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) 2011, 2012 Whamcloud, Inc.
  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         rc = qsd_op_begin(env, qsd, trans, qi, NULL);
 786         if (rc == -EDQUOT || rc == -EINPROGRESS)
 787                 rc = 0;
 788         if (rc)
 789                 return rc;
 790
 791         /* block accounting */
 792         qi->lqi_is_blk = true;
 793
 794         /* more blocks for the new owner ... */
 795         qi->lqi_id.qid_uid = new_id;
 796         qi->lqi_space      = bspace;
 797         rc = qsd_op_begin(env, qsd, trans, qi, NULL);
 798         if (rc == -EDQUOT || rc == -EINPROGRESS)
 799                 rc = 0;
 800         if (rc)
 801                 return rc;
 802
 803         /* and finally less blocks for the current owner */
 804         qi->lqi_id.qid_uid = orig_id;
 805         qi->lqi_space      = -bspace;
 806         rc = qsd_op_begin(env, qsd, trans, qi, NULL);
 807         if (rc == -EDQUOT || rc == -EINPROGRESS)
 808                 rc = 0;
 809         return rc;
 810 }
 811
 812 static int osd_declare_attr_set(const struct lu_env *env,
 813                                 struct dt_object *dt,
 814                                 const struct lu_attr *attr,
 815                                 struct thandle *handle)
 816 {
 817         struct osd_thread_info  *info = osd_oti_get(env);
 818         char                    *buf = osd_oti_get(env)->oti_str;
 819         struct osd_object       *obj = osd_dt_obj(dt);
 820         struct osd_device       *osd = osd_obj2dev(obj);
 821         struct osd_thandle      *oh;
 822         uint64_t                 bspace;
 823         uint32_t                 blksize;
 824         int                      rc;
 825         ENTRY;
 826
 827         if (!dt_object_exists(dt)) {
 828                 /* XXX: sanity check that object creation is declared */
 829                 RETURN(0);
 830         }
 831
 832         LASSERT(handle != NULL);
 833         LASSERT(osd_invariant(obj));
 834
 835         oh = container_of0(handle, struct osd_thandle, ot_super);
 836
 837         LASSERT(obj->oo_sa_hdl != NULL);
 838         dmu_tx_hold_sa(oh->ot_tx, obj->oo_sa_hdl, 0);
 839
 840         sa_object_size(obj->oo_sa_hdl, &blksize, &bspace);
 841         bspace = toqb(bspace * blksize);
 842
 843         if (attr && attr->la_valid & LA_UID) {
 844                 /* account for user inode tracking ZAP update */
 845                 dmu_tx_hold_bonus(oh->ot_tx, osd->od_iusr_oid);
 846                 dmu_tx_hold_zap(oh->ot_tx, osd->od_iusr_oid, TRUE, buf);
 847
 848                 /* quota enforcement for user */
 849                 if (attr->la_uid != obj->oo_attr.la_uid) {
 850                         rc = qsd_transfer(env, osd->od_quota_slave,
 851                                           &oh->ot_quota_trans, USRQUOTA,
 852                                           obj->oo_attr.la_uid, attr->la_uid,
 853                                           bspace, &info->oti_qi);
 854                         if (rc)
 855                                 RETURN(rc);
 856                 }
 857         }
 858         if (attr && attr->la_valid & LA_GID) {
 859                 /* account for user inode tracking ZAP update */
 860                 dmu_tx_hold_bonus(oh->ot_tx, osd->od_igrp_oid);
 861                 dmu_tx_hold_zap(oh->ot_tx, osd->od_igrp_oid, TRUE, buf);
 862
 863                 /* quota enforcement for group */
 864                 if (attr->la_gid != obj->oo_attr.la_gid) {
 865                         rc = qsd_transfer(env, osd->od_quota_slave,
 866                                           &oh->ot_quota_trans, GRPQUOTA,
 867                                           obj->oo_attr.la_gid, attr->la_gid,
 868                                           bspace, &info->oti_qi);
 869                         if (rc)
 870                                 RETURN(rc);
 871                 }
 872         }
 873
 874         RETURN(0);
 875 }
 876
 877 /*
 878  * Set the attributes of an object
 879  *
 880  * The transaction passed to this routine must have
 881  * dmu_tx_hold_bonus(tx, oid) called and then assigned
 882  * to a transaction group.
