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.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2009, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2012, 2016, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  * Lustre is a trademark of Sun Microsystems, Inc.
  31  *
  32  * lustre/osd-zfs/osd_object.c
  33  *
  34  * Author: Alex Zhuravlev <bzzz@whamcloud.com>
  35  * Author: Mike Pershin <tappro@whamcloud.com>
  36  * Author: Johann Lombardi <johann@whamcloud.com>
  37  */
  38
  39 #define DEBUG_SUBSYSTEM S_OSD
  40
  41 #include <lustre_ver.h>
  42 #include <libcfs/libcfs.h>
  43 #include <obd_support.h>
  44 #include <lustre_net.h>
  45 #include <obd.h>
  46 #include <obd_class.h>
  47 #include <lustre_disk.h>
  48 #include <lustre_fid.h>
  49
  50 #include "osd_internal.h"
  51
  52 #include <sys/dnode.h>
  53 #include <sys/dbuf.h>
  54 #include <sys/spa.h>
  55 #include <sys/stat.h>
  56 #include <sys/zap.h>
  57 #include <sys/spa_impl.h>
  58 #include <sys/zfs_znode.h>
  59 #include <sys/dmu_tx.h>
  60 #include <sys/dmu_objset.h>
  61 #include <sys/dsl_prop.h>
  62 #include <sys/sa_impl.h>
  63 #include <sys/txg.h>
  64
  65 char *osd_obj_tag = "osd_object";
  66
  67 static struct dt_object_operations osd_obj_ops;
  68 static struct lu_object_operations osd_lu_obj_ops;
  69 extern struct dt_body_operations osd_body_ops;
  70 static struct dt_object_operations osd_obj_otable_it_ops;
  71
  72 extern struct kmem_cache *osd_object_kmem;
  73
  74 static void
  75 osd_object_sa_fini(struct osd_object *obj)
  76 {
  77         if (obj->oo_sa_hdl) {
  78                 sa_handle_destroy(obj->oo_sa_hdl);
  79                 obj->oo_sa_hdl = NULL;
  80         }
  81 }
  82
  83 static int
  84 osd_object_sa_init(struct osd_object *obj, struct osd_device *o)
  85 {
  86         int rc;
  87
  88         LASSERT(obj->oo_sa_hdl == NULL);
  89         LASSERT(obj->oo_db != NULL);
  90
  91         rc = -sa_handle_get(o->od_os, obj->oo_db->db_object, obj,
  92                             SA_HDL_PRIVATE, &obj->oo_sa_hdl);
  93         if (rc)
  94                 return rc;
  95
  96         /* Cache the xattr object id, valid for the life of the object */
  97         rc = -sa_lookup(obj->oo_sa_hdl, SA_ZPL_XATTR(o), &obj->oo_xattr, 8);
  98         if (rc == -ENOENT) {
  99                 obj->oo_xattr = ZFS_NO_OBJECT;
 100                 rc = 0;
 101         } else if (rc) {
 102                 osd_object_sa_fini(obj);
 103         }
 104
 105         return rc;
 106 }
 107
 108 /*
 109  * Add object to list of dirty objects in tx handle.
 110  */
 111 static void
 112 osd_object_sa_dirty_add(struct osd_object *obj, struct osd_thandle *oh)
 113 {
 114         if (!list_empty(&obj->oo_sa_linkage))
 115                 return;
 116
 117         down(&oh->ot_sa_lock);
 118         write_lock(&obj->oo_attr_lock);
 119         if (likely(list_empty(&obj->oo_sa_linkage)))
 120                 list_add(&obj->oo_sa_linkage, &oh->ot_sa_list);
 121         write_unlock(&obj->oo_attr_lock);
 122         up(&oh->ot_sa_lock);
 123 }
 124
 125 /*
 126  * Release spill block dbuf hold for all dirty SAs.
 127  */
 128 void osd_object_sa_dirty_rele(struct osd_thandle *oh)
 129 {
 130         struct osd_object *obj;
 131
 132         down(&oh->ot_sa_lock);
 133         while (!list_empty(&oh->ot_sa_list)) {
 134                 obj = list_entry(oh->ot_sa_list.next,
 135                                  struct osd_object, oo_sa_linkage);
 136                 sa_spill_rele(obj->oo_sa_hdl);
 137                 write_lock(&obj->oo_attr_lock);
 138                 list_del_init(&obj->oo_sa_linkage);
 139                 write_unlock(&obj->oo_attr_lock);
 140         }
 141         up(&oh->ot_sa_lock);
 142 }
 143
 144 /*
 145  * Update the SA and add the object to the dirty list.
 146  */
 147 int osd_object_sa_update(struct osd_object *obj, sa_attr_type_t type,
 148                          void *buf, uint32_t buflen, struct osd_thandle *oh)
 149 {
 150         int rc;
 151
 152         LASSERT(obj->oo_sa_hdl != NULL);
 153         LASSERT(oh->ot_tx != NULL);
 154
 155         rc = -sa_update(obj->oo_sa_hdl, type, buf, buflen, oh->ot_tx);
 156         osd_object_sa_dirty_add(obj, oh);
 157
 158         return rc;
 159 }
 160
 161 /*
 162  * Bulk update the SA and add the object to the dirty list.
 163  */
 164 static int
 165 osd_object_sa_bulk_update(struct osd_object *obj, sa_bulk_attr_t *attrs,
 166                           int count, struct osd_thandle *oh)
 167 {
 168         int rc;
 169
 170         LASSERT(obj->oo_sa_hdl != NULL);
 171         LASSERT(oh->ot_tx != NULL);
 172
 173         rc = -sa_bulk_update(obj->oo_sa_hdl, attrs, count, oh->ot_tx);
 174         osd_object_sa_dirty_add(obj, oh);
 175
 176         return rc;
 177 }
 178
 179 /*
 180  * Retrieve the attributes of a DMU object
 181  */
 182 int __osd_object_attr_get(const struct lu_env *env, struct osd_device *o,
 183                           struct osd_object *obj, struct lu_attr *la)
 184 {
 185         struct osa_attr *osa = &osd_oti_get(env)->oti_osa;
 186         sa_bulk_attr_t  *bulk = osd_oti_get(env)->oti_attr_bulk;
 187         sa_handle_t     *sa_hdl;
 188         int              cnt = 0;
 189         int              rc;
 190         ENTRY;
 191
 192         LASSERT(obj->oo_db != NULL);
 193
 194         rc = -sa_handle_get(o->od_os, obj->oo_db->db_object, NULL,
 195                             SA_HDL_PRIVATE, &sa_hdl);
 196         if (rc)
 197                 RETURN(rc);
 198
 199         la->la_valid |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE | LA_TYPE |
 200                         LA_SIZE | LA_UID | LA_GID | LA_FLAGS | LA_NLINK;
 201
 202         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(o), NULL, osa->atime, 16);
 203         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(o), NULL, osa->mtime, 16);
 204         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(o), NULL, osa->ctime, 16);
 205         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(o), NULL, &osa->mode, 8);
 206         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(o), NULL, &osa->size, 8);
 207         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(o), NULL, &osa->nlink, 8);
 208         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(o), NULL, &osa->uid, 8);
 209         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(o), NULL, &osa->gid, 8);
 210         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(o), NULL, &osa->flags, 8);
 211         LASSERT(cnt <= ARRAY_SIZE(osd_oti_get(env)->oti_attr_bulk));
 212
 213         rc = -sa_bulk_lookup(sa_hdl, bulk, cnt);
 214         if (rc)
 215                 GOTO(out_sa, rc);
 216
 217         la->la_atime = osa->atime[0];
 218         la->la_mtime = osa->mtime[0];
 219         la->la_ctime = osa->ctime[0];
 220         la->la_mode = osa->mode;
 221         la->la_uid = osa->uid;
 222         la->la_gid = osa->gid;
 223         la->la_nlink = osa->nlink;
 224         la->la_flags = attrs_zfs2fs(osa->flags);
 225         la->la_size = osa->size;
 226
 227         /* Try to get extra flag from LMA. Right now, only LMAI_ORPHAN
 228          * flags is stored in LMA, and it is only for orphan directory */
 229         if (S_ISDIR(la->la_mode) && dt_object_exists(&obj->oo_dt)) {
 230                 struct osd_thread_info *info = osd_oti_get(env);
 231                 struct lustre_mdt_attrs *lma;
 232                 struct lu_buf buf;
 233
 234                 lma = (struct lustre_mdt_attrs *)info->oti_buf;
 235                 buf.lb_buf = lma;
 236                 buf.lb_len = sizeof(info->oti_buf);
 237                 rc = osd_xattr_get(env, &obj->oo_dt, &buf, XATTR_NAME_LMA);
 238                 if (rc > 0) {
 239                         rc = 0;
 240                         lma->lma_incompat = le32_to_cpu(lma->lma_incompat);
 241                         obj->oo_lma_flags =
 242                                 lma_to_lustre_flags(lma->lma_incompat);
 243
 244                 } else if (rc == -ENODATA) {
 245                         rc = 0;
 246                 }
 247         }
 248
 249         if (S_ISCHR(la->la_mode) || S_ISBLK(la->la_mode)) {
 250                 rc = -sa_lookup(sa_hdl, SA_ZPL_RDEV(o), &osa->rdev, 8);
 251                 if (rc)
 252                         GOTO(out_sa, rc);
 253                 la->la_rdev = osa->rdev;
 254                 la->la_valid |= LA_RDEV;
 255         }
 256 out_sa:
 257         sa_handle_destroy(sa_hdl);
 258
 259         RETURN(rc);
 260 }
 261
 262 int __osd_obj2dbuf(const struct lu_env *env, objset_t *os,
 263                    uint64_t oid, dmu_buf_t **dbp)
 264 {
 265         dmu_object_info_t *doi = &osd_oti_get(env)->oti_doi;
 266         int rc;
 267
 268         rc = -sa_buf_hold(os, oid, osd_obj_tag, dbp);
 269         if (rc)
 270                 return rc;
 271
 272         dmu_object_info_from_db(*dbp, doi);
 273         if (unlikely (oid != DMU_USERUSED_OBJECT &&
 274             oid != DMU_GROUPUSED_OBJECT && doi->doi_bonus_type != DMU_OT_SA)) {
 275                 sa_buf_rele(*dbp, osd_obj_tag);
 276                 *dbp = NULL;
 277                 return -EINVAL;
 278         }
 279
 280         LASSERT(*dbp);
 281         LASSERT((*dbp)->db_object == oid);
 282         LASSERT((*dbp)->db_offset == -1);
 283         LASSERT((*dbp)->db_data != NULL);
 284
 285         return 0;
 286 }
 287
 288 /*
 289  * Concurrency: no concurrent access is possible that early in object
 290  * life-cycle.
