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, 2017, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  *
  31  * lustre/osd-zfs/osd_object.c
  32  *
  33  * Author: Alex Zhuravlev <bzzz@whamcloud.com>
  34  * Author: Mike Pershin <tappro@whamcloud.com>
  35  * Author: Johann Lombardi <johann@whamcloud.com>
  36  */
  37
  38 #define DEBUG_SUBSYSTEM S_OSD
  39
  40 #include <libcfs/libcfs.h>
  41 #include <obd_support.h>
  42 #include <lustre_net.h>
  43 #include <obd.h>
  44 #include <obd_class.h>
  45 #include <lustre_disk.h>
  46 #include <lustre_fid.h>
  47
  48 #include "osd_internal.h"
  49
  50 #include <sys/dnode.h>
  51 #include <sys/dbuf.h>
  52 #include <sys/spa.h>
  53 #include <sys/stat.h>
  54 #include <sys/zap.h>
  55 #include <sys/spa_impl.h>
  56 #include <sys/zfs_znode.h>
  57 #include <sys/dmu_tx.h>
  58 #include <sys/dmu_objset.h>
  59 #include <sys/dsl_prop.h>
  60 #include <sys/sa_impl.h>
  61 #include <sys/txg.h>
  62
  63 char *osd_obj_tag = "osd_object";
  64 static int osd_object_sync_delay_us = -1;
  65
  66 static const struct dt_object_operations osd_obj_ops;
  67 static const struct lu_object_operations osd_lu_obj_ops;
  68 static const struct dt_object_operations osd_obj_otable_it_ops;
  69
  70 static void
  71 osd_object_sa_fini(struct osd_object *obj)
  72 {
  73         if (obj->oo_sa_hdl) {
  74                 sa_handle_destroy(obj->oo_sa_hdl);
  75                 obj->oo_sa_hdl = NULL;
  76         }
  77 }
  78
  79 static int
  80 osd_object_sa_init(struct osd_object *obj, struct osd_device *o)
  81 {
  82         int rc;
  83
  84         LASSERT(obj->oo_sa_hdl == NULL);
  85         LASSERT(obj->oo_dn != NULL);
  86
  87         rc = osd_sa_handle_get(obj);
  88         if (rc)
  89                 return rc;
  90
  91         /* Cache the xattr object id, valid for the life of the object */
  92         rc = -sa_lookup(obj->oo_sa_hdl, SA_ZPL_XATTR(o), &obj->oo_xattr, 8);
  93         if (rc == -ENOENT) {
  94                 obj->oo_xattr = ZFS_NO_OBJECT;
  95                 rc = 0;
  96         } else if (rc) {
  97                 osd_object_sa_fini(obj);
  98         }
  99
 100         return rc;
 101 }
 102
 103 /*
 104  * Add object to list of dirty objects in tx handle.
 105  */
 106 void osd_object_sa_dirty_add(struct osd_object *obj, struct osd_thandle *oh)
 107 {
 108         if (!list_empty(&obj->oo_sa_linkage))
 109                 return;
 110
 111         write_lock(&obj->oo_attr_lock);
 112         if (likely(list_empty(&obj->oo_sa_linkage)))
 113                 list_add(&obj->oo_sa_linkage, &oh->ot_sa_list);
 114         write_unlock(&obj->oo_attr_lock);
 115 }
 116
 117 /*
 118  * Release spill block dbuf hold for all dirty SAs.
 119  */
 120 void osd_object_sa_dirty_rele(const struct lu_env *env, struct osd_thandle *oh)
 121 {
 122         struct osd_object *obj;
 123
 124         while (!list_empty(&oh->ot_sa_list)) {
 125                 obj = list_entry(oh->ot_sa_list.next,
 126                                  struct osd_object, oo_sa_linkage);
 127                 write_lock(&obj->oo_attr_lock);
 128                 list_del_init(&obj->oo_sa_linkage);
 129                 write_unlock(&obj->oo_attr_lock);
 130                 if (obj->oo_late_xattr && obj->oo_destroyed == 0) {
 131                         /*
 132                          * take oo_guard to protect oo_sa_xattr buffer
 133                          * from concurrent update by osd_xattr_set()
 134                          */
 135                         LASSERT(oh->ot_assigned != 0);
 136                         down_write(&obj->oo_guard);
 137                         if (obj->oo_destroyed == 0) {
 138                                 if (obj->oo_late_attr_set)
 139                                         __osd_sa_attr_init(env, obj, oh);
 140                                 else if (obj->oo_late_xattr)
 141                                         __osd_sa_xattr_update(env, obj, oh);
 142                         }
 143                         up_write(&obj->oo_guard);
 144                 }
 145                 sa_spill_rele(obj->oo_sa_hdl);
 146         }
 147 }
 148
 149 /*
 150  * Update the SA and add the object to the dirty list.
 151  */
 152 int osd_object_sa_update(struct osd_object *obj, sa_attr_type_t type,
 153                          void *buf, uint32_t buflen, struct osd_thandle *oh)
 154 {
 155         int rc;
 156
 157         LASSERT(obj->oo_sa_hdl != NULL);
 158         LASSERT(oh->ot_tx != NULL);
 159
 160         rc = -sa_update(obj->oo_sa_hdl, type, buf, buflen, oh->ot_tx);
 161         osd_object_sa_dirty_add(obj, oh);
 162
 163         return rc;
 164 }
 165
 166 /*
 167  * Bulk update the SA and add the object to the dirty list.
 168  */
 169 static int
 170 osd_object_sa_bulk_update(struct osd_object *obj, sa_bulk_attr_t *attrs,
 171                           int count, struct osd_thandle *oh)
 172 {
 173         int rc;
 174
 175         LASSERT(obj->oo_sa_hdl != NULL);
 176         LASSERT(oh->ot_tx != NULL);
 177
 178         rc = -sa_bulk_update(obj->oo_sa_hdl, attrs, count, oh->ot_tx);
 179         osd_object_sa_dirty_add(obj, oh);
 180
 181         return rc;
 182 }
 183
 184 /*
 185  * Retrieve the attributes of a DMU object
 186  */
 187 static int __osd_object_attr_get(const struct lu_env *env, struct osd_device *o,
 188                                  struct osd_object *obj, struct lu_attr *la)
 189 {
 190         struct osa_attr *osa = &osd_oti_get(env)->oti_osa;
 191         sa_bulk_attr_t *bulk = osd_oti_get(env)->oti_attr_bulk;
 192         struct lustre_mdt_attrs *lma;
 193         struct lu_buf buf;
 194         int cnt = 0;
 195         int              rc;
 196         ENTRY;
 197
 198         LASSERT(obj->oo_dn != NULL);
 199
 200         la->la_valid |= LA_ATIME | LA_MTIME | LA_CTIME | LA_BTIME | LA_MODE |
 201                         LA_TYPE | LA_SIZE | LA_UID | LA_GID | LA_FLAGS |
 202                         LA_NLINK;
 203
 204         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(o), NULL, osa->atime, 16);
 205         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(o), NULL, osa->mtime, 16);
 206         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(o), NULL, osa->ctime, 16);
 207         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CRTIME(o), NULL, osa->btime, 16);
 208         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(o), NULL, &osa->mode, 8);
 209         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(o), NULL, &osa->size, 8);
 210         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(o), NULL, &osa->nlink, 8);
 211         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(o), NULL, &osa->uid, 8);
 212         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(o), NULL, &osa->gid, 8);
 213         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(o), NULL, &osa->flags, 8);
 214         LASSERT(cnt <= ARRAY_SIZE(osd_oti_get(env)->oti_attr_bulk));
 215
 216         rc = -sa_bulk_lookup(obj->oo_sa_hdl, bulk, cnt);
 217         if (rc)
 218                 GOTO(out_sa, rc);
 219
 220 #ifdef ZFS_PROJINHERIT
 221         if (o->od_projectused_dn && osa->flags & ZFS_PROJID) {
 222                 rc = -sa_lookup(obj->oo_sa_hdl, SA_ZPL_PROJID(o),
 223                                 &osa->projid, 8);
 224                 if (rc)
 225                         GOTO(out_sa, rc);
 226
 227                 la->la_projid = osa->projid;
 228                 la->la_valid |= LA_PROJID;
 229                 obj->oo_with_projid = 1;
 230         } else {
 231                 la->la_projid = ZFS_DEFAULT_PROJID;
 232                 la->la_valid &= ~LA_PROJID;
 233         }
 234 #else
 235         la->la_projid = 0;
 236         la->la_valid &= ~LA_PROJID;
 237 #endif
 238
 239         la->la_atime = osa->atime[0];
 240         la->la_mtime = osa->mtime[0];
 241         la->la_ctime = osa->ctime[0];
 242         la->la_btime = osa->btime[0];
 243         la->la_mode = osa->mode;
 244         la->la_uid = osa->uid;
 245         la->la_gid = osa->gid;
 246         la->la_nlink = osa->nlink;
 247         la->la_flags = attrs_zfs2fs(osa->flags);
 248         la->la_size = osa->size;
 249
 250         /* Try to get extra flags from LMA */
 251         lma = (struct lustre_mdt_attrs *)osd_oti_get(env)->oti_buf;
 252         buf.lb_buf = lma;
 253         buf.lb_len = sizeof(osd_oti_get(env)->oti_buf);
 254         down_read(&obj->oo_guard);
 255         rc = osd_xattr_get_lma(env, obj, &buf);
 256         if (!rc) {
 257                 lma->lma_incompat = le32_to_cpu(lma->lma_incompat);
 258                 obj->oo_lma_flags =
 259                         lma_to_lustre_flags(lma->lma_incompat);
 260         } else if (rc == -ENODATA ||
 261                    !(S_ISDIR(la->la_mode) &&
 262                      dt_object_exists(&obj->oo_dt))) {
 263                 rc = 0;
 264         }
 265         up_read(&obj->oo_guard);
 266
 267         if (S_ISCHR(la->la_mode) || S_ISBLK(la->la_mode)) {
 268                 rc = -sa_lookup(obj->oo_sa_hdl, SA_ZPL_RDEV(o), &osa->rdev, 8);
 269                 if (rc)
 270                         GOTO(out_sa, rc);
