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 <libcfs/libcfs.h>
  42 #include <obd_support.h>
  43 #include <lustre_net.h>
  44 #include <obd.h>
  45 #include <obd_class.h>
  46 #include <lustre_disk.h>
  47 #include <lustre_fid.h>
  48
  49 #include "osd_internal.h"
  50
  51 #include <sys/dnode.h>
  52 #include <sys/dbuf.h>
  53 #include <sys/spa.h>
  54 #include <sys/stat.h>
  55 #include <sys/zap.h>
  56 #include <sys/spa_impl.h>
  57 #include <sys/zfs_znode.h>
  58 #include <sys/dmu_tx.h>
  59 #include <sys/dmu_objset.h>
  60 #include <sys/dsl_prop.h>
  61 #include <sys/sa_impl.h>
  62 #include <sys/txg.h>
  63
  64 char *osd_obj_tag = "osd_object";
  65
  66 static struct dt_object_operations osd_obj_ops;
  67 static struct lu_object_operations osd_lu_obj_ops;
  68 extern struct dt_body_operations osd_body_ops;
  69 static struct dt_object_operations osd_obj_otable_it_ops;
  70
  71 extern struct kmem_cache *osd_object_kmem;
  72
  73 static void
  74 osd_object_sa_fini(struct osd_object *obj)
  75 {
  76         if (obj->oo_sa_hdl) {
  77                 sa_handle_destroy(obj->oo_sa_hdl);
  78                 obj->oo_sa_hdl = NULL;
  79         }
  80 }
  81
  82 static int
  83 osd_object_sa_init(struct osd_object *obj, struct osd_device *o)
  84 {
  85         int rc;
  86
  87         LASSERT(obj->oo_sa_hdl == NULL);
  88         LASSERT(obj->oo_dn != NULL);
  89
  90         rc = osd_sa_handle_get(obj);
  91         if (rc)
  92                 return rc;
  93
  94         /* Cache the xattr object id, valid for the life of the object */
  95         rc = -sa_lookup(obj->oo_sa_hdl, SA_ZPL_XATTR(o), &obj->oo_xattr, 8);
  96         if (rc == -ENOENT) {
  97                 obj->oo_xattr = ZFS_NO_OBJECT;
  98                 rc = 0;
  99         } else if (rc) {
 100                 osd_object_sa_fini(obj);
 101         }
 102
 103         return rc;
 104 }
 105
 106 /*
 107  * Add object to list of dirty objects in tx handle.
 108  */
 109 void osd_object_sa_dirty_add(struct osd_object *obj, struct osd_thandle *oh)
 110 {
 111         if (!list_empty(&obj->oo_sa_linkage))
 112                 return;
 113
 114         write_lock(&obj->oo_attr_lock);
 115         if (likely(list_empty(&obj->oo_sa_linkage)))
 116                 list_add(&obj->oo_sa_linkage, &oh->ot_sa_list);
 117         write_unlock(&obj->oo_attr_lock);
 118 }
 119
 120 /*
 121  * Release spill block dbuf hold for all dirty SAs.
 122  */
 123 void osd_object_sa_dirty_rele(const struct lu_env *env, struct osd_thandle *oh)
 124 {
 125         struct osd_object *obj;
 126
 127         while (!list_empty(&oh->ot_sa_list)) {
 128                 obj = list_entry(oh->ot_sa_list.next,
 129                                  struct osd_object, oo_sa_linkage);
 130                 write_lock(&obj->oo_attr_lock);
 131                 list_del_init(&obj->oo_sa_linkage);
 132                 write_unlock(&obj->oo_attr_lock);
 133                 if (obj->oo_late_xattr) {
 134                         /*
 135                          * take oo_guard to protect oo_sa_xattr buffer
 136                          * from concurrent update by osd_xattr_set()
 137                          */
 138                         LASSERT(oh->ot_assigned != 0);
 139                         down_write(&obj->oo_guard);
 140                         if (obj->oo_late_attr_set)
 141                                 __osd_sa_attr_init(env, obj, oh);
 142                         else if (obj->oo_late_xattr)
 143                                 __osd_sa_xattr_update(env, obj, oh);
 144                         up_write(&obj->oo_guard);
 145                 }
 146                 sa_spill_rele(obj->oo_sa_hdl);
 147         }
 148 }
 149
 150 /*
 151  * Update the SA and add the object to the dirty list.
 152  */
 153 int osd_object_sa_update(struct osd_object *obj, sa_attr_type_t type,
 154                          void *buf, uint32_t buflen, struct osd_thandle *oh)
 155 {
 156         int rc;
 157
 158         LASSERT(obj->oo_sa_hdl != NULL);
 159         LASSERT(oh->ot_tx != NULL);
 160
 161         rc = -sa_update(obj->oo_sa_hdl, type, buf, buflen, oh->ot_tx);
 162         osd_object_sa_dirty_add(obj, oh);
 163
 164         return rc;
 165 }
 166
 167 /*
 168  * Bulk update the SA and add the object to the dirty list.
 169  */
 170 static int
 171 osd_object_sa_bulk_update(struct osd_object *obj, sa_bulk_attr_t *attrs,
 172                           int count, struct osd_thandle *oh)
 173 {
 174         int rc;
 175
 176         LASSERT(obj->oo_sa_hdl != NULL);
 177         LASSERT(oh->ot_tx != NULL);
 178
 179         rc = -sa_bulk_update(obj->oo_sa_hdl, attrs, count, oh->ot_tx);
 180         osd_object_sa_dirty_add(obj, oh);
 181
 182         return rc;
 183 }
 184
 185 /*
 186  * Retrieve the attributes of a DMU object
 187  */
 188 int __osd_object_attr_get(const struct lu_env *env, struct osd_device *o,
 189                           struct osd_object *obj, struct lu_attr *la)
 190 {
 191         struct osa_attr *osa = &osd_oti_get(env)->oti_osa;
 192         sa_bulk_attr_t  *bulk = osd_oti_get(env)->oti_attr_bulk;
 193         int              cnt = 0;
 194         int              rc;
 195         ENTRY;
 196
 197         LASSERT(obj->oo_dn != NULL);
 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(obj->oo_sa_hdl, bulk, cnt);
 214         if (rc)
 215                 GOTO(out_sa, rc);
 216
 217 #ifdef ZFS_PROJINHERIT
 218         if (o->od_projectused_dn && osa->flags & ZFS_PROJID) {
 219                 rc = -sa_lookup(obj->oo_sa_hdl, SA_ZPL_PROJID(o),
 220                                 &osa->projid, 8);
 221                 if (rc)
 222                         GOTO(out_sa, rc);
 223
 224                 la->la_projid = osa->projid;
 225                 la->la_valid |= LA_PROJID;
 226                 obj->oo_with_projid = 1;
 227         } else {
 228                 la->la_projid = ZFS_DEFAULT_PROJID;
 229                 la->la_valid &= ~LA_PROJID;
 230         }
 231 #else
 232         la->la_projid = 0;
 233         la->la_valid &= ~LA_PROJID;
 234 #endif
 235
 236         la->la_atime = osa->atime[0];
 237         la->la_mtime = osa->mtime[0];
 238         la->la_ctime = osa->ctime[0];
 239         la->la_mode = osa->mode;
 240         la->la_uid = osa->uid;
 241         la->la_gid = osa->gid;
 242         la->la_nlink = osa->nlink;
 243         la->la_flags = attrs_zfs2fs(osa->flags);
 244         la->la_size = osa->size;
 245
 246         /* Try to get extra flag from LMA. Right now, only LMAI_ORPHAN
 247          * flags is stored in LMA, and it is only for orphan directory */
 248         if (S_ISDIR(la->la_mode) && dt_object_exists(&obj->oo_dt)) {
 249                 struct osd_thread_info *info = osd_oti_get(env);
 250                 struct lustre_mdt_attrs *lma;
 251                 struct lu_buf buf;
 252
 253                 lma = (struct lustre_mdt_attrs *)info->oti_buf;
 254                 buf.lb_buf = lma;
 255                 buf.lb_len = sizeof(info->oti_buf);
 256                 rc = osd_xattr_get(env, &obj->oo_dt, &buf, XATTR_NAME_LMA);
 257                 if (rc > 0) {
 258                         rc = 0;
 259                         lma->lma_incompat = le32_to_cpu(lma->lma_incompat);
 260                         obj->oo_lma_flags =
 261                                 lma_to_lustre_flags(lma->lma_incompat);
 262
 263                 } else if (rc == -ENODATA) {
 264                         rc = 0;
 265                 }
 266         }
 267
 268         if (S_ISCHR(la->la_mode) || S_ISBLK(la->la_mode)) {
 269                 rc = -sa_lookup(obj->oo_sa_hdl, SA_ZPL_RDEV(o), &osa->rdev, 8);
 270                 if (rc)
 271                         GOTO(out_sa, rc);
 272                 la->la_rdev = osa->rdev;
 273                 la->la_valid |= LA_RDEV;
 274         }
 275 out_sa:
 276
 277         RETURN(rc);
 278 }
 279
 280 int __osd_obj2dnode(objset_t *os, uint64_t oid, dnode_t **dnp)
 281 {
 282         dmu_buf_t *db;
 283         dmu_buf_impl_t *dbi;
 284         int rc;
 285
 286         rc = -dmu_bonus_hold(os, oid, osd_obj_tag, &db);
 287         if (rc)
 288                 return rc;
 289
 290         dbi = (dmu_buf_impl_t *)db;
 291         DB_DNODE_ENTER(dbi);
 292         *dnp = DB_DNODE(dbi);
 293         LASSERT(*dnp != NULL);
 294
 295         return 0;
 296 }
 297
 298 /*
 299  * Concurrency: no concurrent access is possible that early in object
 300  * life-cycle.
