ofd->ofd_namespace,
 934                                                &fti->fti_resid, fiemap,
 935                                                &locked);
 936                         if (rc == 0 && !cfs_list_empty(&locked)) {
 937                                 rc = ofd_fiemap_get(tsi->tsi_env, ofd, fid,
 938                                                     fiemap);
 939                                 unlock_zero_regions(ofd->ofd_namespace,
 940                                                     &locked);
 941                         }
 942                 }
 943         } else if (KEY_IS(KEY_LAST_FID)) {
 944                 struct ofd_device       *ofd = ofd_exp(exp);
 945                 struct ofd_seq          *oseq;
 946                 struct lu_fid           *fid;
 947                 int                      rc;
 948
 949                 req_capsule_extend(tsi->tsi_pill, &RQF_OST_GET_INFO_LAST_FID);
 950                 rc = req_capsule_server_pack(tsi->tsi_pill);
 951                 if (rc)
 952                         RETURN(err_serious(rc));
 953
 954                 fid = req_capsule_client_get(tsi->tsi_pill, &RMF_FID);
 955                 if (fid == NULL)
 956                         RETURN(err_serious(-EPROTO));
 957
 958                 fid_le_to_cpu(&fti->fti_ostid.oi_fid, fid);
 959
 960                 fid = req_capsule_server_get(tsi->tsi_pill, &RMF_FID);
 961                 if (fid == NULL)
 962                         RETURN(-ENOMEM);
 963
 964                 oseq = ofd_seq_load(tsi->tsi_env, ofd,
 965                                     ostid_seq(&fti->fti_ostid));
 966                 if (IS_ERR(oseq))
 967                         RETURN(PTR_ERR(oseq));
 968
 969                 rc = ostid_to_fid(fid, &oseq->os_oi,
 970                                   ofd->ofd_lut.lut_lsd.lsd_osd_index);
 971                 if (rc != 0)
 972                         GOTO(out_put, rc);
 973
 974                 CDEBUG(D_HA, "%s: LAST FID is "DFID"\n", ofd_name(ofd),
 975                        PFID(fid));
 976 out_put:
 977                 ofd_seq_put(tsi->tsi_env, oseq);
 978         } else {
 979                 CERROR("%s: not supported key %s\n", tgt_name(tsi->tsi_tgt),
 980                        (char *)key);
 981                 rc = -EOPNOTSUPP;
 982         }
 983         ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_GET_INFO,
 984                          tsi->tsi_jobid, 1);
 985
 986         RETURN(rc);
 987 }
 988
 989 static int ofd_getattr_hdl(struct tgt_session_info *tsi)
 990 {
 991         struct ofd_thread_info  *fti = tsi2ofd_info(tsi);
 992         struct ofd_device       *ofd = ofd_exp(tsi->tsi_exp);
 993         struct ost_body         *repbody;
 994         struct lustre_handle     lh = { 0 };
 995         struct ofd_object       *fo;
 996         __u64                    flags = 0;
 997         ldlm_mode_t              lock_mode = LCK_PR;
 998         bool                     srvlock;
 999         int                      rc;
1000         ENTRY;
1001
1002         LASSERT(tsi->tsi_ost_body != NULL);
1003
1004         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1005         if (repbody == NULL)
1006                 RETURN(-ENOMEM);
1007
1008         repbody->oa.o_oi = tsi->tsi_ost_body->oa.o_oi;
1009         repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1010
1011         srvlock = tsi->tsi_ost_body->oa.o_valid & OBD_MD_FLFLAGS &&
1012                   tsi->tsi_ost_body->oa.o_flags & OBD_FL_SRVLOCK;
1013
1014         if (srvlock) {
1015                 if (unlikely(tsi->tsi_ost_body->oa.o_flags & OBD_FL_FLUSH))
1016                         lock_mode = LCK_PW;
1017
1018                 rc = tgt_extent_lock(tsi->tsi_tgt->lut_obd->obd_namespace,
1019                                      &tsi->tsi_resid, 0, OBD_OBJECT_EOF, &lh,
1020                                      lock_mode, &flags);
1021                 if (rc != 0)
1022                         RETURN(rc);
1023         }
1024
1025         fo = ofd_object_find_exists(tsi->tsi_env, ofd, &tsi->tsi_fid);
1026         if (IS_ERR(fo))
1027                 GOTO(out, rc = PTR_ERR(fo));
1028
1029         rc = ofd_attr_get(tsi->tsi_env, fo, &fti->fti_attr);
1030         if (rc == 0) {
1031                 __u64    curr_version;
1032
1033                 obdo_from_la(&repbody->oa, &fti->fti_attr,
1034                              OFD_VALID_FLAGS | LA_UID | LA_GID);
1035                 tgt_drop_id(tsi->tsi_exp, &repbody->oa);
1036
1037                 /* Store object version in reply */
1038                 curr_version = dt_version_get(tsi->tsi_env,
1039                                               ofd_object_child(fo));
1040                 if ((__s64)curr_version != -EOPNOTSUPP) {
1041                         repbody->oa.o_valid |= OBD_MD_FLDATAVERSION;
1042                         repbody->oa.o_data_version = curr_version;
1043                 }
1044         }
1045
1046         ofd_object_put(tsi->tsi_env, fo);
1047 out:
1048         if (srvlock)
1049                 tgt_extent_unlock(&lh, lock_mode);
1050
1051         ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_GETATTR,
1052                          tsi->tsi_jobid, 1);
1053
1054         repbody->oa.o_valid |= OBD_MD_FLFLAGS;
1055         repbody->oa.o_flags = OBD_FL_FLUSH;
1056
1057         RETURN(rc);
1058 }
1059
1060 static int ofd_setattr_hdl(struct tgt_session_info *tsi)
1061 {
1062         struct ofd_thread_info  *fti = tsi2ofd_info(tsi);
1063         struct ofd_device       *ofd = ofd_exp(tsi->tsi_exp);
1064         struct ost_body         *body = tsi->tsi_ost_body;
1065         struct ost_body         *repbody;
1066         struct ldlm_resource    *res;
1067         struct ofd_object       *fo;
1068         struct filter_fid       *ff = NULL;
1069         int                      rc = 0;
1070
1071         ENTRY;
1072
1073         LASSERT(body != NULL);
1074
1075         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1076         if (repbody == NULL)
1077                 RETURN(-ENOMEM);
1078
1079         repbody->oa.o_oi = body->oa.o_oi;
1080         repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1081
1082         /* This would be very bad - accidentally truncating a file when
1083          * changing the time or similar - bug 12203. */
1084         if (body->oa.o_valid & OBD_MD_FLSIZE &&
1085             body->oa.o_size != OBD_OBJECT_EOF) {
1086                 static char mdsinum[48];
1087
1088                 if (body->oa.o_valid & OBD_MD_FLFID)
1089                         snprintf(mdsinum, sizeof(mdsinum) - 1,
1090                                  "of parent "DFID, body->oa.o_parent_seq,
1091                                  body->oa.o_parent_oid, 0);
1092                 else
1093                         mdsinum[0] = '\0';
1094
1095                 CERROR("%s: setattr from %s is trying to truncate object "DFID
1096                        " %s\n", ofd_name(ofd), obd_export_nid2str(tsi->tsi_exp),
1097                        PFID(&tsi->tsi_fid), mdsinum);
1098                 RETURN(-EPERM);
1099         }
1100
1101         fo = ofd_object_find_exists(tsi->tsi_env, ofd, &tsi->tsi_fid);
1102         if (IS_ERR(fo))
1103                 GOTO(out, rc = PTR_ERR(fo));
1104
1105         la_from_obdo(&fti->fti_attr, &body->oa, body->oa.o_valid);
1106         fti->fti_attr.la_valid &= ~LA_TYPE;
1107
1108         if (body->oa.o_valid & OBD_MD_FLFID) {
1109                 ff = &fti->fti_mds_fid;
1110                 ofd_prepare_fidea(ff, &body->oa);
1111         }
1112
1113         /* setting objects attributes (including owner/group) */
1114         rc = ofd_attr_set(tsi->tsi_env, fo, &fti->fti_attr, ff);
1115         if (rc != 0)
1116                 GOTO(out_put, rc);
1117
1118         obdo_from_la(&repbody->oa, &fti->fti_attr,
1119                      OFD_VALID_FLAGS | LA_UID | LA_GID);
1120         tgt_drop_id(tsi->tsi_exp, &repbody->oa);
1121
1122         ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_SETATTR,
1123                          tsi->tsi_jobid, 1);
1124         EXIT;
1125 out_put:
1126         ofd_object_put(tsi->tsi_env, fo);
1127 out:
1128         if (rc == 0) {
1129                 /* we do not call this before to avoid lu_object_find() in
1130                  *  ->lvbo_update() holding another reference on the object.
