lustre/osd-zfs/osd_object.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 /*
  31  * Copyright (c) 2012, Intel Corporation.
  32  * Use is subject to license terms.
  33  */
  34 /*
  35  * This file is part of Lustre, http://www.lustre.org/
  36  * Lustre is a trademark of Sun Microsystems, Inc.
  37  *
  38  * lustre/osd-zfs/osd_object.c
  39  *
  40  * Author: Alex Zhuravlev <bzzz@whamcloud.com>
  41  * Author: Mike Pershin <tappro@whamcloud.com>
  42  * Author: Johann Lombardi <johann@whamcloud.com>
  43  */
  44
  45 #ifndef EXPORT_SYMTAB
  46 # define EXPORT_SYMTAB
  47 #endif
  48 #define DEBUG_SUBSYSTEM S_OSD
  49
  50 #include <lustre_ver.h>
  51 #include <libcfs/libcfs.h>
  52 #include <lustre_fsfilt.h>
  53 #include <obd_support.h>
  54 #include <lustre_net.h>
  55 #include <obd.h>
  56 #include <obd_class.h>
  57 #include <lustre_disk.h>
  58 #include <lustre_fid.h>
  59
  60 #include "osd_internal.h"
  61
  62 #include <sys/dnode.h>
  63 #include <sys/dbuf.h>
  64 #include <sys/spa.h>
  65 #include <sys/stat.h>
  66 #include <sys/zap.h>
  67 #include <sys/spa_impl.h>
  68 #include <sys/zfs_znode.h>
  69 #include <sys/dmu_tx.h>
  70 #include <sys/dmu_objset.h>
  71 #include <sys/dsl_prop.h>
  72 #include <sys/sa_impl.h>
  73 #include <sys/txg.h>
  74
  75 static char *osd_obj_tag = "osd_object";
  76
  77 static struct dt_object_operations osd_obj_ops;
  78 static struct lu_object_operations osd_lu_obj_ops;
  79 extern struct dt_body_operations osd_body_ops;
  80 static struct dt_object_operations osd_obj_otable_it_ops;
  81
  82 extern cfs_mem_cache_t *osd_object_kmem;
  83
  84 static void
  85 osd_object_sa_fini(struct osd_object *obj)
  86 {
  87         if (obj->oo_sa_hdl) {
  88                 sa_handle_destroy(obj->oo_sa_hdl);
  89                 obj->oo_sa_hdl = NULL;
  90         }
  91 }
  92
  93 static int
  94 osd_object_sa_init(struct osd_object *obj, udmu_objset_t *uos)
  95 {
  96         int rc;
  97
  98         LASSERT(obj->oo_sa_hdl == NULL);
  99         LASSERT(obj->oo_db != NULL);
 100
 101         rc = -sa_handle_get(uos->os, obj->oo_db->db_object, obj,
 102                             SA_HDL_PRIVATE, &obj->oo_sa_hdl);
 103         if (rc)
 104                 return rc;
 105
 106         /* Cache the xattr object id, valid for the life of the object */
 107         rc = -sa_lookup(obj->oo_sa_hdl, SA_ZPL_XATTR(uos), &obj->oo_xattr, 8);
 108         if (rc == -ENOENT) {
 109                 obj->oo_xattr = ZFS_NO_OBJECT;
 110                 rc = 0;
 111         } else if (rc) {
 112                 osd_object_sa_fini(obj);
 113         }
 114
 115         return rc;
 116 }
 117
 118 /*
 119  * Add object to list of dirty objects in tx handle.
 120  */
 121 static void
 122 osd_object_sa_dirty_add(struct osd_object *obj, struct osd_thandle *oh)
 123 {
 124         if (!cfs_list_empty(&obj->oo_sa_linkage))
 125                 return;
 126
 127         down(&oh->ot_sa_lock);
 128         write_lock(&obj->oo_attr_lock);
 129         if (likely(cfs_list_empty(&obj->oo_sa_linkage)))
 130                 cfs_list_add(&obj->oo_sa_linkage, &oh->ot_sa_list);
 131         write_unlock(&obj->oo_attr_lock);
 132         up(&oh->ot_sa_lock);
 133 }
 134
 135 /*
 136  * Release spill block dbuf hold for all dirty SAs.
 137  */
 138 void osd_object_sa_dirty_rele(struct osd_thandle *oh)
 139 {
 140         struct osd_object *obj;
 141
 142         down(&oh->ot_sa_lock);
 143         while (!cfs_list_empty(&oh->ot_sa_list)) {
 144                 obj = cfs_list_entry(oh->ot_sa_list.next,
 145                                      struct osd_object, oo_sa_linkage);
 146                 sa_spill_rele(obj->oo_sa_hdl);
 147                 write_lock(&obj->oo_attr_lock);
 148                 cfs_list_del_init(&obj->oo_sa_linkage);
 149                 write_unlock(&obj->oo_attr_lock);
 150         }
 151         up(&oh->ot_sa_lock);
 152 }
 153
 154 /*
 155  * Update the SA and add the object to the dirty list.
 156  */
 157 int osd_object_sa_update(struct osd_object *obj, sa_attr_type_t type,
 158                          void *buf, uint32_t buflen, struct osd_thandle *oh)
 159 {
 160         int rc;
 161
 162         LASSERT(obj->oo_sa_hdl != NULL);
 163         LASSERT(oh->ot_tx != NULL);
 164
 165         rc = -sa_update(obj->oo_sa_hdl, type, buf, buflen, oh->ot_tx);
 166         osd_object_sa_dirty_add(obj, oh);
 167
 168         return rc;
 169 }
 170
 171 /*
 172  * Bulk update the SA and add the object to the dirty list.
 173  */
 174 static int
 175 osd_object_sa_bulk_update(struct osd_object *obj, sa_bulk_attr_t *attrs,
 176                           int count, struct osd_thandle *oh)
 177 {
 178         int rc;
 179
 180         LASSERT(obj->oo_sa_hdl != NULL);
 181         LASSERT(oh->ot_tx != NULL);
 182
 183         rc = -sa_bulk_update(obj->oo_sa_hdl, attrs, count, oh->ot_tx);
 184         osd_object_sa_dirty_add(obj, oh);
 185
 186         return rc;
 187 }
 188
 189 /*
 190  * Retrieve the attributes of a DMU object
 191  */
 192 int __osd_object_attr_get(const struct lu_env *env, udmu_objset_t *uos,
 193                           struct osd_object *obj, struct lu_attr *la)
 194 {
 195         struct osa_attr *osa = &osd_oti_get(env)->oti_osa;
 196         sa_handle_t     *sa_hdl;
 197         sa_bulk_attr_t  *bulk;
 198         int              cnt = 0;
 199         int              rc;
 200         ENTRY;
 201
 202         LASSERT(obj->oo_db != NULL);
 203
 204         rc = -sa_handle_get(uos->os, obj->oo_db->db_object, NULL,
 205                             SA_HDL_PRIVATE, &sa_hdl);
 206         if (rc)
 207                 RETURN(rc);
 208
 209         OBD_ALLOC(bulk, sizeof(sa_bulk_attr_t) * 9);
 210         if (bulk == NULL)
 211                 GOTO(out_sa, rc = -ENOMEM);
 212
 213         la->la_valid |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE | LA_TYPE |
 214                         LA_SIZE | LA_UID | LA_GID | LA_FLAGS | LA_NLINK;
 215
 216         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(uos), NULL, osa->atime, 16);
 217         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(uos), NULL, osa->mtime, 16);
 218         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(uos), NULL, osa->ctime, 16);
 219         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(uos), NULL, &osa->mode, 8);
 220         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(uos), NULL, &osa->size, 8);
 221         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(uos), NULL, &osa->nlink, 8);
 222         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(uos), NULL, &osa->uid, 8);
 223         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(uos), NULL, &osa->gid, 8);
 224         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(uos), NULL, &osa->flags, 8);
 225
 226         rc = -sa_bulk_lookup(sa_hdl, bulk, cnt);
 227         if (rc)
 228                 GOTO(out_bulk, rc);
 229
 230         la->la_atime = osa->atime[0];
 231         la->la_mtime = osa->mtime[0];
 232         la->la_ctime = osa->ctime[0];
 233         la->la_mode = osa->mode;
 234         la->la_uid = osa->uid;
 235         la->la_gid = osa->gid;
 236         la->la_nlink = osa->nlink;
 237         la->la_flags = osa->flags;
 238         la->la_size = osa->size;
 239
 240         if (S_ISCHR(la->la_mode) || S_ISBLK(la->la_mode)) {
 241                 rc = -sa_lookup(sa_hdl, SA_ZPL_RDEV(uos), &osa->rdev, 8);
 242                 if (rc)
 243                         GOTO(out_bulk, rc);
 244                 la->la_rdev = osa->rdev;
 245                 la->la_valid |= LA_RDEV;
 246         }
 247 out_bulk:
 248         OBD_FREE(bulk, sizeof(sa_bulk_attr_t) * 9);
 249 out_sa:
 250         sa_handle_destroy(sa_hdl);
 251
 252         RETURN(rc);
 253 }
 254
 255 int __osd_obj2dbuf(const struct lu_env *env, objset_t *os,
 256                    uint64_t oid, dmu_buf_t **dbp, void *tag)
 257 {
 258         dmu_object_info_t *doi = &osd_oti_get(env)->oti_doi;
 259         int rc;
 260
 261         LASSERT(tag);
 262
 263         rc = -sa_buf_hold(os, oid, tag, dbp);
 264         if (rc)
 265                 return rc;
 266
 267         dmu_object_info_from_db(*dbp, doi);
 268         if (unlikely (oid != DMU_USERUSED_OBJECT &&
 269             oid != DMU_GROUPUSED_OBJECT && doi->doi_bonus_type != DMU_OT_SA)) {
 270                 sa_buf_rele(*dbp, tag);
 271                 *dbp = NULL;
 272                 return -EINVAL;
 273         }
 274
 275         LASSERT(*dbp);
 276         LASSERT((*dbp)->db_object == oid);
 277         LASSERT((*dbp)->db_offset == -1);
 278         LASSERT((*dbp)->db_data != NULL);
 279
 280         return 0;
 281 }
 282
 283 /*
 284  * Concurrency: no concurrent access is possible that early in object
 285  * life-cycle.