 271                 la->la_rdev = osa->rdev;
 272                 la->la_valid |= LA_RDEV;
 273         }
 274 out_sa:
 275
 276         RETURN(rc);
 277 }
 278
 279 int __osd_obj2dnode(objset_t *os, uint64_t oid, dnode_t **dnp)
 280 {
 281         dmu_buf_t *db;
 282         dmu_buf_impl_t *dbi;
 283         int rc;
 284
 285         rc = -dmu_bonus_hold(os, oid, osd_obj_tag, &db);
 286         if (rc)
 287                 return rc;
 288
 289         dbi = (dmu_buf_impl_t *)db;
 290         DB_DNODE_ENTER(dbi);
 291         *dnp = DB_DNODE(dbi);
 292         DB_DNODE_EXIT(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                 struct lu_object_header *h;
 312                 struct osd_device *o = osd_dev(d);
 313
 314                 l = &mo->oo_dt.do_lu;
 315                 if (unlikely(o->od_in_init)) {
 316                         OBD_ALLOC_PTR(h);
 317                         if (!h) {
 318                                 OBD_FREE_PTR(mo);
 319                                 return NULL;
 320                         }
 321
 322                         lu_object_header_init(h);
 323                         lu_object_init(l, h, d);
 324                         lu_object_add_top(h, l);
 325                         mo->oo_header = h;
 326                 } else {
 327                         dt_object_init(&mo->oo_dt, NULL, d);
 328                         mo->oo_header = NULL;
 329                 }
 330
 331                 mo->oo_dt.do_ops = &osd_obj_ops;
 332                 l->lo_ops = &osd_lu_obj_ops;
 333                 INIT_LIST_HEAD(&mo->oo_sa_linkage);
 334                 INIT_LIST_HEAD(&mo->oo_unlinked_linkage);
 335                 init_rwsem(&mo->oo_sem);
 336                 init_rwsem(&mo->oo_guard);
 337                 rwlock_init(&mo->oo_attr_lock);
 338                 mo->oo_destroy = OSD_DESTROY_NONE;
 339                 return l;
 340         } else {
 341                 return NULL;
 342         }
 343 }
 344
 345 static void osd_obj_set_blksize(const struct lu_env *env,
 346                                 struct osd_device *osd, struct osd_object *obj)
 347 {
 348         const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
 349         dmu_tx_t *tx;
 350         dnode_t *dn = obj->oo_dn;
 351         uint32_t blksz;
 352         int rc = 0;
 353         ENTRY;
 354
 355         LASSERT(!osd_oti_get(env)->oti_in_trans);
 356
 357         tx = dmu_tx_create(osd->od_os);
 358         if (!tx) {
 359                 CERROR("%s: fail to create tx to set blksize for "DFID"\n",
 360                        osd->od_svname, PFID(fid));
 361                 RETURN_EXIT;
 362         }
 363
 364         dmu_tx_hold_bonus(tx, dn->dn_object);
 365         rc = -dmu_tx_assign(tx, TXG_WAIT);
 366         if (rc) {
 367                 dmu_tx_abort(tx);
 368                 CERROR("%s: fail to assign tx to set blksize for "DFID
 369                        ": rc = %d\n", osd->od_svname, PFID(fid), rc);
 370                 RETURN_EXIT;
 371         }
 372
 373         down_write(&obj->oo_guard);
 374         if (unlikely((1 << dn->dn_datablkshift) >= PAGE_SIZE))
 375                 GOTO(out, rc = 1);
 376
 377         blksz = dn->dn_datablksz;
 378         if (!is_power_of_2(blksz))
 379                 blksz = size_roundup_power2(blksz);
 380
 381         if (blksz > osd->od_max_blksz)
 382                 blksz = osd->od_max_blksz;
 383         else if (blksz < PAGE_SIZE)
 384                 blksz = PAGE_SIZE;
 385         rc = -dmu_object_set_blocksize(osd->od_os, dn->dn_object, blksz, 0, tx);
 386
 387         GOTO(out, rc);
 388
 389 out:
 390         up_write(&obj->oo_guard);
 391         if (rc) {
 392                 dmu_tx_abort(tx);
 393                 if (unlikely(obj->oo_dn->dn_maxblkid > 0))
 394                         rc = 1;
 395                 if (rc < 0)
 396                         CERROR("%s: fail to set blksize for "DFID": rc = %d\n",
 397                                osd->od_svname, PFID(fid), rc);
 398         } else {
 399                 dmu_tx_commit(tx);
 400                 CDEBUG(D_INODE, "%s: set blksize as %u for "DFID"\n",
 401                        osd->od_svname, blksz, PFID(fid));
 402         }
 403 }
 404
 405 /*
 406  * Concurrency: shouldn't matter.
 407  */
 408 static int osd_object_init0(const struct lu_env *env, struct osd_object *obj)
 409 {
 410         struct osd_device       *osd = osd_obj2dev(obj);
 411         const struct lu_fid     *fid = lu_object_fid(&obj->oo_dt.do_lu);
 412         int                      rc = 0;
 413         ENTRY;
 414
 415         LASSERT(obj->oo_dn);
 416
 417         rc = osd_object_sa_init(obj, osd);
 418         if (rc)
 419                 RETURN(rc);
 420
 421         /* cache attrs in object */
 422         rc = __osd_object_attr_get(env, osd, obj, &obj->oo_attr);
 423         if (rc)
 424                 RETURN(rc);
 425
 426         if (likely(!fid_is_acct(fid))) {
 427                 /* no body operations for accounting objects */
 428                 obj->oo_dt.do_body_ops = &osd_body_ops;
 429
 430                 if (S_ISREG(obj->oo_attr.la_mode) &&
 431                     obj->oo_dn->dn_maxblkid == 0 &&
 432                     (1 << obj->oo_dn->dn_datablkshift) < PAGE_SIZE &&
 433                     (fid_is_idif(fid) || fid_is_norm(fid) ||
 434                      fid_is_echo(fid)) &&
 435                     osd->od_is_ost && !osd->od_dt_dev.dd_rdonly)
 436                         osd_obj_set_blksize(env, osd, obj);
 437         }
 438
 439         /*
 440          * initialize object before marking it existing
 441          */
 442         obj->oo_dt.do_lu.lo_header->loh_attr |= obj->oo_attr.la_mode & S_IFMT;
 443
 444         smp_mb();
 445         obj->oo_dt.do_lu.lo_header->loh_attr |= LOHA_EXISTS;
 446
 447         RETURN(0);
 448 }
 449
 450 static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
 451 {
 452         struct osd_thread_info  *info = osd_oti_get(env);
 453         struct lu_buf           buf;
 454         int                     rc;
 455         struct lustre_mdt_attrs *lma;
 456         const struct lu_fid *rfid = lu_object_fid(&obj->oo_dt.do_lu);
 457         ENTRY;
 458
 459         BUILD_BUG_ON(sizeof(info->oti_buf) < sizeof(*lma));
 460         lma = (struct lustre_mdt_attrs *)info->oti_buf;
 461         buf.lb_buf = lma;
 462         buf.lb_len = sizeof(info->oti_buf);
 463
 464         rc = osd_xattr_get(env, &obj->oo_dt, &buf, XATTR_NAME_LMA);
 465         if (rc > 0) {
 466                 rc = 0;
 467                 lustre_lma_swab(lma);
 468                 if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
 469                              CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT))) {
 470                         CWARN("%s: unsupported incompat LMA feature(s) %#x for "
 471                               "fid = "DFID"\n", osd_obj2dev(obj)->od_svname,
 472                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
 473                               PFID(rfid));
 474                         rc = -EOPNOTSUPP;
 475                 } else if (unlikely(!lu_fid_eq(rfid, &lma->lma_self_fid))) {
 476                         CERROR("%s: FID-in-LMA "DFID" does not match the "
 477                               "object self-fid "DFID"\n",
 478                               osd_obj2dev(obj)->od_svname,
 479                               PFID(&lma->lma_self_fid), PFID(rfid));
 480                         rc = -EREMCHG;
 481                 } else {
 482                         struct osd_device *osd = osd_obj2dev(obj);
 483
 484                         if (lma->lma_compat & LMAC_STRIPE_INFO &&
 485                             osd->od_is_ost)
 486                                 obj->oo_pfid_in_lma = 1;
 487                         if (unlikely(lma->lma_incompat & LMAI_REMOTE_PARENT) &&
 488                             osd->od_remote_parent_dir != ZFS_NO_OBJECT)
 489                                 lu_object_set_agent_entry(&obj->oo_dt.do_lu);
 490                 }
 491         } else if (rc == -ENODATA) {
 492                 /* haven't initialize LMA xattr */
 493                 rc = 0;
 494         }
 495
 496         RETURN(rc);
 497 }
 498
 499 /**
 500  * Helper function to retrieve DMU object id from fid for accounting object
 501  */
 502 static dnode_t *osd_quota_fid2dmu(const struct osd_device *osd,
 503                                   const struct lu_fid *fid)
 504 {
 505         dnode_t *dn = NULL;
 506
 507         LASSERT(fid_is_acct(fid));
 508
 509         switch (fid_oid(fid)) {
 510         case ACCT_USER_OID:
 511                 dn = osd->od_userused_dn;
 512                 break;
 513         case ACCT_GROUP_OID:
 514                 dn = osd->od_groupused_dn;
 515                 break;
 516 #ifdef ZFS_PROJINHERIT
 517         case ACCT_PROJECT_OID:
 518                 dn = osd->od_projectused_dn;
 519                 break;
 520 #endif
 521         default:
 522                 break;
 523         }
 524
 525         return dn;
 526 }
 527
 528 /*
 529  * Concurrency: no concurrent access is possible that early in object
 530  * life-cycle.
