lustre/ofd/ofd_dev.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2009, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2012, 2013, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/ofd/ofd.c
  37  *
  38  * Author: Alex Zhuravlev <bzzz@whamcloud.com>
  39  * Author: Mike Pershin <tappro@whamcloud.com>
  40  * Author: Johann Lombardi <johann@whamcloud.com>
  41  */
  42
  43 #define DEBUG_SUBSYSTEM S_FILTER
  44
  45 #include <obd_class.h>
  46 #include <lustre_param.h>
  47 #include <lustre_fid.h>
  48 #include <lustre_lfsck.h>
  49 #include <lustre/lustre_idl.h>
  50 #include <lustre_dlm.h>
  51 #include <lustre_quota.h>
  52
  53 #include "ofd_internal.h"
  54
  55 /* Slab for OFD object allocation */
  56 static struct kmem_cache *ofd_object_kmem;
  57
  58 static struct lu_kmem_descr ofd_caches[] = {
  59         {
  60                 .ckd_cache = &ofd_object_kmem,
  61                 .ckd_name  = "ofd_obj",
  62                 .ckd_size  = sizeof(struct ofd_object)
  63         },
  64         {
  65                 .ckd_cache = NULL
  66         }
  67 };
  68
  69 static int ofd_connect_to_next(const struct lu_env *env, struct ofd_device *m,
  70                                const char *next, struct obd_export **exp)
  71 {
  72         struct obd_connect_data *data = NULL;
  73         struct obd_device       *obd;
  74         int                      rc;
  75         ENTRY;
  76
  77         OBD_ALLOC_PTR(data);
  78         if (data == NULL)
  79                 GOTO(out, rc = -ENOMEM);
  80
  81         obd = class_name2obd(next);
  82         if (obd == NULL) {
  83                 CERROR("%s: can't locate next device: %s\n",
  84                        m->ofd_dt_dev.dd_lu_dev.ld_obd->obd_name, next);
  85                 GOTO(out, rc = -ENOTCONN);
  86         }
  87
  88         data->ocd_connect_flags = OBD_CONNECT_VERSION;
  89         data->ocd_version = LUSTRE_VERSION_CODE;
  90
  91         rc = obd_connect(NULL, exp, obd, &obd->obd_uuid, data, NULL);
  92         if (rc) {
  93                 CERROR("%s: cannot connect to next dev %s: rc = %d\n",
  94                        m->ofd_dt_dev.dd_lu_dev.ld_obd->obd_name, next, rc);
  95                 GOTO(out, rc);
  96         }
  97
  98         m->ofd_dt_dev.dd_lu_dev.ld_site =
  99                 m->ofd_osd_exp->exp_obd->obd_lu_dev->ld_site;
 100         LASSERT(m->ofd_dt_dev.dd_lu_dev.ld_site);
 101         m->ofd_osd = lu2dt_dev(m->ofd_osd_exp->exp_obd->obd_lu_dev);
 102         m->ofd_dt_dev.dd_lu_dev.ld_site->ls_top_dev = &m->ofd_dt_dev.dd_lu_dev;
 103
 104 out:
 105         if (data)
 106                 OBD_FREE_PTR(data);
 107         RETURN(rc);
 108 }
 109
 110 static int ofd_stack_init(const struct lu_env *env,
 111                           struct ofd_device *m, struct lustre_cfg *cfg)
 112 {
 113         const char              *dev = lustre_cfg_string(cfg, 0);
 114         struct lu_device        *d;
 115         struct ofd_thread_info  *info = ofd_info(env);
 116         struct lustre_mount_info *lmi;
 117         int                      rc;
 118         char                    *osdname;
 119
 120         ENTRY;
 121
 122         lmi = server_get_mount(dev);
 123         if (lmi == NULL) {
 124                 CERROR("Cannot get mount info for %s!\n", dev);
 125                 RETURN(-ENODEV);
 126         }
 127
 128         /* find bottom osd */
 129         OBD_ALLOC(osdname, MTI_NAME_MAXLEN);
 130         if (osdname == NULL)
 131                 RETURN(-ENOMEM);
 132
 133         snprintf(osdname, MTI_NAME_MAXLEN, "%s-osd", dev);
 134         rc = ofd_connect_to_next(env, m, osdname, &m->ofd_osd_exp);
 135         OBD_FREE(osdname, MTI_NAME_MAXLEN);
 136         if (rc)
 137                 RETURN(rc);
 138
 139         d = m->ofd_osd_exp->exp_obd->obd_lu_dev;
 140         LASSERT(d);
 141         m->ofd_osd = lu2dt_dev(d);
 142
 143         snprintf(info->fti_u.name, sizeof(info->fti_u.name),
 144                  "%s-osd", lustre_cfg_string(cfg, 0));
 145
 146         RETURN(rc);
 147 }
 148
 149 static void ofd_stack_fini(const struct lu_env *env, struct ofd_device *m,
 150                            struct lu_device *top)
 151 {
 152         struct obd_device       *obd = ofd_obd(m);
 153         struct lustre_cfg_bufs   bufs;
 154         struct lustre_cfg       *lcfg;
 155         char                     flags[3] = "";
 156
 157         ENTRY;
 158
 159         lu_site_purge(env, top->ld_site, ~0);
 160         /* process cleanup, pass mdt obd name to get obd umount flags */
 161         lustre_cfg_bufs_reset(&bufs, obd->obd_name);
 162         if (obd->obd_force)
 163                 strcat(flags, "F");
 164         if (obd->obd_fail)
 165                 strcat(flags, "A");
 166         lustre_cfg_bufs_set_string(&bufs, 1, flags);
 167         lcfg = lustre_cfg_new(LCFG_CLEANUP, &bufs);
 168         if (!lcfg) {
 169                 CERROR("Cannot alloc lcfg!\n");
 170                 RETURN_EXIT;
 171         }
 172
 173         LASSERT(top);
 174         top->ld_ops->ldo_process_config(env, top, lcfg);
 175         lustre_cfg_free(lcfg);
 176
 177         lu_site_purge(env, top->ld_site, ~0);
 178         if (!cfs_hash_is_empty(top->ld_site->ls_obj_hash)) {
 179                 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
 180                 lu_site_print(env, top->ld_site, &msgdata, lu_cdebug_printer);
 181         }
 182
 183         LASSERT(m->ofd_osd_exp);
 184         obd_disconnect(m->ofd_osd_exp);
 185         m->ofd_osd = NULL;
 186
 187         EXIT;
 188 }
 189
 190 /* For interoperability, see mdt_interop_param[]. */
 191 static struct cfg_interop_param ofd_interop_param[] = {
 192         { "ost.quota_type",     NULL },
 193         { NULL }
 194 };
 195
 196 /* Some parameters were moved from ofd to osd and only their
 197  * symlinks were kept in ofd by LU-3106. They are:
 198  * -writehthrough_cache_enable
 199  * -readcache_max_filese
 200  * -read_cache_enable
 201  * -brw_stats
 202  * Since they are not included by the static lprocfs var list,
 203  * a pre-check is added for them to avoid "unknown param" error
 204  * message confuses the customer. If they are matched in this
 205  * check, they will be passed to the osd directly.
 206  */
 207 static bool match_symlink_param(char *param)
 208 {
 209         char *sval;
 210         int paramlen;
 211
 212         if (class_match_param(param, PARAM_OST, &param) == 0) {
 213                 sval = strchr(param, '=');
 214                 if (sval != NULL) {
 215                         paramlen = sval - param;
 216                         if (strncmp(param, "writethrough_cache_enable",
 217                                     paramlen) == 0 ||
 218                             strncmp(param, "readcache_max_filesize",
 219                                     paramlen) == 0 ||
 220                             strncmp(param, "read_cache_enable",
 221                                     paramlen) == 0 ||
 222                             strncmp(param, "brw_stats", paramlen) == 0)
 223                                 return true;
 224                 }
 225         }
 226
 227         return false;
 228 }
 229
 230 /* used by MGS to process specific configurations */
 231 static int ofd_process_config(const struct lu_env *env, struct lu_device *d,
 232                               struct lustre_cfg *cfg)
 233 {
 234         struct ofd_device       *m = ofd_dev(d);
 235         struct dt_device        *dt_next = m->ofd_osd;
 236         struct lu_device        *next = &dt_next->dd_lu_dev;
 237         int                      rc;
 238
 239         ENTRY;
 240
 241         switch (cfg->lcfg_command) {
 242         case LCFG_PARAM: {
 243                 struct lprocfs_static_vars lvars;
 244
 245                 /* For interoperability */
 246                 struct cfg_interop_param   *ptr = NULL;
 247                 struct lustre_cfg          *old_cfg = NULL;
 248                 char                       *param = NULL;
 249
 250                 param = lustre_cfg_string(cfg, 1);
 251                 if (param == NULL) {
 252                         CERROR("param is empty\n");
 253                         rc = -EINVAL;
 254                         break;
 255                 }
 256
 257                 ptr = class_find_old_param(param, ofd_interop_param);
 258                 if (ptr != NULL) {
 259                         if (ptr->new_param == NULL) {
 260                                 rc = 0;