 291  */
 292 struct lu_object *osd_object_alloc(const struct lu_env *env,
 293                                    const struct lu_object_header *hdr,
 294                                    struct lu_device *d)
 295 {
 296         struct osd_object *mo;
 297
 298         OBD_SLAB_ALLOC_PTR_GFP(mo, osd_object_kmem, GFP_NOFS);
 299         if (mo != NULL) {
 300                 struct lu_object *l;
 301
 302                 l = &mo->oo_dt.do_lu;
 303                 dt_object_init(&mo->oo_dt, NULL, d);
 304                 mo->oo_dt.do_ops = &osd_obj_ops;
 305                 l->lo_ops = &osd_lu_obj_ops;
 306                 INIT_LIST_HEAD(&mo->oo_sa_linkage);
 307                 INIT_LIST_HEAD(&mo->oo_unlinked_linkage);
 308                 init_rwsem(&mo->oo_sem);
 309                 init_rwsem(&mo->oo_guard);
 310                 rwlock_init(&mo->oo_attr_lock);
 311                 mo->oo_destroy = OSD_DESTROY_NONE;
 312                 return l;
 313         } else {
 314                 return NULL;
 315         }
 316 }
 317
 318 /*
 319  * Concurrency: shouldn't matter.
 320  */
 321 int osd_object_init0(const struct lu_env *env, struct osd_object *obj)
 322 {
 323         struct osd_device       *osd = osd_obj2dev(obj);
 324         const struct lu_fid     *fid = lu_object_fid(&obj->oo_dt.do_lu);
 325         int                      rc = 0;
 326         ENTRY;
 327
 328         if (obj->oo_db == NULL)
 329                 RETURN(0);
 330
 331         /* object exist */
 332
 333         rc = osd_object_sa_init(obj, osd);
 334         if (rc)
 335                 RETURN(rc);
 336
 337         /* cache attrs in object */
 338         rc = __osd_object_attr_get(env, osd, obj, &obj->oo_attr);
 339         if (rc)
 340                 RETURN(rc);
 341
 342         if (likely(!fid_is_acct(fid)))
 343                 /* no body operations for accounting objects */
 344                 obj->oo_dt.do_body_ops = &osd_body_ops;
 345
 346         /*
 347          * initialize object before marking it existing
 348          */
 349         obj->oo_dt.do_lu.lo_header->loh_attr |= obj->oo_attr.la_mode & S_IFMT;
 350
 351         smp_mb();
 352         obj->oo_dt.do_lu.lo_header->loh_attr |= LOHA_EXISTS;
 353
 354         RETURN(0);
 355 }
 356
 357 static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
 358 {
 359         struct osd_thread_info  *info = osd_oti_get(env);
 360         struct lu_buf           buf;
 361         int                     rc;
 362         struct lustre_mdt_attrs *lma;
 363         ENTRY;
 364
 365         CLASSERT(sizeof(info->oti_buf) >= sizeof(*lma));
 366         lma = (struct lustre_mdt_attrs *)info->oti_buf;
 367         buf.lb_buf = lma;
 368         buf.lb_len = sizeof(info->oti_buf);
 369
 370         rc = osd_xattr_get(env, &obj->oo_dt, &buf, XATTR_NAME_LMA);
 371         if (rc > 0) {
 372                 rc = 0;
 373                 lustre_lma_swab(lma);
 374                 if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
 375                              CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT))) {
 376                         CWARN("%s: unsupported incompat LMA feature(s) %#x for "
 377                               "fid = "DFID"\n", osd_obj2dev(obj)->od_svname,
 378                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
 379                               PFID(lu_object_fid(&obj->oo_dt.do_lu)));
 380                         rc = -EOPNOTSUPP;
 381                 }
 382         } else if (rc == -ENODATA) {
 383                 /* haven't initialize LMA xattr */
 384                 rc = 0;
 385         }
 386
 387         RETURN(rc);
 388 }
 389
 390 /*
 391  * Concurrency: no concurrent access is possible that early in object
 392  * life-cycle.
 393  */
 394 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
 395                            const struct lu_object_conf *conf)
 396 {
 397         struct osd_object       *obj = osd_obj(l);
 398         struct osd_device       *osd = osd_obj2dev(obj);
 399         uint64_t                 oid;
 400         int                      rc;
 401         ENTRY;
 402
 403         LASSERT(osd_invariant(obj));
 404
 405         if (fid_is_otable_it(&l->lo_header->loh_fid)) {
 406                 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
 407                 l->lo_header->loh_attr |= LOHA_EXISTS;
 408                 RETURN(0);
 409         }
 410
 411         if (conf != NULL && conf->loc_flags & LOC_F_NEW)
 412                 GOTO(out, rc = 0);
 413
 414         rc = osd_fid_lookup(env, osd, lu_object_fid(l), &oid);
 415         if (rc == 0) {
 416                 LASSERT(obj->oo_db == NULL);
 417                 rc = __osd_obj2dbuf(env, osd->od_os, oid, &obj->oo_db);
 418                 /* EEXIST will be returned if object is being deleted in ZFS */
 419                 if (rc == -EEXIST) {
 420                         rc = 0;
 421                         GOTO(out, rc);
 422                 }
 423                 if (rc != 0) {
 424                         CERROR("%s: lookup "DFID"/%#llx failed: rc = %d\n",
 425                                osd->od_svname, PFID(lu_object_fid(l)), oid, rc);
 426                         GOTO(out, rc);
 427                 }
 428                 LASSERT(obj->oo_db);
 429                 rc = osd_object_init0(env, obj);
 430                 if (rc != 0)
 431                         GOTO(out, rc);
 432
 433                 rc = osd_check_lma(env, obj);
 434                 if (rc != 0)
 435                         GOTO(out, rc);
 436         } else if (rc == -ENOENT) {
 437                 rc = 0;
 438         }
 439         LASSERT(osd_invariant(obj));
 440 out:
 441         RETURN(rc);
 442 }
 443
 444 /*
 445  * Concurrency: no concurrent access is possible that late in object
 446  * life-cycle.
 447  */
 448 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
 449 {
 450         struct osd_object *obj = osd_obj(l);
 451
 452         LASSERT(osd_invariant(obj));
 453
 454         dt_object_fini(&obj->oo_dt);
 455         OBD_SLAB_FREE_PTR(obj, osd_object_kmem);
 456 }
 457
 458 static int
 459 osd_object_unlinked_add(struct osd_object *obj, struct osd_thandle *oh)
 460 {
 461         int rc = -EBUSY;
 462
 463         LASSERT(obj->oo_destroy == OSD_DESTROY_ASYNC);
 464
 465         /* the object is supposed to be exclusively locked by
 466          * the caller (osd_object_destroy()), while the transaction
 467          * (oh) is per-thread and not shared */
 468         if (likely(list_empty(&obj->oo_unlinked_linkage))) {
 469                 list_add(&obj->oo_unlinked_linkage, &oh->ot_unlinked_list);
 470                 rc = 0;
 471         }
 472
 473         return rc;
 474 }
 475
 476 /* Default to max data size covered by a level-1 indirect block */
 477 static unsigned long osd_sync_destroy_max_size =
 478         1UL << (DN_MAX_INDBLKSHIFT - SPA_BLKPTRSHIFT + SPA_MAXBLOCKSHIFT);
 479 module_param(osd_sync_destroy_max_size, ulong, 0444);
 480 MODULE_PARM_DESC(osd_sync_destroy_max_size, "Maximum object size to use synchronous destroy.");
 481
 482 static inline void
 483 osd_object_set_destroy_type(struct osd_object *obj)
 484 {
 485         /*
 486          * Lock-less OST_WRITE can race with OST_DESTROY, so set destroy type
 487          * only once and use it consistently thereafter.