 301  */
 302 struct lu_object *osd_object_alloc(const struct lu_env *env,
 303                                    const struct lu_object_header *hdr,
 304                                    struct lu_device *d)
 305 {
 306         struct osd_object *mo;
 307
 308         OBD_SLAB_ALLOC_PTR_GFP(mo, osd_object_kmem, GFP_NOFS);
 309         if (mo != NULL) {
 310                 struct lu_object *l;
 311
 312                 l = &mo->oo_dt.do_lu;
 313                 dt_object_init(&mo->oo_dt, NULL, d);
 314                 mo->oo_dt.do_ops = &osd_obj_ops;
 315                 l->lo_ops = &osd_lu_obj_ops;
 316                 INIT_LIST_HEAD(&mo->oo_sa_linkage);
 317                 INIT_LIST_HEAD(&mo->oo_unlinked_linkage);
 318                 init_rwsem(&mo->oo_sem);
 319                 init_rwsem(&mo->oo_guard);
 320                 rwlock_init(&mo->oo_attr_lock);
 321                 mo->oo_destroy = OSD_DESTROY_NONE;
 322                 return l;
 323         } else {
 324                 return NULL;
 325         }
 326 }
 327
 328 static void osd_obj_set_blksize(const struct lu_env *env,
 329                                 struct osd_device *osd, struct osd_object *obj)
 330 {
 331         const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
 332         dmu_tx_t *tx;
 333         dnode_t *dn = obj->oo_dn;
 334         uint32_t blksz;
 335         int rc = 0;
 336         ENTRY;
 337
 338         LASSERT(!osd_oti_get(env)->oti_in_trans);
 339
 340         tx = dmu_tx_create(osd->od_os);
 341         if (!tx) {
 342                 CERROR("%s: fail to create tx to set blksize for "DFID"\n",
 343                        osd->od_svname, PFID(fid));
 344                 RETURN_EXIT;
 345         }
 346
 347         dmu_tx_hold_bonus(tx, dn->dn_object);
 348         rc = -dmu_tx_assign(tx, TXG_WAIT);
 349         if (rc) {
 350                 dmu_tx_abort(tx);
 351                 CERROR("%s: fail to assign tx to set blksize for "DFID
 352                        ": rc = %d\n", osd->od_svname, PFID(fid), rc);
 353                 RETURN_EXIT;
 354         }
 355
 356         down_write(&obj->oo_guard);
 357         if (unlikely((1 << dn->dn_datablkshift) >= PAGE_SIZE))
 358                 GOTO(out, rc = 1);
 359
 360         blksz = dn->dn_datablksz;
 361         if (!is_power_of_2(blksz))
 362                 blksz = size_roundup_power2(blksz);
 363
 364         if (blksz > osd->od_max_blksz)
 365                 blksz = osd->od_max_blksz;
 366         else if (blksz < PAGE_SIZE)
 367                 blksz = PAGE_SIZE;
 368         rc = -dmu_object_set_blocksize(osd->od_os, dn->dn_object, blksz, 0, tx);
 369
 370         GOTO(out, rc);
 371
 372 out:
 373         up_write(&obj->oo_guard);
 374         if (rc) {
 375                 dmu_tx_abort(tx);
 376                 if (unlikely(obj->oo_dn->dn_maxblkid > 0))
 377                         rc = 1;
 378                 if (rc < 0)
 379                         CERROR("%s: fail to set blksize for "DFID": rc = %d\n",
 380                                osd->od_svname, PFID(fid), rc);
 381         } else {
 382                 dmu_tx_commit(tx);
 383                 CDEBUG(D_INODE, "%s: set blksize as %u for "DFID"\n",
 384                        osd->od_svname, blksz, PFID(fid));
 385         }
 386 }
 387
 388 /*
 389  * Concurrency: shouldn't matter.
 390  */
 391 int osd_object_init0(const struct lu_env *env, struct osd_object *obj)
 392 {
 393         struct osd_device       *osd = osd_obj2dev(obj);
 394         const struct lu_fid     *fid = lu_object_fid(&obj->oo_dt.do_lu);
 395         int                      rc = 0;
 396         ENTRY;
 397
 398         LASSERT(obj->oo_dn);
 399
 400         rc = osd_object_sa_init(obj, osd);
 401         if (rc)
 402                 RETURN(rc);
 403
 404         /* cache attrs in object */
 405         rc = __osd_object_attr_get(env, osd, obj, &obj->oo_attr);
 406         if (rc)
 407                 RETURN(rc);
 408
 409         if (likely(!fid_is_acct(fid))) {
 410                 /* no body operations for accounting objects */
 411                 obj->oo_dt.do_body_ops = &osd_body_ops;
 412
 413                 if (S_ISREG(obj->oo_attr.la_mode) &&
 414                     obj->oo_dn->dn_maxblkid == 0 &&
 415                     (1 << obj->oo_dn->dn_datablkshift) < PAGE_SIZE &&
 416                     (fid_is_idif(fid) || fid_is_norm(fid) ||
 417                      fid_is_echo(fid)) &&
 418                     osd->od_is_ost && !osd->od_dt_dev.dd_rdonly)
 419                         osd_obj_set_blksize(env, osd, obj);
 420         }
 421
 422         /*
 423          * initialize object before marking it existing
 424          */
 425         obj->oo_dt.do_lu.lo_header->loh_attr |= obj->oo_attr.la_mode & S_IFMT;
 426
 427         smp_mb();
 428         obj->oo_dt.do_lu.lo_header->loh_attr |= LOHA_EXISTS;
 429
 430         RETURN(0);
 431 }
 432
 433 static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
 434 {
 435         struct osd_thread_info  *info = osd_oti_get(env);
 436         struct lu_buf           buf;
 437         int                     rc;
 438         struct lustre_mdt_attrs *lma;
 439         ENTRY;
 440
 441         CLASSERT(sizeof(info->oti_buf) >= sizeof(*lma));
 442         lma = (struct lustre_mdt_attrs *)info->oti_buf;
 443         buf.lb_buf = lma;
 444         buf.lb_len = sizeof(info->oti_buf);
 445
 446         rc = osd_xattr_get(env, &obj->oo_dt, &buf, XATTR_NAME_LMA);
 447         if (rc > 0) {
 448                 rc = 0;
 449                 lustre_lma_swab(lma);
 450                 if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
 451                              CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT))) {
 452                         CWARN("%s: unsupported incompat LMA feature(s) %#x for "
 453                               "fid = "DFID"\n", osd_obj2dev(obj)->od_svname,
 454                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
 455                               PFID(lu_object_fid(&obj->oo_dt.do_lu)));
 456                         rc = -EOPNOTSUPP;
 457                 }
 458         } else if (rc == -ENODATA) {
 459                 /* haven't initialize LMA xattr */
 460                 rc = 0;
 461         }
 462
 463         RETURN(rc);
 464 }
 465
 466 /**
 467  * Helper function to retrieve DMU object id from fid for accounting object
 468  */
 469 static dnode_t *osd_quota_fid2dmu(const struct osd_device *osd,
 470                                   const struct lu_fid *fid)
 471 {
 472         dnode_t *dn = NULL;
 473
 474         LASSERT(fid_is_acct(fid));
 475
 476         switch (fid_oid(fid)) {
 477         case ACCT_USER_OID:
 478                 dn = osd->od_userused_dn;
 479                 break;
 480         case ACCT_GROUP_OID:
 481                 dn = osd->od_groupused_dn;
 482                 break;
 483 #ifdef ZFS_PROJINHERIT
 484         case ACCT_PROJECT_OID:
 485                 dn = osd->od_projectused_dn;
 486                 break;
 487 #endif
 488         default:
 489                 break;
 490         }
 491
 492         return dn;
 493 }
 494
 495 /*
 496  * Concurrency: no concurrent access is possible that early in object
 497  * life-cycle.
 498  */
 499 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
 500                            const struct lu_object_conf *conf)
 501 {
 502         struct osd_object *obj = osd_obj(l);
 503         struct osd_device *osd = osd_obj2dev(obj);
 504         const struct lu_fid *fid = lu_object_fid(l);
 505         uint64_t oid;
 506         int rc = 0;
 507         ENTRY;
 508
 509         LASSERT(osd_invariant(obj));
 510
 511         if (fid_is_otable_it(&l->lo_header->loh_fid)) {
 512                 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
 513                 l->lo_header->loh_attr |= LOHA_EXISTS;
 514                 RETURN(0);
 515         }
 516
 517         if (conf != NULL && conf->loc_flags & LOC_F_NEW)
 518                 GOTO(out, rc = 0);
 519
 520         if (unlikely(fid_is_acct(fid))) {
 521                 obj->oo_dn = osd_quota_fid2dmu(osd, fid);
 522                 if (obj->oo_dn) {
 523                         obj->oo_dt.do_index_ops = &osd_acct_index_ops;
 524                         l->lo_header->loh_attr |= LOHA_EXISTS;
 525                 }
 526
 527                 GOTO(out, rc = 0);
 528         }
 529
 530         rc = osd_fid_lookup(env, osd, fid, &oid);
 531         if (rc == 0) {
 532                 LASSERT(obj->oo_dn == NULL);
 533                 rc = __osd_obj2dnode(osd->od_os, oid, &obj->oo_dn);
 534                 /* EEXIST will be returned if object is being deleted in ZFS */
 535                 if (rc == -EEXIST) {
 536                         rc = 0;
 537                         GOTO(out, rc);
 538                 }
 539                 if (rc != 0) {
 540                         CERROR("%s: lookup "DFID"/%#llx failed: rc = %d\n",
 541                                osd->od_svname, PFID(lu_object_fid(l)), oid, rc);
 542                         GOTO(out, rc);
 543                 }
 544                 rc = osd_object_init0(env, obj);
 545                 if (rc != 0)
 546                         GOTO(out, rc);
 547
 548                 rc = osd_check_lma(env, obj);
 549                 if (rc != 0)
 550                         GOTO(out, rc);
 551         } else if (rc == -ENOENT) {
 552                 rc = 0;
 553         }
 554         LASSERT(osd_invariant(obj));
 555 out:
 556         RETURN(rc);
 557 }
 558
 559 /*
 560  * Concurrency: no concurrent access is possible that late in object
 561  * life-cycle.