 531  */
 532 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
 533                            const struct lu_object_conf *conf)
 534 {
 535         struct osd_object *obj = osd_obj(l);
 536         struct osd_device *osd = osd_obj2dev(obj);
 537         const struct lu_fid *fid = lu_object_fid(l);
 538         struct lustre_scrub *scrub = &osd->od_scrub;
 539         struct osd_thread_info *info = osd_oti_get(env);
 540         struct luz_direntry *zde = &info->oti_zde;
 541         struct osd_idmap_cache *idc;
 542         char *name = info->oti_str;
 543         uint64_t oid;
 544         int rc = 0;
 545         int rc1;
 546         bool remote = false;
 547         ENTRY;
 548
 549         LASSERT(osd_invariant(obj));
 550
 551         if (fid_is_otable_it(&l->lo_header->loh_fid)) {
 552                 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
 553                 l->lo_header->loh_attr |= LOHA_EXISTS;
 554
 555                 GOTO(out, rc = 0);
 556         }
 557
 558         if (conf && conf->loc_flags & LOC_F_NEW)
 559                 GOTO(out, rc = 0);
 560
 561         if (unlikely(fid_is_acct(fid))) {
 562                 obj->oo_dn = osd_quota_fid2dmu(osd, fid);
 563                 if (obj->oo_dn) {
 564                         obj->oo_dt.do_index_ops = &osd_acct_index_ops;
 565                         l->lo_header->loh_attr |= LOHA_EXISTS;
 566                 }
 567
 568                 GOTO(out, rc = 0);
 569         }
 570
 571         idc = osd_idc_find(env, osd, fid);
 572         if (idc && !idc->oic_remote && idc->oic_dnode != ZFS_NO_OBJECT) {
 573                 oid = idc->oic_dnode;
 574                 goto zget;
 575         }
 576
 577         rc = -ENOENT;
 578         if (!list_empty(&osd->od_scrub.os_inconsistent_items))
 579                 rc = osd_oii_lookup(osd, fid, &oid);
 580
 581         if (rc)
 582                 rc = osd_fid_lookup(env, osd, fid, &oid);
 583
 584         if (rc == -ENOENT) {
 585                 if (likely(!(fid_is_norm(fid) || fid_is_igif(fid)) ||
 586                            fid_is_on_ost(env, osd, fid) ||
 587                            !zfs_test_bit(osd_oi_fid2idx(osd, fid),
 588                                          scrub->os_file.sf_oi_bitmap)))
 589                         GOTO(out, rc = 0);
 590
 591                 rc = -EREMCHG;
 592                 goto trigger;
 593         }
 594
 595         if (rc)
 596                 GOTO(out, rc);
 597
 598 zget:
 599         LASSERT(obj->oo_dn == NULL);
 600
 601         rc = __osd_obj2dnode(osd->od_os, oid, &obj->oo_dn);
 602         /* EEXIST will be returned if object is being deleted in ZFS */
 603         if (rc == -EEXIST)
 604                 GOTO(out, rc = 0);
 605
 606         if (rc) {
 607                 CERROR("%s: lookup "DFID"/%#llx failed: rc = %d\n",
 608                        osd->od_svname, PFID(lu_object_fid(l)), oid, rc);
 609                 GOTO(out, rc);
 610         }
 611
 612         rc = osd_object_init0(env, obj);
 613         if (rc)
 614                 GOTO(out, rc);
 615
 616         if (unlikely(obj->oo_header))
 617                 GOTO(out, rc = 0);
 618
 619         rc = osd_check_lma(env, obj);
 620         if (rc != -EREMCHG)
 621                 GOTO(out, rc);
 622
 623         osd_scrub_refresh_mapping(env, osd, fid, oid, DTO_INDEX_DELETE, true,
 624                                   NULL);
 625
 626 trigger:
 627         /* We still have chance to get the valid dnode: for the object that is
 628          * referenced by remote name entry, the object on the local MDT will be
 629          * linked under the dir /REMOTE_PARENT_DIR with its FID string as name.
 630          *
 631          * During the OI scrub, if we cannot find the OI mapping, we may still
 632          * have change to map the FID to local OID via lookup the dir
 633          * /REMOTE_PARENT_DIR. */
 634         if (!remote && !fid_is_on_ost(env, osd, fid)) {
 635                 osd_fid2str(name, fid, sizeof(info->oti_str));
 636                 rc = osd_zap_lookup(osd, osd->od_remote_parent_dir,
 637                                     NULL, name, 8, 3, (void *)zde);
 638                 if (!rc) {
 639                         oid = zde->lzd_reg.zde_dnode;
 640                         osd_dnode_rele(obj->oo_dn);
 641                         obj->oo_dn = NULL;
 642                         remote = true;
 643                         goto zget;
 644                 }
 645         }
 646
 647         /* The case someone triggered the OI scrub already. */
 648         if (scrub->os_running) {
 649                 if (!rc) {
 650                         LASSERT(remote);
 651
 652                         lu_object_set_agent_entry(l);
 653                         osd_oii_insert(env, osd, fid, oid, false);
 654                 } else {
 655                         rc = -EINPROGRESS;
 656                 }
 657
 658                 GOTO(out, rc);
 659         }
 660
 661         /* The case NOT allow to trigger OI scrub automatically. */
 662         if (osd->od_scrub.os_auto_scrub_interval == AS_NEVER)
 663                 GOTO(out, rc);
 664
 665         /* It is me to trigger the OI scrub. */
 666         rc1 = osd_scrub_start(env, osd, SS_CLEAR_DRYRUN |
 667                               SS_CLEAR_FAILOUT | SS_AUTO_FULL);
 668         CDEBUG_LIMIT(D_LFSCK | D_CONSOLE | D_WARNING,
 669                      "%s: trigger OI scrub by RPC for "DFID"/%#llx: rc = %d\n",
 670                      osd_name(osd), PFID(fid), oid, rc1);
 671         if (!rc) {
 672                 LASSERT(remote);
 673
 674                 lu_object_set_agent_entry(l);
 675                 if (!rc1)
 676                         osd_oii_insert(env, osd, fid, oid, false);
 677         } else {
 678                 if (!rc1)
 679                         rc = -EINPROGRESS;
 680                 else
 681                         rc = -EREMCHG;
 682         }
 683
 684         GOTO(out, rc);
 685
 686 out:
 687         RETURN(rc);
 688 }
 689
 690 /*
 691  * Concurrency: no concurrent access is possible that late in object
 692  * life-cycle.
 693  */
 694 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
 695 {
 696         struct osd_object *obj = osd_obj(l);
 697         struct lu_object_header *h = obj->oo_header;
 698
 699         LASSERT(osd_invariant(obj));
 700
 701         dt_object_fini(&obj->oo_dt);
 702         /* obj doesn't contain an lu_object_header, so we don't need call_rcu */
 703         OBD_SLAB_FREE_PTR(obj, osd_object_kmem);
 704         if (unlikely(h))
 705                 lu_object_header_free(h);
 706 }
 707
 708 static int
 709 osd_object_unlinked_add(struct osd_object *obj, struct osd_thandle *oh)
 710 {
 711         int rc = -EBUSY;
 712
 713         LASSERT(obj->oo_destroy == OSD_DESTROY_ASYNC);
 714
 715         /* the object is supposed to be exclusively locked by
 716          * the caller (osd_destroy()), while the transaction
 717          * (oh) is per-thread and not shared */
 718         if (likely(list_empty(&obj->oo_unlinked_linkage))) {
 719                 list_add(&obj->oo_unlinked_linkage, &oh->ot_unlinked_list);
 720                 rc = 0;
 721         }
 722
 723         return rc;
 724 }
 725
 726 /* Default to max data size covered by a level-1 indirect block */
 727 static unsigned long osd_sync_destroy_max_size =
 728         1UL << (DN_MAX_INDBLKSHIFT - SPA_BLKPTRSHIFT + SPA_MAXBLOCKSHIFT);
 729 module_param(osd_sync_destroy_max_size, ulong, 0444);
 730 MODULE_PARM_DESC(osd_sync_destroy_max_size, "Maximum object size to use synchronous destroy.");
 731
 732 static inline void
 733 osd_object_set_destroy_type(struct osd_object *obj)
 734 {
 735         /*
 736          * Lock-less OST_WRITE can race with OST_DESTROY, so set destroy type
 737          * only once and use it consistently thereafter.