 261                                 CWARN("For interoperability, skip this %s."
 262                                       " It is obsolete.\n", ptr->old_param);
 263                                 break;
 264                         }
 265
 266                         CWARN("Found old param %s, changed it to %s.\n",
 267                               ptr->old_param, ptr->new_param);
 268
 269                         old_cfg = cfg;
 270                         cfg = lustre_cfg_rename(old_cfg, ptr->new_param);
 271                         if (IS_ERR(cfg)) {
 272                                 rc = PTR_ERR(cfg);
 273                                 break;
 274                         }
 275                 }
 276
 277                 if (match_symlink_param(param)) {
 278                         rc = next->ld_ops->ldo_process_config(env, next, cfg);
 279                         break;
 280                 }
 281
 282                 lprocfs_ofd_init_vars(&lvars);
 283                 rc = class_process_proc_param(PARAM_OST, lvars.obd_vars, cfg,
 284                                               d->ld_obd);
 285                 if (rc > 0 || rc == -ENOSYS) {
 286                         CDEBUG(D_CONFIG, "pass param %s down the stack.\n",
 287                                param);
 288                         /* we don't understand; pass it on */
 289                         rc = next->ld_ops->ldo_process_config(env, next, cfg);
 290                 }
 291                 break;
 292         }
 293         case LCFG_SPTLRPC_CONF: {
 294                 rc = -ENOTSUPP;
 295                 break;
 296         }
 297         default:
 298                 /* others are passed further */
 299                 rc = next->ld_ops->ldo_process_config(env, next, cfg);
 300                 break;
 301         }
 302         RETURN(rc);
 303 }
 304
 305 static int ofd_object_init(const struct lu_env *env, struct lu_object *o,
 306                            const struct lu_object_conf *conf)
 307 {
 308         struct ofd_device       *d = ofd_dev(o->lo_dev);
 309         struct lu_device        *under;
 310         struct lu_object        *below;
 311         int                      rc = 0;
 312
 313         ENTRY;
 314
 315         CDEBUG(D_INFO, "object init, fid = "DFID"\n",
 316                PFID(lu_object_fid(o)));
 317
 318         under = &d->ofd_osd->dd_lu_dev;
 319         below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
 320         if (below != NULL)
 321                 lu_object_add(o, below);
 322         else
 323                 rc = -ENOMEM;
 324
 325         RETURN(rc);
 326 }
 327
 328 static void ofd_object_free(const struct lu_env *env, struct lu_object *o)
 329 {
 330         struct ofd_object       *of = ofd_obj(o);
 331         struct lu_object_header *h;
 332
 333         ENTRY;
 334
 335         h = o->lo_header;
 336         CDEBUG(D_INFO, "object free, fid = "DFID"\n",
 337                PFID(lu_object_fid(o)));
 338
 339         lu_object_fini(o);
 340         lu_object_header_fini(h);
 341         OBD_SLAB_FREE_PTR(of, ofd_object_kmem);
 342         EXIT;
 343 }
 344
 345 static int ofd_object_print(const struct lu_env *env, void *cookie,
 346                             lu_printer_t p, const struct lu_object *o)
 347 {
 348         return (*p)(env, cookie, LUSTRE_OST_NAME"-object@%p", o);
 349 }
 350
 351 struct lu_object_operations ofd_obj_ops = {
 352         .loo_object_init        = ofd_object_init,
 353         .loo_object_free        = ofd_object_free,
 354         .loo_object_print       = ofd_object_print
 355 };
 356
 357 static struct lu_object *ofd_object_alloc(const struct lu_env *env,
 358                                           const struct lu_object_header *hdr,
 359                                           struct lu_device *d)
 360 {
 361         struct ofd_object *of;
 362
 363         ENTRY;
 364
 365         OBD_SLAB_ALLOC_PTR_GFP(of, ofd_object_kmem, __GFP_IO);
 366         if (of != NULL) {
 367                 struct lu_object        *o;
 368                 struct lu_object_header *h;
 369
 370                 o = &of->ofo_obj.do_lu;
 371                 h = &of->ofo_header;
 372                 lu_object_header_init(h);
 373                 lu_object_init(o, h, d);
 374                 lu_object_add_top(h, o);
 375                 o->lo_ops = &ofd_obj_ops;
 376                 RETURN(o);
 377         } else {
 378                 RETURN(NULL);
 379         }
 380 }
 381
 382 extern int ost_handle(struct ptlrpc_request *req);
 383
 384 static int ofd_prepare(const struct lu_env *env, struct lu_device *pdev,
 385                        struct lu_device *dev)
 386 {
 387         struct ofd_thread_info          *info;
 388         struct ofd_device               *ofd = ofd_dev(dev);
 389         struct obd_device               *obd = ofd_obd(ofd);
 390         struct lu_device                *next = &ofd->ofd_osd->dd_lu_dev;
 391         struct lfsck_start_param         lsp;
 392         int                              rc;
 393
 394         ENTRY;
 395
 396         rc = lu_env_refill((struct lu_env *)env);
 397         if (rc != 0) {
 398                 CERROR("Failure to refill session: '%d'\n", rc);
 399                 RETURN(rc);
 400         }
 401
 402         info = ofd_info_init(env, NULL);
 403         if (info == NULL)
 404                 RETURN(-EFAULT);
 405
 406         /* initialize lower device */
 407         rc = next->ld_ops->ldo_prepare(env, dev, next);
 408         if (rc != 0)
 409                 RETURN(rc);
 410
 411         rc = lfsck_register(env, ofd->ofd_osd, &ofd->ofd_dt_dev, false);
 412         if (rc != 0) {
 413                 CERROR("%s: failed to initialize lfsck: rc = %d\n",
 414                        obd->obd_name, rc);
 415                 RETURN(rc);
 416         }
 417
 418         lsp.lsp_start = NULL;
 419         lsp.lsp_namespace = ofd->ofd_namespace;
 420         rc = lfsck_start(env, ofd->ofd_osd, &lsp);
 421         if (rc != 0) {
 422                 CWARN("%s: auto trigger paused LFSCK failed: rc = %d\n",
 423                       obd->obd_name, rc);
 424                 rc = 0;
 425         }
 426
 427         target_recovery_init(&ofd->ofd_lut, tgt_request_handle);
 428         LASSERT(obd->obd_no_conn);
 429         spin_lock(&obd->obd_dev_lock);
 430         obd->obd_no_conn = 0;
 431         spin_unlock(&obd->obd_dev_lock);
 432
 433         if (obd->obd_recovering == 0)
 434                 ofd_postrecov(env, ofd);
 435
 436         RETURN(rc);
 437 }
 438
 439 static int ofd_recovery_complete(const struct lu_env *env,
 440                                  struct lu_device *dev)
 441 {
 442         struct ofd_device       *ofd = ofd_dev(dev);
 443         struct lu_device        *next = &ofd->ofd_osd->dd_lu_dev;
 444         int                      rc = 0, max_precreate;
 445
 446         ENTRY;
 447
 448         /* Grant space for object precreation on the self export.
 449          * This initial reserved space (i.e. 10MB for zfs and 280KB for ldiskfs)
 450          * is enough to create 10k objects. More space is then acquired for
 451          * precreation in ofd_grant_create().
 452          */
 453         max_precreate = OST_MAX_PRECREATE * ofd->ofd_dt_conf.ddp_inodespace / 2;
 454         ofd_grant_connect(env, dev->ld_obd->obd_self_export, max_precreate,
 455                           false);
 456         rc = next->ld_ops->ldo_recovery_complete(env, next);
 457         RETURN(rc);
 458 }
 459
 460 static struct lu_device_operations ofd_lu_ops = {
 461         .ldo_object_alloc       = ofd_object_alloc,
 462         .ldo_process_config     = ofd_process_config,
 463         .ldo_recovery_complete  = ofd_recovery_complete,
 464         .ldo_prepare            = ofd_prepare,
 465 };
 466
 467 static int ofd_procfs_init(struct ofd_device *ofd)
 468 {
 469         struct lprocfs_static_vars       lvars;
 470         struct obd_device               *obd = ofd_obd(ofd);
 471         cfs_proc_dir_entry_t            *entry;
 472         int                              rc = 0;
 473
 474         ENTRY;
 475
 476         /* lprocfs must be setup before the ofd so state can be safely added
 477          * to /proc incrementally as the ofd is setup */
 478         lprocfs_ofd_init_vars(&lvars);
 479         rc = lprocfs_obd_setup(obd, lvars.obd_vars);
 480         if (rc) {
 481                 CERROR("%s: lprocfs_obd_setup failed: %d.\n",
 482                        obd->obd_name, rc);
 483                 RETURN(rc);
 484         }
 485
 486         rc = lprocfs_alloc_obd_stats(obd, LPROC_OFD_STATS_LAST);
 487         if (rc) {
 488                 CERROR("%s: lprocfs_alloc_obd_stats failed: %d.\n",
 489                        obd->obd_name, rc);
 490                 GOTO(obd_cleanup, rc);
 491         }
 492
 493         obd->obd_uses_nid_stats = 1;
 494
 495         entry = lprocfs_register("exports", obd->obd_proc_entry, NULL, NULL);
 496         if (IS_ERR(entry)) {
 497                 rc = PTR_ERR(entry);
 498                 CERROR("%s: error %d setting up lprocfs for %s\n",
 499                        obd->obd_name, rc, "exports");
 500                 GOTO(obd_cleanup, rc);
 501         }
 502         obd->obd_proc_exports_entry = entry;
 503
 504         entry = lprocfs_add_simple(obd->obd_proc_exports_entry, "clear",
 505                                    lprocfs_nid_stats_clear_read,
 506                                    lprocfs_nid_stats_clear_write, obd, NULL);
 507         if (IS_ERR(entry)) {
 508                 rc = PTR_ERR(entry);
 509                 CERROR("%s: add proc entry 'clear' failed: %d.\n",
 510                        obd->obd_name, rc);
 511                 GOTO(obd_cleanup, rc);
 512         }
 513
 514         ofd_stats_counter_init(obd->obd_stats);
 515
 516         rc = lprocfs_job_stats_init(obd, LPROC_OFD_STATS_LAST,
 517                                     ofd_stats_counter_init);
 518         if (rc)
 519                 GOTO(remove_entry_clear, rc);
 520         RETURN(0);
 521 remove_entry_clear:
 522         lprocfs_remove_proc_entry("clear", obd->obd_proc_exports_entry);
 523 obd_cleanup:
 524         lprocfs_obd_cleanup(obd);
 525         lprocfs_free_obd_stats(obd);
 526
 527         return rc;
 528 }
 529
 530 static void ofd_procfs_add_brw_stats_symlink(struct ofd_device *ofd)
 531 {
 532         struct obd_device       *obd = ofd_obd(ofd);
 533         struct obd_device       *osd_obd = ofd->ofd_osd_exp->exp_obd;
 534         cfs_proc_dir_entry_t    *osd_root = osd_obd->obd_type->typ_procroot;
 535         cfs_proc_dir_entry_t    *osd_dir;
 536
 537         osd_dir = lprocfs_srch(osd_root, obd->obd_name);
 538         if (osd_dir == NULL)
 539                 return;
 540
 541         if (lprocfs_srch(osd_dir, "brw_stats") != NULL)
 542                 lprocfs_add_symlink("brw_stats", obd->obd_proc_entry,
 543                                     "../../%s/%s/brw_stats",
 544                                     osd_root->name, osd_dir->name);
 545
 546         if (lprocfs_srch(osd_dir, "read_cache_enable") != NULL)
 547                 lprocfs_add_symlink("read_cache_enable", obd->obd_proc_entry,
 548                                     "../../%s/%s/read_cache_enable",
 549                                     osd_root->name, osd_dir->name);
 550
 551         if (lprocfs_srch(osd_dir, "readcache_max_filesize") != NULL)
 552                 lprocfs_add_symlink("readcache_max_filesize",
 553                                     obd->obd_proc_entry,
