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 = attrs_zfs2fs(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         if (fid_is_otable_it(&l->lo_header->loh_fid)) {
 368                 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
 369                 l->lo_header->loh_attr |= LOHA_EXISTS;
 370                 RETURN(0);
 371         }
 372
 373         rc = osd_fid_lookup(env, osd, lu_object_fid(l), &oid);
 374         if (rc == 0) {
 375                 LASSERT(obj->oo_db == NULL);
 376                 rc = __osd_obj2dbuf(env, osd->od_objset.os, oid,
 377                                         &obj->oo_db, osd_obj_tag);
 378                 if (rc == 0) {
 379                         LASSERT(obj->oo_db);
 380                         rc = osd_object_init0(env, obj);
 381                 } else {
 382                         CERROR("%s: lookup "DFID"/"LPX64" failed: rc = %d\n",
 383                                osd->od_svname, PFID(lu_object_fid(l)), oid, rc);
 384                 }
 385         } else if (rc == -ENOENT) {
 386                 rc = 0;
 387         }
 388         LASSERT(osd_invariant(obj));
 389         RETURN(rc);
 390 }
 391
 392 /*
 393  * Concurrency: no concurrent access is possible that late in object
 394  * life-cycle.
 395  */
 396 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
 397 {
 398         struct osd_object *obj = osd_obj(l);
 399
 400         LASSERT(osd_invariant(obj));
 401
 402         dt_object_fini(&obj->oo_dt);
 403         OBD_SLAB_FREE_PTR(obj, osd_object_kmem);
 404 }
 405
 406 static void __osd_declare_object_destroy(const struct lu_env *env,
 407                                          struct osd_object *obj,
 408                                          struct osd_thandle *oh)
 409 {
 410         struct osd_device       *osd = osd_obj2dev(obj);
 411         udmu_objset_t           *uos = &osd->od_objset;
 412         dmu_buf_t               *db = obj->oo_db;
 413         zap_attribute_t         *za = &osd_oti_get(env)->oti_za;
 414         uint64_t                 oid = db->db_object, xid;
 415         dmu_tx_t                *tx = oh->ot_tx;
 416         zap_cursor_t            *zc;
 417         int                      rc = 0;
 418
 419         dmu_tx_hold_free(tx, oid, 0, DMU_OBJECT_END);
 420
 421         /* zap holding xattrs */
 422         if (obj->oo_xattr != ZFS_NO_OBJECT) {
 423                 oid = obj->oo_xattr;
 424
 425                 dmu_tx_hold_free(tx, oid, 0, DMU_OBJECT_END);
 426
 427                 rc = -udmu_zap_cursor_init(&zc, uos, oid, 0);
 428                 if (rc)
 429                         goto out;
 430
 431                 while ((rc = -zap_cursor_retrieve(zc, za)) == 0) {
 432                         BUG_ON(za->za_integer_length != sizeof(uint64_t));
 433                         BUG_ON(za->za_num_integers != 1);
 434
 435                         rc = -zap_lookup(uos->os, obj->oo_xattr, za->za_name,
 436                                          sizeof(uint64_t), 1, &xid);
 437                         if (rc) {
 438                                 CERROR("%s: xattr lookup failed: rc = %d\n",
 439                                        osd->od_svname, rc);
 440                                 goto out_err;
 441                         }
 442                         dmu_tx_hold_free(tx, xid, 0, DMU_OBJECT_END);
 443
 444                         zap_cursor_advance(zc);
 445                 }
 446                 if (rc == -ENOENT)
 447                         rc = 0;
 448 out_err:
 449                 udmu_zap_cursor_fini(zc);
 450         }
 451 out:
 452         if (rc && tx->tx_err == 0)
 453                 tx->tx_err = -rc;
 454 }
 455
 456 static int osd_declare_object_destroy(const struct lu_env *env,
 457                                       struct dt_object *dt,
 458                                       struct thandle *th)
 459 {
 460         char                    *buf = osd_oti_get(env)->oti_str;
 461         const struct lu_fid     *fid = lu_object_fid(&dt->do_lu);
 462         struct osd_object       *obj = osd_dt_obj(dt);
 463         struct osd_device       *osd = osd_obj2dev(obj);
 464         struct osd_thandle      *oh;
 465         uint64_t                 zapid;
 466         int                      rc;
 467         ENTRY;
 468
 469         LASSERT(th != NULL);
 470         LASSERT(dt_object_exists(dt));
 471
 472         oh = container_of0(th, struct osd_thandle, ot_super);
 473         LASSERT(oh->ot_tx != NULL);
 474
 475         /* declare that we'll destroy the object */
 476         __osd_declare_object_destroy(env, obj, oh);
 477
 478         /* declare that we'll remove object from fid-dnode mapping */
 479         zapid = osd_get_name_n_idx(env, osd, fid, buf);
 480         dmu_tx_hold_bonus(oh->ot_tx, zapid);
 481         dmu_tx_hold_zap(oh->ot_tx, zapid, 0, buf);
 482
 483         /* declare that we'll remove object from inode accounting ZAPs */
 484         dmu_tx_hold_bonus(oh->ot_tx, osd->od_iusr_oid);
 485         dmu_tx_hold_zap(oh->ot_tx, osd->od_iusr_oid, 0, buf);
 486         dmu_tx_hold_bonus(oh->ot_tx, osd->od_igrp_oid);
 487         dmu_tx_hold_zap(oh->ot_tx, osd->od_igrp_oid, 0, buf);
 488
 489         /* one less inode */
 490         rc = osd_declare_quota(env, osd, obj->oo_attr.la_uid,
 491                                obj->oo_attr.la_gid, -1, oh, false, NULL, false);
 492         if (rc)
 493                 RETURN(rc);
 494
 495         /* data to be truncated */
 496         rc = osd_declare_quota(env, osd, obj->oo_attr.la_uid,
 497                                obj->oo_attr.la_gid, 0, oh, true, NULL, false);
 498         RETURN(rc);
 499 }
 500
 501 int __osd_object_free(udmu_objset_t *uos, uint64_t oid, dmu_tx_t *tx)
 502 {
 503         LASSERT(uos->objects != 0);
 504         spin_lock(&uos->lock);
 505         uos->objects--;
 506         spin_unlock(&uos->lock);
 507
 508         return -dmu_object_free(uos->os, oid, tx);
 509 }
 510
 511 /*
 512  * Delete a DMU object
 513  *
 514  * The transaction passed to this routine must have
 515  * dmu_tx_hold_free(tx, oid, 0, DMU_OBJECT_END) called
 516  * and then assigned to a transaction group.
 517  *
 518  * This will release db and set it to NULL to prevent further dbuf releases.