 562  */
 563 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
 564 {
 565         struct osd_object *obj = osd_obj(l);
 566
 567         LASSERT(osd_invariant(obj));
 568
 569         dt_object_fini(&obj->oo_dt);
 570         OBD_SLAB_FREE_PTR(obj, osd_object_kmem);
 571 }
 572
 573 static int
 574 osd_object_unlinked_add(struct osd_object *obj, struct osd_thandle *oh)
 575 {
 576         int rc = -EBUSY;
 577
 578         LASSERT(obj->oo_destroy == OSD_DESTROY_ASYNC);
 579
 580         /* the object is supposed to be exclusively locked by
 581          * the caller (osd_destroy()), while the transaction
 582          * (oh) is per-thread and not shared */
 583         if (likely(list_empty(&obj->oo_unlinked_linkage))) {
 584                 list_add(&obj->oo_unlinked_linkage, &oh->ot_unlinked_list);
 585                 rc = 0;
 586         }
 587
 588         return rc;
 589 }
 590
 591 /* Default to max data size covered by a level-1 indirect block */
 592 static unsigned long osd_sync_destroy_max_size =
 593         1UL << (DN_MAX_INDBLKSHIFT - SPA_BLKPTRSHIFT + SPA_MAXBLOCKSHIFT);
 594 module_param(osd_sync_destroy_max_size, ulong, 0444);
 595 MODULE_PARM_DESC(osd_sync_destroy_max_size, "Maximum object size to use synchronous destroy.");
 596
 597 static inline void
 598 osd_object_set_destroy_type(struct osd_object *obj)
 599 {
 600         /*
 601          * Lock-less OST_WRITE can race with OST_DESTROY, so set destroy type
 602          * only once and use it consistently thereafter.
 603          */
 604         down_write(&obj->oo_guard);
 605         if (obj->oo_destroy == OSD_DESTROY_NONE) {
 606                 if (obj->oo_attr.la_size <= osd_sync_destroy_max_size)
 607                         obj->oo_destroy = OSD_DESTROY_SYNC;
 608                 else /* Larger objects are destroyed asynchronously */
 609                         obj->oo_destroy = OSD_DESTROY_ASYNC;
 610         }
 611         up_write(&obj->oo_guard);
 612 }
 613
 614 static int osd_declare_destroy(const struct lu_env *env, struct dt_object *dt,
 615                                struct thandle *th)
 616 {
 617         const struct lu_fid     *fid = lu_object_fid(&dt->do_lu);
 618         struct osd_object       *obj = osd_dt_obj(dt);
 619         struct osd_device       *osd = osd_obj2dev(obj);
 620         struct osd_thandle      *oh;
 621         dnode_t *dn;
 622         int                      rc;
 623         uint64_t                 zapid;
 624         ENTRY;
 625
 626         LASSERT(th != NULL);
 627         LASSERT(dt_object_exists(dt));
 628
 629         oh = container_of0(th, struct osd_thandle, ot_super);
 630         LASSERT(oh->ot_tx != NULL);
 631
 632         /* declare that we'll remove object from fid-dnode mapping */
 633         zapid = osd_get_name_n_idx(env, osd, fid, NULL, 0, &dn);
 634         osd_tx_hold_zap(oh->ot_tx, zapid, dn, FALSE, NULL);
 635
 636         osd_declare_xattrs_destroy(env, obj, oh);
 637
 638         /* one less inode */
 639         rc = osd_declare_quota(env, osd, obj->oo_attr.la_uid,
 640                                obj->oo_attr.la_gid, obj->oo_attr.la_projid,
 641                                -1, oh, NULL, OSD_QID_INODE);
 642         if (rc)
 643                 RETURN(rc);
 644
 645         /* data to be truncated */
 646         rc = osd_declare_quota(env, osd, obj->oo_attr.la_uid,
 647                                obj->oo_attr.la_gid, obj->oo_attr.la_projid,
 648                                0, oh, NULL, OSD_QID_BLK);
 649         if (rc)
 650                 RETURN(rc);
 651
 652         osd_object_set_destroy_type(obj);
 653         if (obj->oo_destroy == OSD_DESTROY_SYNC)
 654                 dmu_tx_hold_free(oh->ot_tx, obj->oo_dn->dn_object,
 655                                  0, DMU_OBJECT_END);
 656         else
 657                 osd_tx_hold_zap(oh->ot_tx, osd->od_unlinked->dn_object,
 658                                 osd->od_unlinked, TRUE, NULL);
 659
 660         /* will help to find FID->ino when this object is being
 661          * added to PENDING/ */
 662         osd_idc_find_and_init(env, osd, obj);
 663
 664         RETURN(0);
 665 }
 666
 667 static int osd_destroy(const struct lu_env *env, struct dt_object *dt,
 668                        struct thandle *th)
 669 {
 670         struct osd_thread_info  *info = osd_oti_get(env);
 671         char                    *buf = info->oti_str;
 672         struct osd_object       *obj = osd_dt_obj(dt);
 673         struct osd_device       *osd = osd_obj2dev(obj);
 674         const struct lu_fid     *fid = lu_object_fid(&dt->do_lu);
 675         struct osd_thandle      *oh;
 676         int                      rc;
 677         uint64_t                 oid, zapid;
 678         dnode_t *zdn;
 679         ENTRY;
 680
 681         down_write(&obj->oo_guard);
 682
 683         if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
 684                 GOTO(out, rc = -ENOENT);
 685
 686         LASSERT(obj->oo_dn != NULL);
 687
 688         oh = container_of0(th, struct osd_thandle, ot_super);
 689         LASSERT(oh != NULL);
 690         LASSERT(oh->ot_tx != NULL);
 691
 692         /* remove obj ref from index dir (it depends) */
 693         zapid = osd_get_name_n_idx(env, osd, fid, buf,
 694                                    sizeof(info->oti_str), &zdn);
 695         rc = osd_zap_remove(osd, zapid, zdn, buf, oh->ot_tx);
 696         if (rc) {
 697                 CERROR("%s: zap_remove(%s) failed: rc = %d\n",
 698                        osd->od_svname, buf, rc);
 699                 GOTO(out, rc);
 700         }
 701
 702         rc = osd_xattrs_destroy(env, obj, oh);
 703         if (rc) {
 704                 CERROR("%s: cannot destroy xattrs for %s: rc = %d\n",
 705                        osd->od_svname, buf, rc);
 706                 GOTO(out, rc);
 707         }
 708
 709         oid = obj->oo_dn->dn_object;
 710         if (unlikely(obj->oo_destroy == OSD_DESTROY_NONE)) {
 711                 /* this may happen if the destroy wasn't declared
 712                  * e.g. when the object is created and then destroyed
 713                  * in the same transaction - we don't need additional
 714                  * space for destroy specifically */
 715                 LASSERT(obj->oo_attr.la_size <= osd_sync_destroy_max_size);
 716                 rc = -dmu_object_free(osd->od_os, oid, oh->ot_tx);
 717                 if (rc)
 718                         CERROR("%s: failed to free %s %llu: rc = %d\n",
 719                                osd->od_svname, buf, oid, rc);
 720         } else if (obj->oo_destroy == OSD_DESTROY_SYNC) {
 721                 rc = -dmu_object_free(osd->od_os, oid, oh->ot_tx);
 722                 if (rc)
 723                         CERROR("%s: failed to free %s %llu: rc = %d\n",
 724                                osd->od_svname, buf, oid, rc);
 725         } else { /* asynchronous destroy */
 726                 char *key = info->oti_key;
 727
 728                 rc = osd_object_unlinked_add(obj, oh);
 729                 if (rc)
 730                         GOTO(out, rc);
 731
 732                 snprintf(key, sizeof(info->oti_key), "%llx", oid);
 733                 rc = osd_zap_add(osd, osd->od_unlinked->dn_object,
 734                                  osd->od_unlinked, key, 8, 1, &oid, oh->ot_tx);
 735                 if (rc)
 736                         CERROR("%s: zap_add_int() failed %s %llu: rc = %d\n",
 737                                osd->od_svname, buf, oid, rc);
 738         }
 739
 740 out:
 741         /* not needed in the cache anymore */
 742         set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
 743         if (rc == 0)
 744                 obj->oo_destroyed = 1;
 745         up_write(&obj->oo_guard);
 746         RETURN (0);
 747 }
 748
 749 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
 750 {
 751         struct osd_object *obj = osd_obj(l);
 752         const struct lu_fid *fid = lu_object_fid(l);
 753
 754         if (obj->oo_dn) {
 755                 if (likely(!fid_is_acct(fid))) {
 756                         osd_object_sa_fini(obj);
 757                         if (obj->oo_sa_xattr) {
 758                                 nvlist_free(obj->oo_sa_xattr);
 759                                 obj->oo_sa_xattr = NULL;
 760                         }
 761                         osd_dnode_rele(obj->oo_dn);
 762                         list_del(&obj->oo_sa_linkage);
 763                 }
 764                 obj->oo_dn = NULL;
 765         }
 766 }
 767
 768 /*
 769  * Concurrency: ->loo_object_release() is called under site spin-lock.
 770  */
 771 static void osd_object_release(const struct lu_env *env,
 772                                struct lu_object *l)
 773 {
 774 }
 775
 776 /*
 777  * Concurrency: shouldn't matter.