1131                  * otherwise concurrent destroy can make the object unavailable
1132                  * for 2nd lu_object_find() waiting for the first reference
1133                  * to go... deadlock! */
1134                 res = ldlm_resource_get(ofd->ofd_namespace, NULL,
1135                                         &tsi->tsi_resid, LDLM_EXTENT, 0);
1136                 if (res != NULL) {
1137                         ldlm_res_lvbo_update(res, NULL, 0);
1138                         ldlm_resource_putref(res);
1139                 }
1140         }
1141         return rc;
1142 }
1143
1144 static int ofd_orphans_destroy(const struct lu_env *env,
1145                                struct obd_export *exp,
1146                                struct ofd_device *ofd, struct obdo *oa)
1147 {
1148         struct ofd_thread_info  *info   = ofd_info(env);
1149         struct lu_fid           *fid    = &info->fti_fid;
1150         struct ost_id           *oi     = &oa->o_oi;
1151         struct ofd_seq          *oseq;
1152         obd_seq                  seq    = ostid_seq(oi);
1153         obd_id                   end_id = ostid_id(oi);
1154         obd_id                   last;
1155         obd_id                   oid;
1156         int                      skip_orphan;
1157         int                      rc     = 0;
1158
1159         ENTRY;
1160
1161         oseq = ofd_seq_get(ofd, seq);
1162         if (oseq == NULL) {
1163                 CERROR("%s: Can not find seq for "DOSTID"\n",
1164                        ofd_name(ofd), POSTID(oi));
1165                 RETURN(-EINVAL);
1166         }
1167
1168         *fid = oi->oi_fid;
1169         last = ofd_seq_last_oid(oseq);
1170         oid = last;
1171
1172         LASSERT(exp != NULL);
1173         skip_orphan = !!(exp_connect_flags(exp) & OBD_CONNECT_SKIP_ORPHAN);
1174
1175         if (OBD_FAIL_CHECK(OBD_FAIL_OST_NODESTROY))
1176                 goto done;
1177
1178         LCONSOLE(D_INFO, "%s: deleting orphan objects from "DOSTID
1179                  " to "DOSTID"\n", ofd_name(ofd), seq, end_id + 1, seq, last);
1180
1181         while (oid > end_id) {
1182                 rc = fid_set_id(fid, oid);
1183                 if (unlikely(rc != 0))
1184                         GOTO(out_put, rc);
1185
1186                 rc = ofd_destroy_by_fid(env, ofd, fid, 1);
1187                 if (rc != 0 && rc != -ENOENT && rc != -ESTALE &&
1188                     likely(rc != -EREMCHG && rc != -EINPROGRESS))
1189                         /* this is pretty fatal... */
1190                         CEMERG("%s: error destroying precreated id "
1191                                DFID": rc = %d\n",
1192                                ofd_name(ofd), PFID(fid), rc);
1193
1194                 oid--;
1195                 if (!skip_orphan) {
1196                         ofd_seq_last_oid_set(oseq, oid);
1197                         /* update last_id on disk periodically so that if we
1198                          * restart * we don't need to re-scan all of the just
1199                          * deleted objects. */
1200                         if ((oid & 511) == 0)
1201                                 ofd_seq_last_oid_write(env, ofd, oseq);
1202                 }
1203         }
1204
1205         CDEBUG(D_HA, "%s: after destroy: set last_id to "DOSTID"\n",
1206                ofd_name(ofd), seq, oid);
1207
1208 done:
1209         if (!skip_orphan) {
1210                 ofd_seq_last_oid_set(oseq, oid);
1211                 rc = ofd_seq_last_oid_write(env, ofd, oseq);
1212         } else {
1213                 /* don't reuse orphan object, return last used objid */
1214                 ostid_set_id(oi, last);
1215                 rc = 0;
1216         }
1217
1218         GOTO(out_put, rc);
1219
1220 out_put:
1221         ofd_seq_put(env, oseq);
1222         return rc;
1223 }
1224
1225 static int ofd_create_hdl(struct tgt_session_info *tsi)
1226 {
1227         struct ost_body         *repbody;
1228         const struct obdo       *oa = &tsi->tsi_ost_body->oa;
1229         struct obdo             *rep_oa;
1230         struct obd_export       *exp = tsi->tsi_exp;
1231         struct ofd_device       *ofd = ofd_exp(exp);
1232         obd_seq                  seq = ostid_seq(&oa->o_oi);
1233         obd_id                   oid = ostid_id(&oa->o_oi);
1234         struct ofd_seq          *oseq;
1235         int                      rc = 0, diff;
1236         int                      sync_trans = 0;
1237
1238         ENTRY;
1239
1240         if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
1241                 RETURN(-EROFS);
1242
1243         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1244         if (repbody == NULL)
1245                 RETURN(-ENOMEM);
1246
1247         down_read(&ofd->ofd_lastid_rwsem);
1248         /* Currently, for safe, we do not distinguish which LAST_ID is broken,
1249          * we may do that in the future.
1250          * Return -ENOSPC until the LAST_ID rebuilt. */
1251         if (unlikely(ofd->ofd_lastid_rebuilding))
1252                 GOTO(out_sem, rc = -ENOSPC);
1253
1254         rep_oa = &repbody->oa;
1255         rep_oa->o_oi = oa->o_oi;
1256
1257         LASSERT(seq >= FID_SEQ_OST_MDT0);
1258         LASSERT(oa->o_valid & OBD_MD_FLGROUP);
1259
1260         CDEBUG(D_INFO, "ofd_create("DOSTID")\n", POSTID(&oa->o_oi));
1261
1262         oseq = ofd_seq_load(tsi->tsi_env, ofd, seq);
1263         if (IS_ERR(oseq)) {
1264                 CERROR("%s: Can't find FID Sequence "LPX64": rc = %ld\n",
1265                        ofd_name(ofd), seq, PTR_ERR(oseq));
1266                 GOTO(out_sem, rc = -EINVAL);
1267         }
1268
1269         if ((oa->o_valid & OBD_MD_FLFLAGS) &&
1270             (oa->o_flags & OBD_FL_RECREATE_OBJS)) {
1271                 if (!ofd_obd(ofd)->obd_recovering ||
1272                     oid > ofd_seq_last_oid(oseq)) {
1273                         CERROR("%s: recreate objid "DOSTID" > last id "LPU64
1274                                "\n", ofd_name(ofd), POSTID(&oa->o_oi),
1275                                ofd_seq_last_oid(oseq));
1276                         GOTO(out_nolock, rc = -EINVAL);
1277                 }
1278                 /* Do nothing here, we re-create objects during recovery
1279                  * upon write replay, see ofd_preprw_write() */
1280                 GOTO(out_nolock, rc = 0);
1281         }
1282         /* former ofd_handle_precreate */
1283         if ((oa->o_valid & OBD_MD_FLFLAGS) &&
1284             (oa->o_flags & OBD_FL_DELORPHAN)) {
1285                 exp->exp_filter_data.fed_lastid_gen = ofd->ofd_lastid_gen;
1286
1287                 /* destroy orphans */
1288                 if (lustre_msg_get_conn_cnt(tgt_ses_req(tsi)->rq_reqmsg) <
1289                     exp->exp_conn_cnt) {
1290                         CERROR("%s: dropping old orphan cleanup request\n",
1291                                ofd_name(ofd));
1292                         GOTO(out_nolock, rc = 0);
1293                 }
1294                 /* This causes inflight precreates to abort and drop lock */
1295                 oseq->os_destroys_in_progress = 1;
1296                 mutex_lock(&oseq->os_create_lock);
1297                 if (!oseq->os_destroys_in_progress) {
1298                         CERROR("%s:["LPU64"] destroys_in_progress already"
1299                                " cleared\n", ofd_name(ofd), seq);
1300                         ostid_set_id(&rep_oa->o_oi, ofd_seq_last_oid(oseq));
1301                         GOTO(out, rc = 0);
1302                 }
1303                 diff = oid - ofd_seq_last_oid(oseq);
1304                 CDEBUG(D_HA, "ofd_last_id() = "LPU64" -> diff = %d\n",
1305                         ofd_seq_last_oid(oseq), diff);
1306                 if (-diff > OST_MAX_PRECREATE) {
1307                         /* FIXME: should reset precreate_next_id on MDS */
1308                         rc = 0;
1309                 } else if (diff < 0) {
1310                         rc = ofd_orphans_destroy(tsi->tsi_env, exp,
1311                                                  ofd, rep_oa);
1312                         oseq->os_destroys_in_progress = 0;