 519  */
 520 static int __osd_object_destroy(const struct lu_env *env,
 521                                 struct osd_object *obj,
 522                                 dmu_tx_t *tx, void *tag)
 523 {
 524         struct osd_device       *osd = osd_obj2dev(obj);
 525         udmu_objset_t           *uos = &osd->od_objset;
 526         uint64_t                 xid;
 527         zap_attribute_t         *za = &osd_oti_get(env)->oti_za;
 528         zap_cursor_t            *zc;
 529         int                      rc;
 530
 531         /* Assert that the transaction has been assigned to a
 532            transaction group. */
 533         LASSERT(tx->tx_txg != 0);
 534
 535         /* zap holding xattrs */
 536         if (obj->oo_xattr != ZFS_NO_OBJECT) {
 537                 rc = -udmu_zap_cursor_init(&zc, uos, obj->oo_xattr, 0);
 538                 if (rc)
 539                         return rc;
 540                 while ((rc = -zap_cursor_retrieve(zc, za)) == 0) {
 541                         BUG_ON(za->za_integer_length != sizeof(uint64_t));
 542                         BUG_ON(za->za_num_integers != 1);
 543
 544                         rc = -zap_lookup(uos->os, obj->oo_xattr, za->za_name,
 545                                          sizeof(uint64_t), 1, &xid);
 546                         if (rc) {
 547                                 CERROR("%s: lookup xattr %s failed: rc = %d\n",
 548                                        osd->od_svname, za->za_name, rc);
 549                                 continue;
 550                         }
 551                         rc = __osd_object_free(uos, xid, tx);
 552                         if (rc)
 553                                 CERROR("%s: fetch xattr %s failed: rc = %d\n",
 554                                        osd->od_svname, za->za_name, rc);
 555                         zap_cursor_advance(zc);
 556                 }
 557                 udmu_zap_cursor_fini(zc);
 558
 559                 rc = __osd_object_free(uos, obj->oo_xattr, tx);
 560                 if (rc)
 561                         CERROR("%s: freeing xattr failed: rc = %d\n",
 562                                osd->od_svname, rc);
 563         }
 564
 565         return __osd_object_free(uos, obj->oo_db->db_object, tx);
 566 }
 567
 568 static int osd_object_destroy(const struct lu_env *env,
 569                               struct dt_object *dt,
 570                               struct thandle *th)
 571 {
 572         char                    *buf = osd_oti_get(env)->oti_str;
 573         struct osd_object       *obj = osd_dt_obj(dt);
 574         struct osd_device       *osd = osd_obj2dev(obj);
 575         const struct lu_fid     *fid = lu_object_fid(&dt->do_lu);
 576         struct osd_thandle      *oh;
 577         int                      rc;
 578         uint64_t                 zapid;
 579         ENTRY;
 580
 581         LASSERT(obj->oo_db != NULL);
 582         LASSERT(dt_object_exists(dt));
 583         LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
 584
 585         oh = container_of0(th, struct osd_thandle, ot_super);
 586         LASSERT(oh != NULL);
 587         LASSERT(oh->ot_tx != NULL);
 588
 589         zapid = osd_get_name_n_idx(env, osd, fid, buf);
 590
 591         /* remove obj ref from index dir (it depends) */
 592         rc = -zap_remove(osd->od_objset.os, zapid, buf, oh->ot_tx);
 593         if (rc) {
 594                 CERROR("%s: zap_remove() failed: rc = %d\n",
 595                        osd->od_svname, rc);
 596                 GOTO(out, rc);
 597         }
 598
 599         /* Remove object from inode accounting. It is not fatal for the destroy
 600          * operation if something goes wrong while updating accounting, but we
 601          * still log an error message to notify the administrator */
 602         rc = -zap_increment_int(osd->od_objset.os, osd->od_iusr_oid,
 603                         obj->oo_attr.la_uid, -1, oh->ot_tx);
 604         if (rc)
 605                 CERROR("%s: failed to remove "DFID" from accounting ZAP for usr"
 606                         " %d: rc = %d\n", osd->od_svname, PFID(fid),
 607                         obj->oo_attr.la_uid, rc);
 608         rc = -zap_increment_int(osd->od_objset.os, osd->od_igrp_oid,
 609                                 obj->oo_attr.la_gid, -1, oh->ot_tx);
 610         if (rc)
 611                 CERROR("%s: failed to remove "DFID" from accounting ZAP for grp"
 612                         " %d: rc = %d\n", osd->od_svname, PFID(fid),
 613                         obj->oo_attr.la_gid, rc);
 614
 615         /* kill object */
 616         rc = __osd_object_destroy(env, obj, oh->ot_tx, osd_obj_tag);
 617         if (rc) {
 618                 CERROR("%s: __osd_object_destroy() failed: rc = %d\n",
 619                        osd->od_svname, rc);
 620                 GOTO(out, rc);
 621         }
 622
 623 out:
 624         /* not needed in the cache anymore */
 625         set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
 626
 627         RETURN (0);
 628 }
 629
 630 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
 631 {
 632         struct osd_object *obj = osd_obj(l);
 633
 634         if (obj->oo_db != NULL) {
 635                 osd_object_sa_fini(obj);
 636                 if (obj->oo_sa_xattr) {
 637                         nvlist_free(obj->oo_sa_xattr);
 638                         obj->oo_sa_xattr = NULL;
 639                 }
 640                 sa_buf_rele(obj->oo_db, osd_obj_tag);
 641                 cfs_list_del(&obj->oo_sa_linkage);
 642                 obj->oo_db = NULL;
 643         }
 644 }
 645
 646 /*
 647  * Concurrency: ->loo_object_release() is called under site spin-lock.
 648  */
 649 static void osd_object_release(const struct lu_env *env,
 650                                struct lu_object *l)
 651 {
 652 }
 653
 654 /*
 655  * Concurrency: shouldn't matter.
 656  */
 657 static int osd_object_print(const struct lu_env *env, void *cookie,
 658                             lu_printer_t p, const struct lu_object *l)
 659 {
 660         struct osd_object *o = osd_obj(l);
 661
 662         return (*p)(env, cookie, LUSTRE_OSD_ZFS_NAME"-object@%p", o);
 663 }
 664
 665 static void osd_object_read_lock(const struct lu_env *env,
 666                                  struct dt_object *dt, unsigned role)
 667 {
 668         struct osd_object *obj = osd_dt_obj(dt);
 669
 670         LASSERT(osd_invariant(obj));
 671
 672         down_read(&obj->oo_sem);
 673 }
 674
 675 static void osd_object_write_lock(const struct lu_env *env,
 676                                   struct dt_object *dt, unsigned role)
 677 {
 678         struct osd_object *obj = osd_dt_obj(dt);
 679
 680         LASSERT(osd_invariant(obj));
 681
 682         down_write(&obj->oo_sem);
 683 }
 684
 685 static void osd_object_read_unlock(const struct lu_env *env,
 686                                    struct dt_object *dt)
 687 {
 688         struct osd_object *obj = osd_dt_obj(dt);
 689
 690         LASSERT(osd_invariant(obj));
 691         up_read(&obj->oo_sem);
 692 }
 693
 694 static void osd_object_write_unlock(const struct lu_env *env,
 695                                     struct dt_object *dt)
 696 {
 697         struct osd_object *obj = osd_dt_obj(dt);
 698
 699         LASSERT(osd_invariant(obj));
 700         up_write(&obj->oo_sem);
 701 }
 702