 738          */
 739         down_write(&obj->oo_guard);
 740         if (obj->oo_destroy == OSD_DESTROY_NONE) {
 741                 if (obj->oo_attr.la_size <= osd_sync_destroy_max_size)
 742                         obj->oo_destroy = OSD_DESTROY_SYNC;
 743                 else /* Larger objects are destroyed asynchronously */
 744                         obj->oo_destroy = OSD_DESTROY_ASYNC;
 745         }
 746         up_write(&obj->oo_guard);
 747 }
 748
 749 static int osd_declare_destroy(const struct lu_env *env, struct dt_object *dt,
 750                                struct thandle *th)
 751 {
 752         const struct lu_fid     *fid = lu_object_fid(&dt->do_lu);
 753         struct osd_object       *obj = osd_dt_obj(dt);
 754         struct osd_device       *osd = osd_obj2dev(obj);
 755         struct osd_thandle      *oh;
 756         dnode_t *dn;
 757         int                      rc;
 758         uint64_t                 zapid;
 759         ENTRY;
 760
 761         LASSERT(th != NULL);
 762         LASSERT(dt_object_exists(dt));
 763
 764         oh = container_of(th, struct osd_thandle, ot_super);
 765         LASSERT(oh->ot_tx != NULL);
 766
 767         dmu_tx_mark_netfree(oh->ot_tx);
 768
 769         /* declare that we'll remove object from fid-dnode mapping */
 770         zapid = osd_get_name_n_idx(env, osd, fid, NULL, 0, &dn);
 771         osd_tx_hold_zap(oh->ot_tx, zapid, dn, FALSE, NULL);
 772
 773         osd_declare_xattrs_destroy(env, obj, oh);
 774
 775         /* one less inode */
 776         rc = osd_declare_quota(env, osd, obj->oo_attr.la_uid,
 777                                obj->oo_attr.la_gid, obj->oo_attr.la_projid,
 778                                -1, oh, NULL, OSD_QID_INODE);
 779         if (rc)
 780                 RETURN(rc);
 781
 782         /* data to be truncated */
 783         rc = osd_declare_quota(env, osd, obj->oo_attr.la_uid,
 784                                obj->oo_attr.la_gid, obj->oo_attr.la_projid,
 785                                0, oh, NULL, OSD_QID_BLK);
 786         if (rc)
 787                 RETURN(rc);
 788
 789         osd_object_set_destroy_type(obj);
 790         if (obj->oo_destroy == OSD_DESTROY_SYNC)
 791                 dmu_tx_hold_free(oh->ot_tx, obj->oo_dn->dn_object,
 792                                  0, DMU_OBJECT_END);
 793         else
 794                 osd_tx_hold_zap(oh->ot_tx, osd->od_unlinked->dn_object,
 795                                 osd->od_unlinked, TRUE, NULL);
 796
 797         /* remove agent entry (if have) from remote parent */
 798         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu))
 799                 osd_tx_hold_zap(oh->ot_tx, osd->od_remote_parent_dir,
 800                                 NULL, FALSE, NULL);
 801
 802         /* will help to find FID->ino when this object is being
 803          * added to PENDING/ */
 804         osd_idc_find_and_init(env, osd, obj);
 805
 806         RETURN(0);
 807 }
 808
 809 static int osd_destroy(const struct lu_env *env, struct dt_object *dt,
 810                        struct thandle *th)
 811 {
 812         struct osd_thread_info  *info = osd_oti_get(env);
 813         char                    *buf = info->oti_str;
 814         struct osd_object       *obj = osd_dt_obj(dt);
 815         struct osd_device       *osd = osd_obj2dev(obj);
 816         const struct lu_fid     *fid = lu_object_fid(&dt->do_lu);
 817         struct osd_thandle      *oh;
 818         int                      rc;
 819         uint64_t                 oid, zapid;
 820         dnode_t *zdn;
 821         ENTRY;
 822
 823         down_write(&obj->oo_guard);
 824
 825         if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
 826                 GOTO(out, rc = -ENOENT);
 827
 828         LASSERT(obj->oo_dn != NULL);
 829
 830         oh = container_of(th, struct osd_thandle, ot_super);
 831         LASSERT(oh != NULL);
 832         LASSERT(oh->ot_tx != NULL);
 833
 834         /* remove obj ref from index dir (it depends) */
 835         zapid = osd_get_name_n_idx(env, osd, fid, buf,
 836                                    sizeof(info->oti_str), &zdn);
 837         rc = osd_xattrs_destroy(env, obj, oh);
 838         if (rc) {
 839                 CERROR("%s: cannot destroy xattrs for %s: rc = %d\n",
 840                        osd->od_svname, buf, rc);
 841                 GOTO(out, rc);
 842         }
 843
 844         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu)) {
 845                 rc = osd_delete_from_remote_parent(env, osd, obj, oh, true);
 846                 if (rc)
 847                         GOTO(out, rc);
 848         }
 849
 850         oid = obj->oo_dn->dn_object;
 851         if (unlikely(obj->oo_destroy == OSD_DESTROY_NONE)) {
 852                 /* this may happen if the destroy wasn't declared
 853                  * e.g. when the object is created and then destroyed
 854                  * in the same transaction - we don't need additional
 855                  * space for destroy specifically */
 856                 LASSERT(obj->oo_attr.la_size <= osd_sync_destroy_max_size);
 857                 rc = -dmu_object_free(osd->od_os, oid, oh->ot_tx);
 858                 if (rc)
 859                         CERROR("%s: failed to free %s/%#llx: rc = %d\n",
 860                                osd->od_svname, buf, oid, rc);
 861         } else if (obj->oo_destroy == OSD_DESTROY_SYNC) {
 862                 rc = -dmu_object_free(osd->od_os, oid, oh->ot_tx);
 863                 if (rc)
 864                         CERROR("%s: failed to free %s/%#llx: rc = %d\n",
 865                                osd->od_svname, buf, oid, rc);
 866         } else { /* asynchronous destroy */
 867                 char *key = info->oti_key;
 868
 869                 rc = osd_object_unlinked_add(obj, oh);
 870                 if (rc)
 871                         GOTO(out, rc);
 872
 873                 snprintf(key, sizeof(info->oti_key), "%llx", oid);
 874                 rc = osd_zap_add(osd, osd->od_unlinked->dn_object,
 875                                  osd->od_unlinked, key, 8, 1, &oid, oh->ot_tx);
 876                 if (rc)
 877                         CERROR("%s: zap_add_int() failed %s/%#llx: rc = %d\n",
 878                                osd->od_svname, buf, oid, rc);
 879         }
 880
 881         /* Remove the OI mapping after the destroy to handle the race with
 882          * OI scrub that may insert missed OI mapping during the interval. */
 883         rc = osd_zap_remove(osd, zapid, zdn, buf, oh->ot_tx);
 884         if (unlikely(rc == -ENOENT))
 885                 rc = 0;
 886         if (rc)
 887                 CERROR("%s: zap_remove(%s) failed: rc = %d\n",
 888                        osd->od_svname, buf, rc);
 889
 890         GOTO(out, rc);
 891
 892 out:
 893         /* not needed in the cache anymore */
 894         set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
 895         if (rc == 0)
 896                 obj->oo_destroyed = 1;
 897         up_write(&obj->oo_guard);
 898         RETURN (0);
 899 }
 900
 901 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
 902 {
 903         struct osd_object *obj = osd_obj(l);
 904         const struct lu_fid *fid = lu_object_fid(l);
 905
 906         if (obj->oo_dn) {
 907                 if (likely(!fid_is_acct(fid))) {
 908                         osd_object_sa_fini(obj);
 909                         if (obj->oo_sa_xattr) {
 910                                 nvlist_free(obj->oo_sa_xattr);
 911                                 obj->oo_sa_xattr = NULL;
 912                         }
 913                         osd_dnode_rele(obj->oo_dn);
 914                         list_del(&obj->oo_sa_linkage);
 915                 }
 916                 obj->oo_dn = NULL;
 917         }
 918 }
 919
 920 /*
 921  * Concurrency: ->loo_object_release() is called under site spin-lock.
 922  */
 923 static void osd_object_release(const struct lu_env *env,
 924                                struct lu_object *l)
 925 {
 926 }
 927
 928 /*
 929  * Concurrency: shouldn't matter.