 778  */
 779 static int osd_object_print(const struct lu_env *env, void *cookie,
 780                             lu_printer_t p, const struct lu_object *l)
 781 {
 782         struct osd_object *o = osd_obj(l);
 783
 784         return (*p)(env, cookie, LUSTRE_OSD_ZFS_NAME"-object@%p", o);
 785 }
 786
 787 static void osd_read_lock(const struct lu_env *env, struct dt_object *dt,
 788                           unsigned role)
 789 {
 790         struct osd_object *obj = osd_dt_obj(dt);
 791
 792         LASSERT(osd_invariant(obj));
 793
 794         down_read_nested(&obj->oo_sem, role);
 795 }
 796
 797 static void osd_write_lock(const struct lu_env *env, struct dt_object *dt,
 798                            unsigned role)
 799 {
 800         struct osd_object *obj = osd_dt_obj(dt);
 801
 802         LASSERT(osd_invariant(obj));
 803
 804         down_write_nested(&obj->oo_sem, role);
 805 }
 806
 807 static void osd_read_unlock(const struct lu_env *env, struct dt_object *dt)
 808 {
 809         struct osd_object *obj = osd_dt_obj(dt);
 810
 811         LASSERT(osd_invariant(obj));
 812         up_read(&obj->oo_sem);
 813 }
 814
 815 static void osd_write_unlock(const struct lu_env *env, struct dt_object *dt)
 816 {
 817         struct osd_object *obj = osd_dt_obj(dt);
 818
 819         LASSERT(osd_invariant(obj));
 820         up_write(&obj->oo_sem);
 821 }
 822
 823 static int osd_write_locked(const struct lu_env *env, struct dt_object *dt)
 824 {
 825         struct osd_object *obj = osd_dt_obj(dt);
 826         int rc = 1;
 827
 828         LASSERT(osd_invariant(obj));
 829
 830         if (down_write_trylock(&obj->oo_sem)) {
 831                 rc = 0;
 832                 up_write(&obj->oo_sem);
 833         }
 834         return rc;
 835 }
 836
 837 static int osd_attr_get(const struct lu_env *env,
 838                         struct dt_object *dt,
 839                         struct lu_attr *attr)
 840 {
 841         struct osd_object       *obj = osd_dt_obj(dt);
 842         uint64_t                 blocks;
 843         uint32_t                 blksize;
 844         int                      rc = 0;
 845
 846         down_read(&obj->oo_guard);
 847
 848         if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
 849                 GOTO(out, rc = -ENOENT);
 850
 851         if (unlikely(fid_is_acct(lu_object_fid(&dt->do_lu))))
 852                 GOTO(out, rc = 0);
 853
 854         LASSERT(osd_invariant(obj));
 855         LASSERT(obj->oo_dn);
 856
 857         read_lock(&obj->oo_attr_lock);
 858         *attr = obj->oo_attr;
 859         if (obj->oo_lma_flags & LUSTRE_ORPHAN_FL)
 860                 attr->la_flags |= LUSTRE_ORPHAN_FL;
 861         read_unlock(&obj->oo_attr_lock);
 862
 863         /* with ZFS_DEBUG zrl_add_debug() called by DB_DNODE_ENTER()
 864          * from within sa_object_size() can block on a mutex, so
 865          * we can't call sa_object_size() holding rwlock */
 866         sa_object_size(obj->oo_sa_hdl, &blksize, &blocks);
 867         /* we do not control size of indices, so always calculate
 868          * it from number of blocks reported by DMU */
 869         if (S_ISDIR(attr->la_mode))
 870                 attr->la_size = 512 * blocks;
 871         /* Block size may be not set; suggest maximal I/O transfers. */
 872         if (blksize == 0)
 873                 blksize = osd_spa_maxblocksize(
 874                         dmu_objset_spa(osd_obj2dev(obj)->od_os));
 875
 876         attr->la_blksize = blksize;
 877         attr->la_blocks = blocks;
 878         attr->la_valid |= LA_BLOCKS | LA_BLKSIZE;
 879
 880 out:
 881         up_read(&obj->oo_guard);
 882         return rc;
 883 }
 884
 885 /* Simple wrapper on top of qsd API which implement quota transfer for osd
 886  * setattr needs. As a reminder, only the root user can change ownership of
 887  * a file, that's why EDQUOT & EINPROGRESS errors are discarded */
 888 static inline int qsd_transfer(const struct lu_env *env,
 889                                struct qsd_instance *qsd,
 890                                struct lquota_trans *trans, int qtype,
 891                                __u64 orig_id, __u64 new_id, __u64 bspace,
 892                                struct lquota_id_info *qi)
 893 {
 894         int     rc;
 895
 896         if (unlikely(qsd == NULL))
 897                 return 0;
 898
 899         LASSERT(qtype >= 0 && qtype < LL_MAXQUOTAS);
 900         qi->lqi_type = qtype;
 901
 902         /* inode accounting */
 903         qi->lqi_is_blk = false;
 904
 905         /* one more inode for the new owner ... */
 906         qi->lqi_id.qid_uid = new_id;
 907         qi->lqi_space      = 1;
 908         rc = qsd_op_begin(env, qsd, trans, qi, NULL);
 909         if (rc == -EDQUOT || rc == -EINPROGRESS)
 910                 rc = 0;
 911         if (rc)
 912                 return rc;
 913
 914         /* and one less inode for the current id */
 915         qi->lqi_id.qid_uid = orig_id;;
 916         qi->lqi_space      = -1;
 917         /* can't get EDQUOT when reducing usage */
 918         rc = qsd_op_begin(env, qsd, trans, qi, NULL);
 919         if (rc == -EINPROGRESS)
 920                 rc = 0;
 921         if (rc)
 922                 return rc;
 923
 924         /* block accounting */
 925         qi->lqi_is_blk = true;
 926
 927         /* more blocks for the new owner ... */
 928         qi->lqi_id.qid_uid = new_id;
 929         qi->lqi_space      = bspace;
 930         rc = qsd_op_begin(env, qsd, trans, qi, NULL);
 931         if (rc == -EDQUOT || rc == -EINPROGRESS)
 932                 rc = 0;
 933         if (rc)
 934                 return rc;
 935
 936         /* and finally less blocks for the current owner */
 937         qi->lqi_id.qid_uid = orig_id;
 938         qi->lqi_space      = -bspace;
 939         rc = qsd_op_begin(env, qsd, trans, qi, NULL);
 940         /* can't get EDQUOT when reducing usage */
 941         if (rc == -EINPROGRESS)
 942                 rc = 0;
 943         return rc;
 944 }
 945
 946 static int osd_declare_attr_set(const struct lu_env *env,
 947                                 struct dt_object *dt,
 948                                 const struct lu_attr *attr,
 949                                 struct thandle *handle)
 950 {
 951         struct osd_thread_info  *info = osd_oti_get(env);
 952         struct osd_object       *obj = osd_dt_obj(dt);
 953         struct osd_device       *osd = osd_obj2dev(obj);
 954         dmu_tx_hold_t           *txh;
 955         struct osd_thandle      *oh;
 956         uint64_t                 bspace;
 957         uint32_t                 blksize;
 958         int                      rc = 0;
 959         bool                     found;
 960         ENTRY;
 961
 962
 963         LASSERT(handle != NULL);
 964         LASSERT(osd_invariant(obj));
 965
 966         oh = container_of0(handle, struct osd_thandle, ot_super);
 967
 968         down_read(&obj->oo_guard);
 969         if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
 970                 GOTO(out, rc = 0);
 971
 972         LASSERT(obj->oo_sa_hdl != NULL);
 973         LASSERT(oh->ot_tx != NULL);
 974         /* regular attributes are part of the bonus buffer */
 975         /* let's check whether this object is already part of
 976          * transaction.. */
 977         found = false;
 978         for (txh = list_head(&oh->ot_tx->tx_holds); txh;
 979              txh = list_next(&oh->ot_tx->tx_holds, txh)) {
 980                 if (txh->txh_dnode == NULL)
 981                         continue;
 982                 if (txh->txh_dnode->dn_object != obj->oo_dn->dn_object)
 983                         continue;
 984                 /* this object is part of the transaction already
 985                  * we don't need to declare bonus again */
 986                 found = true;
 987                 break;
 988         }
 989         if (!found)
 990                 dmu_tx_hold_bonus(oh->ot_tx, obj->oo_dn->dn_object);
 991         if (oh->ot_tx->tx_err != 0)
 992                 GOTO(out, rc = -oh->ot_tx->tx_err);
 993
 994         if (attr && attr->la_valid & LA_FLAGS) {
 995                 /* LMA is usually a part of bonus, no need to declare
 996                  * anything else */
 997         }
 998
 999         if (attr && (attr->la_valid & (LA_UID | LA_GID | LA_PROJID))) {
1000                 sa_object_size(obj->oo_sa_hdl, &blksize, &bspace);
1001                 bspace = toqb(bspace * blksize);
1002         }
1003
1004         if (attr && attr->la_valid & LA_UID) {
1005                 /* quota enforcement for user */
1006                 if (attr->la_uid != obj->oo_attr.la_uid) {
1007                         rc = qsd_transfer(env, osd->od_quota_slave,
1008                                           &oh->ot_quota_trans, USRQUOTA,
1009                                           obj->oo_attr.la_uid, attr->la_uid,
1010                                           bspace, &info->oti_qi);
1011                         if (rc)
1012                                 GOTO(out, rc);
1013                 }
1014         }
1015         if (attr && attr->la_valid & LA_GID) {
1016                 /* quota enforcement for group */
1017                 if (attr->la_gid != obj->oo_attr.la_gid) {
1018                         rc = qsd_transfer(env, osd->od_quota_slave,
1019                                           &oh->ot_quota_trans, GRPQUOTA,
1020                                           obj->oo_attr.la_gid, attr->la_gid,
1021                                           bspace, &info->oti_qi);
1022                         if (rc)
1023                                 GOTO(out, rc);
1024                 }
1025         }
1026 #ifdef ZFS_PROJINHERIT
1027         if (attr && attr->la_valid & LA_PROJID) {
1028                 if (!osd->od_projectused_dn)
1029                         GOTO(out, rc = -EOPNOTSUPP);
1030
1031                 /* Usually, if project quota is upgradable for the device,
1032                  * then the upgrade will be done before or when mount the
1033                  * device. So when we come here, this project should have
1034                  * project ID attribute already (that is zero by default).