 703 static int osd_object_write_locked(const struct lu_env *env,
 704                                    struct dt_object *dt)
 705 {
 706         struct osd_object *obj = osd_dt_obj(dt);
 707         int rc = 1;
 708
 709         LASSERT(osd_invariant(obj));
 710
 711         if (down_write_trylock(&obj->oo_sem)) {
 712                 rc = 0;
 713                 up_write(&obj->oo_sem);
 714         }
 715         return rc;
 716 }
 717
 718 static int osd_attr_get(const struct lu_env *env,
 719                         struct dt_object *dt,
 720                         struct lu_attr *attr,
 721                         struct lustre_capa *capa)
 722 {
 723         struct osd_object       *obj = osd_dt_obj(dt);
 724         uint64_t                 blocks;
 725         uint32_t                 blksize;
 726
 727         LASSERT(dt_object_exists(dt));
 728         LASSERT(osd_invariant(obj));
 729         LASSERT(obj->oo_db);
 730
 731         read_lock(&obj->oo_attr_lock);
 732         *attr = obj->oo_attr;
 733         read_unlock(&obj->oo_attr_lock);
 734
 735         /* with ZFS_DEBUG zrl_add_debug() called by DB_DNODE_ENTER()
 736          * from within sa_object_size() can block on a mutex, so
 737          * we can't call sa_object_size() holding rwlock */
 738         sa_object_size(obj->oo_sa_hdl, &blksize, &blocks);
 739         /* we do not control size of indices, so always calculate
 740          * it from number of blocks reported by DMU */
 741         if (S_ISDIR(attr->la_mode))
 742                 attr->la_size = 512 * blocks;
 743         /* Block size may be not set; suggest maximal I/O transfers. */
 744         if (blksize == 0)
 745                 blksize = 1ULL << SPA_MAXBLOCKSHIFT;
 746
 747         attr->la_blksize = blksize;
 748         attr->la_blocks = blocks;
 749         attr->la_valid |= LA_BLOCKS | LA_BLKSIZE;
 750
 751         return 0;
 752 }
 753
 754 /* Simple wrapper on top of qsd API which implement quota transfer for osd
 755  * setattr needs. As a reminder, only the root user can change ownership of
 756  * a file, that's why EDQUOT & EINPROGRESS errors are discarded */
 757 static inline int qsd_transfer(const struct lu_env *env,
 758                                struct qsd_instance *qsd,
 759                                struct lquota_trans *trans, int qtype,
 760                                __u64 orig_id, __u64 new_id, __u64 bspace,
 761                                struct lquota_id_info *qi)
 762 {
 763         int     rc;
 764
 765         if (unlikely(qsd == NULL))
 766                 return 0;
 767
 768         LASSERT(qtype >= 0 && qtype < MAXQUOTAS);
 769         qi->lqi_type = qtype;
 770
 771         /* inode accounting */
 772         qi->lqi_is_blk = false;
 773
 774         /* one more inode for the new owner ... */
 775         qi->lqi_id.qid_uid = new_id;
 776         qi->lqi_space      = 1;
 777         rc = qsd_op_begin(env, qsd, trans, qi, NULL);
 778         if (rc == -EDQUOT || rc == -EINPROGRESS)
 779                 rc = 0;
 780         if (rc)
 781                 return rc;
 782
 783         /* and one less inode for the current id */
 784         qi->lqi_id.qid_uid = orig_id;;
 785         qi->lqi_space      = -1;
 786         /* can't get EDQUOT when reducing usage */
 787         rc = qsd_op_begin(env, qsd, trans, qi, NULL);
 788         if (rc == -EINPROGRESS)
 789                 rc = 0;
 790         if (rc)
 791                 return rc;
 792
 793         /* block accounting */
 794         qi->lqi_is_blk = true;
 795
 796         /* more blocks for the new owner ... */
 797         qi->lqi_id.qid_uid = new_id;
 798         qi->lqi_space      = bspace;
 799         rc = qsd_op_begin(env, qsd, trans, qi, NULL);
 800         if (rc == -EDQUOT || rc == -EINPROGRESS)
 801                 rc = 0;
 802         if (rc)
 803                 return rc;
 804
 805         /* and finally less blocks for the current owner */
 806         qi->lqi_id.qid_uid = orig_id;
 807         qi->lqi_space      = -bspace;
 808         rc = qsd_op_begin(env, qsd, trans, qi, NULL);
 809         /* can't get EDQUOT when reducing usage */
 810         if (rc == -EINPROGRESS)
 811                 rc = 0;
 812         return rc;
 813 }
 814
 815 static int osd_declare_attr_set(const struct lu_env *env,
 816                                 struct dt_object *dt,
 817                                 const struct lu_attr *attr,
 818                                 struct thandle *handle)
 819 {
 820         struct osd_thread_info  *info = osd_oti_get(env);
 821         char                    *buf = osd_oti_get(env)->oti_str;
 822         struct osd_object       *obj = osd_dt_obj(dt);
 823         struct osd_device       *osd = osd_obj2dev(obj);
 824         struct osd_thandle      *oh;
 825         uint64_t                 bspace;
 826         uint32_t                 blksize;
 827         int                      rc;
 828         ENTRY;
 829
 830         if (!dt_object_exists(dt)) {
 831                 /* XXX: sanity check that object creation is declared */
 832                 RETURN(0);
 833         }
 834
 835         LASSERT(handle != NULL);
 836         LASSERT(osd_invariant(obj));
 837
 838         oh = container_of0(handle, struct osd_thandle, ot_super);
 839
 840         LASSERT(obj->oo_sa_hdl != NULL);
 841         dmu_tx_hold_sa(oh->ot_tx, obj->oo_sa_hdl, 0);
 842
 843         sa_object_size(obj->oo_sa_hdl, &blksize, &bspace);
 844         bspace = toqb(bspace * blksize);
 845
 846         if (attr && attr->la_valid & LA_UID) {
 847                 /* account for user inode tracking ZAP update */
 848                 dmu_tx_hold_bonus(oh->ot_tx, osd->od_iusr_oid);
 849                 dmu_tx_hold_zap(oh->ot_tx, osd->od_iusr_oid, TRUE, buf);
 850
 851                 /* quota enforcement for user */
 852                 if (attr->la_uid != obj->oo_attr.la_uid) {
 853                         rc = qsd_transfer(env, osd->od_quota_slave,
 854                                           &oh->ot_quota_trans, USRQUOTA,
 855                                           obj->oo_attr.la_uid, attr->la_uid,
 856                                           bspace, &info->oti_qi);
 857                         if (rc)
 858                                 RETURN(rc);
 859                 }
 860         }
 861         if (attr && attr->la_valid & LA_GID) {
 862                 /* account for user inode tracking ZAP update */
 863                 dmu_tx_hold_bonus(oh->ot_tx, osd->od_igrp_oid);
 864                 dmu_tx_hold_zap(oh->ot_tx, osd->od_igrp_oid, TRUE, buf);
 865
 866                 /* quota enforcement for group */
 867                 if (attr->la_gid != obj->oo_attr.la_gid) {
 868                         rc = qsd_transfer(env, osd->od_quota_slave,
 869                                           &oh->ot_quota_trans, GRPQUOTA,
 870                                           obj->oo_attr.la_gid, attr->la_gid,
 871                                           bspace, &info->oti_qi);
 872                         if (rc)
 873                                 RETURN(rc);
 874                 }
 875         }
 876
 877         RETURN(0);
 878 }
 879
 880 /*
 881  * Set the attributes of an object
 882  *
 883  * The transaction passed to this routine must have
 884  * dmu_tx_hold_bonus(tx, oid) called and then assigned
 885  * to a transaction group.