 883  */
 884 static int osd_attr_set(const struct lu_env *env, struct dt_object *dt,
 885                         const struct lu_attr *la, struct thandle *handle,
 886                         struct lustre_capa *capa)
 887 {
 888         struct osd_object       *obj = osd_dt_obj(dt);
 889         struct osd_device       *osd = osd_obj2dev(obj);
 890         udmu_objset_t           *uos = &osd->od_objset;
 891         struct osd_thandle      *oh;
 892         struct osa_attr         *osa = &osd_oti_get(env)->oti_osa;
 893         sa_bulk_attr_t          *bulk;
 894         int                      cnt;
 895         int                      rc = 0;
 896
 897         ENTRY;
 898         LASSERT(handle != NULL);
 899         LASSERT(dt_object_exists(dt));
 900         LASSERT(osd_invariant(obj));
 901         LASSERT(obj->oo_sa_hdl);
 902
 903         oh = container_of0(handle, struct osd_thandle, ot_super);
 904         /* Assert that the transaction has been assigned to a
 905            transaction group. */
 906         LASSERT(oh->ot_tx->tx_txg != 0);
 907
 908         if (la->la_valid == 0)
 909                 RETURN(0);
 910
 911         OBD_ALLOC(bulk, sizeof(sa_bulk_attr_t) * 10);
 912         if (bulk == NULL)
 913                 RETURN(-ENOMEM);
 914
 915         /* do both accounting updates outside oo_attr_lock below */
 916         if ((la->la_valid & LA_UID) && (la->la_uid != obj->oo_attr.la_uid)) {
 917                 /* Update user accounting. Failure isn't fatal, but we still
 918                  * log an error message */
 919                 rc = -zap_increment_int(osd->od_objset.os, osd->od_iusr_oid,
 920                                         la->la_uid, 1, oh->ot_tx);
 921                 if (rc)
 922                         CERROR("%s: failed to update accounting ZAP for user "
 923                                 "%d (%d)\n", osd->od_svname, la->la_uid, rc);
 924                 rc = -zap_increment_int(osd->od_objset.os, osd->od_iusr_oid,
 925                                         obj->oo_attr.la_uid, -1, oh->ot_tx);
 926                 if (rc)
 927                         CERROR("%s: failed to update accounting ZAP for user "
 928                                 "%d (%d)\n", osd->od_svname,
 929                                 obj->oo_attr.la_uid, rc);
 930         }
 931         if ((la->la_valid & LA_GID) && (la->la_gid != obj->oo_attr.la_gid)) {
 932                 /* Update group accounting. Failure isn't fatal, but we still
 933                  * log an error message */
 934                 rc = -zap_increment_int(osd->od_objset.os, osd->od_igrp_oid,
 935                                         la->la_gid, 1, oh->ot_tx);
 936                 if (rc)
 937                         CERROR("%s: failed to update accounting ZAP for user "
 938                                 "%d (%d)\n", osd->od_svname, la->la_gid, rc);
 939                 rc = -zap_increment_int(osd->od_objset.os, osd->od_igrp_oid,
 940                                         obj->oo_attr.la_gid, -1, oh->ot_tx);
 941                 if (rc)
 942                         CERROR("%s: failed to update accounting ZAP for user "
 943                                 "%d (%d)\n", osd->od_svname,
 944                                 obj->oo_attr.la_gid, rc);
 945         }
 946
 947         write_lock(&obj->oo_attr_lock);
 948         cnt = 0;
 949         if (la->la_valid & LA_ATIME) {
 950                 osa->atime[0] = obj->oo_attr.la_atime = la->la_atime;
 951                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(uos), NULL,
 952                                  osa->atime, 16);
 953         }
 954         if (la->la_valid & LA_MTIME) {
 955                 osa->mtime[0] = obj->oo_attr.la_mtime = la->la_mtime;
 956                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(uos), NULL,
 957                                  osa->mtime, 16);
 958         }
 959         if (la->la_valid & LA_CTIME) {
 960                 osa->ctime[0] = obj->oo_attr.la_ctime = la->la_ctime;
 961                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(uos), NULL,
 962                                  osa->ctime, 16);
 963         }
 964         if (la->la_valid & LA_MODE) {
 965                 /* mode is stored along with type, so read it first */
 966                 obj->oo_attr.la_mode = (obj->oo_attr.la_mode & S_IFMT) |
 967                         (la->la_mode & ~S_IFMT);
 968                 osa->mode = obj->oo_attr.la_mode;
 969                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(uos), NULL,
 970                                  &osa->mode, 8);
 971         }
 972         if (la->la_valid & LA_SIZE) {
 973                 osa->size = obj->oo_attr.la_size = la->la_size;
 974                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(uos), NULL,
 975                                  &osa->size, 8);
 976         }
 977         if (la->la_valid & LA_NLINK) {
 978                 osa->nlink = obj->oo_attr.la_nlink = la->la_nlink;
 979                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(uos), NULL,
 980                                  &osa->nlink, 8);
 981         }
 982         if (la->la_valid & LA_RDEV) {
 983                 osa->rdev = obj->oo_attr.la_rdev = la->la_rdev;
 984                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_RDEV(uos), NULL,
 985                                  &osa->rdev, 8);
 986         }
 987         if (la->la_valid & LA_FLAGS) {
 988                 osa->flags = obj->oo_attr.la_flags = la->la_flags;
 989                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(uos), NULL,
 990                                  &osa->flags, 8);
 991         }
 992         if (la->la_valid & LA_UID) {
 993                 osa->uid = obj->oo_attr.la_uid = la->la_uid;
 994                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(uos), NULL,
 995                                  &osa->uid, 8);
 996         }
 997         if (la->la_valid & LA_GID) {
 998                 osa->gid = obj->oo_attr.la_gid = la->la_gid;
 999                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(uos), NULL,
1000                                  &osa->gid, 8);
1001         }
1002         obj->oo_attr.la_valid |= la->la_valid;
1003         write_unlock(&obj->oo_attr_lock);
1004
1005         rc = osd_object_sa_bulk_update(obj, bulk, cnt, oh);
1006
1007         OBD_FREE(bulk, sizeof(sa_bulk_attr_t) * 10);
1008         RETURN(rc);
1009 }
1010
1011 /*
1012  * Object creation.