1313                 } else {
1314                         /* XXX: Used by MDS for the first time! */
1315                         oseq->os_destroys_in_progress = 0;
1316                 }
1317         } else {
1318                 if (unlikely(exp->exp_filter_data.fed_lastid_gen !=
1319                              ofd->ofd_lastid_gen)) {
1320                         ofd_obd_disconnect(exp);
1321                         GOTO(out_nolock, rc = -ENOTCONN);
1322                 }
1323
1324                 mutex_lock(&oseq->os_create_lock);
1325                 if (lustre_msg_get_conn_cnt(tgt_ses_req(tsi)->rq_reqmsg) <
1326                     exp->exp_conn_cnt) {
1327                         CERROR("%s: dropping old precreate request\n",
1328                                ofd_name(ofd));
1329                         GOTO(out, rc = 0);
1330                 }
1331                 /* only precreate if seq is 0, IDIF or normal and also o_id
1332                  * must be specfied */
1333                 if ((!fid_seq_is_mdt(seq) && !fid_seq_is_norm(seq) &&
1334                      !fid_seq_is_idif(seq)) || oid == 0) {
1335                         diff = 1; /* shouldn't we create this right now? */
1336                 } else {
1337                         diff = oid - ofd_seq_last_oid(oseq);
1338                         /* Do sync create if the seq is about to used up */
1339                         if (fid_seq_is_idif(seq) || fid_seq_is_mdt0(seq)) {
1340                                 if (unlikely(oid >= IDIF_MAX_OID - 1))
1341                                         sync_trans = 1;
1342                         } else if (fid_seq_is_norm(seq)) {
1343                                 if (unlikely(oid >=
1344                                              LUSTRE_DATA_SEQ_MAX_WIDTH - 1))
1345                                         sync_trans = 1;
1346                         } else {
1347                                 CERROR("%s : invalid o_seq "DOSTID"\n",
1348                                        ofd_name(ofd), POSTID(&oa->o_oi));
1349                                 GOTO(out, rc = -EINVAL);
1350                         }
1351                 }
1352         }
1353         if (diff > 0) {
1354                 cfs_time_t       enough_time = cfs_time_shift(DISK_TIMEOUT);
1355                 obd_id           next_id;
1356                 int              created = 0;
1357                 int              count;
1358
1359                 if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
1360                     !(oa->o_flags & OBD_FL_DELORPHAN)) {
1361                         /* don't enforce grant during orphan recovery */
1362                         rc = ofd_grant_create(tsi->tsi_env,
1363                                               ofd_obd(ofd)->obd_self_export,
1364                                               &diff);
1365                         if (rc) {
1366                                 CDEBUG(D_HA, "%s: failed to acquire grant "
1367                                        "space for precreate (%d): rc = %d\n",
1368                                        ofd_name(ofd), diff, rc);
1369                                 diff = 0;
1370                         }
1371                 }
1372
1373                 /* This can happen if a new OST is formatted and installed
1374                  * in place of an old one at the same index.  Instead of
1375                  * precreating potentially millions of deleted old objects
1376                  * (possibly filling the OST), only precreate the last batch.
1377                  * LFSCK will eventually clean up any orphans. LU-14 */
1378                 if (diff > 5 * OST_MAX_PRECREATE) {
1379                         diff = OST_MAX_PRECREATE / 2;
1380                         LCONSOLE_WARN("%s: precreate FID "DOSTID" is over %u "
1381                                       "larger than the LAST_ID "DOSTID", only "
1382                                       "precreating the last %u objects.\n",
1383                                       ofd_name(ofd), POSTID(&oa->o_oi),
1384                                       5 * OST_MAX_PRECREATE,
1385                                       POSTID(&oseq->os_oi), diff);
1386                         ofd_seq_last_oid_set(oseq, ostid_id(&oa->o_oi) - diff);
1387                 }
1388
1389                 while (diff > 0) {
1390                         next_id = ofd_seq_last_oid(oseq) + 1;
1391                         count = ofd_precreate_batch(ofd, diff);
1392
1393                         CDEBUG(D_HA, "%s: reserve %d objects in group "LPX64
1394                                " at "LPU64"\n", ofd_name(ofd),
1395                                count, seq, next_id);
1396
1397                         if (cfs_time_after(jiffies, enough_time)) {
1398                                 LCONSOLE_WARN("%s: Slow creates, %d/%d objects"
1399                                               " created at a rate of %d/s\n",
1400                                               ofd_name(ofd), created,
1401                                               diff + created,
1402                                               created / DISK_TIMEOUT);
1403                                 break;
1404                         }
1405
1406                         rc = ofd_precreate_objects(tsi->tsi_env, ofd, next_id,
1407                                                    oseq, count, sync_trans);
1408                         if (rc > 0) {
1409                                 created += rc;
1410                                 diff -= rc;
1411                         } else if (rc < 0) {
1412                                 break;
1413                         }
1414                 }
1415                 if (created > 0)
1416                         /* some objects got created, we can return
1417                          * them, even if last creation failed */
1418                         rc = 0;
1419                 else
1420                         CERROR("%s: unable to precreate: rc = %d\n",
1421                                ofd_name(ofd), rc);
1422
1423                 if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
1424                     !(oa->o_flags & OBD_FL_DELORPHAN))
1425                         ofd_grant_commit(tsi->tsi_env,
1426                                          ofd_obd(ofd)->obd_self_export, rc);
1427
1428                 ostid_set_id(&rep_oa->o_oi, ofd_seq_last_oid(oseq));
1429         }
1430         EXIT;
1431         ofd_counter_incr(exp, LPROC_OFD_STATS_CREATE,
1432                          tsi->tsi_jobid, 1);
1433 out:
1434         mutex_unlock(&oseq->os_create_lock);
1435 out_nolock:
1436         if (rc == 0)
1437                 rep_oa->o_valid |= OBD_MD_FLID | OBD_MD_FLGROUP;
1438
1439         ofd_seq_put(tsi->tsi_env, oseq);
1440
1441 out_sem:
1442         up_read(&ofd->ofd_lastid_rwsem);
1443         return rc;
1444 }
1445
1446 static int ofd_destroy_hdl(struct tgt_session_info *tsi)
1447 {
1448         const struct ost_body   *body = tsi->tsi_ost_body;
1449         struct ost_body         *repbody;
1450         struct ofd_device       *ofd = ofd_exp(tsi->tsi_exp);
1451         struct ofd_thread_info  *fti = tsi2ofd_info(tsi);
1452         struct lu_fid           *fid = &fti->fti_fid;
1453         obd_id                   oid;
1454         obd_count                count;
1455         int                      rc = 0;
1456
1457         ENTRY;
1458
1459         if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
1460                 RETURN(-EROFS);
1461
1462         /* This is old case for clients before Lustre 2.4 */
1463         /* If there's a DLM request, cancel the locks mentioned in it */
1464         if (req_capsule_field_present(tsi->tsi_pill, &RMF_DLM_REQ,
1465                                       RCL_CLIENT)) {
1466                 struct ldlm_request *dlm;
1467
1468                 dlm = req_capsule_client_get(tsi->tsi_pill, &RMF_DLM_REQ);
1469                 if (dlm == NULL)
1470                         RETURN(-EFAULT);
1471                 ldlm_request_cancel(tgt_ses_req(tsi), dlm, 0);
1472         }
1473