 886  */
 887 static int osd_attr_set(const struct lu_env *env, struct dt_object *dt,
 888                         const struct lu_attr *la, struct thandle *handle,
 889                         struct lustre_capa *capa)
 890 {
 891         struct osd_object       *obj = osd_dt_obj(dt);
 892         struct osd_device       *osd = osd_obj2dev(obj);
 893         udmu_objset_t           *uos = &osd->od_objset;
 894         struct osd_thandle      *oh;
 895         struct osa_attr         *osa = &osd_oti_get(env)->oti_osa;
 896         sa_bulk_attr_t          *bulk;
 897         int                      cnt;
 898         int                      rc = 0;
 899
 900         ENTRY;
 901         LASSERT(handle != NULL);
 902         LASSERT(dt_object_exists(dt));
 903         LASSERT(osd_invariant(obj));
 904         LASSERT(obj->oo_sa_hdl);
 905
 906         oh = container_of0(handle, struct osd_thandle, ot_super);
 907         /* Assert that the transaction has been assigned to a
 908            transaction group. */
 909         LASSERT(oh->ot_tx->tx_txg != 0);
 910
 911         if (la->la_valid == 0)
 912                 RETURN(0);
 913
 914         OBD_ALLOC(bulk, sizeof(sa_bulk_attr_t) * 10);
 915         if (bulk == NULL)
 916                 RETURN(-ENOMEM);
 917
 918         /* do both accounting updates outside oo_attr_lock below */
 919         if ((la->la_valid & LA_UID) && (la->la_uid != obj->oo_attr.la_uid)) {
 920                 /* Update user accounting. Failure isn't fatal, but we still
 921                  * log an error message */
 922                 rc = -zap_increment_int(osd->od_objset.os, osd->od_iusr_oid,
 923                                         la->la_uid, 1, oh->ot_tx);
 924                 if (rc)
 925                         CERROR("%s: failed to update accounting ZAP for user "
 926                                 "%d (%d)\n", osd->od_svname, la->la_uid, rc);
 927                 rc = -zap_increment_int(osd->od_objset.os, osd->od_iusr_oid,
 928                                         obj->oo_attr.la_uid, -1, oh->ot_tx);
 929                 if (rc)
 930                         CERROR("%s: failed to update accounting ZAP for user "
 931                                 "%d (%d)\n", osd->od_svname,
 932                                 obj->oo_attr.la_uid, rc);
 933         }
 934         if ((la->la_valid & LA_GID) && (la->la_gid != obj->oo_attr.la_gid)) {
 935                 /* Update group accounting. Failure isn't fatal, but we still
 936                  * log an error message */
 937                 rc = -zap_increment_int(osd->od_objset.os, osd->od_igrp_oid,
 938                                         la->la_gid, 1, oh->ot_tx);
 939                 if (rc)
 940                         CERROR("%s: failed to update accounting ZAP for user "
 941                                 "%d (%d)\n", osd->od_svname, la->la_gid, rc);
 942                 rc = -zap_increment_int(osd->od_objset.os, osd->od_igrp_oid,
 943                                         obj->oo_attr.la_gid, -1, oh->ot_tx);
 944                 if (rc)
 945                         CERROR("%s: failed to update accounting ZAP for user "
 946                                 "%d (%d)\n", osd->od_svname,
 947                                 obj->oo_attr.la_gid, rc);
 948         }
 949
 950         write_lock(&obj->oo_attr_lock);
 951         cnt = 0;
 952         if (la->la_valid & LA_ATIME) {
 953                 osa->atime[0] = obj->oo_attr.la_atime = la->la_atime;
 954                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(uos), NULL,
 955                                  osa->atime, 16);
 956         }
 957         if (la->la_valid & LA_MTIME) {
 958                 osa->mtime[0] = obj->oo_attr.la_mtime = la->la_mtime;
 959                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(uos), NULL,
 960                                  osa->mtime, 16);
 961         }
 962         if (la->la_valid & LA_CTIME) {
 963                 osa->ctime[0] = obj->oo_attr.la_ctime = la->la_ctime;
 964                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(uos), NULL,
 965                                  osa->ctime, 16);
 966         }
 967         if (la->la_valid & LA_MODE) {
 968                 /* mode is stored along with type, so read it first */
 969                 obj->oo_attr.la_mode = (obj->oo_attr.la_mode & S_IFMT) |
 970                         (la->la_mode & ~S_IFMT);
 971                 osa->mode = obj->oo_attr.la_mode;
 972                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(uos), NULL,
 973                                  &osa->mode, 8);
 974         }
 975         if (la->la_valid & LA_SIZE) {
 976                 osa->size = obj->oo_attr.la_size = la->la_size;
 977                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(uos), NULL,
 978                                  &osa->size, 8);
 979         }
 980         if (la->la_valid & LA_NLINK) {
 981                 osa->nlink = obj->oo_attr.la_nlink = la->la_nlink;
 982                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(uos), NULL,
 983                                  &osa->nlink, 8);
 984         }
 985         if (la->la_valid & LA_RDEV) {
 986                 osa->rdev = obj->oo_attr.la_rdev = la->la_rdev;
 987                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_RDEV(uos), NULL,
 988                                  &osa->rdev, 8);
 989         }
 990         if (la->la_valid & LA_FLAGS) {
 991                 osa->flags = attrs_fs2zfs(la->la_flags);
 992                 /* many flags are not supported by zfs, so ensure a good cached
 993                  * copy */
 994                 obj->oo_attr.la_flags = attrs_zfs2fs(osa->flags);
 995                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(uos), NULL,
 996                                  &osa->flags, 8);
 997         }
 998         if (la->la_valid & LA_UID) {
 999                 osa->uid = obj->oo_attr.la_uid = la->la_uid;
1000                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(uos), NULL,
1001                                  &osa->uid, 8);
1002         }
1003         if (la->la_valid & LA_GID) {
1004                 osa->gid = obj->oo_attr.la_gid = la->la_gid;
1005                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(uos), NULL,
1006                                  &osa->gid, 8);
1007         }
1008         obj->oo_attr.la_valid |= la->la_valid;
1009         write_unlock(&obj->oo_attr_lock);
1010
1011         rc = osd_object_sa_bulk_update(obj, bulk, cnt, oh);
1012
1013         OBD_FREE(bulk, sizeof(sa_bulk_attr_t) * 10);
1014         RETURN(rc);
1015 }
1016
1017 /*
1018  * Object creation.
1019  *
1020  * XXX temporary solution.