1013  *
1014  * XXX temporary solution.
1015  */
1016
1017 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
1018                         struct dt_object *parent, struct dt_object *child,
1019                         cfs_umode_t child_mode)
1020 {
1021         LASSERT(ah);
1022
1023         memset(ah, 0, sizeof(*ah));
1024         ah->dah_mode = child_mode;
1025 }
1026
1027 static int osd_declare_object_create(const struct lu_env *env,
1028                                      struct dt_object *dt,
1029                                      struct lu_attr *attr,
1030                                      struct dt_allocation_hint *hint,
1031                                      struct dt_object_format *dof,
1032                                      struct thandle *handle)
1033 {
1034         char                    *buf = osd_oti_get(env)->oti_str;
1035         const struct lu_fid     *fid = lu_object_fid(&dt->do_lu);
1036         struct osd_object       *obj = osd_dt_obj(dt);
1037         struct osd_device       *osd = osd_obj2dev(obj);
1038         struct osd_thandle      *oh;
1039         uint64_t                 zapid;
1040         ENTRY;
1041
1042         LASSERT(dof);
1043
1044         switch (dof->dof_type) {
1045                 case DFT_REGULAR:
1046                 case DFT_SYM:
1047                 case DFT_NODE:
1048                         if (obj->oo_dt.do_body_ops == NULL)
1049                                 obj->oo_dt.do_body_ops = &osd_body_ops;
1050                         break;
1051                 default:
1052                         break;
1053         }
1054
1055         LASSERT(handle != NULL);
1056         oh = container_of0(handle, struct osd_thandle, ot_super);
1057         LASSERT(oh->ot_tx != NULL);
1058
1059         switch (dof->dof_type) {
1060                 case DFT_DIR:
1061                 case DFT_INDEX:
1062                         /* for zap create */
1063                         dmu_tx_hold_zap(oh->ot_tx, DMU_NEW_OBJECT, 1, NULL);
1064                         break;
1065                 case DFT_REGULAR:
1066                 case DFT_SYM:
1067                 case DFT_NODE:
1068                         /* first, we'll create new object */
1069                         dmu_tx_hold_bonus(oh->ot_tx, DMU_NEW_OBJECT);
1070                         break;
1071
1072                 default:
1073                         LBUG();
1074                         break;
1075         }
1076
1077         /* and we'll add it to some mapping */
1078         zapid = osd_get_name_n_idx(env, osd, fid, buf);
1079         dmu_tx_hold_bonus(oh->ot_tx, zapid);
1080         dmu_tx_hold_zap(oh->ot_tx, zapid, TRUE, buf);
1081
1082         /* we will also update inode accounting ZAPs */
1083         dmu_tx_hold_bonus(oh->ot_tx, osd->od_iusr_oid);
1084         dmu_tx_hold_zap(oh->ot_tx, osd->od_iusr_oid, TRUE, buf);
1085         dmu_tx_hold_bonus(oh->ot_tx, osd->od_igrp_oid);
1086         dmu_tx_hold_zap(oh->ot_tx, osd->od_igrp_oid, TRUE, buf);
1087
1088         dmu_tx_hold_sa_create(oh->ot_tx, ZFS_SA_BASE_ATTR_SIZE);
1089
1090         __osd_xattr_declare_set(env, obj, sizeof(struct lustre_mdt_attrs),
1091                                 XATTR_NAME_LMA, oh);
1092
1093         RETURN(osd_declare_quota(env, osd, attr->la_uid, attr->la_gid, 1, oh,
1094                                  false, NULL, false));
1095 }
1096
1097 int __osd_attr_init(const struct lu_env *env, udmu_objset_t *uos,
1098                     uint64_t oid, dmu_tx_t *tx, struct lu_attr *la)
1099 {
1100         sa_bulk_attr_t  *bulk;
1101         sa_handle_t     *sa_hdl;
1102         struct osa_attr *osa = &osd_oti_get(env)->oti_osa;
1103         uint64_t         gen;
1104         uint64_t         parent;
1105         uint64_t         crtime[2];
1106         timestruc_t      now;
1107         int              cnt;
1108         int              rc;
1109
1110         gethrestime(&now);
1111         gen = dmu_tx_get_txg(tx);
1112
1113         ZFS_TIME_ENCODE(&now, crtime);
1114         /* XXX: this should be real id of parent for ZPL access, but we have no
1115          * such info in OSD, probably it can be part of dt_object_format */
1116         parent = 0;
1117
1118         osa->atime[0] = la->la_atime;
1119         osa->ctime[0] = la->la_ctime;
1120         osa->mtime[0] = la->la_mtime;
1121         osa->mode = la->la_mode;
1122         osa->uid = la->la_uid;
1123         osa->gid = la->la_gid;
1124         osa->rdev = la->la_rdev;
1125         osa->nlink = la->la_nlink;
1126         osa->flags = la->la_flags;
1127         osa->size  = la->la_size;
1128
1129         /* Now add in all of the "SA" attributes */
1130         rc = -sa_handle_get(uos->os, oid, NULL, SA_HDL_PRIVATE, &sa_hdl);
1131         if (rc)
1132                 return rc;
1133
1134         OBD_ALLOC(bulk, sizeof(sa_bulk_attr_t) * 13);
1135         if (bulk == NULL) {
1136                 rc = -ENOMEM;
1137                 goto out;
1138         }
1139         /*
1140          * we need to create all SA below upon object create.