1474         *fid = body->oa.o_oi.oi_fid;
1475         oid = ostid_id(&body->oa.o_oi);
1476         LASSERT(oid != 0);
1477
1478         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1479
1480         /* check that o_misc makes sense */
1481         if (body->oa.o_valid & OBD_MD_FLOBJCOUNT)
1482                 count = body->oa.o_misc;
1483         else
1484                 count = 1; /* default case - single destroy */
1485
1486         CDEBUG(D_HA, "%s: Destroy object "DOSTID" count %d\n", ofd_name(ofd),
1487                POSTID(&body->oa.o_oi), count);
1488
1489         while (count > 0) {
1490                 int lrc;
1491
1492                 lrc = ofd_destroy_by_fid(tsi->tsi_env, ofd, fid, 0);
1493                 if (lrc == -ENOENT) {
1494                         CDEBUG(D_INODE,
1495                                "%s: destroying non-existent object "DFID"\n",
1496                                ofd_name(ofd), PFID(fid));
1497                         /* rewrite rc with -ENOENT only if it is 0 */
1498                         if (rc == 0)
1499                                 rc = lrc;
1500                 } else if (lrc != 0) {
1501                         CERROR("%s: error destroying object "DFID": %d\n",
1502                                ofd_name(ofd), PFID(fid), lrc);
1503                         rc = lrc;
1504                 }
1505
1506                 count--;
1507                 oid++;
1508                 lrc = fid_set_id(fid, oid);
1509                 if (unlikely(lrc != 0 && count > 0))
1510                         GOTO(out, rc = lrc);
1511         }
1512
1513         ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_DESTROY,
1514                          tsi->tsi_jobid, 1);
1515
1516         GOTO(out, rc);
1517
1518 out:
1519         fid_to_ostid(fid, &repbody->oa.o_oi);
1520         return rc;
1521 }
1522
1523 static int ofd_statfs_hdl(struct tgt_session_info *tsi)
1524 {
1525         struct obd_statfs       *osfs;
1526         int                      rc;
1527
1528         ENTRY;
1529
1530         osfs = req_capsule_server_get(tsi->tsi_pill, &RMF_OBD_STATFS);
1531
1532         rc = ofd_statfs(tsi->tsi_env, tsi->tsi_exp, osfs,
1533                         cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS), 0);
1534         if (rc != 0)
1535                 CERROR("%s: statfs failed: rc = %d\n",
1536                        tgt_name(tsi->tsi_tgt), rc);
1537
1538         if (OBD_FAIL_CHECK(OBD_FAIL_OST_STATFS_EINPROGRESS))
1539                 rc = -EINPROGRESS;
1540
1541         ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_STATFS,
1542                          tsi->tsi_jobid, 1);
1543
1544         RETURN(rc);
1545 }
1546
1547 static int ofd_sync_hdl(struct tgt_session_info *tsi)
1548 {
1549         struct ost_body         *body = tsi->tsi_ost_body;
1550         struct ost_body         *repbody;
1551         struct ofd_thread_info  *fti = tsi2ofd_info(tsi);
1552         struct ofd_device       *ofd = ofd_exp(tsi->tsi_exp);
1553         struct ofd_object       *fo = NULL;
1554         int                      rc = 0;
1555
1556         ENTRY;
1557
1558         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1559
1560         /* if no objid is specified, it means "sync whole filesystem" */
1561         if (!fid_is_zero(&tsi->tsi_fid)) {
1562                 fo = ofd_object_find_exists(tsi->tsi_env, ofd, &tsi->tsi_fid);
1563                 if (IS_ERR(fo))
1564                         RETURN(PTR_ERR(fo));
1565         }
1566
1567         rc = tgt_sync(tsi->tsi_env, tsi->tsi_tgt,
1568                       fo != NULL ? ofd_object_child(fo) : NULL);
1569         if (rc)
1570                 GOTO(put, rc);
1571
1572         ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_SYNC,
1573                          tsi->tsi_jobid, 1);
1574         if (fo == NULL)
1575                 RETURN(0);
1576
1577         repbody->oa.o_oi = body->oa.o_oi;
1578         repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1579
1580         rc = ofd_attr_get(tsi->tsi_env, fo, &fti->fti_attr);
1581         if (rc == 0)
1582                 obdo_from_la(&repbody->oa, &fti->fti_attr,
1583                              OFD_VALID_FLAGS);
1584         else
1585                 /* don't return rc from getattr */
1586                 rc = 0;
1587         EXIT;
1588 put:
1589         if (fo != NULL)
1590                 ofd_object_put(tsi->tsi_env, fo);
1591         return rc;
1592 }
1593
1594 static int ofd_punch_hdl(struct tgt_session_info *tsi)
1595 {
1596         const struct obdo       *oa = &tsi->tsi_ost_body->oa;
1597         struct ost_body         *repbody;
1598         struct ofd_thread_info  *info = tsi2ofd_info(tsi);
1599         struct ldlm_namespace   *ns = tsi->tsi_tgt->lut_obd->obd_namespace;
1600         struct ldlm_resource    *res;
1601         struct ofd_object       *fo;
1602         struct filter_fid       *ff = NULL;
1603         __u64                    flags = 0;
1604         struct lustre_handle     lh = { 0, };
1605         int                      rc;
1606         __u64                    start, end;
1607         bool                     srvlock;
1608
1609         ENTRY;
1610
1611         /* check that we do support OBD_CONNECT_TRUNCLOCK. */
1612         CLASSERT(OST_CONNECT_SUPPORTED & OBD_CONNECT_TRUNCLOCK);
1613
1614         if ((oa->o_valid & (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS)) !=
1615             (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS))
1616                 RETURN(err_serious(-EPROTO));
1617
1618         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1619         if (repbody == NULL)
1620                 RETURN(err_serious(-ENOMEM));
1621
1622         /* punch start,end are passed in o_size,o_blocks throught wire */
1623         start = oa->o_size;
1624         end = oa->o_blocks;
1625
1626         if (end != OBD_OBJECT_EOF) /* Only truncate is supported */
1627                 RETURN(-EPROTO);
1628
1629         /* standard truncate optimization: if file body is completely
1630          * destroyed, don't send data back to the server. */
1631         if (start == 0)
1632                 flags |= LDLM_FL_AST_DISCARD_DATA;
1633
1634         repbody->oa.o_oi = oa->o_oi;
1635         repbody->oa.o_valid = OBD_MD_FLID;
1636
1637         srvlock = oa->o_valid & OBD_MD_FLFLAGS &&
1638                   oa->o_flags & OBD_FL_SRVLOCK;
1639
1640         if (srvlock) {
1641                 rc = tgt_extent_lock(ns, &tsi->tsi_resid, start, end, &lh,
1642                                      LCK_PW, &flags);
1643                 if (rc != 0)
1644                         RETURN(rc);
1645         }
1646
1647         CDEBUG(D_INODE, "calling punch for object "DFID", valid = "LPX64
1648                ", start = "LPD64", end = "LPD64"\n", PFID(&tsi->tsi_fid),
1649                oa->o_valid, start, end);
1650
1651         fo = ofd_object_find_exists(tsi->tsi_env, ofd_exp(tsi->tsi_exp),
1652                                     &tsi->tsi_fid);
1653         if (IS_ERR(fo))
1654                 GOTO(out, rc = PTR_ERR(fo));
1655
1656         la_from_obdo(&info->fti_attr, oa,
1657                      OBD_MD_FLMTIME | OBD_MD_FLATIME | OBD_MD_FLCTIME);
1658         info->fti_attr.la_size = start;
1659         info->fti_attr.la_valid |= LA_SIZE;
1660
1661         if (oa->o_valid & OBD_MD_FLFID) {
1662                 ff = &info->fti_mds_fid;
1663                 ofd_prepare_fidea(ff, oa);
1664         }
1665
1666         rc = ofd_object_punch(tsi->tsi_env, fo, start, end, &info->fti_attr,
1667                               ff, (struct obdo *)oa);
1668         if (rc)
1669                 GOTO(out_put, rc);
1670
1671         ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_PUNCH,
1672                          tsi->tsi_jobid, 1);
1673         EXIT;
1674 out_put:
1675         ofd_object_put(tsi->tsi_env, fo);
1676 out:
1677         if (srvlock)
1678                 tgt_extent_unlock(&lh, LCK_PW);
1679         if (rc == 0) {
1680                 /* we do not call this before to avoid lu_object_find() in
1681                  *  ->lvbo_update() holding another reference on the object.