1021  */
1022
1023 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
1024                         struct dt_object *parent, struct dt_object *child,
1025                         cfs_umode_t child_mode)
1026 {
1027         LASSERT(ah);
1028
1029         memset(ah, 0, sizeof(*ah));
1030         ah->dah_parent = parent;
1031         ah->dah_mode = child_mode;
1032 }
1033
1034 static int osd_declare_object_create(const struct lu_env *env,
1035                                      struct dt_object *dt,
1036                                      struct lu_attr *attr,
1037                                      struct dt_allocation_hint *hint,
1038                                      struct dt_object_format *dof,
1039                                      struct thandle *handle)
1040 {
1041         char                    *buf = osd_oti_get(env)->oti_str;
1042         const struct lu_fid     *fid = lu_object_fid(&dt->do_lu);
1043         struct osd_object       *obj = osd_dt_obj(dt);
1044         struct osd_device       *osd = osd_obj2dev(obj);
1045         struct osd_thandle      *oh;
1046         uint64_t                 zapid;
1047         int                      rc;
1048         ENTRY;
1049
1050         LASSERT(dof);
1051
1052         switch (dof->dof_type) {
1053                 case DFT_REGULAR:
1054                 case DFT_SYM:
1055                 case DFT_NODE:
1056                         if (obj->oo_dt.do_body_ops == NULL)
1057                                 obj->oo_dt.do_body_ops = &osd_body_ops;
1058                         break;
1059                 default:
1060                         break;
1061         }
1062
1063         LASSERT(handle != NULL);
1064         oh = container_of0(handle, struct osd_thandle, ot_super);
1065         LASSERT(oh->ot_tx != NULL);
1066
1067         switch (dof->dof_type) {
1068                 case DFT_DIR:
1069                 case DFT_INDEX:
1070                         /* for zap create */
1071                         dmu_tx_hold_zap(oh->ot_tx, DMU_NEW_OBJECT, 1, NULL);
1072                         break;
1073                 case DFT_REGULAR:
1074                 case DFT_SYM:
1075                 case DFT_NODE:
1076                         /* first, we'll create new object */
1077                         dmu_tx_hold_bonus(oh->ot_tx, DMU_NEW_OBJECT);
1078                         break;
1079
1080                 default:
1081                         LBUG();
1082                         break;
1083         }
1084
1085         /* and we'll add it to some mapping */
1086         zapid = osd_get_name_n_idx(env, osd, fid, buf);
1087         dmu_tx_hold_bonus(oh->ot_tx, zapid);
1088         dmu_tx_hold_zap(oh->ot_tx, zapid, TRUE, buf);
1089
1090         /* we will also update inode accounting ZAPs */
1091         dmu_tx_hold_bonus(oh->ot_tx, osd->od_iusr_oid);
1092         dmu_tx_hold_zap(oh->ot_tx, osd->od_iusr_oid, TRUE, buf);
1093         dmu_tx_hold_bonus(oh->ot_tx, osd->od_igrp_oid);
1094         dmu_tx_hold_zap(oh->ot_tx, osd->od_igrp_oid, TRUE, buf);
1095
1096         dmu_tx_hold_sa_create(oh->ot_tx, ZFS_SA_BASE_ATTR_SIZE);
1097
1098         __osd_xattr_declare_set(env, obj, sizeof(struct lustre_mdt_attrs),
1099                                 XATTR_NAME_LMA, oh);
1100
1101         rc = osd_declare_quota(env, osd, attr->la_uid, attr->la_gid, 1, oh,
1102                                false, NULL, false);
1103         RETURN(rc);
1104 }
1105
1106 int __osd_attr_init(const struct lu_env *env, udmu_objset_t *uos, uint64_t oid,
1107                     dmu_tx_t *tx, struct lu_attr *la, uint64_t parent)
1108 {
1109         sa_bulk_attr_t  *bulk;
1110         sa_handle_t     *sa_hdl;
1111         struct osa_attr *osa = &osd_oti_get(env)->oti_osa;
1112         uint64_t         gen;
1113         uint64_t         crtime[2];
1114         timestruc_t      now;
1115         int              cnt;
1116         int              rc;
1117
1118         gethrestime(&now);
1119         gen = dmu_tx_get_txg(tx);
1120
1121         ZFS_TIME_ENCODE(&now, crtime);
1122
1123         osa->atime[0] = la->la_atime;
1124         osa->ctime[0] = la->la_ctime;
1125         osa->mtime[0] = la->la_mtime;
1126         osa->mode = la->la_mode;
1127         osa->uid = la->la_uid;
1128         osa->gid = la->la_gid;
1129         osa->rdev = la->la_rdev;
1130         osa->nlink = la->la_nlink;
1131         osa->flags = attrs_fs2zfs(la->la_flags);
1132         osa->size  = la->la_size;
1133
1134         /* Now add in all of the "SA" attributes */
1135         rc = -sa_handle_get(uos->os, oid, NULL, SA_HDL_PRIVATE, &sa_hdl);
1136         if (rc)
1137                 return rc;
1138
1139         OBD_ALLOC(bulk, sizeof(sa_bulk_attr_t) * 13);
1140         if (bulk == NULL) {
1141                 rc = -ENOMEM;
1142                 goto out;
1143         }
1144         /*
1145          * we need to create all SA below upon object create.
1146          *
1147          * XXX The attribute order matters since the accounting callback relies
1148          * on static offsets (i.e. SA_*_OFFSET, see zfs_space_delta_cb()) to
1149          * look up the UID/GID attributes. Moreover, the callback does not seem
1150          * to support the spill block.
1151          * We define attributes in the same order as SA_*_OFFSET in order to
1152          * work around the problem. See ORI-610.
1153          */
1154         cnt = 0;
1155         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(uos), NULL, &osa->mode, 8);
1156         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(uos), NULL, &osa->size, 8);
1157         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GEN(uos), NULL, &gen, 8);
1158         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(uos), NULL, &osa->uid, 8);
1159         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(uos), NULL, &osa->gid, 8);
1160         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_PARENT(uos), NULL, &parent, 8);
1161         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(uos), NULL, &osa->flags, 8);
1162         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(uos), NULL, osa->atime, 16);
1163         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(uos), NULL, osa->mtime, 16);
1164         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(uos), NULL, osa->ctime, 16);
1165         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CRTIME(uos), NULL, crtime, 16);
1166         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(uos), NULL, &osa->nlink, 8);
1167         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_RDEV(uos), NULL, &osa->rdev, 8);
1168
1169         rc = -sa_replace_all_by_template(sa_hdl, bulk, cnt, tx);
1170
1171         OBD_FREE(bulk, sizeof(sa_bulk_attr_t) * 13);
1172 out:
1173         sa_handle_destroy(sa_hdl);
1174         return rc;
1175 }
1176
1177 /*
1178  * The transaction passed to this routine must have
1179  * dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT) called and then assigned
1180  * to a transaction group.
1181  */
1182 int __osd_object_create(const struct lu_env *env, udmu_objset_t *uos,
1183                         dmu_buf_t **dbp, dmu_tx_t *tx, struct lu_attr *la,
1184                         uint64_t parent, void *tag)
1185 {
1186         uint64_t oid;
1187         int      rc;
1188
1189         LASSERT(tag);
1190         spin_lock(&uos->lock);
1191         uos->objects++;
1192         spin_unlock(&uos->lock);
1193
1194         /* Assert that the transaction has been assigned to a
1195            transaction group. */
1196         LASSERT(tx->tx_txg != 0);
1197
1198         /* Create a new DMU object. */
1199         oid = dmu_object_alloc(uos->os, DMU_OT_PLAIN_FILE_CONTENTS, 0,
1200                                DMU_OT_SA, DN_MAX_BONUSLEN, tx);
1201         rc = -sa_buf_hold(uos->os, oid, tag, dbp);
1202         if (rc)
1203                 return rc;
1204
1205         LASSERT(la->la_valid & LA_MODE);
1206         la->la_size = 0;
1207         la->la_nlink = 1;
1208
1209         return __osd_attr_init(env, uos, oid, tx, la, parent);
1210 }
1211
1212 /*
1213  * The transaction passed to this routine must have
1214  * dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, ...) called and then assigned
1215  * to a transaction group.
1216  *
1217  * Using ZAP_FLAG_HASH64 will force the ZAP to always be a FAT ZAP.