1141          *
1142          * XXX The attribute order matters since the accounting callback relies
1143          * on static offsets (i.e. SA_*_OFFSET, see zfs_space_delta_cb()) to
1144          * look up the UID/GID attributes. Moreover, the callback does not seem
1145          * to support the spill block.
1146          * We define attributes in the same order as SA_*_OFFSET in order to
1147          * work around the problem. See ORI-610.
1148          */
1149         cnt = 0;
1150         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(uos), NULL, &osa->mode, 8);
1151         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(uos), NULL, &osa->size, 8);
1152         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GEN(uos), NULL, &gen, 8);
1153         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(uos), NULL, &osa->uid, 8);
1154         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(uos), NULL, &osa->gid, 8);
1155         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_PARENT(uos), NULL, &parent, 8);
1156         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(uos), NULL, &osa->flags, 8);
1157         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(uos), NULL, osa->atime, 16);
1158         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(uos), NULL, osa->mtime, 16);
1159         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(uos), NULL, osa->ctime, 16);
1160         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CRTIME(uos), NULL, crtime, 16);
1161         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(uos), NULL, &osa->nlink, 8);
1162         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_RDEV(uos), NULL, &osa->rdev, 8);
1163
1164         rc = -sa_replace_all_by_template(sa_hdl, bulk, cnt, tx);
1165
1166         OBD_FREE(bulk, sizeof(sa_bulk_attr_t) * 13);
1167 out:
1168         sa_handle_destroy(sa_hdl);
1169         return rc;
1170 }
1171
1172 /*
1173  * The transaction passed to this routine must have
1174  * dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT) called and then assigned
1175  * to a transaction group.
1176  */
1177 int __osd_object_create(const struct lu_env *env, udmu_objset_t *uos,
1178                         dmu_buf_t **dbp, dmu_tx_t *tx,
1179                         struct lu_attr *la, void *tag)
1180 {
1181         uint64_t oid;
1182         int      rc;
1183
1184         LASSERT(tag);
1185         spin_lock(&uos->lock);
1186         uos->objects++;
1187         spin_unlock(&uos->lock);
1188
1189         /* Assert that the transaction has been assigned to a
1190            transaction group. */
1191         LASSERT(tx->tx_txg != 0);
1192
1193         /* Create a new DMU object. */
1194         oid = dmu_object_alloc(uos->os, DMU_OT_PLAIN_FILE_CONTENTS, 0,
1195                                DMU_OT_SA, DN_MAX_BONUSLEN, tx);
1196         rc = -sa_buf_hold(uos->os, oid, tag, dbp);
1197         if (rc)
1198                 return rc;
1199
1200         LASSERT(la->la_valid & LA_MODE);
1201         la->la_size = 0;
1202         la->la_nlink = 1;
1203
1204         return __osd_attr_init(env, uos, oid, tx, la);
1205 }
1206
1207 /*
1208  * The transaction passed to this routine must have
1209  * dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, ...) called and then assigned
1210  * to a transaction group.
1211  *
1212  * Using ZAP_FLAG_HASH64 will force the ZAP to always be a FAT ZAP.
1213  * This is fine for directories today, because storing the FID in the dirent
1214  * will also require a FAT ZAP.  If there is a new type of micro ZAP created
1215  * then we might need to re-evaluate the use of this flag and instead do
1216  * a conversion from the different internal ZAP hash formats being used. */
1217 int __osd_zap_create(const struct lu_env *env, udmu_objset_t *uos,
1218                      dmu_buf_t **zap_dbp, dmu_tx_t *tx,
1219                      struct lu_attr *la, void *tag, zap_flags_t flags)
1220 {
1221         uint64_t oid;
1222         int      rc;
1223
1224         LASSERT(tag);
1225
1226         spin_lock(&uos->lock);
1227         uos->objects++;
1228         spin_unlock(&uos->lock);
1229
1230         /* Assert that the transaction has been assigned to a
1231            transaction group. */
1232         LASSERT(tx->tx_txg != 0);
1233
1234         oid = zap_create_flags(uos->os, 0, flags | ZAP_FLAG_HASH64,
1235                                DMU_OT_DIRECTORY_CONTENTS, 12, 12,
1236                                DMU_OT_SA, DN_MAX_BONUSLEN, tx);
1237
1238         rc = -sa_buf_hold(uos->os, oid, tag, zap_dbp);
1239         if (rc)
1240                 return rc;
1241
1242         LASSERT(la->la_valid & LA_MODE);
1243         la->la_size = 2;
1244         la->la_nlink = 1;
1245
1246         return __osd_attr_init(env, uos, oid, tx, la);
1247 }
1248
1249 static dmu_buf_t *osd_mkidx(const struct lu_env *env, struct osd_device *osd,
1250                             struct lu_attr *la, struct osd_thandle *oh)
1251 {
1252         dmu_buf_t *db;
1253         int        rc;
1254
1255         /* Index file should be created as regular file in order not to confuse
1256          * ZPL which could interpret them as directory.
