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