 488          */
 489         down_write(&obj->oo_guard);
 490         if (obj->oo_destroy == OSD_DESTROY_NONE) {
 491                 if (obj->oo_attr.la_size <= osd_sync_destroy_max_size)
 492                         obj->oo_destroy = OSD_DESTROY_SYNC;
 493                 else /* Larger objects are destroyed asynchronously */
 494                         obj->oo_destroy = OSD_DESTROY_ASYNC;
 495         }
 496         up_write(&obj->oo_guard);
 497 }
 498
 499 static int osd_declare_object_destroy(const struct lu_env *env,
 500                                       struct dt_object *dt,
 501                                       struct thandle *th)
 502 {
 503         const struct lu_fid     *fid = lu_object_fid(&dt->do_lu);
 504         struct osd_object       *obj = osd_dt_obj(dt);
 505         struct osd_device       *osd = osd_obj2dev(obj);
 506         struct osd_thandle      *oh;
 507         int                      rc;
 508         uint64_t                 zapid;
 509         ENTRY;
 510
 511         LASSERT(th != NULL);
 512         LASSERT(dt_object_exists(dt));
 513
 514         oh = container_of0(th, struct osd_thandle, ot_super);
 515         LASSERT(oh->ot_tx != NULL);
 516
 517         /* declare that we'll remove object from fid-dnode mapping */
 518         zapid = osd_get_name_n_idx(env, osd, fid, NULL, 0);
 519         dmu_tx_hold_zap(oh->ot_tx, zapid, FALSE, NULL);
 520
 521         osd_declare_xattrs_destroy(env, obj, oh);
 522
 523         /* declare that we'll remove object from inode accounting ZAPs */
 524         dmu_tx_hold_zap(oh->ot_tx, osd->od_iusr_oid, FALSE, NULL);
 525         dmu_tx_hold_zap(oh->ot_tx, osd->od_igrp_oid, FALSE, NULL);
 526
 527         /* one less inode */
 528         rc = osd_declare_quota(env, osd, obj->oo_attr.la_uid,
 529                                obj->oo_attr.la_gid, -1, oh, false, NULL, false);
 530         if (rc)
 531                 RETURN(rc);
 532
 533         /* data to be truncated */
 534         rc = osd_declare_quota(env, osd, obj->oo_attr.la_uid,
 535                                obj->oo_attr.la_gid, 0, oh, true, NULL, false);
 536         if (rc)
 537                 RETURN(rc);
 538
 539         osd_object_set_destroy_type(obj);
 540         if (obj->oo_destroy == OSD_DESTROY_SYNC)
 541                 dmu_tx_hold_free(oh->ot_tx, obj->oo_db->db_object,
 542                                  0, DMU_OBJECT_END);
 543         else
 544                 dmu_tx_hold_zap(oh->ot_tx, osd->od_unlinkedid, TRUE, NULL);
 545
 546         /* will help to find FID->ino when this object is being
 547          * added to PENDING/ */
 548         osd_idc_find_and_init(env, osd, obj);
 549
 550         RETURN(0);
 551 }
 552
 553 static int osd_object_destroy(const struct lu_env *env,
 554                               struct dt_object *dt, struct thandle *th)
 555 {
 556         struct osd_thread_info  *info = osd_oti_get(env);
 557         char                    *buf = info->oti_str;
 558         struct osd_object       *obj = osd_dt_obj(dt);
 559         struct osd_device       *osd = osd_obj2dev(obj);
 560         const struct lu_fid     *fid = lu_object_fid(&dt->do_lu);
 561         struct osd_thandle      *oh;
 562         int                      rc;
 563         uint64_t                 oid, zapid;
 564         ENTRY;
 565
 566         down_write(&obj->oo_guard);
 567
 568         if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
 569                 GOTO(out, rc = -ENOENT);
 570
 571         LASSERT(obj->oo_db != NULL);
 572
 573         oh = container_of0(th, struct osd_thandle, ot_super);
 574         LASSERT(oh != NULL);
 575         LASSERT(oh->ot_tx != NULL);
 576
 577         /* remove obj ref from index dir (it depends) */
 578         zapid = osd_get_name_n_idx(env, osd, fid, buf, sizeof(info->oti_str));
 579         rc = -zap_remove(osd->od_os, zapid, buf, oh->ot_tx);
 580         if (rc) {
 581                 CERROR("%s: zap_remove(%s) failed: rc = %d\n",
 582                        osd->od_svname, buf, rc);
 583                 GOTO(out, rc);
 584         }
 585
 586         rc = osd_xattrs_destroy(env, obj, oh);
 587         if (rc) {
 588                 CERROR("%s: cannot destroy xattrs for %s: rc = %d\n",
 589                        osd->od_svname, buf, rc);
 590                 GOTO(out, rc);
 591         }
 592
 593         /* Remove object from inode accounting. It is not fatal for the destroy
 594          * operation if something goes wrong while updating accounting, but we
 595          * still log an error message to notify the administrator */
 596         rc = -zap_increment_int(osd->od_os, osd->od_iusr_oid,
 597                                 obj->oo_attr.la_uid, -1, oh->ot_tx);
 598         if (rc)
 599                 CERROR("%s: failed to remove "DFID" from accounting ZAP for usr"
 600                        " %d: rc = %d\n", osd->od_svname, PFID(fid),
 601                        obj->oo_attr.la_uid, rc);
 602         rc = -zap_increment_int(osd->od_os, osd->od_igrp_oid,
 603                                 obj->oo_attr.la_gid, -1, oh->ot_tx);
 604         if (rc)
 605                 CERROR("%s: failed to remove "DFID" from accounting ZAP for grp"
 606                        " %d: rc = %d\n", osd->od_svname, PFID(fid),
 607                        obj->oo_attr.la_gid, rc);
 608
 609         oid = obj->oo_db->db_object;
 610         if (unlikely(obj->oo_destroy == OSD_DESTROY_NONE)) {
 611                 /* this may happen if the destroy wasn't declared
 612                  * e.g. when the object is created and then destroyed
 613                  * in the same transaction - we don't need additional
 614                  * space for destroy specifically */
 615                 LASSERT(obj->oo_attr.la_size <= osd_sync_destroy_max_size);
 616                 rc = -dmu_object_free(osd->od_os, oid, oh->ot_tx);
 617                 if (rc)
 618                         CERROR("%s: failed to free %s %llu: rc = %d\n",
 619                                osd->od_svname, buf, oid, rc);
 620         } else if (obj->oo_destroy == OSD_DESTROY_SYNC) {
 621                 rc = -dmu_object_free(osd->od_os, oid, oh->ot_tx);
 622                 if (rc)
 623                         CERROR("%s: failed to free %s %llu: rc = %d\n",
 624                                osd->od_svname, buf, oid, rc);
 625         } else { /* asynchronous destroy */
 626                 rc = osd_object_unlinked_add(obj, oh);
 627                 if (rc)
 628                         GOTO(out, rc);
 629
 630                 rc = -zap_add_int(osd->od_os, osd->od_unlinkedid,
 631                                   oid, oh->ot_tx);
 632                 if (rc)
 633                         CERROR("%s: zap_add_int() failed %s %llu: rc = %d\n",
 634                                osd->od_svname, buf, oid, rc);
 635         }
 636
 637 out:
 638         /* not needed in the cache anymore */
 639         set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
 640         if (rc == 0)
 641                 obj->oo_destroyed = 1;
 642         up_write(&obj->oo_guard);
 643         RETURN (0);
 644 }
 645
 646 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
 647 {
 648         struct osd_object *obj = osd_obj(l);
 649
 650         if (obj->oo_db != NULL) {
 651                 osd_object_sa_fini(obj);
 652                 if (obj->oo_sa_xattr) {
 653                         nvlist_free(obj->oo_sa_xattr);
 654                         obj->oo_sa_xattr = NULL;
 655                 }
 656                 sa_buf_rele(obj->oo_db, osd_obj_tag);
 657                 list_del(&obj->oo_sa_linkage);
 658                 obj->oo_db = NULL;
 659         }
 660 }
 661
 662 /*
 663  * Concurrency: ->loo_object_release() is called under site spin-lock.
 664  */
 665 static void osd_object_release(const struct lu_env *env,
 666                                struct lu_object *l)
 667 {
 668 }
 669
 670 /*
 671  * Concurrency: shouldn't matter.