1218  * This is fine for directories today, because storing the FID in the dirent
1219  * will also require a FAT ZAP.  If there is a new type of micro ZAP created
1220  * then we might need to re-evaluate the use of this flag and instead do
1221  * a conversion from the different internal ZAP hash formats being used. */
1222 int __osd_zap_create(const struct lu_env *env, udmu_objset_t *uos,
1223                      dmu_buf_t **zap_dbp, dmu_tx_t *tx,
1224                      struct lu_attr *la, uint64_t parent,
1225                      void *tag, zap_flags_t flags)
1226 {
1227         uint64_t oid;
1228         int      rc;
1229
1230         LASSERT(tag);
1231
1232         spin_lock(&uos->lock);
1233         uos->objects++;
1234         spin_unlock(&uos->lock);
1235
1236         /* Assert that the transaction has been assigned to a
1237            transaction group. */
1238         LASSERT(tx->tx_txg != 0);
1239
1240         oid = zap_create_flags(uos->os, 0, flags | ZAP_FLAG_HASH64,
1241                                DMU_OT_DIRECTORY_CONTENTS, 12, 12,
1242                                DMU_OT_SA, DN_MAX_BONUSLEN, tx);
1243
1244         rc = -sa_buf_hold(uos->os, oid, tag, zap_dbp);
1245         if (rc)
1246                 return rc;
1247
1248         LASSERT(la->la_valid & LA_MODE);
1249         la->la_size = 2;
1250         la->la_nlink = 1;
1251
1252         return __osd_attr_init(env, uos, oid, tx, la, parent);
1253 }
1254
1255 static dmu_buf_t *osd_mkidx(const struct lu_env *env, struct osd_device *osd,
1256                             struct lu_attr *la, uint64_t parent,
1257                             struct osd_thandle *oh)
1258 {
1259         dmu_buf_t *db;
1260         int        rc;
1261
1262         /* Index file should be created as regular file in order not to confuse
1263          * ZPL which could interpret them as directory.
1264          * We set ZAP_FLAG_UINT64_KEY to let ZFS know than we are going to use
1265          * binary keys */
1266         LASSERT(S_ISREG(la->la_mode));
1267         rc = __osd_zap_create(env, &osd->od_objset, &db, oh->ot_tx, la,
1268                               parent, osd_obj_tag, ZAP_FLAG_UINT64_KEY);
1269         if (rc)
1270                 return ERR_PTR(rc);
1271         return db;
1272 }
1273
1274 static dmu_buf_t *osd_mkdir(const struct lu_env *env, struct osd_device *osd,
1275                             struct lu_attr *la, uint64_t parent,
1276                             struct osd_thandle *oh)
1277 {
1278         dmu_buf_t *db;
1279         int        rc;
1280
1281         LASSERT(S_ISDIR(la->la_mode));
1282         rc = __osd_zap_create(env, &osd->od_objset, &db, oh->ot_tx, la,
1283                               parent, osd_obj_tag, 0);
1284         if (rc)
1285                 return ERR_PTR(rc);
1286         return db;
1287 }
1288
1289 static dmu_buf_t* osd_mkreg(const struct lu_env *env, struct osd_device *osd,
1290                             struct lu_attr *la, uint64_t parent,
1291                             struct osd_thandle *oh)
1292 {
1293         dmu_buf_t *db;
1294         int         rc;
1295
1296         LASSERT(S_ISREG(la->la_mode));
1297         rc = __osd_object_create(env, &osd->od_objset, &db, oh->ot_tx, la,
1298                                  parent, osd_obj_tag);
1299         if (rc)
1300                 return ERR_PTR(rc);
1301
1302         /*
1303          * XXX: a hack, OST to use bigger blocksize. we need
1304          * a method in OSD API to control this from OFD/MDD
1305          */
1306         if (!lu_device_is_md(osd2lu_dev(osd))) {
1307                 rc = -dmu_object_set_blocksize(osd->od_objset.os,
1308                                                db->db_object,
1309                                 128 << 10, 0, oh->ot_tx);
1310                 if (unlikely(rc)) {
1311                         CERROR("%s: can't change blocksize: %d\n",
1312                                osd->od_svname, rc);
1313                         return ERR_PTR(rc);
1314                 }
1315         }
1316
1317         return db;
1318 }
1319
1320 static dmu_buf_t *osd_mksym(const struct lu_env *env, struct osd_device *osd,
1321                             struct lu_attr *la, uint64_t parent,
1322                             struct osd_thandle *oh)
1323 {
1324         dmu_buf_t *db;
1325         int        rc;
1326
1327         LASSERT(S_ISLNK(la->la_mode));
1328         rc = __osd_object_create(env, &osd->od_objset, &db, oh->ot_tx, la,
1329                                  parent, osd_obj_tag);
1330         if (rc)
1331                 return ERR_PTR(rc);
1332         return db;
1333 }
1334
1335 static dmu_buf_t *osd_mknod(const struct lu_env *env, struct osd_device *osd,
1336                             struct lu_attr *la, uint64_t parent,
1337                             struct osd_thandle *oh)
1338 {
1339         dmu_buf_t *db;
1340         int        rc;
1341
1342         la->la_valid = LA_MODE;
1343         if (S_ISCHR(la->la_mode) || S_ISBLK(la->la_mode))
1344                 la->la_valid |= LA_RDEV;
1345
1346         rc = __osd_object_create(env, &osd->od_objset, &db, oh->ot_tx, la,
1347                                  parent, osd_obj_tag);
1348         if (rc)
1349                 return ERR_PTR(rc);
1350         return db;
1351 }
1352
1353 typedef dmu_buf_t *(*osd_obj_type_f)(const struct lu_env *env,
1354                                      struct osd_device *osd,
1355                                      struct lu_attr *la,
1356                                      uint64_t parent,
1357                                      struct osd_thandle *oh);
1358
1359 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
1360 {
1361         osd_obj_type_f result;
1362
1363         switch (type) {
1364         case DFT_DIR:
1365                 result = osd_mkdir;
1366                 break;
1367         case DFT_INDEX:
1368                 result = osd_mkidx;
1369                 break;
1370         case DFT_REGULAR:
1371                 result = osd_mkreg;
1372                 break;
1373         case DFT_SYM:
1374                 result = osd_mksym;
1375                 break;
1376         case DFT_NODE:
1377                 result = osd_mknod;
1378                 break;
1379         default:
1380                 LBUG();
1381                 break;
1382         }
1383         return result;
1384 }
1385
1386 /*
1387  * Primitives for directory (i.e. ZAP) handling
1388  */
1389 static inline int osd_init_lma(const struct lu_env *env, struct osd_object *obj,
1390                                const struct lu_fid *fid, struct osd_thandle *oh)
1391 {
1392         struct osd_thread_info  *info = osd_oti_get(env);
1393         struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
1394         struct lu_buf            buf;
1395         int rc;
1396
1397         lustre_lma_init(lma, fid, 0);
1398         lustre_lma_swab(lma);
1399         buf.lb_buf = lma;
1400         buf.lb_len = sizeof(*lma);
1401
1402         rc = osd_xattr_set_internal(env, obj, &buf, XATTR_NAME_LMA,
1403                                     LU_XATTR_CREATE, oh, BYPASS_CAPA);
1404
1405         return rc;
1406 }
1407
1408 /*
1409  * Concurrency: @dt is write locked.
1410  */
1411 static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
1412                              struct lu_attr *attr,
1413                              struct dt_allocation_hint *hint,
1414                              struct dt_object_format *dof,
1415                              struct thandle *th)
1416 {
1417         struct zpl_direntry     *zde = &osd_oti_get(env)->oti_zde.lzd_reg;
1418         const struct lu_fid     *fid = lu_object_fid(&dt->do_lu);
1419         struct osd_object       *obj = osd_dt_obj(dt);
1420         struct osd_device       *osd = osd_obj2dev(obj);
1421         char                    *buf = osd_oti_get(env)->oti_str;
1422         struct osd_thandle      *oh;
1423         dmu_buf_t               *db;
1424         uint64_t                 zapid;
1425         int                      rc;
1426
1427         ENTRY;
1428
1429         /* concurrent create declarations should not see
1430          * the object inconsistent (db, attr, etc).