 554                                     "../../%s/%s/readcache_max_filesize",
 555                                     osd_root->name, osd_dir->name);
 556
 557         if (lprocfs_srch(osd_dir, "writethrough_cache_enable") != NULL)
 558                 lprocfs_add_symlink("writethrough_cache_enable",
 559                                     obd->obd_proc_entry,
 560                                     "../../%s/%s/writethrough_cache_enable",
 561                                     osd_root->name, osd_dir->name);
 562 }
 563
 564 static void ofd_procfs_fini(struct ofd_device *ofd)
 565 {
 566         struct obd_device *obd = ofd_obd(ofd);
 567
 568         lprocfs_remove_proc_entry("writethrough_cache_enable",
 569                                   obd->obd_proc_entry);
 570         lprocfs_remove_proc_entry("readcache_max_filesize",
 571                                   obd->obd_proc_entry);
 572         lprocfs_remove_proc_entry("read_cache_enable", obd->obd_proc_entry);
 573         lprocfs_remove_proc_entry("brw_stats", obd->obd_proc_entry);
 574         lprocfs_remove_proc_entry("clear", obd->obd_proc_exports_entry);
 575         lprocfs_free_per_client_stats(obd);
 576         lprocfs_obd_cleanup(obd);
 577         lprocfs_free_obd_stats(obd);
 578         lprocfs_job_stats_fini(obd);
 579 }
 580
 581 extern int ost_handle(struct ptlrpc_request *req);
 582
 583 int ofd_fid_fini(const struct lu_env *env, struct ofd_device *ofd)
 584 {
 585         return seq_site_fini(env, &ofd->ofd_seq_site);
 586 }
 587
 588 int ofd_fid_init(const struct lu_env *env, struct ofd_device *ofd)
 589 {
 590         struct seq_server_site  *ss = &ofd->ofd_seq_site;
 591         struct lu_device        *lu = &ofd->ofd_dt_dev.dd_lu_dev;
 592         char                    *obd_name = ofd_name(ofd);
 593         char                    *name = NULL;
 594         int                     rc = 0;
 595
 596         ss = &ofd->ofd_seq_site;
 597         lu->ld_site->ld_seq_site = ss;
 598         ss->ss_lu = lu->ld_site;
 599         ss->ss_node_id = ofd->ofd_lut.lut_lsd.lsd_osd_index;
 600
 601         OBD_ALLOC_PTR(ss->ss_server_seq);
 602         if (ss->ss_server_seq == NULL)
 603                 GOTO(out_free, rc = -ENOMEM);
 604
 605         OBD_ALLOC(name, strlen(obd_name) + 10);
 606         if (!name) {
 607                 OBD_FREE_PTR(ss->ss_server_seq);
 608                 ss->ss_server_seq = NULL;
 609                 GOTO(out_free, rc = -ENOMEM);
 610         }
 611
 612         rc = seq_server_init(env, ss->ss_server_seq, ofd->ofd_osd, obd_name,
 613                              LUSTRE_SEQ_SERVER, ss);
 614         if (rc) {
 615                 CERROR("%s : seq server init error %d\n", obd_name, rc);
 616                 GOTO(out_free, rc);
 617         }
 618         ss->ss_server_seq->lss_space.lsr_index = ss->ss_node_id;
 619
 620         OBD_ALLOC_PTR(ss->ss_client_seq);
 621         if (ss->ss_client_seq == NULL)
 622                 GOTO(out_free, rc = -ENOMEM);
 623
 624         snprintf(name, strlen(obd_name) + 6, "%p-super", obd_name);
 625         rc = seq_client_init(ss->ss_client_seq, NULL, LUSTRE_SEQ_DATA,
 626                              name, NULL);
 627         if (rc) {
 628                 CERROR("%s : seq client init error %d\n", obd_name, rc);
 629                 GOTO(out_free, rc);
 630         }
 631         OBD_FREE(name, strlen(obd_name) + 10);
 632         name = NULL;
 633
 634         rc = seq_server_set_cli(env, ss->ss_server_seq, ss->ss_client_seq);
 635
 636 out_free:
 637         if (rc) {
 638                 if (ss->ss_server_seq) {
 639                         seq_server_fini(ss->ss_server_seq, env);
 640                         OBD_FREE_PTR(ss->ss_server_seq);
 641                         ss->ss_server_seq = NULL;
 642                 }
 643
 644                 if (ss->ss_client_seq) {
 645                         seq_client_fini(ss->ss_client_seq);
 646                         OBD_FREE_PTR(ss->ss_client_seq);
 647                         ss->ss_client_seq = NULL;
 648                 }
 649
 650                 if (name) {
 651                         OBD_FREE(name, strlen(obd_name) + 10);
 652                         name = NULL;
 653                 }
 654         }
 655
 656         return rc;
 657 }
 658
 659 int ofd_set_info_hdl(struct tgt_session_info *tsi)
 660 {
 661         struct ptlrpc_request   *req = tgt_ses_req(tsi);
 662         struct ost_body         *body = NULL, *repbody;
 663         void                    *key, *val = NULL;
 664         int                      keylen, vallen, rc = 0;
 665         bool                     is_grant_shrink;
 666         struct ofd_device       *ofd = ofd_exp(tsi->tsi_exp);
 667
 668         ENTRY;
 669
 670         key = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_KEY);
 671         if (key == NULL) {
 672                 DEBUG_REQ(D_HA, req, "no set_info key");
 673                 RETURN(err_serious(-EFAULT));
 674         }
 675         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_KEY,
 676                                       RCL_CLIENT);
 677
 678         val = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_VAL);
 679         if (val == NULL) {
 680                 DEBUG_REQ(D_HA, req, "no set_info val");
 681                 RETURN(err_serious(-EFAULT));
 682         }
 683         vallen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_VAL,
 684                                       RCL_CLIENT);
 685
 686         is_grant_shrink = KEY_IS(KEY_GRANT_SHRINK);
 687         if (is_grant_shrink)
 688                 /* In this case the value is actually an RMF_OST_BODY, so we
 689                  * transmutate the type of this PTLRPC */
 690                 req_capsule_extend(tsi->tsi_pill, &RQF_OST_SET_GRANT_INFO);
 691
 692         rc = req_capsule_server_pack(tsi->tsi_pill);
 693         if (rc < 0)
 694                 RETURN(rc);
 695
 696         if (is_grant_shrink) {
 697                 body = req_capsule_client_get(tsi->tsi_pill, &RMF_OST_BODY);
 698
 699                 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
 700                 *repbody = *body;
 701
 702                 /** handle grant shrink, similar to a read request */
 703                 ofd_grant_prepare_read(tsi->tsi_env, tsi->tsi_exp,
 704                                        &repbody->oa);
 705         } else if (KEY_IS(KEY_EVICT_BY_NID)) {
 706                 if (vallen > 0)
 707                         obd_export_evict_by_nid(tsi->tsi_exp->exp_obd, val);
 708                 rc = 0;
 709         } else if (KEY_IS(KEY_CAPA_KEY)) {
 710                 rc = ofd_update_capa_key(ofd, val);
 711         } else if (KEY_IS(KEY_SPTLRPC_CONF)) {
 712                 rc = tgt_adapt_sptlrpc_conf(tsi->tsi_tgt, 0);
 713         } else {
 714                 CERROR("%s: Unsupported key %s\n",
 715                        tgt_name(tsi->tsi_tgt), (char *)key);
 716                 rc = -EOPNOTSUPP;
 717         }
 718         ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_SET_INFO,
 719                          tsi->tsi_jobid, 1);
 720
 721         RETURN(rc);
 722 }
 723
 724 static int ofd_fiemap_get(const struct lu_env *env, struct ofd_device *ofd,
 725                           struct lu_fid *fid, struct ll_user_fiemap *fiemap)
 726 {
 727         struct ofd_object       *fo;
 728         int                      rc;
 729
 730         fo = ofd_object_find(env, ofd, fid);
 731         if (IS_ERR(fo)) {
 732                 CERROR("%s: error finding object "DFID"\n",
 733                        ofd_name(ofd), PFID(fid));
 734                 return PTR_ERR(fo);
 735         }
 736
 737         ofd_read_lock(env, fo);
 738         if (ofd_object_exists(fo))
 739                 rc = dt_fiemap_get(env, ofd_object_child(fo), fiemap);
 740         else
 741                 rc = -ENOENT;
 742         ofd_read_unlock(env, fo);
 743         ofd_object_put(env, fo);
 744         return rc;
 745 }
 746
 747 struct locked_region {
 748         cfs_list_t              list;
 749         struct lustre_handle    lh;
 750 };
 751
 752 static int lock_region(struct ldlm_namespace *ns, struct ldlm_res_id *res_id,
 753                        unsigned long long begin, unsigned long long end,
 754                        cfs_list_t *locked)
 755 {
 756         struct locked_region    *region = NULL;
 757         __u64                    flags = 0;
 758         int                      rc;
 759
 760         LASSERT(begin <= end);
 761         OBD_ALLOC_PTR(region);
 762         if (region == NULL)
 763                 return -ENOMEM;
 764
 765         rc = tgt_extent_lock(ns, res_id, begin, end, &region->lh,
 766                              LCK_PR, &flags);
 767         if (rc != 0)
 768                 return rc;
 769
 770         CDEBUG(D_OTHER, "ost lock [%llu,%llu], lh=%p\n", begin, end,
 771                &region->lh);
 772         cfs_list_add(&region->list, locked);
 773
 774         return 0;
 775 }
 776
 777 static int lock_zero_regions(struct ldlm_namespace *ns,
 778                              struct ldlm_res_id *res_id,
 779                              struct ll_user_fiemap *fiemap,
 780                              cfs_list_t *locked)
 781 {
 782         __u64 begin = fiemap->fm_start;
 783         unsigned int i;
 784         int rc = 0;
 785         struct ll_fiemap_extent *fiemap_start = fiemap->fm_extents;
 786
 787         ENTRY;
 788
 789         CDEBUG(D_OTHER, "extents count %u\n", fiemap->fm_mapped_extents);
 790         for (i = 0; i < fiemap->fm_mapped_extents; i++) {
 791                 if (fiemap_start[i].fe_logical > begin) {
 792                         CDEBUG(D_OTHER, "ost lock [%llu,%llu]\n",
 793                                begin, fiemap_start[i].fe_logical);
 794                         rc = lock_region(ns, res_id, begin,
 795                                          fiemap_start[i].fe_logical, locked);
 796                         if (rc)
 797                                 RETURN(rc);
 798                 }
 799
 800                 begin = fiemap_start[i].fe_logical + fiemap_start[i].fe_length;
 801         }
 802
 803         if (begin < (fiemap->fm_start + fiemap->fm_length)) {
 804                 CDEBUG(D_OTHER, "ost lock [%llu,%llu]\n",
 805                        begin, fiemap->fm_start + fiemap->fm_length);
 806                 rc = lock_region(ns, res_id, begin,
 807                                  fiemap->fm_start + fiemap->fm_length, locked);
 808         }
 809
 810         RETURN(rc);
 811 }
 812
 813 static void unlock_zero_regions(struct ldlm_namespace *ns, cfs_list_t *locked)
 814 {
 815         struct locked_region *entry, *temp;
 816
 817         cfs_list_for_each_entry_safe(entry, temp, locked, list) {
 818                 CDEBUG(D_OTHER, "ost unlock lh=%p\n", &entry->lh);
 819                 tgt_extent_unlock(&entry->lh, LCK_PR);
 820                 cfs_list_del(&entry->list);
 821                 OBD_FREE_PTR(entry);
 822         }
 823 }
 824
 825 int ofd_get_info_hdl(struct tgt_session_info *tsi)
 826 {
 827         struct obd_export               *exp = tsi->tsi_exp;
 828         struct ofd_device               *ofd = ofd_exp(exp);
 829         struct ofd_thread_info          *fti = tsi2ofd_info(tsi);
 830         void                            *key;
 831         int                              keylen;
 832         int                              replylen, rc = 0;
 833
 834         ENTRY;
 835
 836         /* this common part for get_info rpc */