1035                  * Otherwise, there was something wrong during the former
1036                  * upgrade, let's return failure to report that.
1037                  *
1038                  * Please note that, different from other attributes, you
1039                  * can NOT simply set the project ID attribute under such
1040                  * case, because adding (NOT change) project ID attribute
1041                  * needs to change the object's attribute layout to match
1042                  * zfs backend quota accounting requirement. */
1043                 if (unlikely(!obj->oo_with_projid))
1044                         GOTO(out, rc = -ENXIO);
1045
1046                 /* quota enforcement for project */
1047                 if (attr->la_projid != obj->oo_attr.la_projid) {
1048                         rc = qsd_transfer(env, osd->od_quota_slave,
1049                                           &oh->ot_quota_trans, PRJQUOTA,
1050                                           obj->oo_attr.la_projid,
1051                                           attr->la_projid, bspace,
1052                                           &info->oti_qi);
1053                         if (rc)
1054                                 GOTO(out, rc);
1055                 }
1056         }
1057 #endif
1058 out:
1059         up_read(&obj->oo_guard);
1060         RETURN(rc);
1061 }
1062
1063 /*
1064  * Set the attributes of an object
1065  *
1066  * The transaction passed to this routine must have
1067  * dmu_tx_hold_bonus(tx, oid) called and then assigned
1068  * to a transaction group.
1069  */
1070 static int osd_attr_set(const struct lu_env *env, struct dt_object *dt,
1071                         const struct lu_attr *la, struct thandle *handle)
1072 {
1073         struct osd_thread_info  *info = osd_oti_get(env);
1074         sa_bulk_attr_t          *bulk = osd_oti_get(env)->oti_attr_bulk;
1075         struct osd_object       *obj = osd_dt_obj(dt);
1076         struct osd_device       *osd = osd_obj2dev(obj);
1077         struct osd_thandle      *oh;
1078         struct osa_attr         *osa = &info->oti_osa;
1079         __u64                    valid = la->la_valid;
1080         int                      cnt;
1081         int                      rc = 0;
1082
1083         ENTRY;
1084
1085         down_read(&obj->oo_guard);
1086         if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
1087                 GOTO(out, rc = -ENOENT);
1088
1089         LASSERT(handle != NULL);
1090         LASSERT(osd_invariant(obj));
1091         LASSERT(obj->oo_sa_hdl);
1092
1093         oh = container_of0(handle, struct osd_thandle, ot_super);
1094         /* Assert that the transaction has been assigned to a
1095            transaction group. */
1096         LASSERT(oh->ot_tx->tx_txg != 0);
1097
1098         /* Only allow set size for regular file */
1099         if (!S_ISREG(dt->do_lu.lo_header->loh_attr))
1100                 valid &= ~(LA_SIZE | LA_BLOCKS);
1101
1102         if (valid & LA_CTIME && la->la_ctime == obj->oo_attr.la_ctime)
1103                 valid &= ~LA_CTIME;
1104
1105         if (valid & LA_MTIME && la->la_mtime == obj->oo_attr.la_mtime)
1106                 valid &= ~LA_MTIME;
1107
1108         if (valid & LA_ATIME && la->la_atime == obj->oo_attr.la_atime)
1109                 valid &= ~LA_ATIME;
1110
1111         if (valid == 0)
1112                 GOTO(out, rc = 0);
1113
1114         if (valid & LA_FLAGS) {
1115                 struct lustre_mdt_attrs *lma;
1116                 struct lu_buf buf;
1117
1118                 if (la->la_flags & LUSTRE_LMA_FL_MASKS) {
1119                         CLASSERT(sizeof(info->oti_buf) >= sizeof(*lma));
1120                         lma = (struct lustre_mdt_attrs *)&info->oti_buf;
1121                         buf.lb_buf = lma;
1122                         buf.lb_len = sizeof(info->oti_buf);
1123                         rc = osd_xattr_get(env, &obj->oo_dt, &buf,
1124                                            XATTR_NAME_LMA);
1125                         if (rc > 0) {
1126                                 lma->lma_incompat =
1127                                         le32_to_cpu(lma->lma_incompat);
1128                                 lma->lma_incompat |=
1129                                         lustre_to_lma_flags(la->la_flags);
1130                                 lma->lma_incompat =
1131                                         cpu_to_le32(lma->lma_incompat);
1132                                 buf.lb_buf = lma;
1133                                 buf.lb_len = sizeof(*lma);
1134                                 rc = osd_xattr_set_internal(env, obj, &buf,
1135                                                             XATTR_NAME_LMA,
1136                                                             LU_XATTR_REPLACE,
1137                                                             oh);
1138                         }
1139                         if (rc < 0) {
1140                                 CWARN("%s: failed to set LMA flags: rc = %d\n",
1141                                        osd->od_svname, rc);
1142                                 GOTO(out, rc);
1143                         }
1144                 }
1145         }
1146
1147         write_lock(&obj->oo_attr_lock);
1148         cnt = 0;
1149
1150         if (valid & LA_PROJID) {
1151 #ifdef ZFS_PROJINHERIT
1152                 /* osd_declare_attr_set() must be called firstly.
1153                  * If osd::od_projectused_dn is not set, then we
1154                  * can not arrive at here. */
1155                 LASSERT(osd->od_projectused_dn);
1156                 LASSERT(obj->oo_with_projid);
1157
1158                 osa->projid = obj->oo_attr.la_projid = la->la_projid;
1159                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_PROJID(osd), NULL,
1160                                  &osa->projid, 8);
1161 #else
1162                 valid &= ~LA_PROJID;
1163 #endif
1164         }
1165
1166         if (valid & LA_ATIME) {
1167                 osa->atime[0] = obj->oo_attr.la_atime = la->la_atime;
1168                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(osd), NULL,
1169                                  osa->atime, 16);
1170         }
1171         if (valid & LA_MTIME) {
1172                 osa->mtime[0] = obj->oo_attr.la_mtime = la->la_mtime;
1173                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(osd), NULL,
1174                                  osa->mtime, 16);
1175         }
1176         if (valid & LA_CTIME) {
1177                 osa->ctime[0] = obj->oo_attr.la_ctime = la->la_ctime;
1178                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(osd), NULL,
1179                                  osa->ctime, 16);
1180         }
1181         if (valid & LA_MODE) {
1182                 /* mode is stored along with type, so read it first */
1183                 obj->oo_attr.la_mode = (obj->oo_attr.la_mode & S_IFMT) |
1184                         (la->la_mode & ~S_IFMT);
1185                 osa->mode = obj->oo_attr.la_mode;
1186                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(osd), NULL,
1187                                  &osa->mode, 8);
1188         }
1189         if (valid & LA_SIZE) {
1190                 osa->size = obj->oo_attr.la_size = la->la_size;
1191                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(osd), NULL,
1192                                  &osa->size, 8);
1193         }
1194         if (valid & LA_NLINK) {
1195                 osa->nlink = obj->oo_attr.la_nlink = la->la_nlink;
1196                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(osd), NULL,
1197                                  &osa->nlink, 8);
1198         }
1199         if (valid & LA_RDEV) {
1200                 osa->rdev = obj->oo_attr.la_rdev = la->la_rdev;
1201                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_RDEV(osd), NULL,
1202                                  &osa->rdev, 8);
1203         }
1204         if (valid & LA_FLAGS) {
1205                 osa->flags = attrs_fs2zfs(la->la_flags);
1206                 /* many flags are not supported by zfs, so ensure a good cached
1207                  * copy */
1208                 obj->oo_attr.la_flags = attrs_zfs2fs(osa->flags);
1209 #ifdef ZFS_PROJINHERIT
1210                 if (obj->oo_with_projid)
1211                         osa->flags |= ZFS_PROJID;
1212 #endif
1213                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(osd), NULL,
1214                                  &osa->flags, 8);
1215         }
1216         if (valid & LA_UID) {
1217                 osa->uid = obj->oo_attr.la_uid = la->la_uid;
1218                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(osd), NULL,
1219                                  &osa->uid, 8);
1220         }
1221         if (valid & LA_GID) {
1222                 osa->gid = obj->oo_attr.la_gid = la->la_gid;
1223                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(osd), NULL,
1224                                  &osa->gid, 8);
1225         }
1226         obj->oo_attr.la_valid |= valid;
1227         write_unlock(&obj->oo_attr_lock);
1228
1229         LASSERT(cnt <= ARRAY_SIZE(osd_oti_get(env)->oti_attr_bulk));
1230         rc = osd_object_sa_bulk_update(obj, bulk, cnt, oh);
1231
1232 out:
1233         up_read(&obj->oo_guard);
1234         RETURN(rc);
1235 }
1236
1237 /*
1238  * Object creation.
1239  *
1240  * XXX temporary solution.