 286  */
 287 struct lu_object *osd_object_alloc(const struct lu_env *env,
 288                                    const struct lu_object_header *hdr,
 289                                    struct lu_device *d)
 290 {
 291         struct osd_object *mo;
 292
 293         OBD_SLAB_ALLOC_PTR_GFP(mo, osd_object_kmem, CFS_ALLOC_IO);
 294         if (mo != NULL) {
 295                 struct lu_object *l;
 296
 297                 l = &mo->oo_dt.do_lu;
 298                 dt_object_init(&mo->oo_dt, NULL, d);
 299                 mo->oo_dt.do_ops = &osd_obj_ops;
 300                 l->lo_ops = &osd_lu_obj_ops;
 301                 CFS_INIT_LIST_HEAD(&mo->oo_sa_linkage);
 302                 init_rwsem(&mo->oo_sem);
 303                 sema_init(&mo->oo_guard, 1);
 304                 rwlock_init(&mo->oo_attr_lock);
 305                 return l;
 306         } else {
 307                 return NULL;
 308         }
 309 }
 310
 311 /*
 312  * Concurrency: shouldn't matter.
 313  */
 314 int osd_object_init0(const struct lu_env *env, struct osd_object *obj)
 315 {
 316         struct osd_device       *osd = osd_obj2dev(obj);
 317         udmu_objset_t           *uos = &osd->od_objset;
 318         const struct lu_fid     *fid  = lu_object_fid(&obj->oo_dt.do_lu);
 319         int                      rc = 0;
 320         ENTRY;
 321
 322         if (obj->oo_db == NULL)
 323                 RETURN(0);
 324
 325         /* object exist */
 326
 327         rc = osd_object_sa_init(obj, uos);
 328         if (rc)
 329                 RETURN(rc);
 330
 331         /* cache attrs in object */
 332         rc = __osd_object_attr_get(env, &osd->od_objset,
 333                                    obj, &obj->oo_attr);
 334         if (rc)
 335                 RETURN(rc);
 336
 337         if (likely(!fid_is_acct(fid)))
 338                 /* no body operations for accounting objects */
 339                 obj->oo_dt.do_body_ops = &osd_body_ops;
 340
 341         /*
 342          * initialize object before marking it existing
 343          */
 344         obj->oo_dt.do_lu.lo_header->loh_attr |= obj->oo_attr.la_mode & S_IFMT;
 345
 346         cfs_mb();
 347         obj->oo_dt.do_lu.lo_header->loh_attr |= LOHA_EXISTS;
 348
 349         RETURN(0);
 350 }
 351
 352 /*
 353  * Concurrency: no concurrent access is possible that early in object
 354  * life-cycle.
 355  */
 356 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
 357                            const struct lu_object_conf *conf)
 358 {
 359         struct osd_object       *obj = osd_obj(l);
 360         struct osd_device       *osd = osd_obj2dev(obj);
 361         uint64_t                 oid;
 362         int                      rc;
 363         ENTRY;
 364
 365         LASSERT(osd_invariant(obj));
 366
 367         rc = osd_fid_lookup(env, osd, lu_object_fid(l), &oid);
 368         if (rc == 0) {
 369                 LASSERT(obj->oo_db == NULL);
 370                 rc = __osd_obj2dbuf(env, osd->od_objset.os, oid,
 371                                         &obj->oo_db, osd_obj_tag);
 372                 if (rc == 0) {
 373                         LASSERT(obj->oo_db);
 374                         rc = osd_object_init0(env, obj);
 375                 } else {
 376                         CERROR("%s: lookup "DFID"/"LPX64" failed: rc = %d\n",
 377                                osd->od_svname, PFID(lu_object_fid(l)), oid, rc);
 378                 }
 379         } else if (rc == -ENOENT) {
 380                 if (fid_is_otable_it(&l->lo_header->loh_fid)) {
 381                         obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
 382                         /* LFSCK iterator object is special without inode */
 383                         l->lo_header->loh_attr |= LOHA_EXISTS;
 384                 }
 385                 rc = 0;
 386         }
 387         LASSERT(osd_invariant(obj));
 388         RETURN(rc);
 389 }
 390
 391 /*
 392  * Concurrency: no concurrent access is possible that late in object
 393  * life-cycle.
 394  */
 395 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
 396 {
 397         struct osd_object *obj = osd_obj(l);
 398
 399         LASSERT(osd_invariant(obj));
 400
 401         dt_object_fini(&obj->oo_dt);
 402         OBD_SLAB_FREE_PTR(obj, osd_object_kmem);
 403 }
 404
 405 static void __osd_declare_object_destroy(const struct lu_env *env,
 406                                          struct osd_object *obj,
 407                                          struct osd_thandle *oh)
 408 {
 409         struct osd_device       *osd = osd_obj2dev(obj);
 410         udmu_objset_t           *uos = &osd->od_objset;
 411         dmu_buf_t               *db = obj->oo_db;
 412         zap_attribute_t         *za = &osd_oti_get(env)->oti_za;
 413         uint64_t                 oid = db->db_object, xid;
 414         dmu_tx_t                *tx = oh->ot_tx;
 415         zap_cursor_t            *zc;
 416         int                      rc = 0;
 417
 418         dmu_tx_hold_free(tx, oid, 0, DMU_OBJECT_END);
 419
 420         /* zap holding xattrs */
 421         if (obj->oo_xattr != ZFS_NO_OBJECT) {
 422                 oid = obj->oo_xattr;
 423
 424                 dmu_tx_hold_free(tx, oid, 0, DMU_OBJECT_END);
 425
 426                 rc = -udmu_zap_cursor_init(&zc, uos, oid, 0);
 427                 if (rc)
 428                         goto out;
 429
 430                 while ((rc = -zap_cursor_retrieve(zc, za)) == 0) {
 431                         BUG_ON(za->za_integer_length != sizeof(uint64_t));
 432                         BUG_ON(za->za_num_integers != 1);
 433
 434                         rc = -zap_lookup(uos->os, obj->oo_xattr, za->za_name,
 435                                          sizeof(uint64_t), 1, &xid);
 436                         if (rc) {
 437                                 CERROR("%s: xattr lookup failed: rc = %d\n",
 438                                        osd->od_svname, rc);
 439                                 goto out_err;
 440                         }
 441                         dmu_tx_hold_free(tx, xid, 0, DMU_OBJECT_END);
 442
 443                         zap_cursor_advance(zc);
 444                 }
 445                 if (rc == -ENOENT)
 446                         rc = 0;
 447 out_err:
 448                 udmu_zap_cursor_fini(zc);
 449         }
 450 out:
 451         if (rc && tx->tx_err == 0)
 452                 tx->tx_err = -rc;
 453 }
 454
 455 static int osd_declare_object_destroy(const struct lu_env *env,
 456                                       struct dt_object *dt,
 457                                       struct thandle *th)
 458 {
 459         char                    *buf = osd_oti_get(env)->oti_str;
 460         const struct lu_fid     *fid = lu_object_fid(&dt->do_lu);
 461         struct osd_object       *obj = osd_dt_obj(dt);
 462         struct osd_device       *osd = osd_obj2dev(obj);
 463         struct osd_thandle      *oh;
 464         uint64_t                 zapid;
 465         int                      rc;
 466         ENTRY;
 467
 468         LASSERT(th != NULL);
 469         LASSERT(dt_object_exists(dt));
 470
 471         oh = container_of0(th, struct osd_thandle, ot_super);
 472         LASSERT(oh->ot_tx != NULL);
 473
 474         /* declare that we'll destroy the object */
 475         __osd_declare_object_destroy(env, obj, oh);
 476
 477         /* declare that we'll remove object from fid-dnode mapping */
 478         zapid = osd_get_name_n_idx(env, osd, fid, buf);
 479         dmu_tx_hold_bonus(oh->ot_tx, zapid);
 480         dmu_tx_hold_zap(oh->ot_tx, zapid, 0, buf);
 481
 482         /* declare that we'll remove object from inode accounting ZAPs */
 483         dmu_tx_hold_bonus(oh->ot_tx, osd->od_iusr_oid);
 484         dmu_tx_hold_zap(oh->ot_tx, osd->od_iusr_oid, 0, buf);
 485         dmu_tx_hold_bonus(oh->ot_tx, osd->od_igrp_oid);
 486         dmu_tx_hold_zap(oh->ot_tx, osd->od_igrp_oid, 0, buf);
 487
 488         /* one less inode */
 489         rc = osd_declare_quota(env, osd, obj->oo_attr.la_uid,
 490                                obj->oo_attr.la_gid, -1, oh, false, NULL, false);
 491         if (rc)
 492                 RETURN(rc);
 493
 494         /* data to be truncated */
 495         rc = osd_declare_quota(env, osd, obj->oo_attr.la_uid,
 496                                obj->oo_attr.la_gid, 0, oh, true, NULL, false);
 497         RETURN(rc);
 498 }
 499
 500 int __osd_object_free(udmu_objset_t *uos, uint64_t oid, dmu_tx_t *tx)
 501 {
 502         LASSERT(uos->objects != 0);
 503         spin_lock(&uos->lock);
 504         uos->objects--;
 505         spin_unlock(&uos->lock);
 506
 507         return -dmu_object_free(uos->os, oid, tx);
 508 }
 509
 510 /*
 511  * Delete a DMU object
 512  *
 513  * The transaction passed to this routine must have
 514  * dmu_tx_hold_free(tx, oid, 0, DMU_OBJECT_END) called
 515  * and then assigned to a transaction group.