 837         key = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_KEY);
 838         if (key == NULL) {
 839                 DEBUG_REQ(D_HA, tgt_ses_req(tsi), "no get_info key");
 840                 RETURN(err_serious(-EPROTO));
 841         }
 842         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_GETINFO_KEY,
 843                                       RCL_CLIENT);
 844
 845         if (KEY_IS(KEY_LAST_ID)) {
 846                 obd_id          *last_id;
 847                 struct ofd_seq  *oseq;
 848
 849                 req_capsule_extend(tsi->tsi_pill, &RQF_OST_GET_INFO_LAST_ID);
 850                 rc = req_capsule_server_pack(tsi->tsi_pill);
 851                 if (rc)
 852                         RETURN(err_serious(rc));
 853
 854                 last_id = req_capsule_server_get(tsi->tsi_pill, &RMF_OBD_ID);
 855
 856                 oseq = ofd_seq_load(tsi->tsi_env, ofd,
 857                                     (obd_seq)exp->exp_filter_data.fed_group);
 858                 if (IS_ERR(oseq))
 859                         rc = -EFAULT;
 860                 else
 861                         *last_id = ofd_seq_last_oid(oseq);
 862                 ofd_seq_put(tsi->tsi_env, oseq);
 863         } else if (KEY_IS(KEY_FIEMAP)) {
 864                 struct ll_fiemap_info_key       *fm_key;
 865                 struct ll_user_fiemap           *fiemap;
 866                 struct lu_fid                   *fid = &fti->fti_fid;
 867
 868                 req_capsule_extend(tsi->tsi_pill, &RQF_OST_GET_INFO_FIEMAP);
 869
 870                 fm_key = req_capsule_client_get(tsi->tsi_pill, &RMF_FIEMAP_KEY);
 871                 rc = tgt_validate_obdo(tsi, &fm_key->oa);
 872                 if (rc)
 873                         RETURN(err_serious(rc));
 874
 875                 replylen = fiemap_count_to_size(fm_key->fiemap.fm_extent_count);
 876                 req_capsule_set_size(tsi->tsi_pill, &RMF_FIEMAP_VAL,
 877                                      RCL_SERVER, replylen);
 878
 879                 rc = req_capsule_server_pack(tsi->tsi_pill);
 880                 if (rc)
 881                         RETURN(err_serious(rc));
 882
 883                 fiemap = req_capsule_server_get(tsi->tsi_pill, &RMF_FIEMAP_VAL);
 884                 if (fiemap == NULL)
 885                         RETURN(-ENOMEM);
 886
 887                 rc = ostid_to_fid(fid, &fm_key->oa.o_oi, 0);
 888                 if (rc != 0)
 889                         RETURN(rc);
 890
 891                 CDEBUG(D_INODE, "get FIEMAP of object "DFID"\n", PFID(fid));
 892
 893                 *fiemap = fm_key->fiemap;
 894                 rc = ofd_fiemap_get(tsi->tsi_env, ofd, fid, fiemap);
 895
 896                 /* LU-3219: Lock the sparse areas to make sure dirty
 897                  * flushed back from client, then call fiemap again. */
 898                 if (fm_key->oa.o_valid & OBD_MD_FLFLAGS &&
 899                     fm_key->oa.o_flags & OBD_FL_SRVLOCK) {
 900                         cfs_list_t locked = CFS_LIST_HEAD_INIT(locked);
 901
 902                         ost_fid_build_resid(fid, &fti->fti_resid);
 903                         rc = lock_zero_regions(ofd->ofd_namespace,
 904                                                &fti->fti_resid, fiemap,
 905                                                &locked);
 906                         if (rc == 0 && !cfs_list_empty(&locked)) {
 907                                 rc = ofd_fiemap_get(tsi->tsi_env, ofd, fid,
 908                                                     fiemap);
 909                                 unlock_zero_regions(ofd->ofd_namespace,
 910                                                     &locked);
 911                         }
 912                 }
 913         } else if (KEY_IS(KEY_LAST_FID)) {
 914                 struct ofd_device       *ofd = ofd_exp(exp);
 915                 struct ofd_seq          *oseq;
 916                 struct lu_fid           *fid;
 917                 int                      rc;
 918
 919                 req_capsule_extend(tsi->tsi_pill, &RQF_OST_GET_INFO_LAST_FID);
 920                 rc = req_capsule_server_pack(tsi->tsi_pill);
 921                 if (rc)
 922                         RETURN(err_serious(rc));
 923
 924                 fid = req_capsule_client_get(tsi->tsi_pill, &RMF_FID);
 925                 if (fid == NULL)
 926                         RETURN(err_serious(-EPROTO));
 927
 928                 fid_le_to_cpu(&fti->fti_ostid.oi_fid, fid);
 929
 930                 fid = req_capsule_server_get(tsi->tsi_pill, &RMF_FID);
 931                 if (fid == NULL)
 932                         RETURN(-ENOMEM);
 933
 934                 oseq = ofd_seq_load(tsi->tsi_env, ofd,
 935                                     ostid_seq(&fti->fti_ostid));
 936                 if (IS_ERR(oseq))
 937                         RETURN(PTR_ERR(oseq));
 938
 939                 rc = ostid_to_fid(fid, &oseq->os_oi,
 940                                   ofd->ofd_lut.lut_lsd.lsd_osd_index);
 941                 if (rc != 0)
 942                         GOTO(out_put, rc);
 943
 944                 CDEBUG(D_HA, "%s: LAST FID is "DFID"\n", ofd_name(ofd),
 945                        PFID(fid));
 946 out_put:
 947                 ofd_seq_put(tsi->tsi_env, oseq);
 948         } else {
 949                 CERROR("%s: not supported key %s\n", tgt_name(tsi->tsi_tgt),
 950                        (char *)key);
 951                 rc = -EOPNOTSUPP;
 952         }
 953         ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_GET_INFO,
 954                          tsi->tsi_jobid, 1);
 955
 956         RETURN(rc);
 957 }
 958
 959 static int ofd_getattr_hdl(struct tgt_session_info *tsi)
 960 {
 961         struct ofd_thread_info  *fti = tsi2ofd_info(tsi);
 962         struct ofd_device       *ofd = ofd_exp(tsi->tsi_exp);
 963         struct ost_body         *repbody;
 964         struct lustre_handle     lh = { 0 };
 965         struct ofd_object       *fo;
 966         __u64                    flags = 0;
 967         ldlm_mode_t              lock_mode = LCK_PR;
 968         bool                     srvlock;
 969         int                      rc;
 970         ENTRY;
 971
 972         LASSERT(tsi->tsi_ost_body != NULL);
 973
 974         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
 975         if (repbody == NULL)
 976                 RETURN(-ENOMEM);
 977
 978         repbody->oa.o_oi = tsi->tsi_ost_body->oa.o_oi;
 979         repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
 980
 981         srvlock = tsi->tsi_ost_body->oa.o_valid & OBD_MD_FLFLAGS &&
 982                   tsi->tsi_ost_body->oa.o_flags & OBD_FL_SRVLOCK;
 983
 984         if (srvlock) {
 985                 if (unlikely(tsi->tsi_ost_body->oa.o_flags & OBD_FL_FLUSH))
 986                         lock_mode = LCK_PW;
 987
 988                 rc = tgt_extent_lock(tsi->tsi_tgt->lut_obd->obd_namespace,
 989                                      &tsi->tsi_resid, 0, OBD_OBJECT_EOF, &lh,
 990                                      lock_mode, &flags);
 991                 if (rc != 0)
 992                         RETURN(rc);
 993         }
 994
 995         fo = ofd_object_find_exists(tsi->tsi_env, ofd, &tsi->tsi_fid);
 996         if (IS_ERR(fo))
 997                 GOTO(out, rc = PTR_ERR(fo));
 998
 999         rc = ofd_attr_get(tsi->tsi_env, fo, &fti->fti_attr);
1000         if (rc == 0) {
1001                 __u64    curr_version;
1002
1003                 obdo_from_la(&repbody->oa, &fti->fti_attr,
1004                              OFD_VALID_FLAGS | LA_UID | LA_GID);
1005                 tgt_drop_id(tsi->tsi_exp, &repbody->oa);
1006
1007                 /* Store object version in reply */
1008                 curr_version = dt_version_get(tsi->tsi_env,
1009                                               ofd_object_child(fo));
1010                 if ((__s64)curr_version != -EOPNOTSUPP) {
1011                         repbody->oa.o_valid |= OBD_MD_FLDATAVERSION;
1012                         repbody->oa.o_data_version = curr_version;
1013                 }
1014         }
1015
1016         ofd_object_put(tsi->tsi_env, fo);
1017 out:
1018         if (srvlock)
1019                 tgt_extent_unlock(&lh, lock_mode);
1020
1021         ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_GETATTR,
1022                          tsi->tsi_jobid, 1);
1023
1024         repbody->oa.o_valid |= OBD_MD_FLFLAGS;
1025         repbody->oa.o_flags = OBD_FL_FLUSH;
1026
1027         RETURN(rc);
1028 }
1029
1030 static int ofd_setattr_hdl(struct tgt_session_info *tsi)
1031 {
1032         struct ofd_thread_info  *fti = tsi2ofd_info(tsi);
1033         struct ofd_device       *ofd = ofd_exp(tsi->tsi_exp);
1034         struct ost_body         *body = tsi->tsi_ost_body;
1035         struct ost_body         *repbody;
1036         struct ldlm_resource    *res;
1037         struct ofd_object       *fo;
1038         struct filter_fid       *ff = NULL;
1039         int                      rc = 0;
1040
1041         ENTRY;
1042
1043         LASSERT(body != NULL);
1044
1045         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1046         if (repbody == NULL)
1047                 RETURN(-ENOMEM);
1048
1049         repbody->oa.o_oi = body->oa.o_oi;
1050         repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1051
1052         /* This would be very bad - accidentally truncating a file when
1053          * changing the time or similar - bug 12203. */
1054         if (body->oa.o_valid & OBD_MD_FLSIZE &&
1055             body->oa.o_size != OBD_OBJECT_EOF) {
1056                 static char mdsinum[48];
1057
1058                 if (body->oa.o_valid & OBD_MD_FLFID)
1059                         snprintf(mdsinum, sizeof(mdsinum) - 1,
1060                                  "of parent "DFID, body->oa.o_parent_seq,
1061                                  body->oa.o_parent_oid, 0);
1062                 else
1063                         mdsinum[0] = '\0';
1064
1065                 CERROR("%s: setattr from %s is trying to truncate object "DFID
1066                        " %s\n", ofd_name(ofd), obd_export_nid2str(tsi->tsi_exp),
1067                        PFID(&tsi->tsi_fid), mdsinum);
1068                 RETURN(-EPERM);
1069         }
1070
1071         fo = ofd_object_find_exists(tsi->tsi_env, ofd, &tsi->tsi_fid);
1072         if (IS_ERR(fo))
1073                 GOTO(out, rc = PTR_ERR(fo));
1074
1075         la_from_obdo(&fti->fti_attr, &body->oa, body->oa.o_valid);
1076         fti->fti_attr.la_valid &= ~LA_TYPE;
1077
1078         if (body->oa.o_valid & OBD_MD_FLFID) {
1079                 ff = &fti->fti_mds_fid;
1080                 ofd_prepare_fidea(ff, &body->oa);
1081         }
1082
1083         /* setting objects attributes (including owner/group) */
1084         rc = ofd_attr_set(tsi->tsi_env, fo, &fti->fti_attr, ff);
1085         if (rc != 0)
1086                 GOTO(out_put, rc);
1087
1088         obdo_from_la(&repbody->oa, &fti->fti_attr,
1089                      OFD_VALID_FLAGS | LA_UID | LA_GID);
1090         tgt_drop_id(tsi->tsi_exp, &repbody->oa);
1091
1092         ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_SETATTR,
1093                          tsi->tsi_jobid, 1);
1094         EXIT;
1095 out_put:
1096         ofd_object_put(tsi->tsi_env, fo);
1097 out:
1098         if (rc == 0) {
1099                 /* we do not call this before to avoid lu_object_find() in
1100                  *  ->lvbo_update() holding another reference on the object.