1431          * in regular cases acquisition should be cheap */
1432         down(&obj->oo_guard);
1433
1434         LASSERT(osd_invariant(obj));
1435         LASSERT(!dt_object_exists(dt));
1436         LASSERT(dof != NULL);
1437
1438         LASSERT(th != NULL);
1439         oh = container_of0(th, struct osd_thandle, ot_super);
1440
1441         /*
1442          * XXX missing: Quote handling.
1443          */
1444
1445         LASSERT(obj->oo_db == NULL);
1446
1447         /* to follow ZFS on-disk format we need
1448          * to initialize parent dnode properly */
1449         zapid = 0;
1450         if (hint && hint->dah_parent)
1451                 zapid = osd_dt_obj(hint->dah_parent)->oo_db->db_object;
1452
1453         db = osd_create_type_f(dof->dof_type)(env, osd, attr, zapid, oh);
1454         if (IS_ERR(db))
1455                 GOTO(out, rc = PTR_ERR(th));
1456
1457         zde->zde_pad = 0;
1458         zde->zde_dnode = db->db_object;
1459         zde->zde_type = IFTODT(attr->la_mode & S_IFMT);
1460
1461         zapid = osd_get_name_n_idx(env, osd, fid, buf);
1462
1463         rc = -zap_add(osd->od_objset.os, zapid, buf, 8, 1, zde, oh->ot_tx);
1464         if (rc)
1465                 GOTO(out, rc);
1466
1467         /* Add new object to inode accounting.
1468          * Errors are not considered as fatal */
1469         rc = -zap_increment_int(osd->od_objset.os, osd->od_iusr_oid,
1470                                 (attr->la_valid & LA_UID) ? attr->la_uid : 0, 1,
1471                                 oh->ot_tx);
1472         if (rc)
1473                 CERROR("%s: failed to add "DFID" to accounting ZAP for usr %d "
1474                         "(%d)\n", osd->od_svname, PFID(fid), attr->la_uid, rc);
1475         rc = -zap_increment_int(osd->od_objset.os, osd->od_igrp_oid,
1476                                 (attr->la_valid & LA_GID) ? attr->la_gid : 0, 1,
1477                                 oh->ot_tx);
1478         if (rc)
1479                 CERROR("%s: failed to add "DFID" to accounting ZAP for grp %d "
1480                         "(%d)\n", osd->od_svname, PFID(fid), attr->la_gid, rc);
1481
1482         /* configure new osd object */
1483         obj->oo_db = db;
1484         rc = osd_object_init0(env, obj);
1485         LASSERT(ergo(rc == 0, dt_object_exists(dt)));
1486         LASSERT(osd_invariant(obj));
1487
1488         rc = osd_init_lma(env, obj, fid, oh);
1489         if (rc) {
1490                 CERROR("%s: can not set LMA on "DFID": rc = %d\n",
1491                        osd->od_svname, PFID(fid), rc);
1492                 /* ignore errors during LMA initialization */
1493                 rc = 0;
1494         }
1495
1496 out:
1497         up(&obj->oo_guard);
1498         RETURN(rc);
1499 }
1500
1501 static int osd_declare_object_ref_add(const struct lu_env *env,
1502                                       struct dt_object *dt,
1503                                       struct thandle *th)
1504 {
1505         return osd_declare_attr_set(env, dt, NULL, th);
1506 }
1507
1508 /*
1509  * Concurrency: @dt is write locked.
1510  */
1511 static int osd_object_ref_add(const struct lu_env *env,
1512                               struct dt_object *dt,
1513                               struct thandle *handle)
1514 {
1515         struct osd_object       *obj = osd_dt_obj(dt);
1516         struct osd_thandle      *oh;
1517         struct osd_device       *osd = osd_obj2dev(obj);
1518         udmu_objset_t           *uos = &osd->od_objset;
1519         uint64_t                 nlink;
1520         int rc;
1521
1522         ENTRY;
1523
1524         LASSERT(osd_invariant(obj));
1525         LASSERT(dt_object_exists(dt));
1526         LASSERT(obj->oo_sa_hdl != NULL);
1527
1528         oh = container_of0(handle, struct osd_thandle, ot_super);
1529
1530         write_lock(&obj->oo_attr_lock);
1531         nlink = ++obj->oo_attr.la_nlink;
1532         write_unlock(&obj->oo_attr_lock);
1533
1534         rc = osd_object_sa_update(obj, SA_ZPL_LINKS(uos), &nlink, 8, oh);
1535         return rc;
1536 }
1537
1538 static int osd_declare_object_ref_del(const struct lu_env *env,
1539                                       struct dt_object *dt,
1540                                       struct thandle *handle)
1541 {
1542         return osd_declare_attr_set(env, dt, NULL, handle);
1543 }
1544
1545 /*
1546  * Concurrency: @dt is write locked.