 672  */
 673 static int osd_object_print(const struct lu_env *env, void *cookie,
 674                             lu_printer_t p, const struct lu_object *l)
 675 {
 676         struct osd_object *o = osd_obj(l);
 677
 678         return (*p)(env, cookie, LUSTRE_OSD_ZFS_NAME"-object@%p", o);
 679 }
 680
 681 static void osd_object_read_lock(const struct lu_env *env,
 682                                  struct dt_object *dt, unsigned role)
 683 {
 684         struct osd_object *obj = osd_dt_obj(dt);
 685
 686         LASSERT(osd_invariant(obj));
 687
 688         down_read_nested(&obj->oo_sem, role);
 689 }
 690
 691 static void osd_object_write_lock(const struct lu_env *env,
 692                                   struct dt_object *dt, unsigned role)
 693 {
 694         struct osd_object *obj = osd_dt_obj(dt);
 695
 696         LASSERT(osd_invariant(obj));
 697
 698         down_write_nested(&obj->oo_sem, role);
 699 }
 700
 701 static void osd_object_read_unlock(const struct lu_env *env,
 702                                    struct dt_object *dt)
 703 {
 704         struct osd_object *obj = osd_dt_obj(dt);
 705
 706         LASSERT(osd_invariant(obj));
 707         up_read(&obj->oo_sem);
 708 }
 709
 710 static void osd_object_write_unlock(const struct lu_env *env,
 711                                     struct dt_object *dt)
 712 {
 713         struct osd_object *obj = osd_dt_obj(dt);
 714
 715         LASSERT(osd_invariant(obj));
 716         up_write(&obj->oo_sem);
 717 }
 718
 719 static int osd_object_write_locked(const struct lu_env *env,
 720                                    struct dt_object *dt)
 721 {
 722         struct osd_object *obj = osd_dt_obj(dt);
 723         int rc = 1;
 724
 725         LASSERT(osd_invariant(obj));
 726
 727         if (down_write_trylock(&obj->oo_sem)) {
 728                 rc = 0;
 729                 up_write(&obj->oo_sem);
 730         }
 731         return rc;
 732 }
 733
 734 static int osd_attr_get(const struct lu_env *env,
 735                         struct dt_object *dt,
 736                         struct lu_attr *attr)
 737 {
 738         struct osd_object       *obj = osd_dt_obj(dt);
 739         uint64_t                 blocks;
 740         uint32_t                 blksize;
 741         int                      rc = 0;
 742
 743         down_read(&obj->oo_guard);
 744
 745         if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
 746                 GOTO(out, rc = -ENOENT);
 747
 748         LASSERT(osd_invariant(obj));
 749         LASSERT(obj->oo_db);
 750
 751         read_lock(&obj->oo_attr_lock);
 752         *attr = obj->oo_attr;
 753         if (obj->oo_lma_flags & LUSTRE_ORPHAN_FL)
 754                 attr->la_flags |= LUSTRE_ORPHAN_FL;
 755         read_unlock(&obj->oo_attr_lock);
 756
 757         /* with ZFS_DEBUG zrl_add_debug() called by DB_DNODE_ENTER()
 758          * from within sa_object_size() can block on a mutex, so
 759          * we can't call sa_object_size() holding rwlock */
 760         sa_object_size(obj->oo_sa_hdl, &blksize, &blocks);
 761         /* we do not control size of indices, so always calculate
 762          * it from number of blocks reported by DMU */
 763         if (S_ISDIR(attr->la_mode))
 764                 attr->la_size = 512 * blocks;
 765         /* Block size may be not set; suggest maximal I/O transfers. */
 766         if (blksize == 0)
 767                 blksize = osd_spa_maxblocksize(
 768                         dmu_objset_spa(osd_obj2dev(obj)->od_os));
 769
 770         attr->la_blksize = blksize;
 771         attr->la_blocks = blocks;
 772         attr->la_valid |= LA_BLOCKS | LA_BLKSIZE;
 773
 774 out:
 775         up_read(&obj->oo_guard);
 776         return rc;
 777 }
 778
 779 /* Simple wrapper on top of qsd API which implement quota transfer for osd
 780  * setattr needs. As a reminder, only the root user can change ownership of
 781  * a file, that's why EDQUOT & EINPROGRESS errors are discarded */
 782 static inline int qsd_transfer(const struct lu_env *env,
 783                                struct qsd_instance *qsd,
 784                                struct lquota_trans *trans, int qtype,
 785                                __u64 orig_id, __u64 new_id, __u64 bspace,
 786                                struct lquota_id_info *qi)
 787 {
 788         int     rc;
 789
 790         if (unlikely(qsd == NULL))
 791                 return 0;
 792
 793         LASSERT(qtype >= 0 && qtype < LL_MAXQUOTAS);
 794         qi->lqi_type = qtype;
 795
 796         /* inode accounting */
 797         qi->lqi_is_blk = false;
 798
 799         /* one more inode for the new owner ... */
 800         qi->lqi_id.qid_uid = new_id;
 801         qi->lqi_space      = 1;
 802         rc = qsd_op_begin(env, qsd, trans, qi, NULL);
 803         if (rc == -EDQUOT || rc == -EINPROGRESS)
 804                 rc = 0;
 805         if (rc)
 806                 return rc;
 807
 808         /* and one less inode for the current id */
 809         qi->lqi_id.qid_uid = orig_id;;
 810         qi->lqi_space      = -1;
 811         /* can't get EDQUOT when reducing usage */
 812         rc = qsd_op_begin(env, qsd, trans, qi, NULL);
 813         if (rc == -EINPROGRESS)
 814                 rc = 0;
 815         if (rc)
 816                 return rc;
 817
 818         /* block accounting */
 819         qi->lqi_is_blk = true;
 820
 821         /* more blocks for the new owner ... */
 822         qi->lqi_id.qid_uid = new_id;
 823         qi->lqi_space      = bspace;
 824         rc = qsd_op_begin(env, qsd, trans, qi, NULL);
 825         if (rc == -EDQUOT || rc == -EINPROGRESS)
 826                 rc = 0;
 827         if (rc)
 828                 return rc;
 829
 830         /* and finally less blocks for the current owner */
 831         qi->lqi_id.qid_uid = orig_id;
 832         qi->lqi_space      = -bspace;
 833         rc = qsd_op_begin(env, qsd, trans, qi, NULL);
 834         /* can't get EDQUOT when reducing usage */
 835         if (rc == -EINPROGRESS)
 836                 rc = 0;
 837         return rc;
 838 }
 839
 840 static int osd_declare_attr_set(const struct lu_env *env,
 841                                 struct dt_object *dt,
 842                                 const struct lu_attr *attr,
 843                                 struct thandle *handle)
 844 {
 845         struct osd_thread_info  *info = osd_oti_get(env);
 846         struct osd_object       *obj = osd_dt_obj(dt);
 847         struct osd_device       *osd = osd_obj2dev(obj);
 848         dmu_tx_hold_t           *txh;
 849         struct osd_thandle      *oh;
 850         uint64_t                 bspace;
 851         uint32_t                 blksize;
 852         int                      rc = 0;
 853         bool                     found;
 854         ENTRY;
 855
 856
 857         LASSERT(handle != NULL);
 858         LASSERT(osd_invariant(obj));
 859
 860         oh = container_of0(handle, struct osd_thandle, ot_super);
 861
 862         down_read(&obj->oo_guard);
 863         if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
 864                 GOTO(out, rc = 0);
 865
 866         LASSERT(obj->oo_sa_hdl != NULL);
 867         LASSERT(oh->ot_tx != NULL);
 868         /* regular attributes are part of the bonus buffer */
 869         /* let's check whether this object is already part of
 870          * transaction.. */
 871         found = false;
 872         for (txh = list_head(&oh->ot_tx->tx_holds); txh;
 873              txh = list_next(&oh->ot_tx->tx_holds, txh)) {
 874                 if (txh->txh_dnode == NULL)
 875                         continue;
 876                 if (txh->txh_dnode->dn_object != obj->oo_db->db_object)
 877                         continue;
 878                 /* this object is part of the transaction already
 879                  * we don't need to declare bonus again */
 880                 found = true;
 881                 break;
 882         }
 883         if (!found)
 884                 dmu_tx_hold_bonus(oh->ot_tx, obj->oo_db->db_object);
 885         if (oh->ot_tx->tx_err != 0)
 886                 GOTO(out, rc = -oh->ot_tx->tx_err);
 887
 888         if (attr && attr->la_valid & LA_FLAGS) {
 889                 /* LMA is usually a part of bonus, no need to declare
 890                  * anything else */
 891         }
 892
 893         if (attr && (attr->la_valid & (LA_UID | LA_GID))) {
 894                 sa_object_size(obj->oo_sa_hdl, &blksize, &bspace);
 895                 bspace = toqb(bspace * blksize);
 896         }
 897
 898         if (attr && attr->la_valid & LA_UID) {
 899                 /* account for user inode tracking ZAP update */
 900                 dmu_tx_hold_zap(oh->ot_tx, osd->od_iusr_oid, FALSE, NULL);
 901
 902                 /* quota enforcement for user */
 903                 if (attr->la_uid != obj->oo_attr.la_uid) {
 904                         rc = qsd_transfer(env, osd->od_quota_slave,
 905                                           &oh->ot_quota_trans, USRQUOTA,
 906                                           obj->oo_attr.la_uid, attr->la_uid,
 907                                           bspace, &info->oti_qi);
 908                         if (rc)
 909                                 GOTO(out, rc);
 910                 }
 911         }
 912         if (attr && attr->la_valid & LA_GID) {
 913                 /* account for user inode tracking ZAP update */
 914                 dmu_tx_hold_zap(oh->ot_tx, osd->od_igrp_oid, FALSE, NULL);
 915
 916                 /* quota enforcement for group */
 917                 if (attr->la_gid != obj->oo_attr.la_gid) {
 918                         rc = qsd_transfer(env, osd->od_quota_slave,
 919                                           &oh->ot_quota_trans, GRPQUOTA,
 920                                           obj->oo_attr.la_gid, attr->la_gid,
 921                                           bspace, &info->oti_qi);
 922                         if (rc)
 923                                 GOTO(out, rc);
 924                 }
 925         }
 926
 927 out:
 928         up_read(&obj->oo_guard);
 929         RETURN(rc);
 930 }
 931
 932 /*
 933  * Set the attributes of an object
 934  *
 935  * The transaction passed to this routine must have
 936  * dmu_tx_hold_bonus(tx, oid) called and then assigned
 937  * to a transaction group.