1682                  * otherwise concurrent destroy can make the object unavailable
1683                  * for 2nd lu_object_find() waiting for the first reference
1684                  * to go... deadlock! */
1685                 res = ldlm_resource_get(ns, NULL, &tsi->tsi_resid,
1686                                         LDLM_EXTENT, 0);
1687                 if (res != NULL) {
1688                         ldlm_res_lvbo_update(res, NULL, 0);
1689                         ldlm_resource_putref(res);
1690                 }
1691         }
1692         return rc;
1693 }
1694
1695 static int ofd_quotactl(struct tgt_session_info *tsi)
1696 {
1697         struct obd_quotactl     *oqctl, *repoqc;
1698         int                      rc;
1699
1700         ENTRY;
1701
1702         oqctl = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_QUOTACTL);
1703         if (oqctl == NULL)
1704                 RETURN(err_serious(-EPROTO));
1705
1706         repoqc = req_capsule_server_get(tsi->tsi_pill, &RMF_OBD_QUOTACTL);
1707         if (repoqc == NULL)
1708                 RETURN(err_serious(-ENOMEM));
1709
1710         /* report success for quota on/off for interoperability with current MDT
1711          * stack */
1712         if (oqctl->qc_cmd == Q_QUOTAON || oqctl->qc_cmd == Q_QUOTAOFF)
1713                 RETURN(0);
1714
1715         *repoqc = *oqctl;
1716         rc = lquotactl_slv(tsi->tsi_env, tsi->tsi_tgt->lut_bottom, repoqc);
1717
1718         ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_QUOTACTL,
1719                          tsi->tsi_jobid, 1);
1720
1721         RETURN(rc);
1722 }
1723
1724 /* High priority request handlers for OFD */
1725
1726 /* prolong locks for the current service time of the corresponding
1727  * portal (= OST_IO_PORTAL)
1728  */
1729 static inline int prolong_timeout(struct ptlrpc_request *req)
1730 {
1731         struct ptlrpc_service_part *svcpt = req->rq_rqbd->rqbd_svcpt;
1732
1733         if (AT_OFF)
1734                 return obd_timeout / 2;
1735
1736         return max(at_est2timeout(at_get(&svcpt->scp_at_estimate)),
1737                    ldlm_timeout);
1738 }
1739
1740 static int ofd_prolong_one_lock(struct tgt_session_info *tsi,
1741                                 struct ldlm_lock *lock,
1742                                 struct ldlm_extent *extent, int timeout)
1743 {
1744
1745         if (lock->l_flags & LDLM_FL_DESTROYED) /* lock already cancelled */
1746                 return 0;
1747
1748         /* XXX: never try to grab resource lock here because we're inside
1749          * exp_bl_list_lock; in ldlm_lockd.c to handle waiting list we take
1750          * res lock and then exp_bl_list_lock. */
1751
1752         if (!(lock->l_flags & LDLM_FL_AST_SENT))
1753                 /* ignore locks not being cancelled */
1754                 return 0;
1755
1756         LDLM_DEBUG(lock, "refreshed for req x"LPU64" ext("LPU64"->"LPU64") "
1757                          "to %ds.\n", tgt_ses_req(tsi)->rq_xid, extent->start,
1758                          extent->end, timeout);
1759
1760         /* OK. this is a possible lock the user holds doing I/O
1761          * let's refresh eviction timer for it */
1762         ldlm_refresh_waiting_lock(lock, timeout);
1763         return 1;
1764 }
1765
1766 static int ofd_prolong_extent_locks(struct tgt_session_info *tsi,
1767                                     __u64 start, __u64 end)
1768 {
1769         struct obd_export       *exp = tsi->tsi_exp;
1770         struct obdo             *oa  = &tsi->tsi_ost_body->oa;
1771         struct ldlm_extent       extent = {
1772                 .start = start,
1773                 .end = end
1774         };
1775         struct ldlm_lock        *lock;
1776         int                      timeout = prolong_timeout(tgt_ses_req(tsi));
1777         int                      lock_count = 0;
1778
1779         ENTRY;
1780
1781         if (oa->o_valid & OBD_MD_FLHANDLE) {
1782                 /* mostly a request should be covered by only one lock, try
1783                  * fast path. */
1784                 lock = ldlm_handle2lock(&oa->o_handle);
1785                 if (lock != NULL) {
1786                         /* Fast path to check if the lock covers the whole IO
1787                          * region exclusively. */
1788                         if (lock->l_granted_mode == LCK_PW &&
1789                             ldlm_extent_contain(&lock->l_policy_data.l_extent,
1790                                                 &extent)) {
1791                                 /* bingo */
1792                                 LASSERT(lock->l_export == exp);
1793                                 lock_count = ofd_prolong_one_lock(tsi, lock,
1794                                                              &extent, timeout);
1795                                 LDLM_LOCK_PUT(lock);
1796                                 RETURN(lock_count);
1797                         }
1798                         LDLM_LOCK_PUT(lock);
1799                 }
1800         }
1801
1802         spin_lock_bh(&exp->exp_bl_list_lock);
1803         list_for_each_entry(lock, &exp->exp_bl_list, l_exp_list) {
1804                 LASSERT(lock->l_flags & LDLM_FL_AST_SENT);
1805                 LASSERT(lock->l_resource->lr_type == LDLM_EXTENT);
1806
1807                 if (!ldlm_res_eq(&tsi->tsi_resid, &lock->l_resource->lr_name))
1808                         continue;
1809
1810                 if (!ldlm_extent_overlap(&lock->l_policy_data.l_extent,
1811                                          &extent))
1812                         continue;
1813
1814                 lock_count += ofd_prolong_one_lock(tsi, lock, &extent, timeout);
1815         }
1816         spin_unlock_bh(&exp->exp_bl_list_lock);
1817
1818         RETURN(lock_count);
1819 }
1820
1821 /**
1822  * Returns 1 if the given PTLRPC matches the given LDLM lock, or 0 if it does
1823  * not.
1824  */
1825 static int ofd_rw_hpreq_lock_match(struct ptlrpc_request *req,
1826                                    struct ldlm_lock *lock)
1827 {
1828         struct niobuf_remote    *rnb;
1829         struct obd_ioobj        *ioo;
1830         ldlm_mode_t              mode;
1831         struct ldlm_extent       ext;
1832         __u32                    opc = lustre_msg_get_opc(req->rq_reqmsg);
1833
1834         ENTRY;
1835
1836         ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
1837         LASSERT(ioo != NULL);
1838
1839         rnb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
1840         LASSERT(rnb != NULL);
1841
1842         ext.start = rnb->offset;
1843         rnb += ioo->ioo_bufcnt - 1;
1844         ext.end = rnb->offset + rnb->len - 1;
1845
1846         LASSERT(lock->l_resource != NULL);
1847         if (!ostid_res_name_eq(&ioo->ioo_oid, &lock->l_resource->lr_name))
1848                 RETURN(0);
1849
1850         mode = LCK_PW;
1851         if (opc == OST_READ)
1852                 mode |= LCK_PR;
1853
1854         if (!(lock->l_granted_mode & mode))
1855                 RETURN(0);
1856
1857         RETURN(ldlm_extent_overlap(&lock->l_policy_data.l_extent, &ext));
1858 }
1859
1860 /**
1861  * High-priority queue request check for whether the given PTLRPC request
1862  * (\a req) is blocking an LDLM lock cancel.
1863  *
1864  * Returns 1 if the given given PTLRPC request (\a req) is blocking an LDLM lock
1865  * cancel, 0 if it is not, and -EFAULT if the request is malformed.
1866  *
1867  * Only OST_READs, OST_WRITEs and OST_PUNCHes go on the h-p RPC queue.  This
1868  * function looks only at OST_READs and OST_WRITEs.
1869  */
1870 static int ofd_rw_hpreq_check(struct ptlrpc_request *req)
1871 {
1872         struct tgt_session_info *tsi;
1873         struct obd_ioobj        *ioo;
1874         struct niobuf_remote    *rnb;
1875         __u64                    start, end;
1876         int                      lock_count;
1877
1878         ENTRY;
1879
1880         /* Don't use tgt_ses_info() to get session info, because lock_match()
1881          * can be called while request has no processing thread yet. */
1882         tsi = lu_context_key_get(&req->rq_session, &tgt_session_key);
1883         LASSERT(tsi != NULL);
1884
1885         /*
1886          * Use LASSERT below because malformed RPCs should have
1887          * been filtered out in tgt_hpreq_handler().