1547  */
1548 static int osd_object_ref_del(const struct lu_env *env,
1549                               struct dt_object *dt,
1550                               struct thandle *handle)
1551 {
1552         struct osd_object       *obj = osd_dt_obj(dt);
1553         struct osd_thandle      *oh;
1554         struct osd_device       *osd = osd_obj2dev(obj);
1555         udmu_objset_t           *uos = &osd->od_objset;
1556         uint64_t                 nlink;
1557         int                      rc;
1558
1559         ENTRY;
1560
1561         LASSERT(osd_invariant(obj));
1562         LASSERT(dt_object_exists(dt));
1563         LASSERT(obj->oo_sa_hdl != NULL);
1564
1565         oh = container_of0(handle, struct osd_thandle, ot_super);
1566         LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
1567
1568         write_lock(&obj->oo_attr_lock);
1569         nlink = --obj->oo_attr.la_nlink;
1570         write_unlock(&obj->oo_attr_lock);
1571
1572         rc = osd_object_sa_update(obj, SA_ZPL_LINKS(uos), &nlink, 8, oh);
1573         return rc;
1574 }
1575
1576 static int capa_is_sane(const struct lu_env *env, struct osd_device *dev,
1577                         struct lustre_capa *capa, struct lustre_capa_key *keys)
1578 {
1579         struct osd_thread_info  *oti = osd_oti_get(env);
1580         struct obd_capa         *oc;
1581         int                      i, rc = 0;
1582         ENTRY;
1583
1584         oc = capa_lookup(dev->od_capa_hash, capa, 0);
1585         if (oc) {
1586                 if (capa_is_expired(oc)) {
1587                         DEBUG_CAPA(D_ERROR, capa, "expired");
1588                         rc = -ESTALE;
1589                 }
1590                 capa_put(oc);
1591                 RETURN(rc);
1592         }
1593
1594         spin_lock(&capa_lock);
1595         for (i = 0; i < 2; i++) {
1596                 if (keys[i].lk_keyid == capa->lc_keyid) {
1597                         oti->oti_capa_key = keys[i];
1598                         break;
1599                 }
1600         }
1601         spin_unlock(&capa_lock);
1602
1603         if (i == 2) {
1604                 DEBUG_CAPA(D_ERROR, capa, "no matched capa key");
1605                 RETURN(-ESTALE);
1606         }
1607
1608         rc = capa_hmac(oti->oti_capa.lc_hmac, capa, oti->oti_capa_key.lk_key);
1609         if (rc)
1610                 RETURN(rc);
1611         if (memcmp(oti->oti_capa.lc_hmac, capa->lc_hmac, sizeof(capa->lc_hmac)))
1612         {
1613                 DEBUG_CAPA(D_ERROR, capa, "HMAC mismatch");
1614                 RETURN(-EACCES);
1615         }
1616
1617         oc = capa_add(dev->od_capa_hash, capa);
1618         capa_put(oc);
1619
1620         RETURN(0);
1621 }
1622
1623 static int osd_object_auth(const struct lu_env *env, struct dt_object *dt,
1624                            struct lustre_capa *capa, __u64 opc)
1625 {
1626         const struct lu_fid     *fid = lu_object_fid(&dt->do_lu);
1627         struct osd_device       *dev = osd_dev(dt->do_lu.lo_dev);
1628         int                      rc;
1629
1630         if (!dev->od_fl_capa)
1631                 return 0;
1632
1633         if (capa == BYPASS_CAPA)
1634                 return 0;
1635
1636         if (!capa) {
1637                 CERROR("no capability is provided for fid "DFID"\n", PFID(fid));
1638                 return -EACCES;
1639         }
1640
1641         if (!lu_fid_eq(fid, &capa->lc_fid)) {
1642                 DEBUG_CAPA(D_ERROR, capa, "fid "DFID" mismatch with",PFID(fid));
1643                 return -EACCES;
1644         }
1645
1646         if (!capa_opc_supported(capa, opc)) {
1647                 DEBUG_CAPA(D_ERROR, capa, "opc "LPX64" not supported by", opc);
1648                 return -EACCES;
1649         }
1650
1651         if ((rc = capa_is_sane(env, dev, capa, dev->od_capa_keys))) {
1652                 DEBUG_CAPA(D_ERROR, capa, "insane (rc %d)", rc);
1653                 return -EACCES;
1654         }
1655
1656         return 0;
1657 }
1658
1659 static struct obd_capa *osd_capa_get(const struct lu_env *env,
1660                                      struct dt_object *dt,
1661                                      struct lustre_capa *old,
1662                                      __u64 opc)
1663 {
1664         struct osd_thread_info  *info = osd_oti_get(env);
1665         const struct lu_fid     *fid = lu_object_fid(&dt->do_lu);
1666         struct osd_object       *obj = osd_dt_obj(dt);
1667         struct osd_device       *dev = osd_obj2dev(obj);
1668         struct lustre_capa_key  *key = &info->oti_capa_key;
1669         struct lustre_capa      *capa = &info->oti_capa;
1670         struct obd_capa         *oc;
1671         int                      rc;
1672         ENTRY;
1673
1674         if (!dev->od_fl_capa)
1675                 RETURN(ERR_PTR(-ENOENT));
1676
1677         LASSERT(dt_object_exists(dt));
1678         LASSERT(osd_invariant(obj));
1679
1680         /* renewal sanity check */
1681         if (old && osd_object_auth(env, dt, old, opc))
1682                 RETURN(ERR_PTR(-EACCES));
1683
1684         capa->lc_fid = *fid;
1685         capa->lc_opc = opc;
1686         capa->lc_uid = 0;
1687         capa->lc_flags = dev->od_capa_alg << 24;
1688         capa->lc_timeout = dev->od_capa_timeout;
1689         capa->lc_expiry = 0;
1690
1691         oc = capa_lookup(dev->od_capa_hash, capa, 1);
1692         if (oc) {
1693                 LASSERT(!capa_is_expired(oc));
1694                 RETURN(oc);
1695         }
1696
1697         spin_lock(&capa_lock);
1698         *key = dev->od_capa_keys[1];
1699         spin_unlock(&capa_lock);
1700
1701         capa->lc_keyid = key->lk_keyid;
1702         capa->lc_expiry = cfs_time_current_sec() + dev->od_capa_timeout;
1703
1704         rc = capa_hmac(capa->lc_hmac, capa, key->lk_key);
1705         if (rc) {
1706                 DEBUG_CAPA(D_ERROR, capa, "HMAC failed: %d for", rc);
1707                 LBUG();
1708                 RETURN(ERR_PTR(rc));
1709         }
1710
1711         oc = capa_add(dev->od_capa_hash, capa);
1712         RETURN(oc);
1713 }
1714
1715 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt)
1716 {
1717         struct osd_device *osd = osd_obj2dev(osd_dt_obj(dt));
1718         ENTRY;
1719
1720         /* XXX: no other option than syncing the whole filesystem until we
1721          * support ZIL */
1722         txg_wait_synced(dmu_objset_pool(osd->od_objset.os), 0ULL);
1723
1724         RETURN(0);
1725 }
1726
1727 static struct dt_object_operations osd_obj_ops = {
1728         .do_read_lock           = osd_object_read_lock,
1729         .do_write_lock          = osd_object_write_lock,
1730         .do_read_unlock         = osd_object_read_unlock,
1731         .do_write_unlock        = osd_object_write_unlock,
1732         .do_write_locked        = osd_object_write_locked,
1733         .do_attr_get            = osd_attr_get,
1734         .do_declare_attr_set    = osd_declare_attr_set,
1735         .do_attr_set            = osd_attr_set,
1736         .do_ah_init             = osd_ah_init,
1737         .do_declare_create      = osd_declare_object_create,
1738         .do_create              = osd_object_create,
1739         .do_declare_destroy     = osd_declare_object_destroy,
1740         .do_destroy             = osd_object_destroy,
1741         .do_index_try           = osd_index_try,
1742         .do_declare_ref_add     = osd_declare_object_ref_add,
1743         .do_ref_add             = osd_object_ref_add,
1744         .do_declare_ref_del     = osd_declare_object_ref_del,
1745         .do_ref_del             = osd_object_ref_del,
1746         .do_xattr_get           = osd_xattr_get,
1747         .do_declare_xattr_set   = osd_declare_xattr_set,
1748         .do_xattr_set           = osd_xattr_set,
1749         .do_declare_xattr_del   = osd_declare_xattr_del,
1750         .do_xattr_del           = osd_xattr_del,
1751         .do_xattr_list          = osd_xattr_list,
1752         .do_capa_get            = osd_capa_get,
1753         .do_object_sync         = osd_object_sync,
1754 };
1755
1756 static struct lu_object_operations osd_lu_obj_ops = {
1757         .loo_object_init        = osd_object_init,
1758         .loo_object_delete      = osd_object_delete,
1759         .loo_object_release     = osd_object_release,
1760         .loo_object_free        = osd_object_free,
1761         .loo_object_print       = osd_object_print,
1762         .loo_object_invariant   = osd_object_invariant,
1763 };
1764
1765 static int osd_otable_it_attr_get(const struct lu_env *env,
1766                                 struct dt_object *dt,
1767                                 struct lu_attr *attr,
1768                                 struct lustre_capa *capa)
1769 {
1770         attr->la_valid = 0;
1771         return 0;
1772 }
1773
1774 static struct dt_object_operations osd_obj_otable_it_ops = {
1775         .do_attr_get    = osd_otable_it_attr_get,
1776         .do_index_try   = osd_index_try,
1777 };
1778