 930  */
 931 static int osd_object_print(const struct lu_env *env, void *cookie,
 932                             lu_printer_t p, const struct lu_object *l)
 933 {
 934         struct osd_object *o = osd_obj(l);
 935
 936         return (*p)(env, cookie, LUSTRE_OSD_ZFS_NAME"-object@%p", o);
 937 }
 938
 939 static void osd_read_lock(const struct lu_env *env, struct dt_object *dt,
 940                           unsigned role)
 941 {
 942         struct osd_object *obj = osd_dt_obj(dt);
 943
 944         LASSERT(osd_invariant(obj));
 945
 946         down_read_nested(&obj->oo_sem, role);
 947 }
 948
 949 static void osd_write_lock(const struct lu_env *env, struct dt_object *dt,
 950                            unsigned role)
 951 {
 952         struct osd_object *obj = osd_dt_obj(dt);
 953
 954         LASSERT(osd_invariant(obj));
 955
 956         down_write_nested(&obj->oo_sem, role);
 957 }
 958
 959 static void osd_read_unlock(const struct lu_env *env, struct dt_object *dt)
 960 {
 961         struct osd_object *obj = osd_dt_obj(dt);
 962
 963         LASSERT(osd_invariant(obj));
 964         up_read(&obj->oo_sem);
 965 }
 966
 967 static void osd_write_unlock(const struct lu_env *env, struct dt_object *dt)
 968 {
 969         struct osd_object *obj = osd_dt_obj(dt);
 970
 971         LASSERT(osd_invariant(obj));
 972         up_write(&obj->oo_sem);
 973 }
 974
 975 static int osd_write_locked(const struct lu_env *env, struct dt_object *dt)
 976 {
 977         struct osd_object *obj = osd_dt_obj(dt);
 978         int rc = 1;
 979
 980         LASSERT(osd_invariant(obj));
 981
 982         if (down_write_trylock(&obj->oo_sem)) {
 983                 rc = 0;
 984                 up_write(&obj->oo_sem);
 985         }
 986         return rc;
 987 }
 988
 989 static int osd_attr_get(const struct lu_env *env, struct dt_object *dt,
 990                         struct lu_attr *attr)
 991 {
 992         struct osd_object *obj = osd_dt_obj(dt);
 993         struct osd_device *osd = osd_obj2dev(obj);
 994         uint64_t blocks;
 995         uint32_t blksize;
 996         int rc = 0;
 997
 998         down_read(&obj->oo_guard);
 999
1000         if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
1001                 GOTO(out, rc = -ENOENT);
1002
1003         if (unlikely(fid_is_acct(lu_object_fid(&dt->do_lu))))
1004                 GOTO(out, rc = 0);
1005
1006         LASSERT(osd_invariant(obj));
1007         LASSERT(obj->oo_dn);
1008
1009         read_lock(&obj->oo_attr_lock);
1010         *attr = obj->oo_attr;
1011         if (obj->oo_lma_flags & LUSTRE_ORPHAN_FL) {
1012                 attr->la_valid |= LA_FLAGS;
1013                 attr->la_flags |= LUSTRE_ORPHAN_FL;
1014         }
1015         if (obj->oo_lma_flags & LUSTRE_ENCRYPT_FL) {
1016                 attr->la_valid |= LA_FLAGS;
1017                 attr->la_flags |= LUSTRE_ENCRYPT_FL;
1018         }
1019         read_unlock(&obj->oo_attr_lock);
1020         if (attr->la_valid & LA_FLAGS && attr->la_flags & LUSTRE_ORPHAN_FL)
1021                 CDEBUG(D_INFO, "%s: set orphan flag on "DFID" (%#llx/%#x)\n",
1022                        osd_obj2dev(obj)->od_svname,
1023                        PFID(lu_object_fid(&dt->do_lu)),
1024                        attr->la_valid, obj->oo_lma_flags);
1025
1026         /* with ZFS_DEBUG zrl_add_debug() called by DB_DNODE_ENTER()
1027          * from within sa_object_size() can block on a mutex, so
1028          * we can't call sa_object_size() holding rwlock */
1029         sa_object_size(obj->oo_sa_hdl, &blksize, &blocks);
1030         /* we do not control size of indices, so always calculate
1031          * it from number of blocks reported by DMU */
1032         if (S_ISDIR(attr->la_mode)) {
1033                 attr->la_size = 512 * blocks;
1034                 rc = -zap_count(osd->od_os, obj->oo_dn->dn_object,
1035                                 &attr->la_dirent_count);
1036         }
1037         /* Block size may be not set; suggest maximal I/O transfers. */
1038         if (blksize == 0)
1039                 blksize = osd_spa_maxblocksize(
1040                         dmu_objset_spa(osd_obj2dev(obj)->od_os));
1041
1042         attr->la_blksize = blksize;
1043         attr->la_blocks = blocks;
1044         attr->la_valid |= LA_BLOCKS | LA_BLKSIZE;
1045
1046 out:
1047         up_read(&obj->oo_guard);
1048         return rc;
1049 }
1050
1051 static int osd_declare_attr_set(const struct lu_env *env,
1052                                 struct dt_object *dt,
1053                                 const struct lu_attr *attr,
1054                                 struct thandle *handle)
1055 {
1056         struct osd_thread_info  *info = osd_oti_get(env);
1057         struct osd_object       *obj = osd_dt_obj(dt);
1058         struct osd_device       *osd = osd_obj2dev(obj);
1059         dmu_tx_hold_t           *txh;
1060         struct osd_thandle      *oh;
1061         uint64_t                 bspace;
1062         uint32_t                 blksize;
1063         int                      rc = 0;
1064         bool                     found;
1065         ENTRY;
1066
1067
1068         LASSERT(handle != NULL);
1069         LASSERT(osd_invariant(obj));
1070
1071         oh = container_of(handle, struct osd_thandle, ot_super);
1072
1073         down_read(&obj->oo_guard);
1074         if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
1075                 GOTO(out_sem, rc = 0);
1076
1077         LASSERT(obj->oo_sa_hdl != NULL);
1078         LASSERT(oh->ot_tx != NULL);
1079         /* regular attributes are part of the bonus buffer */
1080         /* let's check whether this object is already part of
1081          * transaction.. */
1082         found = false;
1083         for (txh = list_head(&oh->ot_tx->tx_holds); txh;
1084              txh = list_next(&oh->ot_tx->tx_holds, txh)) {
1085                 if (txh->txh_dnode == NULL)
1086                         continue;
1087                 if (txh->txh_dnode->dn_object != obj->oo_dn->dn_object)
1088                         continue;
1089                 /* this object is part of the transaction already
1090                  * we don't need to declare bonus again */
1091                 found = true;
1092                 break;
1093         }
1094         if (!found)
1095                 dmu_tx_hold_bonus(oh->ot_tx, obj->oo_dn->dn_object);
1096         if (oh->ot_tx->tx_err != 0)
1097                 GOTO(out_sem, rc = -oh->ot_tx->tx_err);
1098
1099         if (attr && attr->la_valid & LA_FLAGS) {
1100                 /* LMA is usually a part of bonus, no need to declare
1101                  * anything else */
1102         }
1103
1104         if (attr && (attr->la_valid & (LA_UID | LA_GID | LA_PROJID))) {
1105                 sa_object_size(obj->oo_sa_hdl, &blksize, &bspace);
1106                 bspace = toqb(bspace * 512);
1107
1108                 CDEBUG(D_QUOTA,
1109                        "%s: enforce quota on UID %u, GID %u, the quota space is %lld (%u)\n",
1110                        osd->od_svname,
1111                        attr->la_uid, attr->la_gid, bspace, blksize);
1112         }
1113         /* to preserve locking order - qsd_transfer() may need to flush
1114          * currently running transaction when we're out of quota.
1115          */
1116         up_read(&obj->oo_guard);
1117
1118         /* quota enforcement for user */
1119         if (attr && attr->la_valid & LA_UID &&
1120             attr->la_uid != obj->oo_attr.la_uid) {
1121                 rc = qsd_transfer(env, osd_def_qsd(osd),
1122                                   &oh->ot_quota_trans, USRQUOTA,
1123                                   obj->oo_attr.la_uid, attr->la_uid,
1124                                   bspace, &info->oti_qi);
1125                 if (rc)
1126                         GOTO(out, rc);
1127         }
1128
1129         /* quota enforcement for group */
1130         if (attr && attr->la_valid & LA_GID &&
1131             attr->la_gid != obj->oo_attr.la_gid) {
1132                 rc = qsd_transfer(env, osd_def_qsd(osd),
1133                                   &oh->ot_quota_trans, GRPQUOTA,
1134                                   obj->oo_attr.la_gid, attr->la_gid,
1135                                   bspace, &info->oti_qi);
1136                 if (rc)
1137                         GOTO(out, rc);
1138         }
1139 #ifdef ZFS_PROJINHERIT
1140         /* quota enforcement for project */
1141         if (attr && attr->la_valid & LA_PROJID &&
1142             attr->la_projid != obj->oo_attr.la_projid) {
1143                 if (!osd->od_projectused_dn)
1144                         GOTO(out, rc = -EOPNOTSUPP);
1145
1146                 /* Usually, if project quota is upgradable for the
1147                  * device, then the upgrade will be done before or when
1148                  * mount the device. So when we come here, this project
1149                  * should have project ID attribute already (that is
1150                  * zero by default).  Otherwise, there was something
1151                  * wrong during the former upgrade, let's return failure
1152                  * to report that.
1153                  *
1154                  * Please note that, different from other attributes,
1155                  * you can NOT simply set the project ID attribute under
1156                  * such case, because adding (NOT change) project ID
1157                  * attribute needs to change the object's attribute
1158                  * layout to match zfs backend quota accounting
1159                  * requirement.