1257          * We set ZAP_FLAG_UINT64_KEY to let ZFS know than we are going to use
1258          * binary keys */
1259         LASSERT(S_ISREG(la->la_mode));
1260         rc = __osd_zap_create(env, &osd->od_objset, &db, oh->ot_tx, la,
1261                               osd_obj_tag, ZAP_FLAG_UINT64_KEY);
1262         if (rc)
1263                 return ERR_PTR(rc);
1264         return db;
1265 }
1266
1267 static dmu_buf_t *osd_mkdir(const struct lu_env *env, struct osd_device *osd,
1268                             struct lu_attr *la, struct osd_thandle *oh)
1269 {
1270         dmu_buf_t *db;
1271         int        rc;
1272
1273         LASSERT(S_ISDIR(la->la_mode));
1274         rc = __osd_zap_create(env, &osd->od_objset, &db, oh->ot_tx, la,
1275                               osd_obj_tag, 0);
1276         if (rc)
1277                 return ERR_PTR(rc);
1278         return db;
1279 }
1280
1281 static dmu_buf_t* osd_mkreg(const struct lu_env *env, struct osd_device *osd,
1282                             struct lu_attr *la, struct osd_thandle *oh)
1283 {
1284         dmu_buf_t *db;
1285         int         rc;
1286
1287         LASSERT(S_ISREG(la->la_mode));
1288         rc = __osd_object_create(env, &osd->od_objset, &db, oh->ot_tx, la,
1289                                  osd_obj_tag);
1290         if (rc)
1291                 return ERR_PTR(rc);
1292
1293         /*
1294          * XXX: a hack, OST to use bigger blocksize. we need
1295          * a method in OSD API to control this from OFD/MDD
1296          */
1297         if (!lu_device_is_md(osd2lu_dev(osd))) {
1298                 rc = -dmu_object_set_blocksize(osd->od_objset.os,
1299                                                db->db_object,
1300                                 128 << 10, 0, oh->ot_tx);
1301                 if (unlikely(rc)) {
1302                         CERROR("%s: can't change blocksize: %d\n",
1303                                osd->od_svname, rc);
1304                         return ERR_PTR(rc);
1305                 }
1306         }
1307
1308         return db;
1309 }
1310
1311 static dmu_buf_t *osd_mksym(const struct lu_env *env, struct osd_device *osd,
1312                             struct lu_attr *la, struct osd_thandle *oh)
1313 {
1314         dmu_buf_t *db;
1315         int        rc;
1316
1317         LASSERT(S_ISLNK(la->la_mode));
1318         rc = __osd_object_create(env, &osd->od_objset, &db, oh->ot_tx, la,
1319                                  osd_obj_tag);
1320         if (rc)
1321                 return ERR_PTR(rc);
1322         return db;
1323 }
1324
1325 static dmu_buf_t *osd_mknod(const struct lu_env *env, struct osd_device *osd,
1326                             struct lu_attr *la, struct osd_thandle *oh)
1327 {
1328         dmu_buf_t *db;
1329         int        rc;
1330
1331         la->la_valid = LA_MODE;
1332         if (S_ISCHR(la->la_mode) || S_ISBLK(la->la_mode))
1333                 la->la_valid |= LA_RDEV;
1334
1335         rc = __osd_object_create(env, &osd->od_objset, &db, oh->ot_tx, la,
1336                                  osd_obj_tag);
1337         if (rc)
1338                 return ERR_PTR(rc);
1339         return db;
1340 }
1341
1342 typedef dmu_buf_t *(*osd_obj_type_f)(const struct lu_env *env, struct osd_device *osd,
1343                                      struct lu_attr *la, struct osd_thandle *oh);
1344
1345 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
1346 {
1347         osd_obj_type_f result;
1348
1349         switch (type) {
1350         case DFT_DIR:
1351                 result = osd_mkdir;
1352                 break;
1353         case DFT_INDEX:
1354                 result = osd_mkidx;
1355                 break;
1356         case DFT_REGULAR:
1357                 result = osd_mkreg;
1358                 break;
1359         case DFT_SYM:
1360                 result = osd_mksym;
1361                 break;
1362         case DFT_NODE:
1363                 result = osd_mknod;
1364                 break;
1365         default:
1366                 LBUG();
1367                 break;
1368         }
1369         return result;
1370 }
1371
1372 /*
1373  * Primitives for directory (i.e. ZAP) handling
1374  */
1375 static inline int osd_init_lma(const struct lu_env *env, struct osd_object *obj,
1376                                const struct lu_fid *fid, struct osd_thandle *oh)
1377 {
1378         struct osd_thread_info  *info = osd_oti_get(env);
1379         struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
1380         struct lu_buf            buf;
1381         int rc;
1382
1383         lustre_lma_init(lma, fid);
1384         lustre_lma_swab(lma);
1385         buf.lb_buf = lma;
1386         buf.lb_len = sizeof(*lma);
1387
1388         rc = osd_xattr_set_internal(env, obj, &buf, XATTR_NAME_LMA,
1389                                     LU_XATTR_CREATE, oh, BYPASS_CAPA);
1390
1391         return rc;
1392 }
1393
1394 /*
1395  * Concurrency: @dt is write locked.