 516  *
 517  * This will release db and set it to NULL to prevent further dbuf releases.
 518  */
 519 static int __osd_object_destroy(const struct lu_env *env,
 520                                 struct osd_object *obj,
 521                                 dmu_tx_t *tx, void *tag)
 522 {
 523         struct osd_device       *osd = osd_obj2dev(obj);
 524         udmu_objset_t           *uos = &osd->od_objset;
 525         uint64_t                 xid;
 526         zap_attribute_t         *za = &osd_oti_get(env)->oti_za;
 527         zap_cursor_t            *zc;
 528         int                      rc;
 529
 530         /* Assert that the transaction has been assigned to a
 531            transaction group. */
 532         LASSERT(tx->tx_txg != 0);
 533
 534         /* zap holding xattrs */
 535         if (obj->oo_xattr != ZFS_NO_OBJECT) {
 536                 rc = -udmu_zap_cursor_init(&zc, uos, obj->oo_xattr, 0);
 537                 if (rc)
 538                         return rc;
 539                 while ((rc = -zap_cursor_retrieve(zc, za)) == 0) {
 540                         BUG_ON(za->za_integer_length != sizeof(uint64_t));
 541                         BUG_ON(za->za_num_integers != 1);
 542
 543                         rc = -zap_lookup(uos->os, obj->oo_xattr, za->za_name,
 544                                          sizeof(uint64_t), 1, &xid);
 545                         if (rc) {
 546                                 CERROR("%s: lookup xattr %s failed: rc = %d\n",
 547                                        osd->od_svname, za->za_name, rc);
 548                                 continue;
 549                         }
 550                         rc = __osd_object_free(uos, xid, tx);
 551                         if (rc)
 552                                 CERROR("%s: fetch xattr %s failed: rc = %d\n",
 553                                        osd->od_svname, za->za_name, rc);
 554                         zap_cursor_advance(zc);
 555                 }
 556                 udmu_zap_cursor_fini(zc);
 557
 558                 rc = __osd_object_free(uos, obj->oo_xattr, tx);
 559                 if (rc)
 560                         CERROR("%s: freeing xattr failed: rc = %d\n",
 561                                osd->od_svname, rc);
 562         }
 563
 564         return __osd_object_free(uos, obj->oo_db->db_object, tx);
 565 }
 566
 567 static int osd_object_destroy(const struct lu_env *env,
 568                               struct dt_object *dt,
 569                               struct thandle *th)
 570 {
 571         char                    *buf = osd_oti_get(env)->oti_str;
 572         struct osd_object       *obj = osd_dt_obj(dt);
 573         struct osd_device       *osd = osd_obj2dev(obj);
 574         const struct lu_fid     *fid = lu_object_fid(&dt->do_lu);
 575         struct osd_thandle      *oh;
 576         int                      rc;
 577         uint64_t                 zapid;
 578         ENTRY;
 579
 580         LASSERT(obj->oo_db != NULL);
 581         LASSERT(dt_object_exists(dt));
 582         LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
 583
 584         oh = container_of0(th, struct osd_thandle, ot_super);
 585         LASSERT(oh != NULL);
 586         LASSERT(oh->ot_tx != NULL);
 587
 588         zapid = osd_get_name_n_idx(env, osd, fid, buf);
 589
 590         /* remove obj ref from index dir (it depends) */
 591         rc = -zap_remove(osd->od_objset.os, zapid, buf, oh->ot_tx);
 592         if (rc) {
 593                 CERROR("%s: zap_remove() failed: rc = %d\n",
 594                        osd->od_svname, rc);
 595                 GOTO(out, rc);
 596         }
 597
 598         /* Remove object from inode accounting. It is not fatal for the destroy
 599          * operation if something goes wrong while updating accounting, but we
 600          * still log an error message to notify the administrator */
 601         rc = -zap_increment_int(osd->od_objset.os, osd->od_iusr_oid,
 602                         obj->oo_attr.la_uid, -1, oh->ot_tx);
 603         if (rc)
 604                 CERROR("%s: failed to remove "DFID" from accounting ZAP for usr"
 605                         " %d: rc = %d\n", osd->od_svname, PFID(fid),
 606                         obj->oo_attr.la_uid, rc);
 607         rc = -zap_increment_int(osd->od_objset.os, osd->od_igrp_oid,
 608                                 obj->oo_attr.la_gid, -1, oh->ot_tx);
 609         if (rc)
 610                 CERROR("%s: failed to remove "DFID" from accounting ZAP for grp"
 611                         " %d: rc = %d\n", osd->od_svname, PFID(fid),
 612                         obj->oo_attr.la_gid, rc);
 613
 614         /* kill object */
 615         rc = __osd_object_destroy(env, obj, oh->ot_tx, osd_obj_tag);
 616         if (rc) {
 617                 CERROR("%s: __osd_object_destroy() failed: rc = %d\n",
 618                        osd->od_svname, rc);
 619                 GOTO(out, rc);
 620         }
 621
 622 out:
 623         /* not needed in the cache anymore */
 624         set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
 625
 626         RETURN (0);
 627 }
 628
 629 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
 630 {
 631         struct osd_object *obj = osd_obj(l);
 632
 633         if (obj->oo_db != NULL) {
 634                 osd_object_sa_fini(obj);
 635                 if (obj->oo_sa_xattr) {
 636                         nvlist_free(obj->oo_sa_xattr);
 637                         obj->oo_sa_xattr = NULL;
 638                 }
 639                 sa_buf_rele(obj->oo_db, osd_obj_tag);
 640                 cfs_list_del(&obj->oo_sa_linkage);
 641                 obj->oo_db = NULL;
 642         }
 643 }
 644
 645 /*
 646  * Concurrency: ->loo_object_release() is called under site spin-lock.
 647  */
 648 static void osd_object_release(const struct lu_env *env,
 649                                struct lu_object *l)
 650 {
 651 }
 652
 653 /*
 654  * Concurrency: shouldn't matter.