1160                  */
1161                 if (unlikely(!obj->oo_with_projid))
1162                         GOTO(out, rc = -ENXIO);
1163
1164                 rc = qsd_transfer(env, osd_def_qsd(osd),
1165                                   &oh->ot_quota_trans, PRJQUOTA,
1166                                   obj->oo_attr.la_projid,
1167                                   attr->la_projid, bspace,
1168                                   &info->oti_qi);
1169                 if (rc)
1170                         GOTO(out, rc);
1171         }
1172 #endif
1173 out:
1174         RETURN(rc);
1175 out_sem:
1176         up_read(&obj->oo_guard);
1177         RETURN(rc);
1178 }
1179
1180 /*
1181  * Set the attributes of an object
1182  *
1183  * The transaction passed to this routine must have
1184  * dmu_tx_hold_bonus(tx, oid) called and then assigned
1185  * to a transaction group.
1186  */
1187 static int osd_attr_set(const struct lu_env *env, struct dt_object *dt,
1188                         const struct lu_attr *la, struct thandle *handle)
1189 {
1190         struct osd_thread_info  *info = osd_oti_get(env);
1191         sa_bulk_attr_t          *bulk = osd_oti_get(env)->oti_attr_bulk;
1192         struct osd_object       *obj = osd_dt_obj(dt);
1193         struct osd_device       *osd = osd_obj2dev(obj);
1194         struct osd_thandle      *oh;
1195         struct osa_attr         *osa = &info->oti_osa;
1196         __u64                    valid = la->la_valid;
1197         int                      cnt;
1198         int                      rc = 0;
1199
1200         ENTRY;
1201
1202         down_read(&obj->oo_guard);
1203         if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
1204                 GOTO(out, rc = -ENOENT);
1205
1206         LASSERT(handle != NULL);
1207         LASSERT(osd_invariant(obj));
1208         LASSERT(obj->oo_sa_hdl);
1209
1210         oh = container_of(handle, struct osd_thandle, ot_super);
1211         /* Assert that the transaction has been assigned to a
1212            transaction group. */
1213         LASSERT(oh->ot_tx->tx_txg != 0);
1214
1215         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING) && !osd->od_is_ost) {
1216                 struct zpl_direntry *zde = &info->oti_zde.lzd_reg;
1217                 char *buf = info->oti_str;
1218                 dnode_t *zdn = NULL;
1219                 uint64_t zapid;
1220
1221                 zapid = osd_get_name_n_idx(env, osd, lu_object_fid(&dt->do_lu),
1222                                            buf, sizeof(info->oti_str), &zdn);
1223                 rc = osd_zap_lookup(osd, zapid, zdn, buf, 8,
1224                                     sizeof(*zde) / 8, zde);
1225                 if (!rc) {
1226                         zde->zde_dnode -= 1;
1227                         rc = -zap_update(osd->od_os, zapid, buf, 8,
1228                                          sizeof(*zde) / 8, zde, oh->ot_tx);
1229                 }
1230                 if (rc > 0)
1231                         rc = 0;
1232                 GOTO(out, rc);
1233         }
1234
1235         /* Only allow set size for regular file */
1236         if (!S_ISREG(dt->do_lu.lo_header->loh_attr))
1237                 valid &= ~(LA_SIZE | LA_BLOCKS);
1238
1239         if (valid & LA_CTIME && la->la_ctime == obj->oo_attr.la_ctime)
1240                 valid &= ~LA_CTIME;
1241
1242         if (valid & LA_MTIME && la->la_mtime == obj->oo_attr.la_mtime)
1243                 valid &= ~LA_MTIME;
1244
1245         if (valid & LA_ATIME && la->la_atime == obj->oo_attr.la_atime)
1246                 valid &= ~LA_ATIME;
1247
1248         if (valid == 0)
1249                 GOTO(out, rc = 0);
1250
1251         if (valid & LA_FLAGS) {
1252                 struct lustre_mdt_attrs *lma;
1253                 struct lu_buf buf;
1254                 int size = 0;
1255
1256                 if (la->la_flags & LUSTRE_LMA_FL_MASKS) {
1257                         LASSERT(!obj->oo_pfid_in_lma);
1258                         BUILD_BUG_ON(sizeof(info->oti_buf) < sizeof(*lma));
1259                         lma = (struct lustre_mdt_attrs *)&info->oti_buf;
1260                         buf.lb_buf = lma;
1261                         buf.lb_len = sizeof(info->oti_buf);
1262
1263                         /* Please do NOT call osd_xattr_get() directly, that
1264                          * will cause recursive down_read() on oo_guard. */
1265                         rc = osd_xattr_get_internal(env, obj, &buf,
1266                                                     XATTR_NAME_LMA, &size);
1267                         if (!rc && unlikely(size < sizeof(*lma))) {
1268                                 rc = -EINVAL;
1269                         } else if (!rc) {
1270                                 lma->lma_incompat =
1271                                         le32_to_cpu(lma->lma_incompat);
1272                                 lma->lma_incompat |=
1273                                         lustre_to_lma_flags(la->la_flags);
1274                                 lma->lma_incompat =
1275                                         cpu_to_le32(lma->lma_incompat);
1276                                 buf.lb_buf = lma;
1277                                 buf.lb_len = sizeof(*lma);
1278                                 rc = osd_xattr_set_internal(env, obj, &buf,
1279                                                             XATTR_NAME_LMA,
1280                                                             LU_XATTR_REPLACE,
1281                                                             oh);
1282                         }
1283                         if (rc < 0) {
1284                                 CWARN("%s: failed to set LMA flags: rc = %d\n",
1285                                        osd->od_svname, rc);
1286                                 GOTO(out, rc);
1287                         } else {
1288                                 obj->oo_lma_flags =
1289                                         la->la_flags & LUSTRE_LMA_FL_MASKS;
1290                         }
1291                 }
1292         }
1293
1294         write_lock(&obj->oo_attr_lock);
1295         cnt = 0;
1296
1297         if (valid & LA_PROJID) {
1298 #ifdef ZFS_PROJINHERIT
1299                 if (osd->od_projectused_dn) {
1300                         LASSERT(obj->oo_with_projid);
1301
1302                         osa->projid = obj->oo_attr.la_projid = la->la_projid;
1303                         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_PROJID(osd), NULL,
1304                                          &osa->projid, 8);
1305                 } else
1306 #endif
1307                         valid &= ~LA_PROJID;
1308         }
1309
1310         if (valid & LA_ATIME) {
1311                 osa->atime[0] = obj->oo_attr.la_atime = la->la_atime;
1312                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(osd), NULL,
1313                                  osa->atime, 16);
1314         }
1315         if (valid & LA_MTIME) {
1316                 osa->mtime[0] = obj->oo_attr.la_mtime = la->la_mtime;
1317                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(osd), NULL,
1318                                  osa->mtime, 16);
1319         }
1320         if (valid & LA_CTIME) {
1321                 osa->ctime[0] = obj->oo_attr.la_ctime = la->la_ctime;
1322                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(osd), NULL,
1323                                  osa->ctime, 16);
1324         }
1325         if (valid & LA_MODE) {
1326                 /* mode is stored along with type, so read it first */
1327                 obj->oo_attr.la_mode = (obj->oo_attr.la_mode & S_IFMT) |
1328                         (la->la_mode & ~S_IFMT);
1329                 osa->mode = obj->oo_attr.la_mode;
1330                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(osd), NULL,
1331                                  &osa->mode, 8);
1332         }
1333         if (valid & LA_SIZE) {
1334                 osa->size = obj->oo_attr.la_size = la->la_size;
1335                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(osd), NULL,
1336                                  &osa->size, 8);
1337         }
1338         if (valid & LA_NLINK) {
1339                 osa->nlink = obj->oo_attr.la_nlink = la->la_nlink;
1340                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(osd), NULL,
1341                                  &osa->nlink, 8);
1342         }
1343         if (valid & LA_RDEV) {
1344                 osa->rdev = obj->oo_attr.la_rdev = la->la_rdev;
1345                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_RDEV(osd), NULL,
1346                                  &osa->rdev, 8);
1347         }
1348         if (valid & LA_FLAGS) {
1349                 osa->flags = attrs_fs2zfs(la->la_flags);
1350                 /* many flags are not supported by zfs, so ensure a good cached
1351                  * copy */
1352                 obj->oo_attr.la_flags = attrs_zfs2fs(osa->flags);
1353 #ifdef ZFS_PROJINHERIT
1354                 if (obj->oo_with_projid && osd->od_projectused_dn)
1355                         osa->flags |= ZFS_PROJID;
1356 #endif
1357                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(osd), NULL,
1358                                  &osa->flags, 8);
1359         }
1360         if (valid & LA_UID) {
1361                 osa->uid = obj->oo_attr.la_uid = la->la_uid;
1362                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(osd), NULL,
1363                                  &osa->uid, 8);
1364         }
1365         if (valid & LA_GID) {
1366                 osa->gid = obj->oo_attr.la_gid = la->la_gid;
1367                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(osd), NULL,
1368                                  &osa->gid, 8);
1369         }
1370         obj->oo_attr.la_valid |= valid;
1371         write_unlock(&obj->oo_attr_lock);
1372
1373         LASSERT(cnt <= ARRAY_SIZE(osd_oti_get(env)->oti_attr_bulk));
1374         rc = osd_object_sa_bulk_update(obj, bulk, cnt, oh);
1375
1376 out:
1377         up_read(&obj->oo_guard);
1378         RETURN(rc);
1379 }
1380
1381 /*
1382  * Object creation.