1396  */
1397 static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
1398                              struct lu_attr *attr,
1399                              struct dt_allocation_hint *hint,
1400                              struct dt_object_format *dof,
1401                              struct thandle *th)
1402 {
1403         struct zpl_direntry     *zde = &osd_oti_get(env)->oti_zde.lzd_reg;
1404         const struct lu_fid     *fid = lu_object_fid(&dt->do_lu);
1405         struct osd_object       *obj = osd_dt_obj(dt);
1406         struct osd_device       *osd = osd_obj2dev(obj);
1407         char                    *buf = osd_oti_get(env)->oti_str;
1408         struct osd_thandle      *oh;
1409         dmu_buf_t               *db;
1410         uint64_t                 zapid;
1411         int                      rc;
1412
1413         ENTRY;
1414
1415         /* concurrent create declarations should not see
1416          * the object inconsistent (db, attr, etc).
1417          * in regular cases acquisition should be cheap */
1418         down(&obj->oo_guard);
1419
1420         LASSERT(osd_invariant(obj));
1421         LASSERT(!dt_object_exists(dt));
1422         LASSERT(dof != NULL);
1423
1424         LASSERT(th != NULL);
1425         oh = container_of0(th, struct osd_thandle, ot_super);
1426
1427         /*
1428          * XXX missing: Quote handling.
1429          */
1430
1431         LASSERT(obj->oo_db == NULL);
1432
1433         db = osd_create_type_f(dof->dof_type)(env, osd, attr, oh);
1434         if (IS_ERR(db))
1435                 GOTO(out, rc = PTR_ERR(th));
1436
1437         zde->zde_pad = 0;
1438         zde->zde_dnode = db->db_object;
1439         zde->zde_type = IFTODT(attr->la_mode & S_IFMT);
1440
1441         zapid = osd_get_name_n_idx(env, osd, fid, buf);
1442
1443         rc = -zap_add(osd->od_objset.os, zapid, buf, 8, 1, zde, oh->ot_tx);
1444         if (rc)
1445                 GOTO(out, rc);
1446
1447         /* Add new object to inode accounting.
1448          * Errors are not considered as fatal */
1449         rc = -zap_increment_int(osd->od_objset.os, osd->od_iusr_oid,
1450                                 (attr->la_valid & LA_UID) ? attr->la_uid : 0, 1,
1451                                 oh->ot_tx);
1452         if (rc)
1453                 CERROR("%s: failed to add "DFID" to accounting ZAP for usr %d "
1454                         "(%d)\n", osd->od_svname, PFID(fid), attr->la_uid, rc);
1455         rc = -zap_increment_int(osd->od_objset.os, osd->od_igrp_oid,
1456                                 (attr->la_valid & LA_GID) ? attr->la_gid : 0, 1,
1457                                 oh->ot_tx);
1458         if (rc)
1459                 CERROR("%s: failed to add "DFID" to accounting ZAP for grp %d "
1460                         "(%d)\n", osd->od_svname, PFID(fid), attr->la_gid, rc);
1461
1462         /* configure new osd object */
1463         obj->oo_db = db;
1464         rc = osd_object_init0(env, obj);
1465         LASSERT(ergo(rc == 0, dt_object_exists(dt)));
1466         LASSERT(osd_invariant(obj));
1467
1468         rc = osd_init_lma(env, obj, fid, oh);
1469         if (rc) {
1470                 CERROR("%s: can not set LMA on "DFID": rc = %d\n",
1471                        osd->od_svname, PFID(fid), rc);
1472                 /* ignore errors during LMA initialization */
1473                 rc = 0;
1474         }
1475
1476 out:
1477         up(&obj->oo_guard);
1478         RETURN(rc);
1479 }
1480
1481 static int osd_declare_object_ref_add(const struct lu_env *env,
1482                                       struct dt_object *dt,
1483                                       struct thandle *th)
1484 {
1485         return osd_declare_attr_set(env, dt, NULL, th);
1486 }
1487
1488 /*
1489  * Concurrency: @dt is write locked.
1490  */
1491 static int osd_object_ref_add(const struct lu_env *env,
1492                               struct dt_object *dt,
1493                               struct thandle *handle)
1494 {
1495         struct osd_object       *obj = osd_dt_obj(dt);
1496         struct osd_thandle      *oh;
1497         struct osd_device       *osd = osd_obj2dev(obj);
1498         udmu_objset_t           *uos = &osd->od_objset;
1499         uint64_t                 nlink;
1500         int rc;
1501
1502         ENTRY;
1503
1504         LASSERT(osd_invariant(obj));
1505         LASSERT(dt_object_exists(dt));
1506         LASSERT(obj->oo_sa_hdl != NULL);
1507
1508         oh = container_of0(handle, struct osd_thandle, ot_super);
1509
1510         write_lock(&obj->oo_attr_lock);
1511         nlink = ++obj->oo_attr.la_nlink;
1512         write_unlock(&obj->oo_attr_lock);
1513
1514         rc = osd_object_sa_update(obj, SA_ZPL_LINKS(uos), &nlink, 8, oh);
1515         return rc;
1516 }
1517
1518 static int osd_declare_object_ref_del(const struct lu_env *env,
1519                                       struct dt_object *dt,
1520                                       struct thandle *handle)
1521 {
1522         return osd_declare_attr_set(env, dt, NULL, handle);
1523 }
1524
1525 /*
1526  * Concurrency: @dt is write locked.