 655  */
 656 static int osd_object_print(const struct lu_env *env, void *cookie,
 657                             lu_printer_t p, const struct lu_object *l)
 658 {
 659         struct osd_object *o = osd_obj(l);
 660
 661         return (*p)(env, cookie, LUSTRE_OSD_ZFS_NAME"-object@%p", o);
 662 }
 663
 664 static void osd_object_read_lock(const struct lu_env *env,
 665                                  struct dt_object *dt, unsigned role)
 666 {
 667         struct osd_object *obj = osd_dt_obj(dt);
 668
 669         LASSERT(osd_invariant(obj));
 670
 671         down_read(&obj->oo_sem);
 672 }
 673
 674 static void osd_object_write_lock(const struct lu_env *env,
 675                                   struct dt_object *dt, unsigned role)
 676 {
 677         struct osd_object *obj = osd_dt_obj(dt);
 678
 679         LASSERT(osd_invariant(obj));
 680
 681         down_write(&obj->oo_sem);
 682 }
 683
 684 static void osd_object_read_unlock(const struct lu_env *env,
 685                                    struct dt_object *dt)
 686 {
 687         struct osd_object *obj = osd_dt_obj(dt);
 688
 689         LASSERT(osd_invariant(obj));
 690         up_read(&obj->oo_sem);
 691 }
 692
 693 static void osd_object_write_unlock(const struct lu_env *env,
 694                                     struct dt_object *dt)
 695 {
 696         struct osd_object *obj = osd_dt_obj(dt);
 697
 698         LASSERT(osd_invariant(obj));
 699         up_write(&obj->oo_sem);
 700 }
 701
 702 static int osd_object_write_locked(const struct lu_env *env,
 703                                    struct dt_object *dt)
 704 {
 705         struct osd_object *obj = osd_dt_obj(dt);
 706         int rc = 1;
 707
 708         LASSERT(osd_invariant(obj));
 709
 710         if (down_write_trylock(&obj->oo_sem)) {
 711                 rc = 0;
 712                 up_write(&obj->oo_sem);
 713         }
 714         return rc;
 715 }
 716
 717 static int osd_attr_get(const struct lu_env *env,
 718                         struct dt_object *dt,
 719                         struct lu_attr *attr,
 720                         struct lustre_capa *capa)
 721 {
 722         struct osd_object       *obj = osd_dt_obj(dt);
 723         uint64_t                 blocks;
 724         uint32_t                 blksize;
 725
 726         LASSERT(dt_object_exists(dt));
 727         LASSERT(osd_invariant(obj));
 728         LASSERT(obj->oo_db);
 729
 730         read_lock(&obj->oo_attr_lock);
 731         *attr = obj->oo_attr;
 732         read_unlock(&obj->oo_attr_lock);
 733
 734         /* with ZFS_DEBUG zrl_add_debug() called by DB_DNODE_ENTER()
 735          * from within sa_object_size() can block on a mutex, so
 736          * we can't call sa_object_size() holding rwlock */
 737         sa_object_size(obj->oo_sa_hdl, &blksize, &blocks);
 738         /* we do not control size of indices, so always calculate
 739          * it from number of blocks reported by DMU */
 740         if (S_ISDIR(attr->la_mode))
 741                 attr->la_size = 512 * blocks;
 742         /* Block size may be not set; suggest maximal I/O transfers. */
 743         if (blksize == 0)
 744                 blksize = 1ULL << SPA_MAXBLOCKSHIFT;
 745
 746         attr->la_blksize = blksize;
 747         attr->la_blocks = blocks;
 748         attr->la_valid |= LA_BLOCKS | LA_BLKSIZE;
 749
 750         return 0;
 751 }
 752
 753 /* Simple wrapper on top of qsd API which implement quota transfer for osd
 754  * setattr needs. As a reminder, only the root user can change ownership of
 755  * a file, that's why EDQUOT & EINPROGRESS errors are discarded */
 756 static inline int qsd_transfer(const struct lu_env *env,
 757                                struct qsd_instance *qsd,
 758                                struct lquota_trans *trans, int qtype,
 759                                __u64 orig_id, __u64 new_id, __u64 bspace,
 760                                struct lquota_id_info *qi)
 761 {
 762         int     rc;
 763
 764         if (unlikely(qsd == NULL))
 765                 return 0;
 766
 767         LASSERT(qtype >= 0 && qtype < MAXQUOTAS);
 768         qi->lqi_type = qtype;
 769
 770         /* inode accounting */
 771         qi->lqi_is_blk = false;
 772
 773         /* one more inode for the new owner ... */
 774         qi->lqi_id.qid_uid = new_id;
 775         qi->lqi_space      = 1;
 776         rc = qsd_op_begin(env, qsd, trans, qi, NULL);
 777         if (rc == -EDQUOT || rc == -EINPROGRESS)
 778                 rc = 0;
 779         if (rc)
 780                 return rc;
 781
 782         /* and one less inode for the current id */
 783         qi->lqi_id.qid_uid = orig_id;;
 784         qi->lqi_space      = -1;
 785         /* can't get EDQUOT when reducing usage */
 786         rc = qsd_op_begin(env, qsd, trans, qi, NULL);
 787         if (rc == -EINPROGRESS)
 788                 rc = 0;
 789         if (rc)
 790                 return rc;
 791
 792         /* block accounting */
 793         qi->lqi_is_blk = true;
 794
 795         /* more blocks for the new owner ... */
 796         qi->lqi_id.qid_uid = new_id;
 797         qi->lqi_space      = bspace;
 798         rc = qsd_op_begin(env, qsd, trans, qi, NULL);
 799         if (rc == -EDQUOT || rc == -EINPROGRESS)
 800                 rc = 0;
 801         if (rc)
 802                 return rc;
 803
 804         /* and finally less blocks for the current owner */
 805         qi->lqi_id.qid_uid = orig_id;
 806         qi->lqi_space      = -bspace;
 807         rc = qsd_op_begin(env, qsd, trans, qi, NULL);
 808         /* can't get EDQUOT when reducing usage */
 809         if (rc == -EINPROGRESS)
 810                 rc = 0;
 811         return rc;
 812 }
 813
 814 static int osd_declare_attr_set(const struct lu_env *env,
 815                                 struct dt_object *dt,
 816                                 const struct lu_attr *attr,
 817                                 struct thandle *handle)
 818 {
 819         struct osd_thread_info  *info = osd_oti_get(env);
 820         char                    *buf = osd_oti_get(env)->oti_str;
 821         struct osd_object       *obj = osd_dt_obj(dt);
 822         struct osd_device       *osd = osd_obj2dev(obj);
 823         struct osd_thandle      *oh;
 824         uint64_t                 bspace;
 825         uint32_t                 blksize;
 826         int                      rc;
 827         ENTRY;
 828
 829         if (!dt_object_exists(dt)) {
 830                 /* XXX: sanity check that object creation is declared */
 831                 RETURN(0);
 832         }
 833
 834         LASSERT(handle != NULL);
 835         LASSERT(osd_invariant(obj));
 836
 837         oh = container_of0(handle, struct osd_thandle, ot_super);
 838
 839         LASSERT(obj->oo_sa_hdl != NULL);
 840         dmu_tx_hold_sa(oh->ot_tx, obj->oo_sa_hdl, 0);
 841
 842         sa_object_size(obj->oo_sa_hdl, &blksize, &bspace);
 843         bspace = toqb(bspace * blksize);
 844
 845         if (attr && attr->la_valid & LA_UID) {
 846                 /* account for user inode tracking ZAP update */
 847                 dmu_tx_hold_bonus(oh->ot_tx, osd->od_iusr_oid);
 848                 dmu_tx_hold_zap(oh->ot_tx, osd->od_iusr_oid, TRUE, buf);
 849
 850                 /* quota enforcement for user */
 851                 if (attr->la_uid != obj->oo_attr.la_uid) {
 852                         rc = qsd_transfer(env, osd->od_quota_slave,
 853                                           &oh->ot_quota_trans, USRQUOTA,
 854                                           obj->oo_attr.la_uid, attr->la_uid,
 855                                           bspace, &info->oti_qi);
 856                         if (rc)
 857                                 RETURN(rc);
 858                 }
 859         }
 860         if (attr && attr->la_valid & LA_GID) {
 861                 /* account for user inode tracking ZAP update */
 862                 dmu_tx_hold_bonus(oh->ot_tx, osd->od_igrp_oid);
 863                 dmu_tx_hold_zap(oh->ot_tx, osd->od_igrp_oid, TRUE, buf);
 864
 865                 /* quota enforcement for group */
 866                 if (attr->la_gid != obj->oo_attr.la_gid) {
 867                         rc = qsd_transfer(env, osd->od_quota_slave,
 868                                           &oh->ot_quota_trans, GRPQUOTA,
 869                                           obj->oo_attr.la_gid, attr->la_gid,
 870                                           bspace, &info->oti_qi);
 871                         if (rc)
 872                                 RETURN(rc);
 873                 }
 874         }
 875
 876         RETURN(0);
 877 }
 878
 879 /*
 880  * Set the attributes of an object
 881  *
 882  * The transaction passed to this routine must have
 883  * dmu_tx_hold_bonus(tx, oid) called and then assigned
 884  * to a transaction group.