1383  *
1384  * XXX temporary solution.
1385  */
1386
1387 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
1388                         struct dt_object *parent, struct dt_object *child,
1389                         umode_t child_mode)
1390 {
1391         LASSERT(ah);
1392
1393         ah->dah_parent = parent;
1394         ah->dah_mode = child_mode;
1395
1396         if (parent != NULL && !dt_object_remote(parent)) {
1397                 /* will help to find FID->ino at dt_insert("..") */
1398                 struct osd_object *pobj = osd_dt_obj(parent);
1399
1400                 osd_idc_find_and_init(env, osd_obj2dev(pobj), pobj);
1401         }
1402 }
1403
1404 static int osd_declare_create(const struct lu_env *env, struct dt_object *dt,
1405                               struct lu_attr *attr,
1406                               struct dt_allocation_hint *hint,
1407                               struct dt_object_format *dof,
1408                               struct thandle *handle)
1409 {
1410         const struct lu_fid     *fid = lu_object_fid(&dt->do_lu);
1411         struct osd_object       *obj = osd_dt_obj(dt);
1412         struct osd_device       *osd = osd_obj2dev(obj);
1413         struct osd_thandle      *oh;
1414         uint64_t                 zapid;
1415         dnode_t                 *dn;
1416         int                      rc, dnode_size;
1417         ENTRY;
1418
1419         LASSERT(dof);
1420
1421         switch (dof->dof_type) {
1422                 case DFT_REGULAR:
1423                 case DFT_SYM:
1424                 case DFT_NODE:
1425                         if (obj->oo_dt.do_body_ops == NULL)
1426                                 obj->oo_dt.do_body_ops = &osd_body_ops;
1427                         break;
1428                 default:
1429                         break;
1430         }
1431
1432         LASSERT(handle != NULL);
1433         oh = container_of(handle, struct osd_thandle, ot_super);
1434         LASSERT(oh->ot_tx != NULL);
1435
1436         /* this is the minimum set of EAs on every Lustre object */
1437         obj->oo_ea_in_bonus = OSD_BASE_EA_IN_BONUS;
1438         /* reserve 32 bytes for extra stuff like ACLs */
1439         dnode_size = size_roundup_power2(obj->oo_ea_in_bonus + 32);
1440
1441         switch (dof->dof_type) {
1442                 case DFT_DIR:
1443                         dt->do_index_ops = &osd_dir_ops;
1444                         fallthrough;
1445                 case DFT_INDEX:
1446                         /* for zap create */
1447                         dmu_tx_hold_zap(oh->ot_tx, DMU_NEW_OBJECT, FALSE, NULL);
1448                         dmu_tx_hold_sa_create(oh->ot_tx, dnode_size);
1449                         break;
1450                 case DFT_REGULAR:
1451                 case DFT_SYM:
1452                 case DFT_NODE:
1453                         /* first, we'll create new object */
1454                         dmu_tx_hold_sa_create(oh->ot_tx, dnode_size);
1455                         break;
1456
1457                 default:
1458                         LBUG();
1459                         break;
1460         }
1461
1462         /* and we'll add it to some mapping */
1463         zapid = osd_get_name_n_idx(env, osd, fid, NULL, 0, &dn);
1464         osd_tx_hold_zap(oh->ot_tx, zapid, dn, TRUE, NULL);
1465
1466         /* will help to find FID->ino mapping at dt_insert() */
1467         osd_idc_find_and_init(env, osd, obj);
1468
1469         rc = osd_declare_quota(env, osd, attr->la_uid, attr->la_gid,
1470                                attr->la_projid, 1, oh, NULL, OSD_QID_INODE);
1471
1472         RETURN(rc);
1473 }
1474
1475 int __osd_attr_init(const struct lu_env *env, struct osd_device *osd,
1476                     struct osd_object *obj, sa_handle_t *sa_hdl, dmu_tx_t *tx,
1477                     struct lu_attr *la, uint64_t parent,
1478                     nvlist_t *xattr)
1479 {
1480         sa_bulk_attr_t *bulk = osd_oti_get(env)->oti_attr_bulk;
1481         struct osa_attr *osa = &osd_oti_get(env)->oti_osa;
1482         uint64_t gen;
1483         inode_timespec_t now;
1484         int cnt;
1485         int rc;
1486         char *dxattr = NULL;
1487         size_t sa_size;
1488
1489
1490         LASSERT(sa_hdl);
1491
1492         gen = dmu_tx_get_txg(tx);
1493         gethrestime(&now);
1494         ZFS_TIME_ENCODE(&now, osa->btime);
1495
1496         osa->atime[0] = la->la_atime;
1497         osa->ctime[0] = la->la_ctime;
1498         osa->mtime[0] = la->la_mtime;
1499         osa->mode = la->la_mode;
1500         osa->uid = la->la_uid;
1501         osa->gid = la->la_gid;
1502         osa->rdev = la->la_rdev;
1503         osa->nlink = la->la_nlink;
1504         if (la->la_valid & LA_FLAGS)
1505                 osa->flags = attrs_fs2zfs(la->la_flags);
1506         else
1507                 osa->flags = 0;
1508         osa->size  = la->la_size;
1509 #ifdef ZFS_PROJINHERIT
1510         if (osd->od_projectused_dn) {
1511                 if (la->la_valid & LA_PROJID)
1512                         osa->projid = la->la_projid;
1513                 else
1514                         osa->projid = ZFS_DEFAULT_PROJID;
1515                 osa->flags |= ZFS_PROJID;
1516                 if (obj)
1517                         obj->oo_with_projid = 1;
1518         } else {
1519                 osa->flags &= ~ZFS_PROJID;
1520         }
1521 #endif
1522
1523         /*
1524          * we need to create all SA below upon object create.
1525          *
1526          * XXX The attribute order matters since the accounting callback relies
1527          * on static offsets (i.e. SA_*_OFFSET, see zfs_space_delta_cb()) to
1528          * look up the UID/GID/PROJID attributes. Moreover, the callback does
1529          * not seem to support the spill block.
1530          * We define attributes in the same order as SA_*_OFFSET in order to
1531          * work around the problem. See ORI-610.
1532          */
1533         cnt = 0;
1534         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(osd), NULL, &osa->mode, 8);
1535         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(osd), NULL, &osa->size, 8);
1536         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GEN(osd), NULL, &gen, 8);
1537         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(osd), NULL, &osa->uid, 8);
1538         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(osd), NULL, &osa->gid, 8);
1539         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_PARENT(osd), NULL, &parent, 8);
1540         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(osd), NULL, &osa->flags, 8);
1541         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(osd), NULL, osa->atime, 16);
1542         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(osd), NULL, osa->mtime, 16);
1543         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(osd), NULL, osa->ctime, 16);
1544         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CRTIME(osd), NULL, osa->btime, 16);
1545         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(osd), NULL, &osa->nlink, 8);
1546 #ifdef ZFS_PROJINHERIT
1547         if (osd->od_projectused_dn)
1548                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_PROJID(osd), NULL,
1549                                  &osa->projid, 8);
1550 #endif
1551         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_RDEV(osd), NULL, &osa->rdev, 8);
1552         LASSERT(cnt <= ARRAY_SIZE(osd_oti_get(env)->oti_attr_bulk));
1553
1554         if (xattr) {
1555                 rc = -nvlist_size(xattr, &sa_size, NV_ENCODE_XDR);
1556                 LASSERT(rc == 0);
1557
1558                 dxattr = osd_zio_buf_alloc(sa_size);
1559                 LASSERT(dxattr);
1560
1561                 rc = -nvlist_pack(xattr, &dxattr, &sa_size,
1562                                 NV_ENCODE_XDR, KM_SLEEP);
1563                 LASSERT(rc == 0);
1564
1565                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_DXATTR(osd),
1566                                 NULL, dxattr, sa_size);
1567         }
1568
1569         rc = -sa_replace_all_by_template(sa_hdl, bulk, cnt, tx);
1570         if (dxattr)
1571                 osd_zio_buf_free(dxattr, sa_size);
1572
1573         return rc;
1574 }
1575
1576 int osd_find_new_dnode(const struct lu_env *env, dmu_tx_t *tx,
1577                        uint64_t oid, dnode_t **dnp)
1578 {
1579         dmu_tx_hold_t *txh;
1580         int rc = 0;
1581
1582         /* take dnode_t from tx to save on dnode#->dnode_t lookup */
1583         for (txh = list_tail(&tx->tx_holds); txh;
1584              txh = list_prev(&tx->tx_holds, txh)) {
1585                 dnode_t *dn = txh->txh_dnode;
1586                 dmu_buf_impl_t *db;
1587
1588                 if (dn == NULL)
1589                         continue;
1590                 if (dn->dn_object != oid)
1591                         continue;
1592                 db = dn->dn_bonus;
1593                 if (db == NULL) {
1594                         rw_enter(&dn->dn_struct_rwlock, RW_WRITER);
1595                         if (dn->dn_bonus == NULL)
1596                                 dbuf_create_bonus(dn);
1597                         rw_exit(&dn->dn_struct_rwlock);
1598                 }
1599                 db = dn->dn_bonus;
1600                 LASSERT(db);
1601                 LASSERT(dn->dn_handle);
1602                 DB_DNODE_ENTER(db);
1603                 if (zfs_refcount_add(&db->db_holds, osd_obj_tag) == 1) {
1604                         zfs_refcount_add(&dn->dn_holds, osd_obj_tag);
1605                         atomic_inc_32(&dn->dn_dbufs_count);
1606                 }
1607                 *dnp = dn;
1608                 DB_DNODE_EXIT(db);
1609                 dbuf_read(db, NULL, DB_RF_MUST_SUCCEED | DB_RF_NOPREFETCH);
1610                 break;
1611         }
1612
1613         if (unlikely(*dnp == NULL))
1614                 rc = __osd_obj2dnode(tx->tx_objset, oid, dnp);
1615
1616         return rc;
1617 }
1618
1619 #ifdef HAVE_DMU_OBJECT_ALLOC_DNSIZE
1620 int osd_find_dnsize(struct osd_device *osd, int ea_in_bonus)
1621 {
1622         int dnsize;
1623
1624         if (osd->od_dnsize == ZFS_DNSIZE_AUTO) {
1625                 dnsize = DNODE_MIN_SIZE;
1626                 do {
1627                         if (DN_BONUS_SIZE(dnsize) >= ea_in_bonus + 32)
1628                                 break;
1629                         dnsize <<= 1;
1630                 } while (dnsize < DNODE_MAX_SIZE);
1631                 if (dnsize > DNODE_MAX_SIZE)
1632                         dnsize = DNODE_MAX_SIZE;
1633         } else if (osd->od_dnsize == ZFS_DNSIZE_1K) {
1634                 dnsize = 1024;
1635         } else if (osd->od_dnsize == ZFS_DNSIZE_2K) {
1636                 dnsize = 2048;
1637         } else if (osd->od_dnsize == ZFS_DNSIZE_4K) {
1638                 dnsize = 4096;
1639         } else if (osd->od_dnsize == ZFS_DNSIZE_8K) {
1640                 dnsize = 8192;
1641         } else if (osd->od_dnsize == ZFS_DNSIZE_16K) {
1642                 dnsize = 16384;
1643         } else {
1644                 dnsize = DNODE_MIN_SIZE;
1645         }
1646         return dnsize;
1647 }
1648 #endif
1649
1650 /*
1651  * The transaction passed to this routine must have
1652  * dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT) called and then assigned
1653  * to a transaction group.