1101                  * otherwise concurrent destroy can make the object unavailable
1102                  * for 2nd lu_object_find() waiting for the first reference
1103                  * to go... deadlock! */
1104                 res = ldlm_resource_get(ofd->ofd_namespace, NULL,
1105                                         &tsi->tsi_resid, LDLM_EXTENT, 0);
1106                 if (res != NULL) {
1107                         ldlm_res_lvbo_update(res, NULL, 0);
1108                         ldlm_resource_putref(res);
1109                 }
1110         }
1111         return rc;
1112 }
1113
1114 static int ofd_create_hdl(struct tgt_session_info *tsi)
1115 {
1116         struct ost_body         *repbody;
1117         const struct obdo       *oa = &tsi->tsi_ost_body->oa;
1118         struct obdo             *rep_oa;
1119         struct ofd_device       *ofd = ofd_exp(tsi->tsi_exp);
1120         obd_seq                  seq = ostid_seq(&oa->o_oi);
1121         obd_id                   oid = ostid_id(&oa->o_oi);
1122         struct ofd_seq          *oseq;
1123         int                      rc = 0, diff;
1124         int                      sync_trans = 0;
1125
1126         ENTRY;
1127
1128         if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
1129                 RETURN(-EROFS);
1130
1131         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1132         if (repbody == NULL)
1133                 RETURN(-ENOMEM);
1134
1135         rep_oa = &repbody->oa;
1136         rep_oa->o_oi = oa->o_oi;
1137
1138         LASSERT(seq >= FID_SEQ_OST_MDT0);
1139         LASSERT(oa->o_valid & OBD_MD_FLGROUP);
1140
1141         CDEBUG(D_INFO, "ofd_create("DOSTID")\n", POSTID(&oa->o_oi));
1142
1143         oseq = ofd_seq_load(tsi->tsi_env, ofd, seq);
1144         if (IS_ERR(oseq)) {
1145                 CERROR("%s: Can't find FID Sequence "LPX64": rc = %ld\n",
1146                        ofd_name(ofd), seq, PTR_ERR(oseq));
1147                 RETURN(-EINVAL);
1148         }
1149
1150         if ((oa->o_valid & OBD_MD_FLFLAGS) &&
1151             (oa->o_flags & OBD_FL_RECREATE_OBJS)) {
1152                 if (!ofd_obd(ofd)->obd_recovering ||
1153                     oid > ofd_seq_last_oid(oseq)) {
1154                         CERROR("%s: recreate objid "DOSTID" > last id "LPU64
1155                                "\n", ofd_name(ofd), POSTID(&oa->o_oi),
1156                                ofd_seq_last_oid(oseq));
1157                         GOTO(out_nolock, rc = -EINVAL);
1158                 }
1159                 /* Do nothing here, we re-create objects during recovery
1160                  * upon write replay, see ofd_preprw_write() */
1161                 GOTO(out_nolock, rc = 0);
1162         }
1163         /* former ofd_handle_precreate */
1164         if ((oa->o_valid & OBD_MD_FLFLAGS) &&
1165             (oa->o_flags & OBD_FL_DELORPHAN)) {
1166                 /* destroy orphans */
1167                 if (lustre_msg_get_conn_cnt(tgt_ses_req(tsi)->rq_reqmsg) <
1168                     tsi->tsi_exp->exp_conn_cnt) {
1169                         CERROR("%s: dropping old orphan cleanup request\n",
1170                                ofd_name(ofd));
1171                         GOTO(out_nolock, rc = 0);
1172                 }
1173                 /* This causes inflight precreates to abort and drop lock */
1174                 oseq->os_destroys_in_progress = 1;
1175                 mutex_lock(&oseq->os_create_lock);
1176                 if (!oseq->os_destroys_in_progress) {
1177                         CERROR("%s:["LPU64"] destroys_in_progress already"
1178                                " cleared\n", ofd_name(ofd), seq);
1179                         ostid_set_id(&rep_oa->o_oi, ofd_seq_last_oid(oseq));
1180                         GOTO(out, rc = 0);
1181                 }
1182                 diff = oid - ofd_seq_last_oid(oseq);
1183                 CDEBUG(D_HA, "ofd_last_id() = "LPU64" -> diff = %d\n",
1184                         ofd_seq_last_oid(oseq), diff);
1185                 if (-diff > OST_MAX_PRECREATE) {
1186                         /* FIXME: should reset precreate_next_id on MDS */
1187                         rc = 0;
1188                 } else if (diff < 0) {
1189                         rc = ofd_orphans_destroy(tsi->tsi_env, tsi->tsi_exp,
1190                                                  ofd, rep_oa);
1191                         oseq->os_destroys_in_progress = 0;
1192                 } else {
1193                         /* XXX: Used by MDS for the first time! */
1194                         oseq->os_destroys_in_progress = 0;
1195                 }
1196         } else {
1197                 mutex_lock(&oseq->os_create_lock);
1198                 if (lustre_msg_get_conn_cnt(tgt_ses_req(tsi)->rq_reqmsg) <
1199                     tsi->tsi_exp->exp_conn_cnt) {
1200                         CERROR("%s: dropping old precreate request\n",
1201                                ofd_name(ofd));
1202                         GOTO(out, rc = 0);
1203                 }
1204                 /* only precreate if seq is 0, IDIF or normal and also o_id
1205                  * must be specfied */
1206                 if ((!fid_seq_is_mdt(seq) && !fid_seq_is_norm(seq) &&
1207                      !fid_seq_is_idif(seq)) || oid == 0) {
1208                         diff = 1; /* shouldn't we create this right now? */
1209                 } else {
1210                         diff = oid - ofd_seq_last_oid(oseq);
1211                         /* Do sync create if the seq is about to used up */
1212                         if (fid_seq_is_idif(seq) || fid_seq_is_mdt0(seq)) {
1213                                 if (unlikely(oid >= IDIF_MAX_OID - 1))
1214                                         sync_trans = 1;
1215                         } else if (fid_seq_is_norm(seq)) {
1216                                 if (unlikely(oid >=
1217                                              LUSTRE_DATA_SEQ_MAX_WIDTH - 1))
1218                                         sync_trans = 1;
1219                         } else {
1220                                 CERROR("%s : invalid o_seq "DOSTID"\n",
1221                                        ofd_name(ofd), POSTID(&oa->o_oi));
1222                                 GOTO(out, rc = -EINVAL);
1223                         }
1224                 }
1225         }
1226         if (diff > 0) {
1227                 cfs_time_t       enough_time = cfs_time_shift(DISK_TIMEOUT);
1228                 obd_id           next_id;
1229                 int              created = 0;
1230                 int              count;
1231
1232                 if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
1233                     !(oa->o_flags & OBD_FL_DELORPHAN)) {
1234                         /* don't enforce grant during orphan recovery */
1235                         rc = ofd_grant_create(tsi->tsi_env,
1236                                               ofd_obd(ofd)->obd_self_export,
1237                                               &diff);
1238                         if (rc) {
1239                                 CDEBUG(D_HA, "%s: failed to acquire grant "
1240                                        "space for precreate (%d): rc = %d\n",
1241                                        ofd_name(ofd), diff, rc);
1242                                 diff = 0;
1243                         }
1244                 }
1245
1246                 /* This can happen if a new OST is formatted and installed
1247                  * in place of an old one at the same index.  Instead of
1248                  * precreating potentially millions of deleted old objects
1249                  * (possibly filling the OST), only precreate the last batch.