1527  */
1528 static int osd_object_ref_del(const struct lu_env *env,
1529                               struct dt_object *dt,
1530                               struct thandle *handle)
1531 {
1532         struct osd_object       *obj = osd_dt_obj(dt);
1533         struct osd_thandle      *oh;
1534         struct osd_device       *osd = osd_obj2dev(obj);
1535         udmu_objset_t           *uos = &osd->od_objset;
1536         uint64_t                 nlink;
1537         int                      rc;
1538
1539         ENTRY;
1540
1541         LASSERT(osd_invariant(obj));
1542         LASSERT(dt_object_exists(dt));
1543         LASSERT(obj->oo_sa_hdl != NULL);
1544
1545         oh = container_of0(handle, struct osd_thandle, ot_super);
1546         LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
1547
1548         write_lock(&obj->oo_attr_lock);
1549         nlink = --obj->oo_attr.la_nlink;
1550         write_unlock(&obj->oo_attr_lock);
1551
1552         rc = osd_object_sa_update(obj, SA_ZPL_LINKS(uos), &nlink, 8, oh);
1553         return rc;
1554 }
1555
1556 static int capa_is_sane(const struct lu_env *env, struct osd_device *dev,
1557                         struct lustre_capa *capa, struct lustre_capa_key *keys)
1558 {
1559         struct osd_thread_info  *oti = osd_oti_get(env);
1560         struct obd_capa         *oc;
1561         int                      i, rc = 0;
1562         ENTRY;
1563
1564         oc = capa_lookup(dev->od_capa_hash, capa, 0);
1565         if (oc) {
1566                 if (capa_is_expired(oc)) {
1567                         DEBUG_CAPA(D_ERROR, capa, "expired");
1568                         rc = -ESTALE;
1569                 }
1570                 capa_put(oc);
1571                 RETURN(rc);
1572         }
1573
1574         spin_lock(&capa_lock);
1575         for (i = 0; i < 2; i++) {
1576                 if (keys[i].lk_keyid == capa->lc_keyid) {
1577                         oti->oti_capa_key = keys[i];
1578                         break;
1579                 }
1580         }
1581         spin_unlock(&capa_lock);
1582
1583         if (i == 2) {
1584                 DEBUG_CAPA(D_ERROR, capa, "no matched capa key");
1585                 RETURN(-ESTALE);
1586         }
1587
1588         rc = capa_hmac(oti->oti_capa.lc_hmac, capa, oti->oti_capa_key.lk_key);
1589         if (rc)
1590                 RETURN(rc);
1591         if (memcmp(oti->oti_capa.lc_hmac, capa->lc_hmac, sizeof(capa->lc_hmac)))
1592         {
1593                 DEBUG_CAPA(D_ERROR, capa, "HMAC mismatch");
1594                 RETURN(-EACCES);
1595         }
1596
1597         oc = capa_add(dev->od_capa_hash, capa);
1598         capa_put(oc);
1599
1600         RETURN(0);
1601 }
1602
1603 static int osd_object_auth(const struct lu_env *env, struct dt_object *dt,
1604                            struct lustre_capa *capa, __u64 opc)
1605 {
1606         const struct lu_fid     *fid = lu_object_fid(&dt->do_lu);
1607         struct osd_device       *dev = osd_dev(dt->do_lu.lo_dev);
1608         int                      rc;
1609
1610         if (!dev->od_fl_capa)
1611                 return 0;
1612
1613         if (capa == BYPASS_CAPA)
1614                 return 0;
1615
1616         if (!capa) {
1617                 CERROR("no capability is provided for fid "DFID"\n", PFID(fid));
1618                 return -EACCES;
1619         }
1620
1621         if (!lu_fid_eq(fid, &capa->lc_fid)) {
1622                 DEBUG_CAPA(D_ERROR, capa, "fid "DFID" mismatch with",PFID(fid));
1623                 return -EACCES;
1624         }
1625
1626         if (!capa_opc_supported(capa, opc)) {
1627                 DEBUG_CAPA(D_ERROR, capa, "opc "LPX64" not supported by", opc);
1628                 return -EACCES;
1629         }
1630
1631         if ((rc = capa_is_sane(env, dev, capa, dev->od_capa_keys))) {
1632                 DEBUG_CAPA(D_ERROR, capa, "insane (rc %d)", rc);
1633                 return -EACCES;
1634         }
1635
1636         return 0;
1637 }
1638
1639 static struct obd_capa *osd_capa_get(const struct lu_env *env,
1640                                      struct dt_object *dt,
1641                                      struct lustre_capa *old,
1642                                      __u64 opc)
1643 {
1644         struct osd_thread_info  *info = osd_oti_get(env);
1645         const struct lu_fid     *fid = lu_object_fid(&dt->do_lu);
1646         struct osd_object       *obj = osd_dt_obj(dt);
1647         struct osd_device       *dev = osd_obj2dev(obj);