1241  */
1242
1243 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
1244                         struct dt_object *parent, struct dt_object *child,
1245                         umode_t child_mode)
1246 {
1247         LASSERT(ah);
1248
1249         ah->dah_parent = parent;
1250         ah->dah_mode = child_mode;
1251
1252         if (parent != NULL && !dt_object_remote(parent)) {
1253                 /* will help to find FID->ino at dt_insert("..") */
1254                 struct osd_object *pobj = osd_dt_obj(parent);
1255
1256                 osd_idc_find_and_init(env, osd_obj2dev(pobj), pobj);
1257         }
1258 }
1259
1260 static int osd_declare_create(const struct lu_env *env, struct dt_object *dt,
1261                               struct lu_attr *attr,
1262                               struct dt_allocation_hint *hint,
1263                               struct dt_object_format *dof,
1264                               struct thandle *handle)
1265 {
1266         const struct lu_fid     *fid = lu_object_fid(&dt->do_lu);
1267         struct osd_object       *obj = osd_dt_obj(dt);
1268         struct osd_device       *osd = osd_obj2dev(obj);
1269         struct osd_thandle      *oh;
1270         uint64_t                 zapid;
1271         dnode_t                 *dn;
1272         int                      rc, dnode_size;
1273         ENTRY;
1274
1275         LASSERT(dof);
1276
1277         switch (dof->dof_type) {
1278                 case DFT_REGULAR:
1279                 case DFT_SYM:
1280                 case DFT_NODE:
1281                         if (obj->oo_dt.do_body_ops == NULL)
1282                                 obj->oo_dt.do_body_ops = &osd_body_ops;
1283                         break;
1284                 default:
1285                         break;
1286         }
1287
1288         LASSERT(handle != NULL);
1289         oh = container_of0(handle, struct osd_thandle, ot_super);
1290         LASSERT(oh->ot_tx != NULL);
1291
1292         /* this is the minimum set of EAs on every Lustre object */
1293         obj->oo_ea_in_bonus = ZFS_SA_BASE_ATTR_SIZE +
1294                                 sizeof(__u64) + /* VBR VERSION */
1295                                 sizeof(struct lustre_mdt_attrs); /* LMA */
1296         /* reserve 32 bytes for extra stuff like ACLs */
1297         dnode_size = size_roundup_power2(obj->oo_ea_in_bonus + 32);
1298
1299         switch (dof->dof_type) {
1300                 case DFT_DIR:
1301                         dt->do_index_ops = &osd_dir_ops;
1302                 case DFT_INDEX:
1303                         /* for zap create */
1304                         dmu_tx_hold_zap(oh->ot_tx, DMU_NEW_OBJECT, FALSE, NULL);
1305                         dmu_tx_hold_sa_create(oh->ot_tx, dnode_size);
1306                         break;
1307                 case DFT_REGULAR:
1308                 case DFT_SYM:
1309                 case DFT_NODE:
1310                         /* first, we'll create new object */
1311                         dmu_tx_hold_sa_create(oh->ot_tx, dnode_size);
1312                         break;
1313
1314                 default:
1315                         LBUG();
1316                         break;
1317         }
1318
1319         /* and we'll add it to some mapping */
1320         zapid = osd_get_name_n_idx(env, osd, fid, NULL, 0, &dn);
1321         osd_tx_hold_zap(oh->ot_tx, zapid, dn, TRUE, NULL);
1322
1323         /* will help to find FID->ino mapping at dt_insert() */
1324         osd_idc_find_and_init(env, osd, obj);
1325
1326         rc = osd_declare_quota(env, osd, attr->la_uid, attr->la_gid,
1327                                attr->la_projid, 1, oh, NULL, OSD_QID_INODE);
1328
1329         RETURN(rc);
1330 }
1331
1332 int __osd_attr_init(const struct lu_env *env, struct osd_device *osd,
1333                     struct osd_object *obj, sa_handle_t *sa_hdl, dmu_tx_t *tx,
1334                     struct lu_attr *la, uint64_t parent,
1335                     nvlist_t *xattr)
1336 {
1337         sa_bulk_attr_t  *bulk = osd_oti_get(env)->oti_attr_bulk;
1338         struct osa_attr *osa = &osd_oti_get(env)->oti_osa;
1339         uint64_t         gen;
1340         uint64_t         crtime[2];
1341         timestruc_t      now;
1342         int              cnt;
1343         int              rc;
1344         char *dxattr = NULL;
1345         size_t sa_size;
1346
1347
1348         LASSERT(sa_hdl);
1349
1350         gen = dmu_tx_get_txg(tx);
1351         gethrestime(&now);
1352         ZFS_TIME_ENCODE(&now, crtime);
1353
1354         osa->atime[0] = la->la_atime;
1355         osa->ctime[0] = la->la_ctime;
1356         osa->mtime[0] = la->la_mtime;
1357         osa->mode = la->la_mode;
1358         osa->uid = la->la_uid;
1359         osa->gid = la->la_gid;
1360         osa->rdev = la->la_rdev;
1361         osa->nlink = la->la_nlink;
1362         if (la->la_valid & LA_FLAGS)
1363                 osa->flags = attrs_fs2zfs(la->la_flags);
1364         else
1365                 osa->flags = 0;
1366         osa->size  = la->la_size;
1367 #ifdef ZFS_PROJINHERIT
1368         if (osd->od_projectused_dn) {
1369                 if (la->la_valid & LA_PROJID)
1370                         osa->projid = la->la_projid;
1371                 else
1372                         osa->projid = ZFS_DEFAULT_PROJID;
1373                 osa->flags |= ZFS_PROJID;
1374                 if (obj)
1375                         obj->oo_with_projid = 1;
1376         } else {
1377                 osa->flags &= ~ZFS_PROJID;
1378         }
1379 #endif
1380
1381         /*
1382          * we need to create all SA below upon object create.
1383          *
1384          * XXX The attribute order matters since the accounting callback relies
1385          * on static offsets (i.e. SA_*_OFFSET, see zfs_space_delta_cb()) to
1386          * look up the UID/GID/PROJID attributes. Moreover, the callback does
1387          * not seem to support the spill block.
1388          * We define attributes in the same order as SA_*_OFFSET in order to
1389          * work around the problem. See ORI-610.
1390          */
1391         cnt = 0;
1392         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(osd), NULL, &osa->mode, 8);
1393         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(osd), NULL, &osa->size, 8);
1394         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GEN(osd), NULL, &gen, 8);
1395         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(osd), NULL, &osa->uid, 8);
1396         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(osd), NULL, &osa->gid, 8);
1397         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_PARENT(osd), NULL, &parent, 8);
1398         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(osd), NULL, &osa->flags, 8);
1399         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(osd), NULL, osa->atime, 16);
1400         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(osd), NULL, osa->mtime, 16);
1401         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(osd), NULL, osa->ctime, 16);
1402         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CRTIME(osd), NULL, crtime, 16);
1403         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(osd), NULL, &osa->nlink, 8);
1404 #ifdef ZFS_PROJINHERIT
1405         if (osd->od_projectused_dn)
1406                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_PROJID(osd), NULL,
1407                                  &osa->projid, 8);
1408 #endif
1409         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_RDEV(osd), NULL, &osa->rdev, 8);
1410         LASSERT(cnt <= ARRAY_SIZE(osd_oti_get(env)->oti_attr_bulk));
1411
1412         if (xattr) {
1413                 rc = -nvlist_size(xattr, &sa_size, NV_ENCODE_XDR);
1414                 LASSERT(rc == 0);
1415
1416                 dxattr = osd_zio_buf_alloc(sa_size);
1417                 LASSERT(dxattr);
1418
1419                 rc = -nvlist_pack(xattr, &dxattr, &sa_size,
1420                                 NV_ENCODE_XDR, KM_SLEEP);
1421                 LASSERT(rc == 0);
1422
1423                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_DXATTR(osd),
1424                                 NULL, dxattr, sa_size);
1425         }
1426
1427         rc = -sa_replace_all_by_template(sa_hdl, bulk, cnt, tx);
1428         if (dxattr)
1429                 osd_zio_buf_free(dxattr, sa_size);
1430
1431         return rc;
1432 }
1433
1434 static int osd_find_new_dnode(const struct lu_env *env, dmu_tx_t *tx,
1435                               uint64_t oid, dnode_t **dnp)
1436 {
1437         dmu_tx_hold_t *txh;
1438         int rc = 0;
1439
1440         /* take dnode_t from tx to save on dnode#->dnode_t lookup */
1441         for (txh = list_tail(&tx->tx_holds); txh;
1442              txh = list_prev(&tx->tx_holds, txh)) {
1443                 dnode_t *dn = txh->txh_dnode;
1444                 dmu_buf_impl_t *db;
1445
1446                 if (dn == NULL)
1447                         continue;
1448                 if (dn->dn_object != oid)
1449                         continue;
1450                 db = dn->dn_bonus;
1451                 if (db == NULL) {
1452                         rw_enter(&dn->dn_struct_rwlock, RW_WRITER);
1453                         if (dn->dn_bonus == NULL)
1454                                 dbuf_create_bonus(dn);
1455                         rw_exit(&dn->dn_struct_rwlock);
1456                 }
1457                 db = dn->dn_bonus;
1458                 LASSERT(db);
1459                 LASSERT(dn->dn_handle);
1460                 DB_DNODE_ENTER(db);
1461                 if (refcount_add(&db->db_holds, osd_obj_tag) == 1) {
1462                         refcount_add(&dn->dn_holds, tag);
1463                         atomic_inc_32(&dn->dn_dbufs_count);
1464                 }
1465                 *dnp = dn;
1466                 dbuf_read(db, NULL, DB_RF_MUST_SUCCEED | DB_RF_NOPREFETCH);
1467                 break;
1468         }
1469
1470         if (unlikely(*dnp == NULL))
1471                 rc = __osd_obj2dnode(tx->tx_objset, oid, dnp);
1472
1473         return rc;
1474 }
1475
1476 #ifdef HAVE_DMU_OBJECT_ALLOC_DNSIZE
1477 static int osd_find_dnsize(struct osd_object *obj)
1478 {
1479         struct osd_device *osd = osd_obj2dev(obj);
1480         int dnsize;
1481
1482         if (osd->od_dnsize == ZFS_DNSIZE_AUTO) {
1483                 dnsize = DNODE_MIN_SIZE;
1484                 do {
1485                         if (DN_BONUS_SIZE(dnsize) >= obj->oo_ea_in_bonus + 32)
1486                                 break;
1487                         dnsize <<= 1;
1488                 } while (dnsize < DNODE_MAX_SIZE);
1489                 if (dnsize > DNODE_MAX_SIZE)
1490                         dnsize = DNODE_MAX_SIZE;
1491         } else if (osd->od_dnsize == ZFS_DNSIZE_1K) {
1492                 dnsize = 1024;
1493         } else if (osd->od_dnsize == ZFS_DNSIZE_2K) {
1494                 dnsize = 2048;
1495         } else if (osd->od_dnsize == ZFS_DNSIZE_4K) {
1496                 dnsize = 4096;
1497         } else if (osd->od_dnsize == ZFS_DNSIZE_8K) {
1498                 dnsize = 8192;
1499         } else if (osd->od_dnsize == ZFS_DNSIZE_16K) {
1500                 dnsize = 16384;
1501         } else {
1502                 dnsize = DNODE_MIN_SIZE;
1503         }
1504         return dnsize;
1505 }
1506 #else
1507 static int inline osd_find_dnsize(struct osd_object *obj)
1508 {
1509         return DN_MAX_BONUSLEN;
1510 }
1511 #endif
1512
1513 /*
1514  * The transaction passed to this routine must have
1515  * dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT) called and then assigned
1516  * to a transaction group.