1888          */
1889         ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
1890         LASSERT(ioo != NULL);
1891
1892         rnb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
1893         LASSERT(rnb != NULL);
1894         LASSERT(!(rnb->flags & OBD_BRW_SRVLOCK));
1895
1896         start = rnb->offset;
1897         rnb += ioo->ioo_bufcnt - 1;
1898         end = rnb->offset + rnb->len - 1;
1899
1900         DEBUG_REQ(D_RPCTRACE, req, "%s %s: refresh rw locks: "DFID
1901                                    " ("LPU64"->"LPU64")\n",
1902                   tgt_name(tsi->tsi_tgt), current->comm,
1903                   PFID(&tsi->tsi_fid), start, end);
1904
1905         lock_count = ofd_prolong_extent_locks(tsi, start, end);
1906
1907         CDEBUG(D_DLMTRACE, "%s: refreshed %u locks timeout for req %p.\n",
1908                tgt_name(tsi->tsi_tgt), lock_count, req);
1909
1910         RETURN(lock_count > 0);
1911 }
1912
1913 static void ofd_rw_hpreq_fini(struct ptlrpc_request *req)
1914 {
1915         ofd_rw_hpreq_check(req);
1916 }
1917
1918 /**
1919  * Like tgt_rw_hpreq_lock_match(), but for OST_PUNCH RPCs.
1920  */
1921 static int ofd_punch_hpreq_lock_match(struct ptlrpc_request *req,
1922                                       struct ldlm_lock *lock)
1923 {
1924         struct tgt_session_info *tsi;
1925
1926         /* Don't use tgt_ses_info() to get session info, because lock_match()
1927          * can be called while request has no processing thread yet. */
1928         tsi = lu_context_key_get(&req->rq_session, &tgt_session_key);
1929         LASSERT(tsi != NULL);
1930
1931         LASSERT(tsi->tsi_ost_body != NULL);
1932         if (tsi->tsi_ost_body->oa.o_valid & OBD_MD_FLHANDLE &&
1933             tsi->tsi_ost_body->oa.o_handle.cookie == lock->l_handle.h_cookie)
1934                 return 1;
1935
1936         return 0;
1937 }
1938
1939 /**
1940  * Like ost_rw_hpreq_check(), but for OST_PUNCH RPCs.
1941  */
1942 static int ofd_punch_hpreq_check(struct ptlrpc_request *req)
1943 {
1944         struct tgt_session_info *tsi;
1945         struct obdo             *oa;
1946         int                      lock_count;
1947
1948         ENTRY;
1949
1950         /* Don't use tgt_ses_info() to get session info, because lock_match()
1951          * can be called while request has no processing thread yet. */
1952         tsi = lu_context_key_get(&req->rq_session, &tgt_session_key);
1953         LASSERT(tsi != NULL);
1954         oa = &tsi->tsi_ost_body->oa;
1955
1956         LASSERT(!(oa->o_valid & OBD_MD_FLFLAGS &&
1957                   oa->o_flags & OBD_FL_SRVLOCK));
1958
1959         CDEBUG(D_DLMTRACE,
1960                "%s: refresh locks: "LPU64"/"LPU64" ("LPU64"->"LPU64")\n",
1961                tgt_name(tsi->tsi_tgt), tsi->tsi_resid.name[0],
1962                tsi->tsi_resid.name[1], oa->o_size, oa->o_blocks);
1963
1964         lock_count = ofd_prolong_extent_locks(tsi, oa->o_size, oa->o_blocks);
1965
1966         CDEBUG(D_DLMTRACE, "%s: refreshed %u locks timeout for req %p.\n",
1967                tgt_name(tsi->tsi_tgt), lock_count, req);
1968
1969         RETURN(lock_count > 0);
1970 }
1971
1972 static void ofd_punch_hpreq_fini(struct ptlrpc_request *req)
1973 {
1974         ofd_punch_hpreq_check(req);
1975 }
1976
1977 struct ptlrpc_hpreq_ops ofd_hpreq_rw = {
1978         .hpreq_lock_match       = ofd_rw_hpreq_lock_match,
1979         .hpreq_check            = ofd_rw_hpreq_check,
1980         .hpreq_fini             = ofd_rw_hpreq_fini
1981 };
1982
1983 struct ptlrpc_hpreq_ops ofd_hpreq_punch = {
1984         .hpreq_lock_match       = ofd_punch_hpreq_lock_match,
1985         .hpreq_check            = ofd_punch_hpreq_check,
1986         .hpreq_fini             = ofd_punch_hpreq_fini
1987 };
1988
1989 /** Assign high priority operations to the IO requests */
1990 static void ofd_hp_brw(struct tgt_session_info *tsi)
1991 {
1992         struct niobuf_remote    *rnb;
1993         struct obd_ioobj        *ioo;
1994
1995         ENTRY;
1996
1997         ioo = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_IOOBJ);
1998         LASSERT(ioo != NULL); /* must exist after request preprocessing */
1999         if (ioo->ioo_bufcnt > 0) {
2000                 rnb = req_capsule_client_get(tsi->tsi_pill, &RMF_NIOBUF_REMOTE);
2001                 LASSERT(rnb != NULL); /* must exist after request preprocessing */
2002
2003                 /* no high priority if server lock is needed */
2004                 if (rnb->flags & OBD_BRW_SRVLOCK)
2005                         return;
2006         }
2007         tgt_ses_req(tsi)->rq_ops = &ofd_hpreq_rw;
2008 }
2009
2010 static void ofd_hp_punch(struct tgt_session_info *tsi)
2011 {
2012         LASSERT(tsi->tsi_ost_body != NULL); /* must exists if we are here */
2013         /* no high-priority if server lock is needed */
2014         if (tsi->tsi_ost_body->oa.o_valid & OBD_MD_FLFLAGS &&
2015             tsi->tsi_ost_body->oa.o_flags & OBD_FL_SRVLOCK)
2016                 return;
2017         tgt_ses_req(tsi)->rq_ops = &ofd_hpreq_punch;
2018 }
2019
2020 #define OBD_FAIL_OST_READ_NET   OBD_FAIL_OST_BRW_NET
2021 #define OBD_FAIL_OST_WRITE_NET  OBD_FAIL_OST_BRW_NET
2022 #define OST_BRW_READ    OST_READ
2023 #define OST_BRW_WRITE   OST_WRITE
2024
2025 static struct tgt_handler ofd_tgt_handlers[] = {
2026 TGT_RPC_HANDLER(OST_FIRST_OPC,
2027                 0,                      OST_CONNECT,    tgt_connect,
2028                 &RQF_CONNECT, LUSTRE_OBD_VERSION),
2029 TGT_RPC_HANDLER(OST_FIRST_OPC,
2030                 0,                      OST_DISCONNECT, tgt_disconnect,
2031                 &RQF_OST_DISCONNECT, LUSTRE_OBD_VERSION),
2032 TGT_RPC_HANDLER(OST_FIRST_OPC,
2033                 0,                      OST_SET_INFO,   ofd_set_info_hdl,
2034                 &RQF_OBD_SET_INFO, LUSTRE_OST_VERSION),
2035 TGT_OST_HDL(0,                          OST_GET_INFO,   ofd_get_info_hdl),
2036 TGT_OST_HDL(HABEO_CORPUS| HABEO_REFERO, OST_GETATTR,    ofd_getattr_hdl),
2037 TGT_OST_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR,
2038                                         OST_SETATTR,    ofd_setattr_hdl),
2039 TGT_OST_HDL(0           | HABEO_REFERO | MUTABOR,
2040                                         OST_CREATE,     ofd_create_hdl),
2041 TGT_OST_HDL(0           | HABEO_REFERO | MUTABOR,
2042                                         OST_DESTROY,    ofd_destroy_hdl),
2043 TGT_OST_HDL(0           | HABEO_REFERO, OST_STATFS,     ofd_statfs_hdl),
2044 TGT_OST_HDL_HP(HABEO_CORPUS| HABEO_REFERO,
2045                                         OST_BRW_READ,   tgt_brw_read,
2046                                                         ofd_hp_brw),
2047 /* don't set CORPUS flag for brw_write because -ENOENT may be valid case */
2048 TGT_OST_HDL_HP(HABEO_CORPUS| MUTABOR,   OST_BRW_WRITE,  tgt_brw_write,
2049                                                         ofd_hp_brw),
2050 TGT_OST_HDL_HP(HABEO_CORPUS| HABEO_REFERO | MUTABOR,
2051                                         OST_PUNCH,      ofd_punch_hdl,