1648         struct lustre_capa_key  *key = &info->oti_capa_key;
1649         struct lustre_capa      *capa = &info->oti_capa;
1650         struct obd_capa         *oc;
1651         int                      rc;
1652         ENTRY;
1653
1654         if (!dev->od_fl_capa)
1655                 RETURN(ERR_PTR(-ENOENT));
1656
1657         LASSERT(dt_object_exists(dt));
1658         LASSERT(osd_invariant(obj));
1659
1660         /* renewal sanity check */
1661         if (old && osd_object_auth(env, dt, old, opc))
1662                 RETURN(ERR_PTR(-EACCES));
1663
1664         capa->lc_fid = *fid;
1665         capa->lc_opc = opc;
1666         capa->lc_uid = 0;
1667         capa->lc_flags = dev->od_capa_alg << 24;
1668         capa->lc_timeout = dev->od_capa_timeout;
1669         capa->lc_expiry = 0;
1670
1671         oc = capa_lookup(dev->od_capa_hash, capa, 1);
1672         if (oc) {
1673                 LASSERT(!capa_is_expired(oc));
1674                 RETURN(oc);
1675         }
1676
1677         spin_lock(&capa_lock);
1678         *key = dev->od_capa_keys[1];
1679         spin_unlock(&capa_lock);
1680
1681         capa->lc_keyid = key->lk_keyid;
1682         capa->lc_expiry = cfs_time_current_sec() + dev->od_capa_timeout;
1683
1684         rc = capa_hmac(capa->lc_hmac, capa, key->lk_key);
1685         if (rc) {
1686                 DEBUG_CAPA(D_ERROR, capa, "HMAC failed: %d for", rc);
1687                 LBUG();
1688                 RETURN(ERR_PTR(rc));
1689         }
1690
1691         oc = capa_add(dev->od_capa_hash, capa);
1692         RETURN(oc);
1693 }
1694
1695 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt)
1696 {
1697         struct osd_device *osd = osd_obj2dev(osd_dt_obj(dt));
1698         ENTRY;
1699
1700         /* XXX: no other option than syncing the whole filesystem until we
1701          * support ZIL */
1702         txg_wait_synced(dmu_objset_pool(osd->od_objset.os), 0ULL);
1703
1704         RETURN(0);
1705 }
1706
1707 static struct dt_object_operations osd_obj_ops = {
1708         .do_read_lock           = osd_object_read_lock,
1709         .do_write_lock          = osd_object_write_lock,
1710         .do_read_unlock         = osd_object_read_unlock,
1711         .do_write_unlock        = osd_object_write_unlock,
1712         .do_write_locked        = osd_object_write_locked,
1713         .do_attr_get            = osd_attr_get,
1714         .do_declare_attr_set    = osd_declare_attr_set,
1715         .do_attr_set            = osd_attr_set,
1716         .do_ah_init             = osd_ah_init,
1717         .do_declare_create      = osd_declare_object_create,
1718         .do_create              = osd_object_create,
1719         .do_declare_destroy     = osd_declare_object_destroy,
1720         .do_destroy             = osd_object_destroy,
1721         .do_index_try           = osd_index_try,
1722         .do_declare_ref_add     = osd_declare_object_ref_add,
1723         .do_ref_add             = osd_object_ref_add,
1724         .do_declare_ref_del     = osd_declare_object_ref_del,
1725         .do_ref_del             = osd_object_ref_del,
1726         .do_xattr_get           = osd_xattr_get,
1727         .do_declare_xattr_set   = osd_declare_xattr_set,
1728         .do_xattr_set           = osd_xattr_set,
1729         .do_declare_xattr_del   = osd_declare_xattr_del,
1730         .do_xattr_del           = osd_xattr_del,
1731         .do_xattr_list          = osd_xattr_list,
1732         .do_capa_get            = osd_capa_get,
1733         .do_object_sync         = osd_object_sync,
1734 };
1735
1736 static struct lu_object_operations osd_lu_obj_ops = {
1737         .loo_object_init        = osd_object_init,
1738         .loo_object_delete      = osd_object_delete,
1739         .loo_object_release     = osd_object_release,
1740         .loo_object_free        = osd_object_free,
1741         .loo_object_print       = osd_object_print,
1742         .loo_object_invariant   = osd_object_invariant,
1743 };
1744
1745 static int osd_otable_it_attr_get(const struct lu_env *env,
1746                                 struct dt_object *dt,
1747                                 struct lu_attr *attr,
1748                                 struct lustre_capa *capa)
1749 {
1750         attr->la_valid = 0;
1751         return 0;
1752 }
1753
1754 static struct dt_object_operations osd_obj_otable_it_ops = {
1755         .do_attr_get    = osd_otable_it_attr_get,
1756         .do_index_try   = osd_index_try,
1757 };
1758