1250                  * LFSCK will eventually clean up any orphans. LU-14 */
1251                 if (diff > 5 * OST_MAX_PRECREATE) {
1252                         diff = OST_MAX_PRECREATE / 2;
1253                         LCONSOLE_WARN("%s: precreate FID "DOSTID" is over %u "
1254                                       "larger than the LAST_ID "DOSTID", only "
1255                                       "precreating the last %u objects.\n",
1256                                       ofd_name(ofd), POSTID(&oa->o_oi),
1257                                       5 * OST_MAX_PRECREATE,
1258                                       POSTID(&oseq->os_oi), diff);
1259                         ofd_seq_last_oid_set(oseq, ostid_id(&oa->o_oi) - diff);
1260                 }
1261
1262                 while (diff > 0) {
1263                         next_id = ofd_seq_last_oid(oseq) + 1;
1264                         count = ofd_precreate_batch(ofd, diff);
1265
1266                         CDEBUG(D_HA, "%s: reserve %d objects in group "LPX64
1267                                " at "LPU64"\n", ofd_name(ofd),
1268                                count, seq, next_id);
1269
1270                         if (cfs_time_after(jiffies, enough_time)) {
1271                                 LCONSOLE_WARN("%s: Slow creates, %d/%d objects"
1272                                               " created at a rate of %d/s\n",
1273                                               ofd_name(ofd), created,
1274                                               diff + created,
1275                                               created / DISK_TIMEOUT);
1276                                 break;
1277                         }
1278
1279                         rc = ofd_precreate_objects(tsi->tsi_env, ofd, next_id,
1280                                                    oseq, count, sync_trans);
1281                         if (rc > 0) {
1282                                 created += rc;
1283                                 diff -= rc;
1284                         } else if (rc < 0) {
1285                                 break;
1286                         }
1287                 }
1288                 if (created > 0)
1289                         /* some objects got created, we can return
1290                          * them, even if last creation failed */
1291                         rc = 0;
1292                 else
1293                         CERROR("%s: unable to precreate: rc = %d\n",
1294                                ofd_name(ofd), rc);
1295
1296                 if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
1297                     !(oa->o_flags & OBD_FL_DELORPHAN))
1298                         ofd_grant_commit(tsi->tsi_env,
1299                                          ofd_obd(ofd)->obd_self_export, rc);
1300
1301                 ostid_set_id(&rep_oa->o_oi, ofd_seq_last_oid(oseq));
1302         }
1303         EXIT;
1304         ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_CREATE,
1305                          tsi->tsi_jobid, 1);
1306 out:
1307         mutex_unlock(&oseq->os_create_lock);
1308 out_nolock:
1309         if (rc == 0)
1310                 rep_oa->o_valid |= OBD_MD_FLID | OBD_MD_FLGROUP;
1311
1312         ofd_seq_put(tsi->tsi_env, oseq);
1313         return rc;
1314 }
1315
1316 static int ofd_destroy_hdl(struct tgt_session_info *tsi)
1317 {
1318         const struct ost_body   *body = tsi->tsi_ost_body;
1319         struct ost_body         *repbody;
1320         struct ofd_device       *ofd = ofd_exp(tsi->tsi_exp);
1321         struct ofd_thread_info  *fti = tsi2ofd_info(tsi);
1322         obd_count                count;
1323         int                      rc = 0;
1324
1325         ENTRY;
1326
1327         if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
1328                 RETURN(-EROFS);
1329
1330         /* This is old case for clients before Lustre 2.4 */
1331         /* If there's a DLM request, cancel the locks mentioned in it */
1332         if (req_capsule_field_present(tsi->tsi_pill, &RMF_DLM_REQ,
1333                                       RCL_CLIENT)) {
1334                 struct ldlm_request *dlm;
1335
1336                 dlm = req_capsule_client_get(tsi->tsi_pill, &RMF_DLM_REQ);
1337                 if (dlm == NULL)
1338                         RETURN(-EFAULT);
1339                 ldlm_request_cancel(tgt_ses_req(tsi), dlm, 0);
1340         }
1341
1342         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1343         repbody->oa.o_oi = body->oa.o_oi;
1344
1345         /* check that o_misc makes sense */
1346         if (body->oa.o_valid & OBD_MD_FLOBJCOUNT)
1347                 count = body->oa.o_misc;
1348         else
1349                 count = 1; /* default case - single destroy */
1350
1351         CDEBUG(D_HA, "%s: Destroy object "DOSTID" count %d\n", ofd_name(ofd),
1352                POSTID(&body->oa.o_oi), count);
1353         while (count > 0) {
1354                 int lrc;
1355
1356                 lrc = ostid_to_fid(&fti->fti_fid, &repbody->oa.o_oi, 0);
1357                 if (lrc != 0) {
1358                         if (rc == 0)
1359                                 rc = lrc;
1360                         GOTO(out, rc);
1361                 }
1362                 lrc = ofd_destroy_by_fid(tsi->tsi_env, ofd, &fti->fti_fid, 0);
1363                 if (lrc == -ENOENT) {
1364                         CDEBUG(D_INODE,
1365                                "%s: destroying non-existent object "DFID"\n",
1366                                ofd_name(ofd), PFID(&fti->fti_fid));
1367                         /* rewrite rc with -ENOENT only if it is 0 */
1368                         if (rc == 0)
1369                                 rc = lrc;
1370                 } else if (lrc != 0) {
1371                         CERROR("%s: error destroying object "DFID": %d\n",
1372                                ofd_name(ofd), PFID(&fti->fti_fid),
1373                                rc);
1374                         rc = lrc;
1375                 }
1376                 count--;
1377                 ostid_inc_id(&repbody->oa.o_oi);
1378         }
1379
1380         ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_DESTROY,
1381                          tsi->tsi_jobid, 1);
1382 out:
1383         RETURN(rc);
1384 }
1385
1386 static int ofd_statfs_hdl(struct tgt_session_info *tsi)
1387 {
1388         struct obd_statfs       *osfs;
1389         int                      rc;
1390
1391         ENTRY;
1392
1393         osfs = req_capsule_server_get(tsi->tsi_pill, &RMF_OBD_STATFS);
1394
1395         rc = ofd_statfs(tsi->tsi_env, tsi->tsi_exp, osfs,
1396                         cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS), 0);
1397         if (rc != 0)
1398                 CERROR("%s: statfs failed: rc = %d\n",
1399                        tgt_name(tsi->tsi_tgt), rc);
1400
1401         if (OBD_FAIL_CHECK(OBD_FAIL_OST_STATFS_EINPROGRESS))
1402                 rc = -EINPROGRESS;
1403
1404         ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_STATFS,
1405                          tsi->tsi_jobid, 1);
1406
1407         RETURN(rc);
1408 }
1409
1410 static int ofd_sync_hdl(struct tgt_session_info *tsi)
1411 {
1412         struct ost_body         *body = tsi->tsi_ost_body;
1413         struct ost_body         *repbody;
1414         struct ofd_thread_info  *fti = tsi2ofd_info(tsi);
1415         struct ofd_device       *ofd = ofd_exp(tsi->tsi_exp);
1416         struct ofd_object       *fo = NULL;
1417         int                      rc = 0;
1418
1419         ENTRY;
1420
1421         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1422
1423         /* if no objid is specified, it means "sync whole filesystem" */
1424         if (!fid_is_zero(&tsi->tsi_fid)) {
1425                 fo = ofd_object_find_exists(tsi->tsi_env, ofd, &tsi->tsi_fid);
1426                 if (IS_ERR(fo))
1427                         RETURN(PTR_ERR(fo));
1428         }
1429
1430         rc = tgt_sync(tsi->tsi_env, tsi->tsi_tgt,
1431                       fo != NULL ? ofd_object_child(fo) : NULL);
1432         if (rc)
1433                 GOTO(put, rc);
1434
1435         ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_SYNC,
1436                          tsi->tsi_jobid, 1);
1437         if (fo == NULL)
1438                 RETURN(0);
1439
1440         repbody->oa.o_oi = body->oa.o_oi;
1441         repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1442
1443         rc = ofd_attr_get(tsi->tsi_env, fo, &fti->fti_attr);
1444         if (rc == 0)
1445                 obdo_from_la(&repbody->oa, &fti->fti_attr,
1446                              OFD_VALID_FLAGS);
1447         else
1448                 /* don't return rc from getattr */
1449                 rc = 0;
1450         EXIT;
1451 put:
1452         if (fo != NULL)
1453                 ofd_object_put(tsi->tsi_env, fo);
1454         return rc;
1455 }
1456
1457 static int ofd_punch_hdl(struct tgt_session_info *tsi)
1458 {
1459         const struct obdo       *oa = &tsi->tsi_ost_body->oa;
1460         struct ost_body         *repbody;
1461         struct ofd_thread_info  *info = tsi2ofd_info(tsi);
1462         struct ldlm_namespace   *ns = tsi->tsi_tgt->lut_obd->obd_namespace;
1463         struct ldlm_resource    *res;
1464         struct ofd_object       *fo;
1465         struct filter_fid       *ff = NULL;
1466         __u64                    flags = 0;
1467         struct lustre_handle     lh = { 0, };
1468         int                      rc;
1469         __u64                    start, end;
1470         bool                     srvlock;
1471
1472         ENTRY;
1473
1474         /* check that we do support OBD_CONNECT_TRUNCLOCK. */
1475         CLASSERT(OST_CONNECT_SUPPORTED & OBD_CONNECT_TRUNCLOCK);
1476
1477         if ((oa->o_valid & (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS)) !=
1478             (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS))
1479                 RETURN(err_serious(-EPROTO));
1480
1481         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
1482         if (repbody == NULL)
1483                 RETURN(err_serious(-ENOMEM));
1484
1485         /* punch start,end are passed in o_size,o_blocks throught wire */
1486         start = oa->o_size;
1487         end = oa->o_blocks;
1488
1489         if (end != OBD_OBJECT_EOF) /* Only truncate is supported */
1490                 RETURN(-EPROTO);
1491
1492         /* standard truncate optimization: if file body is completely
1493          * destroyed, don't send data back to the server. */
1494         if (start == 0)
1495                 flags |= LDLM_FL_AST_DISCARD_DATA;
1496
1497         repbody->oa.o_oi = oa->o_oi;
1498         repbody->oa.o_valid = OBD_MD_FLID;
1499
1500         srvlock = oa->o_valid & OBD_MD_FLFLAGS &&
1501                   oa->o_flags & OBD_FL_SRVLOCK;
1502
1503         if (srvlock) {
1504                 rc = tgt_extent_lock(ns, &tsi->tsi_resid, start, end, &lh,
1505                                      LCK_PW, &flags);
1506                 if (rc != 0)
1507                         RETURN(rc);
1508         }
1509
1510         CDEBUG(D_INODE, "calling punch for object "DFID", valid = "LPX64
1511                ", start = "LPD64", end = "LPD64"\n", PFID(&tsi->tsi_fid),
1512                oa->o_valid, start, end);
1513
1514         fo = ofd_object_find_exists(tsi->tsi_env, ofd_exp(tsi->tsi_exp),
1515                                     &tsi->tsi_fid);
1516         if (IS_ERR(fo))
1517                 GOTO(out, rc = PTR_ERR(fo));
1518
1519         la_from_obdo(&info->fti_attr, oa,
1520                      OBD_MD_FLMTIME | OBD_MD_FLATIME | OBD_MD_FLCTIME);
1521         info->fti_attr.la_size = start;
1522         info->fti_attr.la_valid |= LA_SIZE;
1523
1524         if (oa->o_valid & OBD_MD_FLFID) {
1525                 ff = &info->fti_mds_fid;
1526                 ofd_prepare_fidea(ff, oa);
1527         }
1528
1529         rc = ofd_object_punch(tsi->tsi_env, fo, start, end, &info->fti_attr,
1530                               ff);
1531         if (rc)
1532                 GOTO(out_put, rc);
1533
1534         ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_PUNCH,
1535                          tsi->tsi_jobid, 1);
1536         EXIT;
1537 out_put:
1538         ofd_object_put(tsi->tsi_env, fo);
1539 out:
1540         if (srvlock)
1541                 tgt_extent_unlock(&lh, LCK_PW);
1542         if (rc == 0) {
1543                 /* we do not call this before to avoid lu_object_find() in
1544                  *  ->lvbo_update() holding another reference on the object.