2052                                                         ofd_hp_punch),
2053 TGT_OST_HDL(HABEO_CORPUS| HABEO_REFERO, OST_SYNC,       ofd_sync_hdl),
2054 TGT_OST_HDL(0           | HABEO_REFERO, OST_QUOTACTL,   ofd_quotactl),
2055 };
2056
2057 static struct tgt_opc_slice ofd_common_slice[] = {
2058         {
2059                 .tos_opc_start  = OST_FIRST_OPC,
2060                 .tos_opc_end    = OST_LAST_OPC,
2061                 .tos_hs         = ofd_tgt_handlers
2062         },
2063         {
2064                 .tos_opc_start  = OBD_FIRST_OPC,
2065                 .tos_opc_end    = OBD_LAST_OPC,
2066                 .tos_hs         = tgt_obd_handlers
2067         },
2068         {
2069                 .tos_opc_start  = LDLM_FIRST_OPC,
2070                 .tos_opc_end    = LDLM_LAST_OPC,
2071                 .tos_hs         = tgt_dlm_handlers
2072         },
2073         {
2074                 .tos_opc_start  = OUT_UPDATE_FIRST_OPC,
2075                 .tos_opc_end    = OUT_UPDATE_LAST_OPC,
2076                 .tos_hs         = tgt_out_handlers
2077         },
2078         {
2079                 .tos_opc_start  = SEQ_FIRST_OPC,
2080                 .tos_opc_end    = SEQ_LAST_OPC,
2081                 .tos_hs         = seq_handlers
2082         },
2083         {
2084                 .tos_opc_start  = LFSCK_FIRST_OPC,
2085                 .tos_opc_end    = LFSCK_LAST_OPC,
2086                 .tos_hs         = tgt_lfsck_handlers
2087         },
2088         {
2089                 .tos_hs         = NULL
2090         }
2091 };
2092
2093 static int ofd_init0(const struct lu_env *env, struct ofd_device *m,
2094                      struct lu_device_type *ldt, struct lustre_cfg *cfg)
2095 {
2096         const char              *dev = lustre_cfg_string(cfg, 0);
2097         struct ofd_thread_info  *info = NULL;
2098         struct obd_device       *obd;
2099         struct obd_statfs       *osfs;
2100         int                      rc;
2101
2102         ENTRY;
2103
2104         obd = class_name2obd(dev);
2105         if (obd == NULL) {
2106                 CERROR("Cannot find obd with name %s\n", dev);
2107                 RETURN(-ENODEV);
2108         }
2109
2110         rc = lu_env_refill((struct lu_env *)env);
2111         if (rc != 0)
2112                 RETURN(rc);
2113
2114         obd->u.obt.obt_magic = OBT_MAGIC;
2115
2116         m->ofd_fmd_max_num = OFD_FMD_MAX_NUM_DEFAULT;
2117         m->ofd_fmd_max_age = OFD_FMD_MAX_AGE_DEFAULT;
2118
2119         spin_lock_init(&m->ofd_flags_lock);
2120         m->ofd_raid_degraded = 0;
2121         m->ofd_syncjournal = 0;
2122         ofd_slc_set(m);
2123         m->ofd_grant_compat_disable = 0;
2124         m->ofd_soft_sync_limit = OFD_SOFT_SYNC_LIMIT_DEFAULT;
2125
2126         /* statfs data */
2127         spin_lock_init(&m->ofd_osfs_lock);
2128         m->ofd_osfs_age = cfs_time_shift_64(-1000);
2129         m->ofd_osfs_unstable = 0;
2130         m->ofd_statfs_inflight = 0;
2131         m->ofd_osfs_inflight = 0;
2132
2133         /* grant data */
2134         spin_lock_init(&m->ofd_grant_lock);
2135         m->ofd_tot_dirty = 0;
2136         m->ofd_tot_granted = 0;
2137         m->ofd_tot_pending = 0;
2138         m->ofd_seq_count = 0;
2139         init_waitqueue_head(&m->ofd_inconsistency_thread.t_ctl_waitq);
2140         INIT_LIST_HEAD(&m->ofd_inconsistency_list);
2141         spin_lock_init(&m->ofd_inconsistency_lock);
2142
2143         spin_lock_init(&m->ofd_batch_lock);
2144         rwlock_init(&obd->u.filter.fo_sptlrpc_lock);
2145         sptlrpc_rule_set_init(&obd->u.filter.fo_sptlrpc_rset);
2146         init_rwsem(&m->ofd_lastid_rwsem);
2147
2148         obd->u.filter.fo_fl_oss_capa = 0;
2149         CFS_INIT_LIST_HEAD(&obd->u.filter.fo_capa_keys);
2150         obd->u.filter.fo_capa_hash = init_capa_hash();
2151         if (obd->u.filter.fo_capa_hash == NULL)
2152                 RETURN(-ENOMEM);
2153
2154         m->ofd_dt_dev.dd_lu_dev.ld_ops = &ofd_lu_ops;
2155         m->ofd_dt_dev.dd_lu_dev.ld_obd = obd;
2156         /* set this lu_device to obd, because error handling need it */
2157         obd->obd_lu_dev = &m->ofd_dt_dev.dd_lu_dev;
2158
2159         rc = ofd_procfs_init(m);
2160         if (rc) {
2161                 CERROR("Can't init ofd lprocfs, rc %d\n", rc);
2162                 RETURN(rc);
2163         }
2164
2165         /* No connection accepted until configurations will finish */
2166         spin_lock(&obd->obd_dev_lock);
2167         obd->obd_no_conn = 1;
2168         spin_unlock(&obd->obd_dev_lock);
2169         obd->obd_replayable = 1;
2170         if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
2171                 char *str = lustre_cfg_string(cfg, 4);
2172
2173                 if (strchr(str, 'n')) {
2174                         CWARN("%s: recovery disabled\n", obd->obd_name);
2175                         obd->obd_replayable = 0;
2176                 }
2177         }
2178
2179         info = ofd_info_init(env, NULL);
2180         if (info == NULL)
2181                 RETURN(-EFAULT);
2182
2183         rc = ofd_stack_init(env, m, cfg);
2184         if (rc) {
2185                 CERROR("Can't init device stack, rc %d\n", rc);
2186                 GOTO(err_fini_proc, rc);
2187         }
2188
2189         ofd_procfs_add_brw_stats_symlink(m);
2190
2191         /* populate cached statfs data */
2192         osfs = &ofd_info(env)->fti_u.osfs;
2193         rc = ofd_statfs_internal(env, m, osfs, 0, NULL);
2194         if (rc != 0) {
2195                 CERROR("%s: can't get statfs data, rc %d\n", obd->obd_name, rc);
2196                 GOTO(err_fini_stack, rc);
2197         }
2198         if (!IS_PO2(osfs->os_bsize)) {
2199                 CERROR("%s: blocksize (%d) is not a power of 2\n",
2200                                 obd->obd_name, osfs->os_bsize);
2201                 GOTO(err_fini_stack, rc = -EPROTO);
2202         }
2203         m->ofd_blockbits = fls(osfs->os_bsize) - 1;
2204
2205         m->ofd_precreate_batch = OFD_PRECREATE_BATCH_DEFAULT;
2206         if (osfs->os_bsize * osfs->os_blocks < OFD_PRECREATE_SMALL_FS)
2207                 m->ofd_precreate_batch = OFD_PRECREATE_BATCH_SMALL;
2208
2209         snprintf(info->fti_u.name, sizeof(info->fti_u.name), "%s-%s",
2210                  "filter"/*LUSTRE_OST_NAME*/, obd->obd_uuid.uuid);
2211         m->ofd_namespace = ldlm_namespace_new(obd, info->fti_u.name,
2212                                               LDLM_NAMESPACE_SERVER,
2213                                               LDLM_NAMESPACE_GREEDY,
2214                                               LDLM_NS_TYPE_OST);
2215         if (m->ofd_namespace == NULL)
2216                 GOTO(err_fini_stack, rc = -ENOMEM);
2217         /* set obd_namespace for compatibility with old code */
2218         obd->obd_namespace = m->ofd_namespace;
2219         ldlm_register_intent(m->ofd_namespace, ofd_intent_policy);
2220         m->ofd_namespace->ns_lvbo = &ofd_lvbo;
2221         m->ofd_namespace->ns_lvbp = m;
2222