 938  */
 939 static int osd_attr_set(const struct lu_env *env, struct dt_object *dt,
 940                         const struct lu_attr *la, struct thandle *handle)
 941 {
 942         struct osd_thread_info  *info = osd_oti_get(env);
 943         sa_bulk_attr_t          *bulk = osd_oti_get(env)->oti_attr_bulk;
 944         struct osd_object       *obj = osd_dt_obj(dt);
 945         struct osd_device       *osd = osd_obj2dev(obj);
 946         struct osd_thandle      *oh;
 947         struct osa_attr         *osa = &info->oti_osa;
 948         __u64                    valid = la->la_valid;
 949         int                      cnt;
 950         int                      rc = 0;
 951
 952         ENTRY;
 953
 954         down_read(&obj->oo_guard);
 955         if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
 956                 GOTO(out, rc = -ENOENT);
 957
 958         LASSERT(handle != NULL);
 959         LASSERT(osd_invariant(obj));
 960         LASSERT(obj->oo_sa_hdl);
 961
 962         oh = container_of0(handle, struct osd_thandle, ot_super);
 963         /* Assert that the transaction has been assigned to a
 964            transaction group. */
 965         LASSERT(oh->ot_tx->tx_txg != 0);
 966
 967         /* Only allow set size for regular file */
 968         if (!S_ISREG(dt->do_lu.lo_header->loh_attr))
 969                 valid &= ~(LA_SIZE | LA_BLOCKS);
 970
 971         if (valid & LA_CTIME && la->la_ctime == obj->oo_attr.la_ctime)
 972                 valid &= ~LA_CTIME;
 973
 974         if (valid & LA_MTIME && la->la_mtime == obj->oo_attr.la_mtime)
 975                 valid &= ~LA_MTIME;
 976
 977         if (valid & LA_ATIME && la->la_atime == obj->oo_attr.la_atime)
 978                 valid &= ~LA_ATIME;
 979
 980         if (valid == 0)
 981                 GOTO(out, rc = 0);
 982
 983         if (valid & LA_FLAGS) {
 984                 struct lustre_mdt_attrs *lma;
 985                 struct lu_buf buf;
 986
 987                 if (la->la_flags & LUSTRE_LMA_FL_MASKS) {
 988                         CLASSERT(sizeof(info->oti_buf) >= sizeof(*lma));
 989                         lma = (struct lustre_mdt_attrs *)&info->oti_buf;
 990                         buf.lb_buf = lma;
 991                         buf.lb_len = sizeof(info->oti_buf);
 992                         rc = osd_xattr_get(env, &obj->oo_dt, &buf,
 993                                            XATTR_NAME_LMA);
 994                         if (rc > 0) {
 995                                 lma->lma_incompat =
 996                                         le32_to_cpu(lma->lma_incompat);
 997                                 lma->lma_incompat |=
 998                                         lustre_to_lma_flags(la->la_flags);
 999                                 lma->lma_incompat =
1000                                         cpu_to_le32(lma->lma_incompat);
1001                                 buf.lb_buf = lma;
1002                                 buf.lb_len = sizeof(*lma);
1003                                 rc = osd_xattr_set_internal(env, obj, &buf,
1004                                                             XATTR_NAME_LMA,
1005                                                             LU_XATTR_REPLACE,
1006                                                             oh);
1007                         }
1008                         if (rc < 0) {
1009                                 CWARN("%s: failed to set LMA flags: rc = %d\n",
1010                                        osd->od_svname, rc);
1011                                 RETURN(rc);
1012                         }
1013                 }
1014         }
1015
1016         /* do both accounting updates outside oo_attr_lock below */
1017         if ((valid & LA_UID) && (la->la_uid != obj->oo_attr.la_uid)) {
1018                 /* Update user accounting. Failure isn't fatal, but we still
1019                  * log an error message */
1020                 rc = -zap_increment_int(osd->od_os, osd->od_iusr_oid,
1021                                         la->la_uid, 1, oh->ot_tx);
1022                 if (rc)
1023                         CERROR("%s: failed to update accounting ZAP for user "
1024                                 "%d (%d)\n", osd->od_svname, la->la_uid, rc);
1025                 rc = -zap_increment_int(osd->od_os, osd->od_iusr_oid,
1026                                         obj->oo_attr.la_uid, -1, oh->ot_tx);
1027                 if (rc)
1028                         CERROR("%s: failed to update accounting ZAP for user "
1029                                 "%d (%d)\n", osd->od_svname,
1030                                 obj->oo_attr.la_uid, rc);
1031         }
1032         if ((valid & LA_GID) && (la->la_gid != obj->oo_attr.la_gid)) {
1033                 /* Update group accounting. Failure isn't fatal, but we still
1034                  * log an error message */
1035                 rc = -zap_increment_int(osd->od_os, osd->od_igrp_oid,
1036                                         la->la_gid, 1, oh->ot_tx);
1037                 if (rc)
1038                         CERROR("%s: failed to update accounting ZAP for user "
1039                                 "%d (%d)\n", osd->od_svname, la->la_gid, rc);
1040                 rc = -zap_increment_int(osd->od_os, osd->od_igrp_oid,
1041                                         obj->oo_attr.la_gid, -1, oh->ot_tx);
1042                 if (rc)
1043                         CERROR("%s: failed to update accounting ZAP for user "
1044                                 "%d (%d)\n", osd->od_svname,
1045                                 obj->oo_attr.la_gid, rc);
1046         }
1047
1048         write_lock(&obj->oo_attr_lock);
1049         cnt = 0;
1050         if (valid & LA_ATIME) {
1051                 osa->atime[0] = obj->oo_attr.la_atime = la->la_atime;
1052                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(osd), NULL,
1053                                  osa->atime, 16);
1054         }
1055         if (valid & LA_MTIME) {
1056                 osa->mtime[0] = obj->oo_attr.la_mtime = la->la_mtime;
1057                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(osd), NULL,
1058                                  osa->mtime, 16);
1059         }
1060         if (valid & LA_CTIME) {
1061                 osa->ctime[0] = obj->oo_attr.la_ctime = la->la_ctime;
1062                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(osd), NULL,
1063                                  osa->ctime, 16);
1064         }
1065         if (valid & LA_MODE) {
1066                 /* mode is stored along with type, so read it first */
1067                 obj->oo_attr.la_mode = (obj->oo_attr.la_mode & S_IFMT) |
1068                         (la->la_mode & ~S_IFMT);
1069                 osa->mode = obj->oo_attr.la_mode;
1070                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(osd), NULL,
1071                                  &osa->mode, 8);
1072         }
1073         if (valid & LA_SIZE) {
1074                 osa->size = obj->oo_attr.la_size = la->la_size;
1075                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(osd), NULL,
1076                                  &osa->size, 8);
1077         }
1078         if (valid & LA_NLINK) {
1079                 osa->nlink = obj->oo_attr.la_nlink = la->la_nlink;
1080                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(osd), NULL,
1081                                  &osa->nlink, 8);
1082         }
1083         if (valid & LA_RDEV) {
1084                 osa->rdev = obj->oo_attr.la_rdev = la->la_rdev;
1085                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_RDEV(osd), NULL,
1086                                  &osa->rdev, 8);
1087         }
1088         if (valid & LA_FLAGS) {
1089                 osa->flags = attrs_fs2zfs(la->la_flags);
1090                 /* many flags are not supported by zfs, so ensure a good cached
1091                  * copy */
1092                 obj->oo_attr.la_flags = attrs_zfs2fs(osa->flags);
1093                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(osd), NULL,
1094                                  &osa->flags, 8);
1095         }
1096         if (valid & LA_UID) {
1097                 osa->uid = obj->oo_attr.la_uid = la->la_uid;
1098                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(osd), NULL,
1099                                  &osa->uid, 8);
1100         }
1101         if (valid & LA_GID) {
1102                 osa->gid = obj->oo_attr.la_gid = la->la_gid;
1103                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(osd), NULL,
1104                                  &osa->gid, 8);
1105         }
1106         obj->oo_attr.la_valid |= valid;
1107         write_unlock(&obj->oo_attr_lock);
1108
1109         LASSERT(cnt <= ARRAY_SIZE(osd_oti_get(env)->oti_attr_bulk));
1110         rc = osd_object_sa_bulk_update(obj, bulk, cnt, oh);
1111
1112 out:
1113         up_read(&obj->oo_guard);
1114         RETURN(rc);
1115 }
1116
1117 /*
1118  * Object creation.