1517  */
1518 int __osd_object_create(const struct lu_env *env, struct osd_object *obj,
1519                         dnode_t **dnp, dmu_tx_t *tx, struct lu_attr *la)
1520 {
1521         struct osd_device   *osd = osd_obj2dev(obj);
1522         const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
1523         dmu_object_type_t    type = DMU_OT_PLAIN_FILE_CONTENTS;
1524         uint64_t oid;
1525
1526         /* Use DMU_OTN_UINT8_METADATA for local objects so their data blocks
1527          * would get an additional ditto copy */
1528         if (unlikely(S_ISREG(la->la_mode) &&
1529                      fid_seq_is_local_file(fid_seq(fid))))
1530                 type = DMU_OTN_UINT8_METADATA;
1531
1532         /* Create a new DMU object using the default dnode size. */
1533         oid = osd_dmu_object_alloc(osd->od_os, type, 0,
1534                                    osd_find_dnsize(obj), tx);
1535
1536         LASSERT(la->la_valid & LA_MODE);
1537         la->la_size = 0;
1538         la->la_nlink = 1;
1539
1540         return osd_find_new_dnode(env, tx, oid, dnp);
1541 }
1542
1543 /*
1544  * The transaction passed to this routine must have
1545  * dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, ...) called and then assigned
1546  * to a transaction group.
1547  *
1548  * Using ZAP_FLAG_HASH64 will force the ZAP to always be a FAT ZAP.
1549  * This is fine for directories today, because storing the FID in the dirent
1550  * will also require a FAT ZAP.  If there is a new type of micro ZAP created
1551  * then we might need to re-evaluate the use of this flag and instead do
1552  * a conversion from the different internal ZAP hash formats being used. */
1553 int __osd_zap_create(const struct lu_env *env, struct osd_device *osd,
1554                      dnode_t **dnp, dmu_tx_t *tx, struct lu_attr *la,
1555                      unsigned dnsize, zap_flags_t flags)
1556 {
1557         uint64_t oid;
1558
1559         /* Assert that the transaction has been assigned to a
1560            transaction group. */
1561         LASSERT(tx->tx_txg != 0);
1562         *dnp = NULL;
1563
1564         oid = osd_zap_create_flags(osd->od_os, 0, flags | ZAP_FLAG_HASH64,
1565                                    DMU_OT_DIRECTORY_CONTENTS,
1566                                    14, /* == ZFS fzap_default_blockshift */
1567                                    DN_MAX_INDBLKSHIFT, /* indirect blockshift */
1568                                    dnsize, tx);
1569
1570         la->la_size = 2;
1571         la->la_nlink = 1;
1572
1573         return osd_find_new_dnode(env, tx, oid, dnp);
1574 }
1575
1576 static dnode_t *osd_mkidx(const struct lu_env *env, struct osd_object *obj,
1577                           struct lu_attr *la, struct osd_thandle *oh)
1578 {
1579         dnode_t *dn;
1580         int rc;
1581
1582         /* Index file should be created as regular file in order not to confuse
1583          * ZPL which could interpret them as directory.
1584          * We set ZAP_FLAG_UINT64_KEY to let ZFS know than we are going to use
1585          * binary keys */
1586         LASSERT(S_ISREG(la->la_mode));
1587         rc = __osd_zap_create(env, osd_obj2dev(obj), &dn, oh->ot_tx, la,
1588                               osd_find_dnsize(obj), ZAP_FLAG_UINT64_KEY);
1589         if (rc)
1590                 return ERR_PTR(rc);
1591         return dn;
1592 }
1593
1594 static dnode_t *osd_mkdir(const struct lu_env *env, struct osd_object *obj,
1595                           struct lu_attr *la, struct osd_thandle *oh)
1596 {
1597         dnode_t *dn;
1598         int rc;
1599
1600         LASSERT(S_ISDIR(la->la_mode));
1601         rc = __osd_zap_create(env, osd_obj2dev(obj), &dn, oh->ot_tx, la,
1602                               osd_find_dnsize(obj), 0);
1603         if (rc)
1604                 return ERR_PTR(rc);
1605         return dn;
1606 }
1607
1608 static dnode_t *osd_mkreg(const struct lu_env *env, struct osd_object *obj,
1609                           struct lu_attr *la, struct osd_thandle *oh)
1610 {
1611         const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
1612         struct osd_device *osd = osd_obj2dev(obj);
1613         dnode_t *dn;
1614         int rc;
1615
1616         LASSERT(S_ISREG(la->la_mode));
1617         rc = __osd_object_create(env, obj, &dn, oh->ot_tx, la);
1618         if (rc)
1619                 return ERR_PTR(rc);
1620
1621         if ((fid_is_idif(fid) || fid_is_norm(fid) || fid_is_echo(fid))) {
1622                 /* The minimum block size must be at least page size otherwise
1623                  * it will break the assumption in tgt_thread_big_cache where
1624                  * the array size is PTLRPC_MAX_BRW_PAGES. It will also affect
1625                  * RDMA due to subpage transfer size */
1626                 rc = -dmu_object_set_blocksize(osd->od_os, dn->dn_object,
1627                                                PAGE_SIZE, 0, oh->ot_tx);
1628                 if (unlikely(rc)) {
1629                         CERROR("%s: can't change blocksize: %d\n",
1630                                osd->od_svname, rc);
1631                         return ERR_PTR(rc);
1632                 }
1633         }
1634
1635         return dn;
1636 }
1637
1638 static dnode_t *osd_mksym(const struct lu_env *env, struct osd_object *obj,
1639                           struct lu_attr *la, struct osd_thandle *oh)
1640 {
1641         dnode_t *dn;
1642         int rc;
1643
1644         LASSERT(S_ISLNK(la->la_mode));
1645         rc = __osd_object_create(env, obj, &dn, oh->ot_tx, la);
1646         if (rc)
1647                 return ERR_PTR(rc);
1648         return dn;
1649 }
1650
1651 static dnode_t *osd_mknod(const struct lu_env *env, struct osd_object *obj,
1652                           struct lu_attr *la, struct osd_thandle *oh)
1653 {
1654         dnode_t *dn;
1655         int rc;
1656
1657         if (S_ISCHR(la->la_mode) || S_ISBLK(la->la_mode))
1658                 la->la_valid |= LA_RDEV;
1659
1660         rc = __osd_object_create(env, obj, &dn, oh->ot_tx, la);
1661         if (rc)
1662                 return ERR_PTR(rc);
1663         return dn;
1664 }
1665
1666 typedef dnode_t *(*osd_obj_type_f)(const struct lu_env *env,
1667                                    struct osd_object *obj,
1668                                    struct lu_attr *la,
1669                                    struct osd_thandle *oh);
1670
1671 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
1672 {
1673         osd_obj_type_f result;
1674
1675         switch (type) {
1676         case DFT_DIR:
1677                 result = osd_mkdir;
1678                 break;
1679         case DFT_INDEX:
1680                 result = osd_mkidx;
1681                 break;
1682         case DFT_REGULAR:
1683                 result = osd_mkreg;
1684                 break;
1685         case DFT_SYM:
1686                 result = osd_mksym;
1687                 break;
1688         case DFT_NODE:
1689                 result = osd_mknod;
1690                 break;
1691         default:
1692                 LBUG();
1693                 break;
1694         }
1695         return result;
1696 }
1697
1698 /*
1699  * Concurrency: @dt is write locked.
1700  */
1701 static int osd_create(const struct lu_env *env, struct dt_object *dt,
1702                       struct lu_attr *attr, struct dt_allocation_hint *hint,
1703                       struct dt_object_format *dof, struct thandle *th)
1704 {
1705         struct osd_thread_info  *info = osd_oti_get(env);
1706         struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
1707         struct zpl_direntry     *zde = &info->oti_zde.lzd_reg;
1708         const struct lu_fid     *fid = lu_object_fid(&dt->do_lu);
1709         struct osd_object       *obj = osd_dt_obj(dt);
1710         struct osd_device       *osd = osd_obj2dev(obj);
1711         char                    *buf = info->oti_str;
1712         struct osd_thandle      *oh;
1713         dnode_t *dn = NULL, *zdn = NULL;
1714         uint64_t                 zapid, parent = 0;
1715         int                      rc;
1716
1717         ENTRY;
1718
1719         LASSERT(!fid_is_acct(fid));
1720
1721         /* concurrent create declarations should not see
1722          * the object inconsistent (db, attr, etc).