 885  */
 886 static int osd_attr_set(const struct lu_env *env, struct dt_object *dt,
 887                         const struct lu_attr *la, struct thandle *handle,
 888                         struct lustre_capa *capa)
 889 {
 890         struct osd_object       *obj = osd_dt_obj(dt);
 891         struct osd_device       *osd = osd_obj2dev(obj);
 892         udmu_objset_t           *uos = &osd->od_objset;
 893         struct osd_thandle      *oh;
 894         struct osa_attr         *osa = &osd_oti_get(env)->oti_osa;
 895         sa_bulk_attr_t          *bulk;
 896         int                      cnt;
 897         int                      rc = 0;
 898
 899         ENTRY;
 900         LASSERT(handle != NULL);
 901         LASSERT(dt_object_exists(dt));
 902         LASSERT(osd_invariant(obj));
 903         LASSERT(obj->oo_sa_hdl);
 904
 905         oh = container_of0(handle, struct osd_thandle, ot_super);
 906         /* Assert that the transaction has been assigned to a
 907            transaction group. */
 908         LASSERT(oh->ot_tx->tx_txg != 0);
 909
 910         if (la->la_valid == 0)
 911                 RETURN(0);
 912
 913         OBD_ALLOC(bulk, sizeof(sa_bulk_attr_t) * 10);
 914         if (bulk == NULL)
 915                 RETURN(-ENOMEM);
 916
 917         /* do both accounting updates outside oo_attr_lock below */
 918         if ((la->la_valid & LA_UID) && (la->la_uid != obj->oo_attr.la_uid)) {
 919                 /* Update user accounting. Failure isn't fatal, but we still
 920                  * log an error message */
 921                 rc = -zap_increment_int(osd->od_objset.os, osd->od_iusr_oid,
 922                                         la->la_uid, 1, oh->ot_tx);
 923                 if (rc)
 924                         CERROR("%s: failed to update accounting ZAP for user "
 925                                 "%d (%d)\n", osd->od_svname, la->la_uid, rc);
 926                 rc = -zap_increment_int(osd->od_objset.os, osd->od_iusr_oid,
 927                                         obj->oo_attr.la_uid, -1, oh->ot_tx);
 928                 if (rc)
 929                         CERROR("%s: failed to update accounting ZAP for user "
 930                                 "%d (%d)\n", osd->od_svname,
 931                                 obj->oo_attr.la_uid, rc);
 932         }
 933         if ((la->la_valid & LA_GID) && (la->la_gid != obj->oo_attr.la_gid)) {
 934                 /* Update group accounting. Failure isn't fatal, but we still
 935                  * log an error message */
 936                 rc = -zap_increment_int(osd->od_objset.os, osd->od_igrp_oid,
 937                                         la->la_gid, 1, oh->ot_tx);
 938                 if (rc)
 939                         CERROR("%s: failed to update accounting ZAP for user "
 940                                 "%d (%d)\n", osd->od_svname, la->la_gid, rc);
 941                 rc = -zap_increment_int(osd->od_objset.os, osd->od_igrp_oid,
 942                                         obj->oo_attr.la_gid, -1, oh->ot_tx);
 943                 if (rc)
 944                         CERROR("%s: failed to update accounting ZAP for user "
 945                                 "%d (%d)\n", osd->od_svname,
 946                                 obj->oo_attr.la_gid, rc);
 947         }
 948
 949         write_lock(&obj->oo_attr_lock);
 950         cnt = 0;
 951         if (la->la_valid & LA_ATIME) {
 952                 osa->atime[0] = obj->oo_attr.la_atime = la->la_atime;
 953                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(uos), NULL,
 954                                  osa->atime, 16);
 955         }
 956         if (la->la_valid & LA_MTIME) {
 957                 osa->mtime[0] = obj->oo_attr.la_mtime = la->la_mtime;
 958                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(uos), NULL,
 959                                  osa->mtime, 16);
 960         }
 961         if (la->la_valid & LA_CTIME) {
 962                 osa->ctime[0] = obj->oo_attr.la_ctime = la->la_ctime;
 963                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(uos), NULL,
 964                                  osa->ctime, 16);
 965         }
 966         if (la->la_valid & LA_MODE) {
 967                 /* mode is stored along with type, so read it first */
 968                 obj->oo_attr.la_mode = (obj->oo_attr.la_mode & S_IFMT) |
 969                         (la->la_mode & ~S_IFMT);
 970                 osa->mode = obj->oo_attr.la_mode;
 971                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(uos), NULL,
 972                                  &osa->mode, 8);
 973         }
 974         if (la->la_valid & LA_SIZE) {
 975                 osa->size = obj->oo_attr.la_size = la->la_size;
 976                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(uos), NULL,
 977                                  &osa->size, 8);
 978         }
 979         if (la->la_valid & LA_NLINK) {
 980                 osa->nlink = obj->oo_attr.la_nlink = la->la_nlink;
 981                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(uos), NULL,
 982                                  &osa->nlink, 8);
 983         }
 984         if (la->la_valid & LA_RDEV) {
 985                 osa->rdev = obj->oo_attr.la_rdev = la->la_rdev;
 986                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_RDEV(uos), NULL,
 987                                  &osa->rdev, 8);
 988         }
 989         if (la->la_valid & LA_FLAGS) {
 990                 osa->flags = obj->oo_attr.la_flags = la->la_flags;
 991                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(uos), NULL,
 992                                  &osa->flags, 8);
 993         }
 994         if (la->la_valid & LA_UID) {
 995                 osa->uid = obj->oo_attr.la_uid = la->la_uid;
 996                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(uos), NULL,
 997                                  &osa->uid, 8);
 998         }
 999         if (la->la_valid & LA_GID) {
1000                 osa->gid = obj->oo_attr.la_gid = la->la_gid;
1001                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(uos), NULL,
1002                                  &osa->gid, 8);
1003         }
1004         obj->oo_attr.la_valid |= la->la_valid;
1005         write_unlock(&obj->oo_attr_lock);
1006
1007         rc = osd_object_sa_bulk_update(obj, bulk, cnt, oh);
1008
1009         OBD_FREE(bulk, sizeof(sa_bulk_attr_t) * 10);
1010         RETURN(rc);
1011 }
1012
1013 /*
1014  * Object creation.
1015  *
1016  * XXX temporary solution.
1017  */
1018
1019 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
1020                         struct dt_object *parent, struct dt_object *child,
1021                         cfs_umode_t child_mode)
1022 {
1023         LASSERT(ah);
1024
1025         memset(ah, 0, sizeof(*ah));
1026         ah->dah_mode = child_mode;
1027 }
1028
1029 static int osd_declare_object_create(const struct lu_env *env,
1030                                      struct dt_object *dt,
1031                                      struct lu_attr *attr,
1032                                      struct dt_allocation_hint *hint,
1033                                      struct dt_object_format *dof,
1034                                      struct thandle *handle)
1035 {
1036         char                    *buf = osd_oti_get(env)->oti_str;
1037         const struct lu_fid     *fid = lu_object_fid(&dt->do_lu);
1038         struct osd_object       *obj = osd_dt_obj(dt);
1039         struct osd_device       *osd = osd_obj2dev(obj);
1040         struct osd_thandle      *oh;
1041         uint64_t                 zapid;
1042         int                      rc;
1043         ENTRY;
1044
1045         LASSERT(dof);
1046
1047         switch (dof->dof_type) {
1048                 case DFT_REGULAR:
1049                 case DFT_SYM:
1050                 case DFT_NODE:
1051                         if (obj->oo_dt.do_body_ops == NULL)
1052                                 obj->oo_dt.do_body_ops = &osd_body_ops;
1053                         break;
1054                 default:
1055                         break;
1056         }
1057
1058         LASSERT(handle != NULL);
1059         oh = container_of0(handle, struct osd_thandle, ot_super);
1060         LASSERT(oh->ot_tx != NULL);
1061
1062         switch (dof->dof_type) {
1063                 case DFT_DIR:
1064                 case DFT_INDEX:
1065                         /* for zap create */
1066                         dmu_tx_hold_zap(oh->ot_tx, DMU_NEW_OBJECT, 1, NULL);
1067                         break;
1068                 case DFT_REGULAR:
1069                 case DFT_SYM:
1070                 case DFT_NODE:
1071                         /* first, we'll create new object */
1072                         dmu_tx_hold_bonus(oh->ot_tx, DMU_NEW_OBJECT);
1073                         break;
1074
1075                 default:
1076                         LBUG();
1077                         break;
1078         }
1079
1080         /* and we'll add it to some mapping */
1081         zapid = osd_get_name_n_idx(env, osd, fid, buf);
1082         dmu_tx_hold_bonus(oh->ot_tx, zapid);
1083         dmu_tx_hold_zap(oh->ot_tx, zapid, TRUE, buf);
1084
1085         /* we will also update inode accounting ZAPs */
1086         dmu_tx_hold_bonus(oh->ot_tx, osd->od_iusr_oid);
1087         dmu_tx_hold_zap(oh->ot_tx, osd->od_iusr_oid, TRUE, buf);
1088         dmu_tx_hold_bonus(oh->ot_tx, osd->od_igrp_oid);
1089         dmu_tx_hold_zap(oh->ot_tx, osd->od_igrp_oid, TRUE, buf);
1090
1091         dmu_tx_hold_sa_create(oh->ot_tx, ZFS_SA_BASE_ATTR_SIZE);
1092
1093         __osd_xattr_declare_set(env, obj, sizeof(struct lustre_mdt_attrs),
1094                                 XATTR_NAME_LMA, oh);
1095
1096         rc = osd_declare_quota(env, osd, attr->la_uid, attr->la_gid, 1, oh,
1097                                false, NULL, false);
1098         RETURN(rc);
1099 }
1100
1101 int __osd_attr_init(const struct lu_env *env, udmu_objset_t *uos,
1102                     uint64_t oid, dmu_tx_t *tx, struct lu_attr *la)
1103 {
1104         sa_bulk_attr_t  *bulk;
1105         sa_handle_t     *sa_hdl;
1106         struct osa_attr *osa = &osd_oti_get(env)->oti_osa;
1107         uint64_t         gen;
1108         uint64_t         parent;
1109         uint64_t         crtime[2];
1110         timestruc_t      now;
1111         int              cnt;
1112         int              rc;
1113
1114         gethrestime(&now);
1115         gen = dmu_tx_get_txg(tx);
1116
1117         ZFS_TIME_ENCODE(&now, crtime);
1118         /* XXX: this should be real id of parent for ZPL access, but we have no
1119          * such info in OSD, probably it can be part of dt_object_format */
1120         parent = 0;
1121
1122         osa->atime[0] = la->la_atime;
1123         osa->ctime[0] = la->la_ctime;
1124         osa->mtime[0] = la->la_mtime;
1125         osa->mode = la->la_mode;
1126         osa->uid = la->la_uid;
1127         osa->gid = la->la_gid;
1128         osa->rdev = la->la_rdev;
1129         osa->nlink = la->la_nlink;
1130         osa->flags = la->la_flags;
1131         osa->size  = la->la_size;
1132
1133         /* Now add in all of the "SA" attributes */
1134         rc = -sa_handle_get(uos->os, oid, NULL, SA_HDL_PRIVATE, &sa_hdl);
1135         if (rc)
1136                 return rc;
1137
1138         OBD_ALLOC(bulk, sizeof(sa_bulk_attr_t) * 13);
1139         if (bulk == NULL) {
1140                 rc = -ENOMEM;
1141                 goto out;
1142         }
1143         /*
1144          * we need to create all SA below upon object create.