1119  *
1120  * XXX temporary solution.
1121  */
1122
1123 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
1124                         struct dt_object *parent, struct dt_object *child,
1125                         umode_t child_mode)
1126 {
1127         LASSERT(ah);
1128
1129         ah->dah_parent = parent;
1130         ah->dah_mode = child_mode;
1131
1132         if (parent != NULL && !dt_object_remote(parent)) {
1133                 /* will help to find FID->ino at dt_insert("..") */
1134                 struct osd_object *pobj = osd_dt_obj(parent);
1135
1136                 osd_idc_find_and_init(env, osd_obj2dev(pobj), pobj);
1137         }
1138 }
1139
1140 static int osd_declare_object_create(const struct lu_env *env,
1141                                      struct dt_object *dt,
1142                                      struct lu_attr *attr,
1143                                      struct dt_allocation_hint *hint,
1144                                      struct dt_object_format *dof,
1145                                      struct thandle *handle)
1146 {
1147         const struct lu_fid     *fid = lu_object_fid(&dt->do_lu);
1148         struct osd_object       *obj = osd_dt_obj(dt);
1149         struct osd_device       *osd = osd_obj2dev(obj);
1150         struct osd_thandle      *oh;
1151         uint64_t                 zapid;
1152         int                      rc, dnode_size;
1153         ENTRY;
1154
1155         LASSERT(dof);
1156
1157         switch (dof->dof_type) {
1158                 case DFT_REGULAR:
1159                 case DFT_SYM:
1160                 case DFT_NODE:
1161                         if (obj->oo_dt.do_body_ops == NULL)
1162                                 obj->oo_dt.do_body_ops = &osd_body_ops;
1163                         break;
1164                 default:
1165                         break;
1166         }
1167
1168         LASSERT(handle != NULL);
1169         oh = container_of0(handle, struct osd_thandle, ot_super);
1170         LASSERT(oh->ot_tx != NULL);
1171
1172         /* this is the minimum set of EAs on every Lustre object */
1173         obj->oo_ea_in_bonus = ZFS_SA_BASE_ATTR_SIZE +
1174                                 sizeof(__u64) + /* VBR VERSION */
1175                                 sizeof(struct lustre_mdt_attrs); /* LMA */
1176         /* reserve 32 bytes for extra stuff like ACLs */
1177         dnode_size = size_roundup_power2(obj->oo_ea_in_bonus + 32);
1178
1179         switch (dof->dof_type) {
1180                 case DFT_DIR:
1181                         dt->do_index_ops = &osd_dir_ops;
1182                 case DFT_INDEX:
1183                         /* for zap create */
1184                         dmu_tx_hold_zap(oh->ot_tx, DMU_NEW_OBJECT, FALSE, NULL);
1185                         dmu_tx_hold_sa_create(oh->ot_tx, dnode_size);
1186                         break;
1187                 case DFT_REGULAR:
1188                 case DFT_SYM:
1189                 case DFT_NODE:
1190                         /* first, we'll create new object */
1191                         dmu_tx_hold_sa_create(oh->ot_tx, dnode_size);
1192                         break;
1193
1194                 default:
1195                         LBUG();
1196                         break;
1197         }
1198
1199         /* and we'll add it to some mapping */
1200         zapid = osd_get_name_n_idx(env, osd, fid, NULL, 0);
1201         dmu_tx_hold_zap(oh->ot_tx, zapid, TRUE, NULL);
1202
1203         /* we will also update inode accounting ZAPs */
1204         dmu_tx_hold_zap(oh->ot_tx, osd->od_iusr_oid, FALSE, NULL);
1205         dmu_tx_hold_zap(oh->ot_tx, osd->od_igrp_oid, FALSE, NULL);
1206
1207         /* will help to find FID->ino mapping at dt_insert() */
1208         osd_idc_find_and_init(env, osd, obj);
1209
1210         rc = osd_declare_quota(env, osd, attr->la_uid, attr->la_gid, 1, oh,
1211                                false, NULL, false);
1212
1213         RETURN(rc);
1214 }
1215
1216 int __osd_attr_init(const struct lu_env *env, struct osd_device *osd,
1217                     sa_handle_t *sa_hdl, dmu_tx_t *tx,
1218                     struct lu_attr *la, uint64_t parent)
1219 {
1220         sa_bulk_attr_t  *bulk = osd_oti_get(env)->oti_attr_bulk;
1221         struct osa_attr *osa = &osd_oti_get(env)->oti_osa;
1222         uint64_t         gen;
1223         uint64_t         crtime[2];
1224         timestruc_t      now;
1225         int              cnt;
1226         int              rc;
1227
1228         LASSERT(sa_hdl);
1229
1230         gen = dmu_tx_get_txg(tx);
1231         gethrestime(&now);
1232         ZFS_TIME_ENCODE(&now, crtime);
1233
1234         osa->atime[0] = la->la_atime;
1235         osa->ctime[0] = la->la_ctime;
1236         osa->mtime[0] = la->la_mtime;
1237         osa->mode = la->la_mode;
1238         osa->uid = la->la_uid;
1239         osa->gid = la->la_gid;
1240         osa->rdev = la->la_rdev;
1241         osa->nlink = la->la_nlink;
1242         osa->flags = attrs_fs2zfs(la->la_flags);
1243         osa->size  = la->la_size;
1244
1245         /*
1246          * we need to create all SA below upon object create.
1247          *
1248          * XXX The attribute order matters since the accounting callback relies
1249          * on static offsets (i.e. SA_*_OFFSET, see zfs_space_delta_cb()) to
1250          * look up the UID/GID attributes. Moreover, the callback does not seem
1251          * to support the spill block.
1252          * We define attributes in the same order as SA_*_OFFSET in order to
1253          * work around the problem. See ORI-610.
1254          */
1255         cnt = 0;
1256         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(osd), NULL, &osa->mode, 8);
1257         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(osd), NULL, &osa->size, 8);
1258         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GEN(osd), NULL, &gen, 8);
1259         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(osd), NULL, &osa->uid, 8);
1260         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(osd), NULL, &osa->gid, 8);
1261         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_PARENT(osd), NULL, &parent, 8);
1262         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(osd), NULL, &osa->flags, 8);
1263         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(osd), NULL, osa->atime, 16);
1264         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(osd), NULL, osa->mtime, 16);
1265         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(osd), NULL, osa->ctime, 16);
1266         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CRTIME(osd), NULL, crtime, 16);
1267         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(osd), NULL, &osa->nlink, 8);
1268         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_RDEV(osd), NULL, &osa->rdev, 8);
1269         LASSERT(cnt <= ARRAY_SIZE(osd_oti_get(env)->oti_attr_bulk));
1270
1271         rc = -sa_replace_all_by_template(sa_hdl, bulk, cnt, tx);
1272
1273         return rc;
1274 }
1275
1276 /*
1277  * The transaction passed to this routine must have
1278  * dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT) called and then assigned
1279  * to a transaction group.
1280  */
1281 int __osd_object_create(const struct lu_env *env, struct osd_object *obj,
1282                         dmu_buf_t **dbp, dmu_tx_t *tx, struct lu_attr *la)
1283 {
1284         uint64_t             oid;
1285         int                  rc;
1286         struct osd_device   *osd = osd_obj2dev(obj);
1287         const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
1288         dmu_object_type_t    type = DMU_OT_PLAIN_FILE_CONTENTS;
1289
1290         /* Use DMU_OTN_UINT8_METADATA for local objects so their data blocks
1291          * would get an additional ditto copy */
1292         if (unlikely(S_ISREG(la->la_mode) &&
1293                      fid_seq_is_local_file(fid_seq(fid))))
1294                 type = DMU_OTN_UINT8_METADATA;
1295
1296         /* Create a new DMU object using the default dnode size. */
1297         oid = osd_dmu_object_alloc(osd->od_os, type, 0, 0, tx);
1298         rc = -sa_buf_hold(osd->od_os, oid, osd_obj_tag, dbp);
1299         LASSERTF(rc == 0, "sa_buf_hold %llu failed: %d\n", oid, rc);
1300
1301         LASSERT(la->la_valid & LA_MODE);
1302         la->la_size = 0;
1303         la->la_nlink = 1;
1304
1305         return 0;
1306 }
1307
1308 /*
1309  * The transaction passed to this routine must have
1310  * dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, ...) called and then assigned
1311  * to a transaction group.
1312  *
1313  * Using ZAP_FLAG_HASH64 will force the ZAP to always be a FAT ZAP.