2223         ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
2224                            "filter_ldlm_cb_client", &obd->obd_ldlm_client);
2225
2226         dt_conf_get(env, m->ofd_osd, &m->ofd_dt_conf);
2227
2228         /* Allow at most ddp_grant_reserved% of the available filesystem space
2229          * to be granted to clients, so that any errors in the grant overhead
2230          * calculations do not allow granting more space to clients than can be
2231          * written. Assumes that in aggregate the grant overhead calculations do
2232          * not have more than ddp_grant_reserved% estimation error in them. */
2233         m->ofd_grant_ratio =
2234                 ofd_grant_ratio_conv(m->ofd_dt_conf.ddp_grant_reserved);
2235
2236         rc = tgt_init(env, &m->ofd_lut, obd, m->ofd_osd, ofd_common_slice,
2237                       OBD_FAIL_OST_ALL_REQUEST_NET,
2238                       OBD_FAIL_OST_ALL_REPLY_NET);
2239         if (rc)
2240                 GOTO(err_free_ns, rc);
2241
2242         rc = ofd_fs_setup(env, m, obd);
2243         if (rc)
2244                 GOTO(err_fini_lut, rc);
2245
2246         rc = ofd_start_inconsistency_verification_thread(m);
2247         if (rc != 0)
2248                 GOTO(err_fini_fs, rc);
2249
2250         RETURN(0);
2251
2252 err_fini_fs:
2253         ofd_fs_cleanup(env, m);
2254 err_fini_lut:
2255         tgt_fini(env, &m->ofd_lut);
2256 err_free_ns:
2257         ldlm_namespace_free(m->ofd_namespace, 0, obd->obd_force);
2258         obd->obd_namespace = m->ofd_namespace = NULL;
2259 err_fini_stack:
2260         ofd_stack_fini(env, m, &m->ofd_osd->dd_lu_dev);
2261 err_fini_proc:
2262         ofd_procfs_fini(m);
2263         return rc;
2264 }
2265
2266 static void ofd_fini(const struct lu_env *env, struct ofd_device *m)
2267 {
2268         struct obd_device       *obd = ofd_obd(m);
2269         struct lu_device        *d   = &m->ofd_dt_dev.dd_lu_dev;
2270         struct lfsck_stop        stop;
2271
2272         stop.ls_status = LS_PAUSED;
2273         stop.ls_flags = 0;
2274         lfsck_stop(env, m->ofd_osd, &stop);
2275         target_recovery_fini(obd);
2276         obd_exports_barrier(obd);
2277         obd_zombie_barrier();
2278
2279         tgt_fini(env, &m->ofd_lut);
2280         ofd_stop_inconsistency_verification_thread(m);
2281         lfsck_degister(env, m->ofd_osd);
2282         ofd_fs_cleanup(env, m);
2283
2284         ofd_free_capa_keys(m);
2285         cleanup_capa_hash(obd->u.filter.fo_capa_hash);
2286
2287         if (m->ofd_namespace != NULL) {
2288                 ldlm_namespace_free(m->ofd_namespace, NULL,
2289                                     d->ld_obd->obd_force);
2290                 d->ld_obd->obd_namespace = m->ofd_namespace = NULL;
2291         }
2292
2293         ofd_stack_fini(env, m, &m->ofd_dt_dev.dd_lu_dev);
2294         ofd_procfs_fini(m);
2295         LASSERT(atomic_read(&d->ld_ref) == 0);
2296         server_put_mount(obd->obd_name);
2297         EXIT;
2298 }
2299
2300 static struct lu_device *ofd_device_fini(const struct lu_env *env,
2301                                          struct lu_device *d)
2302 {
2303         ENTRY;
2304         ofd_fini(env, ofd_dev(d));
2305         RETURN(NULL);
2306 }
2307
2308 static struct lu_device *ofd_device_free(const struct lu_env *env,
2309                                          struct lu_device *d)
2310 {
2311         struct ofd_device *m = ofd_dev(d);
2312
2313         dt_device_fini(&m->ofd_dt_dev);
2314         OBD_FREE_PTR(m);
2315         RETURN(NULL);
2316 }
2317
2318 static struct lu_device *ofd_device_alloc(const struct lu_env *env,
2319                                           struct lu_device_type *t,
2320                                           struct lustre_cfg *cfg)
2321 {
2322         struct ofd_device *m;
2323         struct lu_device  *l;
2324         int                rc;
2325
2326         OBD_ALLOC_PTR(m);
2327         if (m == NULL)
2328                 return ERR_PTR(-ENOMEM);
2329
2330         l = &m->ofd_dt_dev.dd_lu_dev;
2331         dt_device_init(&m->ofd_dt_dev, t);
2332         rc = ofd_init0(env, m, t, cfg);
2333         if (rc != 0) {
2334                 ofd_device_free(env, l);
2335                 l = ERR_PTR(rc);
2336         }
2337
2338         return l;
2339 }
2340
2341 /* thread context key constructor/destructor */
2342 LU_KEY_INIT_FINI(ofd, struct ofd_thread_info);
2343
2344 static void ofd_key_exit(const struct lu_context *ctx,
2345                          struct lu_context_key *key, void *data)
2346 {
2347         struct ofd_thread_info *info = data;
2348
2349         info->fti_env = NULL;
2350         info->fti_exp = NULL;
2351
2352         info->fti_xid = 0;
2353         info->fti_pre_version = 0;
2354         info->fti_used = 0;
2355
2356         memset(&info->fti_attr, 0, sizeof info->fti_attr);
2357 }
2358
2359 struct lu_context_key ofd_thread_key = {
2360         .lct_tags = LCT_DT_THREAD,
2361         .lct_init = ofd_key_init,
2362         .lct_fini = ofd_key_fini,
2363         .lct_exit = ofd_key_exit
2364 };
2365
2366 /* type constructor/destructor: mdt_type_init, mdt_type_fini */
2367 LU_TYPE_INIT_FINI(ofd, &ofd_thread_key);
2368
2369 static struct lu_device_type_operations ofd_device_type_ops = {
2370         .ldto_init              = ofd_type_init,
2371         .ldto_fini              = ofd_type_fini,
2372
2373         .ldto_start             = ofd_type_start,
2374         .ldto_stop              = ofd_type_stop,
2375
2376         .ldto_device_alloc      = ofd_device_alloc,
2377         .ldto_device_free       = ofd_device_free,
2378         .ldto_device_fini       = ofd_device_fini
2379 };
2380
2381 static struct lu_device_type ofd_device_type = {
2382         .ldt_tags       = LU_DEVICE_DT,
2383         .ldt_name       = LUSTRE_OST_NAME,
2384         .ldt_ops        = &ofd_device_type_ops,
2385         .ldt_ctx_tags   = LCT_DT_THREAD
2386 };
2387
2388 int __init ofd_init(void)
2389 {
2390         struct lprocfs_static_vars      lvars;
2391         int                             rc;
2392
2393         rc = lu_kmem_init(ofd_caches);
2394         if (rc)
2395                 return rc;
2396
2397         rc = ofd_fmd_init();
2398         if (rc) {
2399                 lu_kmem_fini(ofd_caches);
2400                 return(rc);
2401         }
2402
2403         lprocfs_ofd_init_vars(&lvars);
2404
2405         rc = class_register_type(&ofd_obd_ops, NULL, NULL,
2406 #ifndef HAVE_ONLY_PROCFS_SEQ
2407                                 lvars.module_vars,
2408 #endif
2409                                 LUSTRE_OST_NAME, &ofd_device_type);
2410         return rc;
2411 }
2412
2413 void __exit ofd_exit(void)
2414 {
2415         ofd_fmd_exit();
2416         lu_kmem_fini(ofd_caches);
2417         class_unregister_type(LUSTRE_OST_NAME);
2418 }
2419
2420 MODULE_AUTHOR("Whamcloud, Inc. <http://www.whamcloud.com/>");
2421 MODULE_DESCRIPTION("Lustre Object Filtering Device");
2422 MODULE_LICENSE("GPL");
2423
2424 module_init(ofd_init);
2425 module_exit(ofd_exit);