1145          *
1146          * XXX The attribute order matters since the accounting callback relies
1147          * on static offsets (i.e. SA_*_OFFSET, see zfs_space_delta_cb()) to
1148          * look up the UID/GID attributes. Moreover, the callback does not seem
1149          * to support the spill block.
1150          * We define attributes in the same order as SA_*_OFFSET in order to
1151          * work around the problem. See ORI-610.
1152          */
1153         cnt = 0;
1154         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(uos), NULL, &osa->mode, 8);
1155         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(uos), NULL, &osa->size, 8);
1156         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GEN(uos), NULL, &gen, 8);
1157         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(uos), NULL, &osa->uid, 8);
1158         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(uos), NULL, &osa->gid, 8);
1159         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_PARENT(uos), NULL, &parent, 8);
1160         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(uos), NULL, &osa->flags, 8);
1161         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(uos), NULL, osa->atime, 16);
1162         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(uos), NULL, osa->mtime, 16);
1163         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(uos), NULL, osa->ctime, 16);
1164         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CRTIME(uos), NULL, crtime, 16);
1165         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(uos), NULL, &osa->nlink, 8);
1166         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_RDEV(uos), NULL, &osa->rdev, 8);
1167
1168         rc = -sa_replace_all_by_template(sa_hdl, bulk, cnt, tx);
1169
1170         OBD_FREE(bulk, sizeof(sa_bulk_attr_t) * 13);
1171 out:
1172         sa_handle_destroy(sa_hdl);
1173         return rc;
1174 }
1175
1176 /*
1177  * The transaction passed to this routine must have
1178  * dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT) called and then assigned
1179  * to a transaction group.
1180  */
1181 int __osd_object_create(const struct lu_env *env, udmu_objset_t *uos,
1182                         dmu_buf_t **dbp, dmu_tx_t *tx,
1183                         struct lu_attr *la, void *tag)
1184 {
1185         uint64_t oid;
1186         int      rc;
1187
1188         LASSERT(tag);
1189         spin_lock(&uos->lock);
1190         uos->objects++;
1191         spin_unlock(&uos->lock);
1192
1193         /* Assert that the transaction has been assigned to a
1194            transaction group. */
1195         LASSERT(tx->tx_txg != 0);
1196
1197         /* Create a new DMU object. */
1198         oid = dmu_object_alloc(uos->os, DMU_OT_PLAIN_FILE_CONTENTS, 0,
1199                                DMU_OT_SA, DN_MAX_BONUSLEN, tx);
1200         rc = -sa_buf_hold(uos->os, oid, tag, dbp);
1201         if (rc)
1202                 return rc;
1203
1204         LASSERT(la->la_valid & LA_MODE);
1205         la->la_size = 0;
1206         la->la_nlink = 1;
1207
1208         return __osd_attr_init(env, uos, oid, tx, la);
1209 }
1210
1211 /*
1212  * The transaction passed to this routine must have
1213  * dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, ...) called and then assigned
1214  * to a transaction group.
1215  *
1216  * Using ZAP_FLAG_HASH64 will force the ZAP to always be a FAT ZAP.
1217  * This is fine for directories today, because storing the FID in the dirent
1218  * will also require a FAT ZAP.  If there is a new type of micro ZAP created
1219  * then we might need to re-evaluate the use of this flag and instead do
1220  * a conversion from the different internal ZAP hash formats being used. */
1221 int __osd_zap_create(const struct lu_env *env, udmu_objset_t *uos,
1222                      dmu_buf_t **zap_dbp, dmu_tx_t *tx,
1223                      struct lu_attr *la, void *tag, zap_flags_t flags)
1224 {
1225         uint64_t oid;
1226         int      rc;
1227
1228         LASSERT(tag);
1229
1230         spin_lock(&uos->lock);
1231         uos->objects++;
1232         spin_unlock(&uos->lock);
1233
1234         /* Assert that the transaction has been assigned to a
1235            transaction group. */
1236         LASSERT(tx->tx_txg != 0);
1237
1238         oid = zap_create_flags(uos->os, 0, flags | ZAP_FLAG_HASH64,
1239                                DMU_OT_DIRECTORY_CONTENTS, 12, 12,
1240                                DMU_OT_SA, DN_MAX_BONUSLEN, tx);
1241
1242         rc = -sa_buf_hold(uos->os, oid, tag, zap_dbp);
1243         if (rc)
1244                 return rc;
1245
1246         LASSERT(la->la_valid & LA_MODE);
1247         la->la_size = 2;
1248         la->la_nlink = 1;
1249
1250         return __osd_attr_init(env, uos, oid, tx, la);
1251 }
1252
1253 static dmu_buf_t *osd_mkidx(const struct lu_env *env, struct osd_device *osd,
1254                             struct lu_attr *la, struct osd_thandle *oh)
1255 {
1256         dmu_buf_t *db;
1257         int        rc;
1258
1259         /* Index file should be created as regular file in order not to confuse
1260          * ZPL which could interpret them as directory.