1545                  * otherwise concurrent destroy can make the object unavailable
1546                  * for 2nd lu_object_find() waiting for the first reference
1547                  * to go... deadlock! */
1548                 res = ldlm_resource_get(ns, NULL, &tsi->tsi_resid,
1549                                         LDLM_EXTENT, 0);
1550                 if (res != NULL) {
1551                         ldlm_res_lvbo_update(res, NULL, 0);
1552                         ldlm_resource_putref(res);
1553                 }
1554         }
1555         return rc;
1556 }
1557
1558
1559 static int ofd_quotactl(struct tgt_session_info *tsi)
1560 {
1561         struct obd_quotactl     *oqctl, *repoqc;
1562         int                      rc;
1563
1564         ENTRY;
1565
1566         oqctl = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_QUOTACTL);
1567         if (oqctl == NULL)
1568                 RETURN(err_serious(-EPROTO));
1569
1570         repoqc = req_capsule_server_get(tsi->tsi_pill, &RMF_OBD_QUOTACTL);
1571         if (repoqc == NULL)
1572                 RETURN(err_serious(-ENOMEM));
1573
1574         /* report success for quota on/off for interoperability with current MDT
1575          * stack */
1576         if (oqctl->qc_cmd == Q_QUOTAON || oqctl->qc_cmd == Q_QUOTAOFF)
1577                 RETURN(0);
1578
1579         *repoqc = *oqctl;
1580         rc = lquotactl_slv(tsi->tsi_env, tsi->tsi_tgt->lut_bottom, repoqc);
1581
1582         ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_QUOTACTL,
1583                          tsi->tsi_jobid, 1);
1584
1585         RETURN(rc);
1586 }
1587
1588 #define OBD_FAIL_OST_READ_NET   OBD_FAIL_OST_BRW_NET
1589 #define OBD_FAIL_OST_WRITE_NET  OBD_FAIL_OST_BRW_NET
1590 #define OST_BRW_READ    OST_READ
1591 #define OST_BRW_WRITE   OST_WRITE
1592
1593 static struct tgt_handler ofd_tgt_handlers[] = {
1594 TGT_RPC_HANDLER(OST_FIRST_OPC,
1595                 0,                      OST_CONNECT,    tgt_connect,
1596                 &RQF_CONNECT, LUSTRE_OBD_VERSION),
1597 TGT_RPC_HANDLER(OST_FIRST_OPC,
1598                 0,                      OST_DISCONNECT, tgt_disconnect,
1599                 &RQF_OST_DISCONNECT, LUSTRE_OBD_VERSION),
1600 TGT_RPC_HANDLER(OST_FIRST_OPC,
1601                 0,                      OST_SET_INFO,   ofd_set_info_hdl,
1602                 &RQF_OBD_SET_INFO, LUSTRE_OST_VERSION),
1603 TGT_OST_HDL(0,                          OST_GET_INFO,   ofd_get_info_hdl),
1604 TGT_OST_HDL(HABEO_CORPUS| HABEO_REFERO, OST_GETATTR,    ofd_getattr_hdl),
1605 TGT_OST_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR,
1606                                         OST_SETATTR,    ofd_setattr_hdl),
1607 TGT_OST_HDL(0           | HABEO_REFERO | MUTABOR,
1608                                         OST_CREATE,     ofd_create_hdl),
1609 TGT_OST_HDL(0           | HABEO_REFERO | MUTABOR,
1610                                         OST_DESTROY,    ofd_destroy_hdl),
1611 TGT_OST_HDL(0           | HABEO_REFERO, OST_STATFS,     ofd_statfs_hdl),
1612 TGT_OST_HDL(HABEO_CORPUS| HABEO_REFERO, OST_BRW_READ,   tgt_brw_read),
1613 /* don't set CORPUS flag for brw_write because -ENOENT may be valid case */
1614 TGT_OST_HDL(MUTABOR,                    OST_BRW_WRITE,  tgt_brw_write),
1615 TGT_OST_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR,
1616                                         OST_PUNCH,      ofd_punch_hdl),
1617 TGT_OST_HDL(HABEO_CORPUS| HABEO_REFERO, OST_SYNC,       ofd_sync_hdl),
1618 TGT_OST_HDL(0           | HABEO_REFERO, OST_QUOTACTL,   ofd_quotactl),
1619 };
1620
1621 static struct tgt_opc_slice ofd_common_slice[] = {
1622         {
1623                 .tos_opc_start  = OST_FIRST_OPC,
1624                 .tos_opc_end    = OST_LAST_OPC,
1625                 .tos_hs         = ofd_tgt_handlers
1626         },
1627         {
1628                 .tos_opc_start  = OBD_FIRST_OPC,
1629                 .tos_opc_end    = OBD_LAST_OPC,
1630                 .tos_hs         = tgt_obd_handlers
1631         },
1632         {
1633                 .tos_opc_start  = LDLM_FIRST_OPC,
1634                 .tos_opc_end    = LDLM_LAST_OPC,
1635                 .tos_hs         = tgt_dlm_handlers
1636         },
1637         {
1638                 .tos_opc_start  = UPDATE_OBJ,
1639                 .tos_opc_end    = UPDATE_LAST_OPC,
1640                 .tos_hs         = tgt_out_handlers
1641         },
1642         {
1643                 .tos_opc_start  = SEQ_FIRST_OPC,
1644                 .tos_opc_end    = SEQ_LAST_OPC,
1645                 .tos_hs         = seq_handlers
1646         },
1647         {
1648                 .tos_hs         = NULL
1649         }
1650 };
1651
1652 static int ofd_init0(const struct lu_env *env, struct ofd_device *m,
1653                      struct lu_device_type *ldt, struct lustre_cfg *cfg)
1654 {
1655         const char              *dev = lustre_cfg_string(cfg, 0);
1656         struct ofd_thread_info  *info = NULL;
1657         struct obd_device       *obd;
1658         struct obd_statfs       *osfs;
1659         int                      rc;
1660
1661         ENTRY;
1662
1663         obd = class_name2obd(dev);
1664         if (obd == NULL) {
1665                 CERROR("Cannot find obd with name %s\n", dev);
1666                 RETURN(-ENODEV);
1667         }
1668
1669         rc = lu_env_refill((struct lu_env *)env);
1670         if (rc != 0)
1671                 RETURN(rc);
1672
1673         obd->u.obt.obt_magic = OBT_MAGIC;
1674
1675         m->ofd_fmd_max_num = OFD_FMD_MAX_NUM_DEFAULT;
1676         m->ofd_fmd_max_age = OFD_FMD_MAX_AGE_DEFAULT;
1677
1678         spin_lock_init(&m->ofd_flags_lock);
1679         m->ofd_raid_degraded = 0;
1680         m->ofd_syncjournal = 0;
1681         ofd_slc_set(m);
1682         m->ofd_grant_compat_disable = 0;
1683         m->ofd_soft_sync_limit = OFD_SOFT_SYNC_LIMIT_DEFAULT;
1684
1685         /* statfs data */
1686         spin_lock_init(&m->ofd_osfs_lock);
1687         m->ofd_osfs_age = cfs_time_shift_64(-1000);
1688         m->ofd_osfs_unstable = 0;
1689         m->ofd_statfs_inflight = 0;
1690         m->ofd_osfs_inflight = 0;
1691
1692         /* grant data */
1693         spin_lock_init(&m->ofd_grant_lock);
1694         m->ofd_tot_dirty = 0;
1695         m->ofd_tot_granted = 0;
1696         m->ofd_tot_pending = 0;
1697         m->ofd_seq_count = 0;
1698
1699         spin_lock_init(&m->ofd_batch_lock);
1700         rwlock_init(&obd->u.filter.fo_sptlrpc_lock);
1701         sptlrpc_rule_set_init(&obd->u.filter.fo_sptlrpc_rset);
1702
1703         obd->u.filter.fo_fl_oss_capa = 0;
1704         CFS_INIT_LIST_HEAD(&obd->u.filter.fo_capa_keys);
1705         obd->u.filter.fo_capa_hash = init_capa_hash();
1706         if (obd->u.filter.fo_capa_hash == NULL)
1707                 RETURN(-ENOMEM);
1708
1709         m->ofd_dt_dev.dd_lu_dev.ld_ops = &ofd_lu_ops;
1710         m->ofd_dt_dev.dd_lu_dev.ld_obd = obd;
1711         /* set this lu_device to obd, because error handling need it */
1712         obd->obd_lu_dev = &m->ofd_dt_dev.dd_lu_dev;
1713
1714         rc = ofd_procfs_init(m);
1715         if (rc) {
1716                 CERROR("Can't init ofd lprocfs, rc %d\n", rc);
1717                 RETURN(rc);
1718         }
1719
1720         /* No connection accepted until configurations will finish */
1721         spin_lock(&obd->obd_dev_lock);
1722         obd->obd_no_conn = 1;
1723         spin_unlock(&obd->obd_dev_lock);
1724         obd->obd_replayable = 1;
1725         if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
1726                 char *str = lustre_cfg_string(cfg, 4);
1727
1728                 if (strchr(str, 'n')) {
1729                         CWARN("%s: recovery disabled\n", obd->obd_name);
1730                         obd->obd_replayable = 0;
1731                 }
1732         }
1733
1734         info = ofd_info_init(env, NULL);
1735         if (info == NULL)
1736                 RETURN(-EFAULT);
1737
1738         rc = ofd_stack_init(env, m, cfg);
1739         if (rc) {
1740                 CERROR("Can't init device stack, rc %d\n", rc);
1741                 GOTO(err_fini_proc, rc);
1742         }
1743
1744         ofd_procfs_add_brw_stats_symlink(m);
1745
1746         /* populate cached statfs data */