1314  * This is fine for directories today, because storing the FID in the dirent
1315  * will also require a FAT ZAP.  If there is a new type of micro ZAP created
1316  * then we might need to re-evaluate the use of this flag and instead do
1317  * a conversion from the different internal ZAP hash formats being used. */
1318 int __osd_zap_create(const struct lu_env *env, struct osd_device *osd,
1319                      dmu_buf_t **zap_dbp, dmu_tx_t *tx,
1320                      struct lu_attr *la, zap_flags_t flags)
1321 {
1322         uint64_t oid;
1323         int      rc;
1324
1325         /* Assert that the transaction has been assigned to a
1326            transaction group. */
1327         LASSERT(tx->tx_txg != 0);
1328
1329         oid = osd_zap_create_flags(osd->od_os, 0, flags | ZAP_FLAG_HASH64,
1330                                    DMU_OT_DIRECTORY_CONTENTS,
1331                                    14, /* == ZFS fzap_default_blockshift */
1332                                    DN_MAX_INDBLKSHIFT, /* indirect blockshift */
1333                                    0, tx);
1334
1335         rc = -sa_buf_hold(osd->od_os, oid, osd_obj_tag, zap_dbp);
1336         if (rc)
1337                 return rc;
1338
1339         la->la_size = 2;
1340         la->la_nlink = 1;
1341
1342         return 0;
1343 }
1344
1345 static dmu_buf_t *osd_mkidx(const struct lu_env *env, struct osd_object *obj,
1346                             struct lu_attr *la, struct osd_thandle *oh)
1347 {
1348         dmu_buf_t *db;
1349         int        rc;
1350
1351         /* Index file should be created as regular file in order not to confuse
1352          * ZPL which could interpret them as directory.
1353          * We set ZAP_FLAG_UINT64_KEY to let ZFS know than we are going to use
1354          * binary keys */
1355         LASSERT(S_ISREG(la->la_mode));
1356         rc = __osd_zap_create(env, osd_obj2dev(obj), &db, oh->ot_tx, la,
1357                               ZAP_FLAG_UINT64_KEY);
1358         if (rc)
1359                 return ERR_PTR(rc);
1360         return db;
1361 }
1362
1363 static dmu_buf_t *osd_mkdir(const struct lu_env *env, struct osd_object *obj,
1364                             struct lu_attr *la, struct osd_thandle *oh)
1365 {
1366         dmu_buf_t *db;
1367         int        rc;
1368
1369         LASSERT(S_ISDIR(la->la_mode));
1370         rc = __osd_zap_create(env, osd_obj2dev(obj), &db, oh->ot_tx, la, 0);
1371         if (rc)
1372                 return ERR_PTR(rc);
1373         return db;
1374 }
1375
1376 static dmu_buf_t *osd_mkreg(const struct lu_env *env, struct osd_object *obj,
1377                             struct lu_attr *la, struct osd_thandle *oh)
1378 {
1379         const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
1380         dmu_buf_t           *db;
1381         int                  rc;
1382         struct osd_device *osd = osd_obj2dev(obj);
1383
1384         LASSERT(S_ISREG(la->la_mode));
1385         rc = __osd_object_create(env, obj, &db, oh->ot_tx, la);
1386         if (rc)
1387                 return ERR_PTR(rc);
1388
1389         if ((fid_is_idif(fid) || fid_is_norm(fid) || fid_is_echo(fid)) &&
1390             osd->od_is_ost) {
1391                 /* The minimum block size must be at least page size otherwise
1392                  * it will break the assumption in tgt_thread_big_cache where
1393                  * the array size is PTLRPC_MAX_BRW_PAGES. It will also affect
1394                  * RDMA due to subpage transfer size */
1395                 rc = -dmu_object_set_blocksize(osd->od_os, db->db_object,
1396                                                PAGE_SIZE, 0, oh->ot_tx);
1397                 if (unlikely(rc)) {
1398                         CERROR("%s: can't change blocksize: %d\n",
1399                                osd->od_svname, rc);
1400                         return ERR_PTR(rc);
1401                 }
1402         }
1403
1404         return db;
1405 }
1406
1407 static dmu_buf_t *osd_mksym(const struct lu_env *env, struct osd_object *obj,
1408                             struct lu_attr *la, struct osd_thandle *oh)
1409 {
1410         dmu_buf_t *db;
1411         int        rc;
1412
1413         LASSERT(S_ISLNK(la->la_mode));
1414         rc = __osd_object_create(env, obj, &db, oh->ot_tx, la);
1415         if (rc)
1416                 return ERR_PTR(rc);
1417         return db;
1418 }
1419
1420 static dmu_buf_t *osd_mknod(const struct lu_env *env, struct osd_object *obj,
1421                             struct lu_attr *la, struct osd_thandle *oh)
1422 {
1423         dmu_buf_t *db;
1424         int        rc;
1425
1426         if (S_ISCHR(la->la_mode) || S_ISBLK(la->la_mode))
1427                 la->la_valid |= LA_RDEV;
1428
1429         rc = __osd_object_create(env, obj, &db, oh->ot_tx, la);
1430         if (rc)
1431                 return ERR_PTR(rc);
1432         return db;
1433 }
1434
1435 typedef dmu_buf_t *(*osd_obj_type_f)(const struct lu_env *env,
1436                                      struct osd_object *obj,
1437                                      struct lu_attr *la,
1438                                      struct osd_thandle *oh);
1439
1440 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
1441 {
1442         osd_obj_type_f result;
1443
1444         switch (type) {
1445         case DFT_DIR:
1446                 result = osd_mkdir;
1447                 break;
1448         case DFT_INDEX:
1449                 result = osd_mkidx;
1450                 break;
1451         case DFT_REGULAR:
1452                 result = osd_mkreg;
1453                 break;
1454         case DFT_SYM:
1455                 result = osd_mksym;
1456                 break;
1457         case DFT_NODE:
1458                 result = osd_mknod;
1459                 break;
1460         default:
1461                 LBUG();
1462                 break;
1463         }
1464         return result;
1465 }
1466
1467 /*
1468  * Concurrency: @dt is write locked.
1469  */
1470 static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
1471                              struct lu_attr *attr,
1472                              struct dt_allocation_hint *hint,
1473                              struct dt_object_format *dof,
1474                              struct thandle *th)
1475 {
1476         struct osd_thread_info  *info = osd_oti_get(env);
1477         struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
1478         struct zpl_direntry     *zde = &info->oti_zde.lzd_reg;
1479         const struct lu_fid     *fid = lu_object_fid(&dt->do_lu);
1480         struct osd_object       *obj = osd_dt_obj(dt);
1481         struct osd_device       *osd = osd_obj2dev(obj);
1482         char                    *buf = info->oti_str;
1483         struct osd_thandle      *oh;
1484         dmu_buf_t               *db = NULL;
1485         uint64_t                 zapid, parent = 0;
1486         int                      rc;
1487
1488         ENTRY;
1489
1490         /* concurrent create declarations should not see
1491          * the object inconsistent (db, attr, etc).
1492          * in regular cases acquisition should be cheap */
1493         down_write(&obj->oo_guard);
1494
1495         if (unlikely(dt_object_exists(dt)))
1496                 GOTO(out, rc = -EEXIST);
1497
1498         LASSERT(osd_invariant(obj));
1499         LASSERT(dof != NULL);
1500
1501         LASSERT(th != NULL);
1502         oh = container_of0(th, struct osd_thandle, ot_super);
1503
1504         /*
1505          * XXX missing: Quote handling.
1506          */
1507
1508         LASSERT(obj->oo_db == NULL);
1509
1510         /* to follow ZFS on-disk format we need
1511          * to initialize parent dnode properly */
1512         if (hint != NULL && hint->dah_parent != NULL &&
1513             !dt_object_remote(hint->dah_parent))
1514                 parent = osd_dt_obj(hint->dah_parent)->oo_db->db_object;
1515
1516         /* we may fix some attributes, better do not change the source */
1517         obj->oo_attr = *attr;
1518         obj->oo_attr.la_valid |= LA_SIZE | LA_NLINK | LA_TYPE;
1519
1520         db = osd_create_type_f(dof->dof_type)(env, obj, &obj->oo_attr, oh);
1521         if (IS_ERR(db)) {
1522                 rc = PTR_ERR(db);
1523                 db = NULL;
1524                 GOTO(out, rc);
1525         }
1526
1527         zde->zde_pad = 0;
1528         zde->zde_dnode = db->db_object;
1529         zde->zde_type = IFTODT(attr->la_mode & S_IFMT);
1530
1531         zapid = osd_get_name_n_idx(env, osd, fid, buf, sizeof(info->oti_str));
1532
1533         rc = -zap_add(osd->od_os, zapid, buf, 8, 1, zde, oh->ot_tx);
1534         if (rc)
1535                 GOTO(out, rc);
1536
1537         /* Now add in all of the "SA" attributes */
1538         rc = -sa_handle_get(osd->od_os, db->db_object, NULL,
1539                             SA_HDL_PRIVATE, &obj->oo_sa_hdl);
1540         if (rc)
1541                 GOTO(out, rc);
1542
1543         /* configure new osd object */
1544         obj->oo_db = db;
1545         parent = parent != 0 ? parent : zapid;
1546         rc = __osd_attr_init(env, osd, obj->oo_sa_hdl, oh->ot_tx,
1547                              &obj->oo_attr, parent);
1548         if (rc)
1549                 GOTO(out, rc);
1550
1551         /* XXX: oo_lma_flags */
1552         obj->oo_dt.do_lu.lo_header->loh_attr |= obj->oo_attr.la_mode & S_IFMT;
1553         smp_mb();
1554         obj->oo_dt.do_lu.lo_header->loh_attr |= LOHA_EXISTS;
1555         if (likely(!fid_is_acct(lu_object_fid(&obj->oo_dt.do_lu))))
1556                 /* no body operations for accounting objects */
1557                 obj->oo_dt.do_body_ops = &osd_body_ops;
1558
1559         rc = -nvlist_alloc(&obj->oo_sa_xattr, NV_UNIQUE_NAME, KM_SLEEP);
1560         if (rc)
1561                 GOTO(out, rc);
1562
1563         /* initialize LMA */
1564         lustre_lma_init(lma, lu_object_fid(&obj->oo_dt.do_lu), 0, 0);
1565         lustre_lma_swab(lma);
1566         rc = -nvlist_add_byte_array(obj->oo_sa_xattr, XATTR_NAME_LMA,
1567                                     (uchar_t *)lma, sizeof(*lma));
1568         if (rc)
1569                 GOTO(out, rc);
1570         rc = __osd_sa_xattr_update(env, obj, oh);
1571         if (rc)
1572                 GOTO(out, rc);
1573         osd_idc_find_and_init(env, osd, obj);
1574
1575         /* Add new object to inode accounting.