1261          * We set ZAP_FLAG_UINT64_KEY to let ZFS know than we are going to use
1262          * binary keys */
1263         LASSERT(S_ISREG(la->la_mode));
1264         rc = __osd_zap_create(env, &osd->od_objset, &db, oh->ot_tx, la,
1265                               osd_obj_tag, ZAP_FLAG_UINT64_KEY);
1266         if (rc)
1267                 return ERR_PTR(rc);
1268         return db;
1269 }
1270
1271 static dmu_buf_t *osd_mkdir(const struct lu_env *env, struct osd_device *osd,
1272                             struct lu_attr *la, struct osd_thandle *oh)
1273 {
1274         dmu_buf_t *db;
1275         int        rc;
1276
1277         LASSERT(S_ISDIR(la->la_mode));
1278         rc = __osd_zap_create(env, &osd->od_objset, &db, oh->ot_tx, la,
1279                               osd_obj_tag, 0);
1280         if (rc)
1281                 return ERR_PTR(rc);
1282         return db;
1283 }
1284
1285 static dmu_buf_t* osd_mkreg(const struct lu_env *env, struct osd_device *osd,
1286                             struct lu_attr *la, struct osd_thandle *oh)
1287 {
1288         dmu_buf_t *db;
1289         int         rc;
1290
1291         LASSERT(S_ISREG(la->la_mode));
1292         rc = __osd_object_create(env, &osd->od_objset, &db, oh->ot_tx, la,
1293                                  osd_obj_tag);
1294         if (rc)
1295                 return ERR_PTR(rc);
1296
1297         /*
1298          * XXX: a hack, OST to use bigger blocksize. we need
1299          * a method in OSD API to control this from OFD/MDD
1300          */
1301         if (!lu_device_is_md(osd2lu_dev(osd))) {
1302                 rc = -dmu_object_set_blocksize(osd->od_objset.os,
1303                                                db->db_object,
1304                                 128 << 10, 0, oh->ot_tx);
1305                 if (unlikely(rc)) {
1306                         CERROR("%s: can't change blocksize: %d\n",
1307                                osd->od_svname, rc);
1308                         return ERR_PTR(rc);
1309                 }
1310         }
1311
1312         return db;
1313 }
1314
1315 static dmu_buf_t *osd_mksym(const struct lu_env *env, struct osd_device *osd,
1316                             struct lu_attr *la, struct osd_thandle *oh)
1317 {
1318         dmu_buf_t *db;
1319         int        rc;
1320
1321         LASSERT(S_ISLNK(la->la_mode));
1322         rc = __osd_object_create(env, &osd->od_objset, &db, oh->ot_tx, la,
1323                                  osd_obj_tag);
1324         if (rc)
1325                 return ERR_PTR(rc);
1326         return db;
1327 }
1328
1329 static dmu_buf_t *osd_mknod(const struct lu_env *env, struct osd_device *osd,
1330                             struct lu_attr *la, struct osd_thandle *oh)
1331 {
1332         dmu_buf_t *db;
1333         int        rc;
1334
1335         la->la_valid = LA_MODE;
1336         if (S_ISCHR(la->la_mode) || S_ISBLK(la->la_mode))
1337                 la->la_valid |= LA_RDEV;
1338
1339         rc = __osd_object_create(env, &osd->od_objset, &db, oh->ot_tx, la,
1340                                  osd_obj_tag);
1341         if (rc)
1342                 return ERR_PTR(rc);
1343         return db;
1344 }
1345
1346 typedef dmu_buf_t *(*osd_obj_type_f)(const struct lu_env *env, struct osd_device *osd,
1347                                      struct lu_attr *la, struct osd_thandle *oh);
1348
1349 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
1350 {
1351         osd_obj_type_f result;
1352
1353         switch (type) {
1354         case DFT_DIR:
1355                 result = osd_mkdir;
1356                 break;
1357         case DFT_INDEX:
1358                 result = osd_mkidx;
1359                 break;
1360         case DFT_REGULAR:
1361                 result = osd_mkreg;
1362                 break;
1363         case DFT_SYM:
1364                 result = osd_mksym;
1365                 break;
1366         case DFT_NODE:
1367                 result = osd_mknod;
1368                 break;
1369         default:
1370                 LBUG();
1371                 break;
1372         }
1373         return result;
1374 }
1375
1376 /*
1377  * Primitives for directory (i.e. ZAP) handling
1378  */
1379 static inline int osd_init_lma(const struct lu_env *env, struct osd_object *obj,
1380                                const struct lu_fid *fid, struct osd_thandle *oh)
1381 {
1382         struct osd_thread_info  *info = osd_oti_get(env);
1383         struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
1384         struct lu_buf            buf;
1385         int rc;
1386
1387         lustre_lma_init(lma, fid);
1388         lustre_lma_swab(lma);
1389         buf.lb_buf = lma;
1390         buf.lb_len = sizeof(*lma);
1391
1392         rc = osd_xattr_set_internal(env, obj, &buf, XATTR_NAME_LMA,
1393                                     LU_XATTR_CREATE, oh, BYPASS_CAPA);
1394
1395         return rc;
1396 }
1397
1398 /*
1399  * Concurrency: @dt is write locked.
1400  */
1401 static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
1402                              struct lu_attr *attr,
1403                              struct dt_allocation_hint *hint,
1404                              struct dt_object_format *dof,
1405                              struct thandle *th)
1406 {
1407         struct zpl_direntry     *zde = &osd_oti_get(env)->oti_zde.lzd_reg;
1408         const struct lu_fid     *fid = lu_object_fid(&dt->do_lu);
1409         struct osd_object       *obj = osd_dt_obj(dt);
1410         struct osd_device       *osd = osd_obj2dev(obj);
1411         char                    *buf = osd_oti_get(env)->oti_str;
1412         struct osd_thandle      *oh;
1413         dmu_buf_t               *db;
1414         uint64_t                 zapid;
1415         int                      rc;
1416
1417         ENTRY;
1418
1419         /* concurrent create declarations should not see
1420          * the object inconsistent (db, attr, etc).
1421          * in regular cases acquisition should be cheap */
1422         down(&obj->oo_guard);
1423
1424         LASSERT(osd_invariant(obj));
1425         LASSERT(!dt_object_exists(dt));
1426         LASSERT(dof != NULL);
1427
1428         LASSERT(th != NULL);
1429         oh = container_of0(th, struct osd_thandle, ot_super);
1430
1431         /*
1432          * XXX missing: Quote handling.
1433          */
1434
1435         LASSERT(obj->oo_db == NULL);
1436
1437         db = osd_create_type_f(dof->dof_type)(env, osd, attr, oh);
1438         if (IS_ERR(db))
1439                 GOTO(out, rc = PTR_ERR(th));
1440
1441         zde->zde_pad = 0;
1442         zde->zde_dnode = db->db_object;
1443         zde->zde_type = IFTODT(attr->la_mode & S_IFMT);
1444
1445         zapid = osd_get_name_n_idx(env, osd, fid, buf);
1446
1447         rc = -zap_add(osd->od_objset.os, zapid, buf, 8, 1, zde, oh->ot_tx);
1448         if (rc)
1449                 GOTO(out, rc);
1450
1451         /* Add new object to inode accounting.
1452          * Errors are not considered as fatal */
1453         rc = -zap_increment_int(osd->od_objset.os, osd->od_iusr_oid,
1454                                 (attr->la_valid & LA_UID) ? attr->la_uid : 0, 1,
1455                                 oh->ot_tx);
1456         if (rc)
1457                 CERROR("%s: failed to add "DFID" to accounting ZAP for usr %d "
1458                         "(%d)\n", osd->od_svname, PFID(fid), attr->la_uid, rc);
1459         rc = -zap_increment_int(osd->od_objset.os, osd->od_igrp_oid,
1460                                 (attr->la_valid & LA_GID) ? attr->la_gid : 0, 1,
1461                                 oh->ot_tx);
1462         if (rc)
1463                 CERROR("%s: failed to add "DFID" to accounting ZAP for grp %d "
1464                         "(%d)\n", osd->od_svname, PFID(fid), attr->la_gid, rc);
1465
1466         /* configure new osd object */
1467         obj->oo_db = db;
1468         rc = osd_object_init0(env, obj);
1469         LASSERT(ergo(rc == 0, dt_object_exists(dt)));
1470         LASSERT(osd_invariant(obj));
1471
1472         rc = osd_init_lma(env, obj, fid, oh);
1473         if (rc) {
1474                 CERROR("%s: can not set LMA on "DFID": rc = %d\n",
1475                        osd->od_svname, PFID(fid), rc);
1476                 /* ignore errors during LMA initialization */
1477                 rc = 0;
1478         }
1479
1480 out:
1481         up(&obj->oo_guard);
1482         RETURN(rc);
1483 }
1484
1485 static int osd_declare_object_ref_add(const struct lu_env *env,
1486                                       struct dt_object *dt,
1487                                       struct thandle *th)
1488 {
1489         return osd_declare_attr_set(env, dt, NULL, th);
1490 }
1491
1492 /*
1493  * Concurrency: @dt is write locked.
1494  */
1495 static int osd_object_ref_add(const struct lu_env *env,
1496                               struct dt_object *dt,
1497                               struct thandle *handle)
1498 {
1499         struct osd_object       *obj = osd_dt_obj(dt);
1500         struct osd_thandle      *oh;
1501         struct osd_device       *osd = osd_obj2dev(obj);
1502         udmu_objset_t           *uos = &osd->od_objset;
1503         uint64_t                 nlink;
1504         int rc;
1505
1506         ENTRY;
1507
1508         LASSERT(osd_invariant(obj));
1509         LASSERT(dt_object_exists(dt));
1510         LASSERT(obj->oo_sa_hdl != NULL);
1511
1512         oh = container_of0(handle, struct osd_thandle, ot_super);
1513
1514         write_lock(&obj->oo_attr_lock);
1515         nlink = ++obj->oo_attr.la_nlink;
1516         write_unlock(&obj->oo_attr_lock);
1517
1518         rc = osd_object_sa_update(obj, SA_ZPL_LINKS(uos), &nlink, 8, oh);
1519         return rc;
1520 }
1521
1522 static int osd_declare_object_ref_del(const struct lu_env *env,
1523                                       struct dt_object *dt,
1524                                       struct thandle *handle)
1525 {
1526         return osd_declare_attr_set(env, dt, NULL, handle);
1527 }
1528
1529 /*
1530  * Concurrency: @dt is write locked.