1747         osfs = &ofd_info(env)->fti_u.osfs;
1748         rc = ofd_statfs_internal(env, m, osfs, 0, NULL);
1749         if (rc != 0) {
1750                 CERROR("%s: can't get statfs data, rc %d\n", obd->obd_name, rc);
1751                 GOTO(err_fini_stack, rc);
1752         }
1753         if (!IS_PO2(osfs->os_bsize)) {
1754                 CERROR("%s: blocksize (%d) is not a power of 2\n",
1755                                 obd->obd_name, osfs->os_bsize);
1756                 GOTO(err_fini_stack, rc = -EPROTO);
1757         }
1758         m->ofd_blockbits = fls(osfs->os_bsize) - 1;
1759
1760         m->ofd_precreate_batch = OFD_PRECREATE_BATCH_DEFAULT;
1761         if (osfs->os_bsize * osfs->os_blocks < OFD_PRECREATE_SMALL_FS)
1762                 m->ofd_precreate_batch = OFD_PRECREATE_BATCH_SMALL;
1763
1764         snprintf(info->fti_u.name, sizeof(info->fti_u.name), "%s-%s",
1765                  "filter"/*LUSTRE_OST_NAME*/, obd->obd_uuid.uuid);
1766         m->ofd_namespace = ldlm_namespace_new(obd, info->fti_u.name,
1767                                               LDLM_NAMESPACE_SERVER,
1768                                               LDLM_NAMESPACE_GREEDY,
1769                                               LDLM_NS_TYPE_OST);
1770         if (m->ofd_namespace == NULL)
1771                 GOTO(err_fini_stack, rc = -ENOMEM);
1772         /* set obd_namespace for compatibility with old code */
1773         obd->obd_namespace = m->ofd_namespace;
1774         ldlm_register_intent(m->ofd_namespace, ofd_intent_policy);
1775         m->ofd_namespace->ns_lvbo = &ofd_lvbo;
1776         m->ofd_namespace->ns_lvbp = m;
1777
1778         ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
1779                            "filter_ldlm_cb_client", &obd->obd_ldlm_client);
1780
1781         dt_conf_get(env, m->ofd_osd, &m->ofd_dt_conf);
1782
1783         /* Allow at most ddp_grant_reserved% of the available filesystem space
1784          * to be granted to clients, so that any errors in the grant overhead
1785          * calculations do not allow granting more space to clients than can be
1786          * written. Assumes that in aggregate the grant overhead calculations do
1787          * not have more than ddp_grant_reserved% estimation error in them. */
1788         m->ofd_grant_ratio =
1789                 ofd_grant_ratio_conv(m->ofd_dt_conf.ddp_grant_reserved);
1790
1791         rc = tgt_init(env, &m->ofd_lut, obd, m->ofd_osd, ofd_common_slice,
1792                       OBD_FAIL_OST_ALL_REQUEST_NET,
1793                       OBD_FAIL_OST_ALL_REPLY_NET);
1794         if (rc)
1795                 GOTO(err_free_ns, rc);
1796
1797         rc = ofd_fs_setup(env, m, obd);
1798         if (rc)
1799                 GOTO(err_fini_lut, rc);
1800
1801         RETURN(0);
1802 err_fini_lut:
1803         tgt_fini(env, &m->ofd_lut);
1804 err_free_ns:
1805         ldlm_namespace_free(m->ofd_namespace, 0, obd->obd_force);
1806         obd->obd_namespace = m->ofd_namespace = NULL;
1807 err_fini_stack:
1808         ofd_stack_fini(env, m, &m->ofd_osd->dd_lu_dev);
1809 err_fini_proc:
1810         ofd_procfs_fini(m);
1811         return rc;
1812 }
1813
1814 static void ofd_fini(const struct lu_env *env, struct ofd_device *m)
1815 {
1816         struct obd_device *obd = ofd_obd(m);
1817         struct lu_device  *d = &m->ofd_dt_dev.dd_lu_dev;
1818
1819         lfsck_stop(env, m->ofd_osd, true);
1820         lfsck_degister(env, m->ofd_osd);
1821         target_recovery_fini(obd);
1822         obd_exports_barrier(obd);
1823         obd_zombie_barrier();
1824
1825         tgt_fini(env, &m->ofd_lut);
1826         ofd_fs_cleanup(env, m);
1827
1828         ofd_free_capa_keys(m);
1829         cleanup_capa_hash(obd->u.filter.fo_capa_hash);
1830
1831         if (m->ofd_namespace != NULL) {
1832                 ldlm_namespace_free(m->ofd_namespace, NULL,
1833                                     d->ld_obd->obd_force);
1834                 d->ld_obd->obd_namespace = m->ofd_namespace = NULL;
1835         }
1836
1837         ofd_stack_fini(env, m, &m->ofd_dt_dev.dd_lu_dev);
1838         ofd_procfs_fini(m);
1839         LASSERT(cfs_atomic_read(&d->ld_ref) == 0);
1840         server_put_mount(obd->obd_name);
1841         EXIT;
1842 }
1843
1844 static struct lu_device *ofd_device_fini(const struct lu_env *env,
1845                                          struct lu_device *d)
1846 {
1847         ENTRY;
1848         ofd_fini(env, ofd_dev(d));
1849         RETURN(NULL);
1850 }
1851
1852 static struct lu_device *ofd_device_free(const struct lu_env *env,
1853                                          struct lu_device *d)
1854 {
1855         struct ofd_device *m = ofd_dev(d);
1856
1857         dt_device_fini(&m->ofd_dt_dev);
1858         OBD_FREE_PTR(m);
1859         RETURN(NULL);
1860 }
1861
1862 static struct lu_device *ofd_device_alloc(const struct lu_env *env,
1863                                           struct lu_device_type *t,
1864                                           struct lustre_cfg *cfg)
1865 {
1866         struct ofd_device *m;
1867         struct lu_device  *l;
1868         int                rc;
1869
1870         OBD_ALLOC_PTR(m);
1871         if (m == NULL)
1872                 return ERR_PTR(-ENOMEM);
1873
1874         l = &m->ofd_dt_dev.dd_lu_dev;
1875         dt_device_init(&m->ofd_dt_dev, t);
1876         rc = ofd_init0(env, m, t, cfg);
1877         if (rc != 0) {
1878                 ofd_device_free(env, l);
1879                 l = ERR_PTR(rc);
1880         }
1881
1882         return l;
1883 }
1884
1885 /* thread context key constructor/destructor */
1886 LU_KEY_INIT_FINI(ofd, struct ofd_thread_info);
1887
1888 static void ofd_key_exit(const struct lu_context *ctx,
1889                          struct lu_context_key *key, void *data)
1890 {
1891         struct ofd_thread_info *info = data;
1892
1893         info->fti_env = NULL;
1894         info->fti_exp = NULL;
1895
1896         info->fti_xid = 0;
1897         info->fti_pre_version = 0;
1898         info->fti_used = 0;
1899
1900         memset(&info->fti_attr, 0, sizeof info->fti_attr);
1901 }
1902
1903 struct lu_context_key ofd_thread_key = {
1904         .lct_tags = LCT_DT_THREAD,
1905         .lct_init = ofd_key_init,
1906         .lct_fini = ofd_key_fini,
1907         .lct_exit = ofd_key_exit
1908 };
1909
1910 /* type constructor/destructor: mdt_type_init, mdt_type_fini */
1911 LU_TYPE_INIT_FINI(ofd, &ofd_thread_key);
1912
1913 static struct lu_device_type_operations ofd_device_type_ops = {
1914         .ldto_init              = ofd_type_init,
1915         .ldto_fini              = ofd_type_fini,
1916
1917         .ldto_start             = ofd_type_start,
1918         .ldto_stop              = ofd_type_stop,
1919
1920         .ldto_device_alloc      = ofd_device_alloc,
1921         .ldto_device_free       = ofd_device_free,
1922         .ldto_device_fini       = ofd_device_fini
1923 };
1924
1925 static struct lu_device_type ofd_device_type = {
1926         .ldt_tags       = LU_DEVICE_DT,
1927         .ldt_name       = LUSTRE_OST_NAME,
1928         .ldt_ops        = &ofd_device_type_ops,
1929         .ldt_ctx_tags   = LCT_DT_THREAD
1930 };
1931
1932 int __init ofd_init(void)
1933 {
1934         struct lprocfs_static_vars      lvars;
1935         int                             rc;
1936
1937         rc = lu_kmem_init(ofd_caches);
1938         if (rc)
1939                 return rc;
1940
1941         rc = ofd_fmd_init();
1942         if (rc) {
1943                 lu_kmem_fini(ofd_caches);
1944                 return(rc);
1945         }
1946
1947         lprocfs_ofd_init_vars(&lvars);
1948
1949         rc = class_register_type(&ofd_obd_ops, NULL, NULL,
1950 #ifndef HAVE_ONLY_PROCFS_SEQ
1951                                 lvars.module_vars,
1952 #endif
1953                                 LUSTRE_OST_NAME, &ofd_device_type);
1954         return rc;
1955 }
1956
1957 void __exit ofd_exit(void)
1958 {
1959         ofd_fmd_exit();
1960         lu_kmem_fini(ofd_caches);
1961         class_unregister_type(LUSTRE_OST_NAME);
1962 }
1963
1964 MODULE_AUTHOR("Whamcloud, Inc. <http://www.whamcloud.com/>");
1965 MODULE_DESCRIPTION("Lustre Object Filtering Device");
1966 MODULE_LICENSE("GPL");
1967
1968 module_init(ofd_init);
1969 module_exit(ofd_exit);