1723          * in regular cases acquisition should be cheap */
1724         down_write(&obj->oo_guard);
1725
1726         if (unlikely(dt_object_exists(dt)))
1727                 GOTO(out, rc = -EEXIST);
1728
1729         LASSERT(osd_invariant(obj));
1730         LASSERT(dof != NULL);
1731
1732         LASSERT(th != NULL);
1733         oh = container_of0(th, struct osd_thandle, ot_super);
1734
1735         LASSERT(obj->oo_dn == NULL);
1736
1737         /* to follow ZFS on-disk format we need
1738          * to initialize parent dnode properly */
1739         if (hint != NULL && hint->dah_parent != NULL &&
1740             !dt_object_remote(hint->dah_parent))
1741                 parent = osd_dt_obj(hint->dah_parent)->oo_dn->dn_object;
1742
1743         /* we may fix some attributes, better do not change the source */
1744         obj->oo_attr = *attr;
1745         obj->oo_attr.la_valid |= LA_SIZE | LA_NLINK | LA_TYPE;
1746
1747 #ifdef ZFS_PROJINHERIT
1748         if (osd->od_projectused_dn) {
1749                 if (!(obj->oo_attr.la_valid & LA_PROJID))
1750                         obj->oo_attr.la_projid = ZFS_DEFAULT_PROJID;
1751                 obj->oo_with_projid = 1;
1752         }
1753 #endif
1754
1755         dn = osd_create_type_f(dof->dof_type)(env, obj, &obj->oo_attr, oh);
1756         if (IS_ERR(dn)) {
1757                 rc = PTR_ERR(dn);
1758                 dn = NULL;
1759                 GOTO(out, rc);
1760         }
1761
1762         zde->zde_pad = 0;
1763         zde->zde_dnode = dn->dn_object;
1764         zde->zde_type = IFTODT(attr->la_mode & S_IFMT);
1765
1766         zapid = osd_get_name_n_idx(env, osd, fid, buf,
1767                                    sizeof(info->oti_str), &zdn);
1768         rc = osd_zap_add(osd, zapid, zdn, buf, 8, 1, zde, oh->ot_tx);
1769         if (rc)
1770                 GOTO(out, rc);
1771         obj->oo_dn = dn;
1772         /* Now add in all of the "SA" attributes */
1773         rc = osd_sa_handle_get(obj);
1774         if (rc)
1775                 GOTO(out, rc);
1776
1777         rc = -nvlist_alloc(&obj->oo_sa_xattr, NV_UNIQUE_NAME, KM_SLEEP);
1778         if (rc)
1779                 GOTO(out, rc);
1780
1781         /* initialize LMA */
1782         lustre_lma_init(lma, fid, 0, 0);
1783         lustre_lma_swab(lma);
1784         rc = -nvlist_add_byte_array(obj->oo_sa_xattr, XATTR_NAME_LMA,
1785                                     (uchar_t *)lma, sizeof(*lma));
1786         if (rc)
1787                 GOTO(out, rc);
1788
1789         /* configure new osd object */
1790         obj->oo_parent = parent != 0 ? parent : zapid;
1791         obj->oo_late_attr_set = 1;
1792         rc = __osd_sa_xattr_schedule_update(env, obj, oh);
1793         if (rc)
1794                 GOTO(out, rc);
1795
1796         /* XXX: oo_lma_flags */
1797         obj->oo_dt.do_lu.lo_header->loh_attr |= obj->oo_attr.la_mode & S_IFMT;
1798         if (likely(!fid_is_acct(lu_object_fid(&obj->oo_dt.do_lu))))
1799                 /* no body operations for accounting objects */
1800                 obj->oo_dt.do_body_ops = &osd_body_ops;
1801
1802         osd_idc_find_and_init(env, osd, obj);
1803
1804 out:
1805         if (unlikely(rc && dn)) {
1806                 dmu_object_free(osd->od_os, dn->dn_object, oh->ot_tx);
1807                 osd_dnode_rele(dn);
1808                 obj->oo_dn = NULL;
1809         } else if (!rc) {
1810                 obj->oo_dt.do_lu.lo_header->loh_attr |= LOHA_EXISTS;
1811         }
1812         up_write(&obj->oo_guard);
1813         RETURN(rc);
1814 }
1815
1816 static int osd_declare_ref_add(const struct lu_env *env, struct dt_object *dt,
1817                                struct thandle *th)
1818 {
1819         return osd_declare_attr_set(env, dt, NULL, th);
1820 }
1821
1822 /*
1823  * Concurrency: @dt is write locked.
1824  */
1825 static int osd_ref_add(const struct lu_env *env, struct dt_object *dt,
1826                        struct thandle *handle)
1827 {
1828         struct osd_object       *obj = osd_dt_obj(dt);
1829         struct osd_thandle      *oh;
1830         struct osd_device       *osd = osd_obj2dev(obj);
1831         uint64_t                 nlink;
1832         int rc;
1833
1834         ENTRY;
1835
1836         down_read(&obj->oo_guard);
1837         if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
1838                 GOTO(out, rc = -ENOENT);
1839
1840         LASSERT(osd_invariant(obj));
1841         LASSERT(obj->oo_sa_hdl != NULL);
1842
1843         oh = container_of0(handle, struct osd_thandle, ot_super);
1844
1845         write_lock(&obj->oo_attr_lock);
1846         nlink = ++obj->oo_attr.la_nlink;
1847         write_unlock(&obj->oo_attr_lock);
1848
1849         rc = osd_object_sa_update(obj, SA_ZPL_LINKS(osd), &nlink, 8, oh);
1850
1851 out:
1852         up_read(&obj->oo_guard);
1853         RETURN(rc);
1854 }
1855
1856 static int osd_declare_ref_del(const struct lu_env *env, struct dt_object *dt,
1857                                struct thandle *handle)
1858 {
1859         return osd_declare_attr_set(env, dt, NULL, handle);
1860 }
1861
1862 /*
1863  * Concurrency: @dt is write locked.
1864  */
1865 static int osd_ref_del(const struct lu_env *env, struct dt_object *dt,
1866                        struct thandle *handle)
1867 {
1868         struct osd_object       *obj = osd_dt_obj(dt);
1869         struct osd_thandle      *oh;
1870         struct osd_device       *osd = osd_obj2dev(obj);
1871         uint64_t                 nlink;
1872         int                      rc;
1873
1874         ENTRY;
1875
1876         down_read(&obj->oo_guard);
1877
1878         if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
1879                 GOTO(out, rc = -ENOENT);
1880
1881         LASSERT(osd_invariant(obj));
1882         LASSERT(obj->oo_sa_hdl != NULL);
1883
1884         oh = container_of0(handle, struct osd_thandle, ot_super);
1885         LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
1886
1887         write_lock(&obj->oo_attr_lock);
1888         nlink = --obj->oo_attr.la_nlink;
1889         write_unlock(&obj->oo_attr_lock);
1890
1891         rc = osd_object_sa_update(obj, SA_ZPL_LINKS(osd), &nlink, 8, oh);
1892
1893 out:
1894         up_read(&obj->oo_guard);
1895         RETURN(rc);
1896 }
1897
1898 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
1899                            __u64 start, __u64 end)
1900 {
1901         struct osd_device *osd = osd_obj2dev(osd_dt_obj(dt));
1902         ENTRY;
1903
1904         /* XXX: no other option than syncing the whole filesystem until we
1905          * support ZIL.  If the object tracked the txg that it was last
1906          * modified in, it could pass that txg here instead of "0".  Maybe
1907          * the changes are already committed, so no wait is needed at all? */
1908         if (!osd->od_dt_dev.dd_rdonly)
1909                 txg_wait_synced(dmu_objset_pool(osd->od_os), 0ULL);
1910
1911         RETURN(0);
1912 }
1913
1914 static int osd_invalidate(const struct lu_env *env, struct dt_object *dt)
1915 {
1916         return 0;
1917 }
1918
1919 static struct dt_object_operations osd_obj_ops = {
1920         .do_read_lock           = osd_read_lock,
1921         .do_write_lock          = osd_write_lock,
1922         .do_read_unlock         = osd_read_unlock,
1923         .do_write_unlock        = osd_write_unlock,
1924         .do_write_locked        = osd_write_locked,
1925         .do_attr_get            = osd_attr_get,
1926         .do_declare_attr_set    = osd_declare_attr_set,
1927         .do_attr_set            = osd_attr_set,
1928         .do_ah_init             = osd_ah_init,
1929         .do_declare_create      = osd_declare_create,
1930         .do_create              = osd_create,
1931         .do_declare_destroy     = osd_declare_destroy,
1932         .do_destroy             = osd_destroy,
1933         .do_index_try           = osd_index_try,
1934         .do_declare_ref_add     = osd_declare_ref_add,
1935         .do_ref_add             = osd_ref_add,
1936         .do_declare_ref_del     = osd_declare_ref_del,
1937         .do_ref_del             = osd_ref_del,
1938         .do_xattr_get           = osd_xattr_get,
1939         .do_declare_xattr_set   = osd_declare_xattr_set,
1940         .do_xattr_set           = osd_xattr_set,
1941         .do_declare_xattr_del   = osd_declare_xattr_del,
1942         .do_xattr_del           = osd_xattr_del,
1943         .do_xattr_list          = osd_xattr_list,
1944         .do_object_sync         = osd_object_sync,
1945         .do_invalidate          = osd_invalidate,
1946 };
1947
1948 static struct lu_object_operations osd_lu_obj_ops = {
1949         .loo_object_init        = osd_object_init,
1950         .loo_object_delete      = osd_object_delete,
1951         .loo_object_release     = osd_object_release,
1952         .loo_object_free        = osd_object_free,
1953         .loo_object_print       = osd_object_print,
1954         .loo_object_invariant   = osd_object_invariant,
1955 };
1956
1957 static int osd_otable_it_attr_get(const struct lu_env *env,
1958                                 struct dt_object *dt,
1959                                 struct lu_attr *attr)
1960 {
1961         attr->la_valid = 0;
1962         return 0;
1963 }
1964
1965 static struct dt_object_operations osd_obj_otable_it_ops = {
1966         .do_attr_get            = osd_otable_it_attr_get,
1967         .do_index_try           = osd_index_try,
1968 };