1576          * Errors are not considered as fatal */
1577         rc = -zap_increment_int(osd->od_os, osd->od_iusr_oid,
1578                                 (attr->la_valid & LA_UID) ? attr->la_uid : 0, 1,
1579                                 oh->ot_tx);
1580         if (rc)
1581                 CERROR("%s: failed to add "DFID" to accounting ZAP for usr %d "
1582                         "(%d)\n", osd->od_svname, PFID(fid), attr->la_uid, rc);
1583         rc = -zap_increment_int(osd->od_os, osd->od_igrp_oid,
1584                                 (attr->la_valid & LA_GID) ? attr->la_gid : 0, 1,
1585                                 oh->ot_tx);
1586         if (rc)
1587                 CERROR("%s: failed to add "DFID" to accounting ZAP for grp %d "
1588                         "(%d)\n", osd->od_svname, PFID(fid), attr->la_gid, rc);
1589
1590 out:
1591         if (unlikely(rc && db)) {
1592                 dmu_object_free(osd->od_os, db->db_object, oh->ot_tx);
1593                 sa_buf_rele(db, osd_obj_tag);
1594                 obj->oo_db = NULL;
1595         }
1596         up_write(&obj->oo_guard);
1597         RETURN(rc);
1598 }
1599
1600 static int osd_declare_object_ref_add(const struct lu_env *env,
1601                                       struct dt_object *dt,
1602                                       struct thandle *th)
1603 {
1604         return osd_declare_attr_set(env, dt, NULL, th);
1605 }
1606
1607 /*
1608  * Concurrency: @dt is write locked.
1609  */
1610 static int osd_object_ref_add(const struct lu_env *env,
1611                               struct dt_object *dt,
1612                               struct thandle *handle)
1613 {
1614         struct osd_object       *obj = osd_dt_obj(dt);
1615         struct osd_thandle      *oh;
1616         struct osd_device       *osd = osd_obj2dev(obj);
1617         uint64_t                 nlink;
1618         int rc;
1619
1620         ENTRY;
1621
1622         down_read(&obj->oo_guard);
1623         if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
1624                 GOTO(out, rc = -ENOENT);
1625
1626         LASSERT(osd_invariant(obj));
1627         LASSERT(obj->oo_sa_hdl != NULL);
1628
1629         oh = container_of0(handle, struct osd_thandle, ot_super);
1630
1631         write_lock(&obj->oo_attr_lock);
1632         nlink = ++obj->oo_attr.la_nlink;
1633         write_unlock(&obj->oo_attr_lock);
1634
1635         rc = osd_object_sa_update(obj, SA_ZPL_LINKS(osd), &nlink, 8, oh);
1636
1637 out:
1638         up_read(&obj->oo_guard);
1639         RETURN(rc);
1640 }
1641
1642 static int osd_declare_object_ref_del(const struct lu_env *env,
1643                                       struct dt_object *dt,
1644                                       struct thandle *handle)
1645 {
1646         return osd_declare_attr_set(env, dt, NULL, handle);
1647 }
1648
1649 /*
1650  * Concurrency: @dt is write locked.
1651  */
1652 static int osd_object_ref_del(const struct lu_env *env,
1653                               struct dt_object *dt,
1654                               struct thandle *handle)
1655 {
1656         struct osd_object       *obj = osd_dt_obj(dt);
1657         struct osd_thandle      *oh;
1658         struct osd_device       *osd = osd_obj2dev(obj);
1659         uint64_t                 nlink;
1660         int                      rc;
1661
1662         ENTRY;
1663
1664         down_read(&obj->oo_guard);
1665
1666         if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
1667                 GOTO(out, rc = -ENOENT);
1668
1669         LASSERT(osd_invariant(obj));
1670         LASSERT(obj->oo_sa_hdl != NULL);
1671
1672         oh = container_of0(handle, struct osd_thandle, ot_super);
1673         LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
1674
1675         write_lock(&obj->oo_attr_lock);
1676         nlink = --obj->oo_attr.la_nlink;
1677         write_unlock(&obj->oo_attr_lock);
1678
1679         rc = osd_object_sa_update(obj, SA_ZPL_LINKS(osd), &nlink, 8, oh);
1680
1681 out:
1682         up_read(&obj->oo_guard);
1683         RETURN(rc);
1684 }
1685
1686 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
1687                            __u64 start, __u64 end)
1688 {
1689         struct osd_device *osd = osd_obj2dev(osd_dt_obj(dt));
1690         ENTRY;
1691
1692         /* XXX: no other option than syncing the whole filesystem until we
1693          * support ZIL.  If the object tracked the txg that it was last
1694          * modified in, it could pass that txg here instead of "0".  Maybe
1695          * the changes are already committed, so no wait is needed at all? */
1696         txg_wait_synced(dmu_objset_pool(osd->od_os), 0ULL);
1697
1698         RETURN(0);
1699 }
1700
1701 static int osd_invalidate(const struct lu_env *env, struct dt_object *dt)
1702 {
1703         return 0;
1704 }
1705
1706 static struct dt_object_operations osd_obj_ops = {
1707         .do_read_lock           = osd_object_read_lock,
1708         .do_write_lock          = osd_object_write_lock,
1709         .do_read_unlock         = osd_object_read_unlock,
1710         .do_write_unlock        = osd_object_write_unlock,
1711         .do_write_locked        = osd_object_write_locked,
1712         .do_attr_get            = osd_attr_get,
1713         .do_declare_attr_set    = osd_declare_attr_set,
1714         .do_attr_set            = osd_attr_set,
1715         .do_ah_init             = osd_ah_init,
1716         .do_declare_create      = osd_declare_object_create,
1717         .do_create              = osd_object_create,
1718         .do_declare_destroy     = osd_declare_object_destroy,
1719         .do_destroy             = osd_object_destroy,
1720         .do_index_try           = osd_index_try,
1721         .do_declare_ref_add     = osd_declare_object_ref_add,
1722         .do_ref_add             = osd_object_ref_add,
1723         .do_declare_ref_del     = osd_declare_object_ref_del,
1724         .do_ref_del             = osd_object_ref_del,
1725         .do_xattr_get           = osd_xattr_get,
1726         .do_declare_xattr_set   = osd_declare_xattr_set,
1727         .do_xattr_set           = osd_xattr_set,
1728         .do_declare_xattr_del   = osd_declare_xattr_del,
1729         .do_xattr_del           = osd_xattr_del,
1730         .do_xattr_list          = osd_xattr_list,
1731         .do_object_sync         = osd_object_sync,
1732         .do_invalidate          = osd_invalidate,
1733 };
1734
1735 static struct lu_object_operations osd_lu_obj_ops = {
1736         .loo_object_init        = osd_object_init,
1737         .loo_object_delete      = osd_object_delete,
1738         .loo_object_release     = osd_object_release,
1739         .loo_object_free        = osd_object_free,
1740         .loo_object_print       = osd_object_print,
1741         .loo_object_invariant   = osd_object_invariant,
1742 };
1743
1744 static int osd_otable_it_attr_get(const struct lu_env *env,
1745                                 struct dt_object *dt,
1746                                 struct lu_attr *attr)
1747 {
1748         attr->la_valid = 0;
1749         return 0;
1750 }
1751
1752 static struct dt_object_operations osd_obj_otable_it_ops = {
1753         .do_attr_get    = osd_otable_it_attr_get,
1754         .do_index_try   = osd_index_try,
1755 };