1531  */
1532 static int osd_object_ref_del(const struct lu_env *env,
1533                               struct dt_object *dt,
1534                               struct thandle *handle)
1535 {
1536         struct osd_object       *obj = osd_dt_obj(dt);
1537         struct osd_thandle      *oh;
1538         struct osd_device       *osd = osd_obj2dev(obj);
1539         udmu_objset_t           *uos = &osd->od_objset;
1540         uint64_t                 nlink;
1541         int                      rc;
1542
1543         ENTRY;
1544
1545         LASSERT(osd_invariant(obj));
1546         LASSERT(dt_object_exists(dt));
1547         LASSERT(obj->oo_sa_hdl != NULL);
1548
1549         oh = container_of0(handle, struct osd_thandle, ot_super);
1550         LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
1551
1552         write_lock(&obj->oo_attr_lock);
1553         nlink = --obj->oo_attr.la_nlink;
1554         write_unlock(&obj->oo_attr_lock);
1555
1556         rc = osd_object_sa_update(obj, SA_ZPL_LINKS(uos), &nlink, 8, oh);
1557         return rc;
1558 }
1559
1560 static int capa_is_sane(const struct lu_env *env, struct osd_device *dev,
1561                         struct lustre_capa *capa, struct lustre_capa_key *keys)
1562 {
1563         struct osd_thread_info  *oti = osd_oti_get(env);
1564         struct obd_capa         *oc;
1565         int                      i, rc = 0;
1566         ENTRY;
1567
1568         oc = capa_lookup(dev->od_capa_hash, capa, 0);
1569         if (oc) {
1570                 if (capa_is_expired(oc)) {
1571                         DEBUG_CAPA(D_ERROR, capa, "expired");
1572                         rc = -ESTALE;
1573                 }
1574                 capa_put(oc);
1575                 RETURN(rc);
1576         }
1577
1578         spin_lock(&capa_lock);
1579         for (i = 0; i < 2; i++) {
1580                 if (keys[i].lk_keyid == capa->lc_keyid) {
1581                         oti->oti_capa_key = keys[i];
1582                         break;
1583                 }
1584         }
1585         spin_unlock(&capa_lock);
1586
1587         if (i == 2) {
1588                 DEBUG_CAPA(D_ERROR, capa, "no matched capa key");
1589                 RETURN(-ESTALE);
1590         }
1591
1592         rc = capa_hmac(oti->oti_capa.lc_hmac, capa, oti->oti_capa_key.lk_key);
1593         if (rc)
1594                 RETURN(rc);
1595         if (memcmp(oti->oti_capa.lc_hmac, capa->lc_hmac, sizeof(capa->lc_hmac)))
1596         {
1597                 DEBUG_CAPA(D_ERROR, capa, "HMAC mismatch");
1598                 RETURN(-EACCES);
1599         }
1600
1601         oc = capa_add(dev->od_capa_hash, capa);
1602         capa_put(oc);
1603
1604         RETURN(0);
1605 }
1606
1607 static int osd_object_auth(const struct lu_env *env, struct dt_object *dt,
1608                            struct lustre_capa *capa, __u64 opc)
1609 {
1610         const struct lu_fid     *fid = lu_object_fid(&dt->do_lu);
1611         struct osd_device       *dev = osd_dev(dt->do_lu.lo_dev);
1612         int                      rc;
1613
1614         if (!dev->od_fl_capa)
1615                 return 0;
1616
1617         if (capa == BYPASS_CAPA)
1618                 return 0;
1619
1620         if (!capa) {
1621                 CERROR("no capability is provided for fid "DFID"\n", PFID(fid));
1622                 return -EACCES;
1623         }
1624
1625         if (!lu_fid_eq(fid, &capa->lc_fid)) {
1626                 DEBUG_CAPA(D_ERROR, capa, "fid "DFID" mismatch with",PFID(fid));
1627                 return -EACCES;
1628         }
1629
1630         if (!capa_opc_supported(capa, opc)) {
1631                 DEBUG_CAPA(D_ERROR, capa, "opc "LPX64" not supported by", opc);
1632                 return -EACCES;
1633         }
1634
1635         if ((rc = capa_is_sane(env, dev, capa, dev->od_capa_keys))) {
1636                 DEBUG_CAPA(D_ERROR, capa, "insane (rc %d)", rc);
1637                 return -EACCES;
1638         }
1639
1640         return 0;
1641 }
1642
1643 static struct obd_capa *osd_capa_get(const struct lu_env *env,
1644                                      struct dt_object *dt,
1645                                      struct lustre_capa *old,
1646                                      __u64 opc)
1647 {
1648         struct osd_thread_info  *info = osd_oti_get(env);
1649         const struct lu_fid     *fid = lu_object_fid(&dt->do_lu);
1650         struct osd_object       *obj = osd_dt_obj(dt);
1651         struct osd_device       *dev = osd_obj2dev(obj);
1652         struct lustre_capa_key  *key = &info->oti_capa_key;
1653         struct lustre_capa      *capa = &info->oti_capa;
1654         struct obd_capa         *oc;
1655         int                      rc;
1656         ENTRY;
1657
1658         if (!dev->od_fl_capa)
1659                 RETURN(ERR_PTR(-ENOENT));
1660
1661         LASSERT(dt_object_exists(dt));
1662         LASSERT(osd_invariant(obj));
1663
1664         /* renewal sanity check */
1665         if (old && osd_object_auth(env, dt, old, opc))
1666                 RETURN(ERR_PTR(-EACCES));
1667
1668         capa->lc_fid = *fid;
1669         capa->lc_opc = opc;
1670         capa->lc_uid = 0;
1671         capa->lc_flags = dev->od_capa_alg << 24;
1672         capa->lc_timeout = dev->od_capa_timeout;
1673         capa->lc_expiry = 0;
1674
1675         oc = capa_lookup(dev->od_capa_hash, capa, 1);
1676         if (oc) {
1677                 LASSERT(!capa_is_expired(oc));
1678                 RETURN(oc);
1679         }
1680
1681         spin_lock(&capa_lock);
1682         *key = dev->od_capa_keys[1];
1683         spin_unlock(&capa_lock);
1684
1685         capa->lc_keyid = key->lk_keyid;
1686         capa->lc_expiry = cfs_time_current_sec() + dev->od_capa_timeout;
1687
1688         rc = capa_hmac(capa->lc_hmac, capa, key->lk_key);
1689         if (rc) {
1690                 DEBUG_CAPA(D_ERROR, capa, "HMAC failed: %d for", rc);
1691                 LBUG();
1692                 RETURN(ERR_PTR(rc));
1693         }
1694
1695         oc = capa_add(dev->od_capa_hash, capa);
1696         RETURN(oc);
1697 }
1698
1699 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt)
1700 {
1701         struct osd_device *osd = osd_obj2dev(osd_dt_obj(dt));
1702         ENTRY;
1703
1704         /* XXX: no other option than syncing the whole filesystem until we
1705          * support ZIL */
1706         txg_wait_synced(dmu_objset_pool(osd->od_objset.os), 0ULL);
1707
1708         RETURN(0);
1709 }
1710
1711 static struct dt_object_operations osd_obj_ops = {
1712         .do_read_lock           = osd_object_read_lock,
1713         .do_write_lock          = osd_object_write_lock,
1714         .do_read_unlock         = osd_object_read_unlock,
1715         .do_write_unlock        = osd_object_write_unlock,
1716         .do_write_locked        = osd_object_write_locked,
1717         .do_attr_get            = osd_attr_get,
1718         .do_declare_attr_set    = osd_declare_attr_set,
1719         .do_attr_set            = osd_attr_set,
1720         .do_ah_init             = osd_ah_init,
1721         .do_declare_create      = osd_declare_object_create,
1722         .do_create              = osd_object_create,
1723         .do_declare_destroy     = osd_declare_object_destroy,
1724         .do_destroy             = osd_object_destroy,
1725         .do_index_try           = osd_index_try,
1726         .do_declare_ref_add     = osd_declare_object_ref_add,
1727         .do_ref_add             = osd_object_ref_add,
1728         .do_declare_ref_del     = osd_declare_object_ref_del,
1729         .do_ref_del             = osd_object_ref_del,
1730         .do_xattr_get           = osd_xattr_get,
1731         .do_declare_xattr_set   = osd_declare_xattr_set,
1732         .do_xattr_set           = osd_xattr_set,
1733         .do_declare_xattr_del   = osd_declare_xattr_del,
1734         .do_xattr_del           = osd_xattr_del,
1735         .do_xattr_list          = osd_xattr_list,
1736         .do_capa_get            = osd_capa_get,
1737         .do_object_sync         = osd_object_sync,
1738 };
1739
1740 static struct lu_object_operations osd_lu_obj_ops = {
1741         .loo_object_init        = osd_object_init,
1742         .loo_object_delete      = osd_object_delete,
1743         .loo_object_release     = osd_object_release,
1744         .loo_object_free        = osd_object_free,
1745         .loo_object_print       = osd_object_print,
1746         .loo_object_invariant   = osd_object_invariant,
1747 };
1748
1749 static int osd_otable_it_attr_get(const struct lu_env *env,
1750                                 struct dt_object *dt,
1751                                 struct lu_attr *attr,
1752                                 struct lustre_capa *capa)
1753 {
1754         attr->la_valid = 0;
1755         return 0;
1756 }
1757
1758 static struct dt_object_operations osd_obj_otable_it_ops = {
1759         .do_attr_get    = osd_otable_it_attr_get,
1760         .do_index_try   = osd_index_try,
1761 };
1762