1654  */
1655 int __osd_object_create(const struct lu_env *env, struct osd_device *osd,
1656                         struct osd_object *obj, const struct lu_fid *fid,
1657                         dnode_t **dnp, dmu_tx_t *tx, struct lu_attr *la)
1658 {
1659         dmu_object_type_t type = DMU_OT_PLAIN_FILE_CONTENTS;
1660         uint64_t oid;
1661         int size;
1662
1663         /* Use DMU_OTN_UINT8_METADATA for local objects so their data blocks
1664          * would get an additional ditto copy */
1665         if (unlikely(S_ISREG(la->la_mode) &&
1666                      fid_seq_is_local_file(fid_seq(fid))))
1667                 type = DMU_OTN_UINT8_METADATA;
1668
1669         /* Create a new DMU object using the default dnode size. */
1670         if (obj)
1671                 size = obj->oo_ea_in_bonus;
1672         else
1673                 size = OSD_BASE_EA_IN_BONUS;
1674         oid = osd_dmu_object_alloc(osd->od_os, type, 0,
1675                                    osd_find_dnsize(osd, size), tx);
1676
1677         LASSERT(la->la_valid & LA_MODE);
1678         la->la_size = 0;
1679         la->la_nlink = 1;
1680
1681         return osd_find_new_dnode(env, tx, oid, dnp);
1682 }
1683
1684 /*
1685  * The transaction passed to this routine must have
1686  * dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, ...) called and then assigned
1687  * to a transaction group.
1688  *
1689  * Using ZAP_FLAG_HASH64 will force the ZAP to always be a FAT ZAP.
1690  * This is fine for directories today, because storing the FID in the dirent
1691  * will also require a FAT ZAP.  If there is a new type of micro ZAP created
1692  * then we might need to re-evaluate the use of this flag and instead do
1693  * a conversion from the different internal ZAP hash formats being used. */
1694 int __osd_zap_create(const struct lu_env *env, struct osd_device *osd,
1695                      dnode_t **dnp, dmu_tx_t *tx, struct lu_attr *la,
1696                      unsigned dnsize, zap_flags_t flags)
1697 {
1698         uint64_t oid;
1699
1700         /* Assert that the transaction has been assigned to a
1701            transaction group. */
1702         LASSERT(tx->tx_txg != 0);
1703         *dnp = NULL;
1704
1705         oid = osd_zap_create_flags(osd->od_os, 0, flags | ZAP_FLAG_HASH64,
1706                                    DMU_OT_DIRECTORY_CONTENTS,
1707                                    14, /* == ZFS fzap_default_blockshift */
1708                                    DN_MAX_INDBLKSHIFT, /* indirect blockshift */
1709                                    dnsize, tx);
1710
1711         la->la_size = 2;
1712         la->la_nlink = 1;
1713
1714         return osd_find_new_dnode(env, tx, oid, dnp);
1715 }
1716
1717 static dnode_t *osd_mkidx(const struct lu_env *env, struct osd_object *obj,
1718                           struct lu_attr *la, struct osd_thandle *oh)
1719 {
1720         struct osd_device *osd = osd_obj2dev(obj);
1721         dnode_t *dn;
1722         int rc;
1723
1724         /* Index file should be created as regular file in order not to confuse
1725          * ZPL which could interpret them as directory.
1726          * We set ZAP_FLAG_UINT64_KEY to let ZFS know than we are going to use
1727          * binary keys */
1728         LASSERT(S_ISREG(la->la_mode));
1729         rc = __osd_zap_create(env, osd, &dn, oh->ot_tx, la,
1730                 osd_find_dnsize(osd, obj->oo_ea_in_bonus), ZAP_FLAG_UINT64_KEY);
1731         if (rc)
1732                 return ERR_PTR(rc);
1733         return dn;
1734 }
1735
1736 static dnode_t *osd_mkdir(const struct lu_env *env, struct osd_object *obj,
1737                           struct lu_attr *la, struct osd_thandle *oh)
1738 {
1739         struct osd_device *osd = osd_obj2dev(obj);
1740         dnode_t *dn;
1741         int rc;
1742
1743         LASSERT(S_ISDIR(la->la_mode));
1744         rc = __osd_zap_create(env, osd, &dn, oh->ot_tx, la,
1745                               osd_find_dnsize(osd, obj->oo_ea_in_bonus), 0);
1746         if (rc)
1747                 return ERR_PTR(rc);
1748         return dn;
1749 }
1750
1751 static dnode_t *osd_mkreg(const struct lu_env *env, struct osd_object *obj,
1752                           struct lu_attr *la, struct osd_thandle *oh)
1753 {
1754         const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
1755         struct osd_device *osd = osd_obj2dev(obj);
1756         dnode_t *dn;
1757         int rc;
1758
1759         LASSERT(S_ISREG(la->la_mode));
1760         rc = __osd_object_create(env, osd, obj, fid, &dn, oh->ot_tx, la);
1761         if (rc)
1762                 return ERR_PTR(rc);
1763
1764         if ((fid_is_idif(fid) || fid_is_norm(fid) || fid_is_echo(fid))) {
1765                 /* The minimum block size must be at least page size otherwise
1766                  * it will break the assumption in tgt_thread_big_cache where
1767                  * the array size is PTLRPC_MAX_BRW_PAGES. It will also affect
1768                  * RDMA due to subpage transfer size */
1769                 rc = -dmu_object_set_blocksize(osd->od_os, dn->dn_object,
1770                                                PAGE_SIZE, 0, oh->ot_tx);
1771                 if (unlikely(rc)) {
1772                         CERROR("%s: can't change blocksize: %d\n",
1773                                osd->od_svname, rc);
1774                         return ERR_PTR(rc);
1775                 }
1776         } else if ((fid_is_llog(fid))) {
1777                 rc = -dmu_object_set_blocksize(osd->od_os, dn->dn_object,
1778                                                LLOG_MIN_CHUNK_SIZE, 0, oh->ot_tx);
1779                 if (unlikely(rc)) {
1780                         CERROR("%s: can't change blocksize: %d\n",
1781                                osd->od_svname, rc);
1782                         return ERR_PTR(rc);
1783                 }
1784         }
1785
1786         return dn;
1787 }
1788
1789 static dnode_t *osd_mksym(const struct lu_env *env, struct osd_object *obj,
1790                           struct lu_attr *la, struct osd_thandle *oh)
1791 {
1792         dnode_t *dn;
1793         int rc;
1794
1795         LASSERT(S_ISLNK(la->la_mode));
1796         rc = __osd_object_create(env, osd_obj2dev(obj), obj,
1797                                  lu_object_fid(&obj->oo_dt.do_lu),
1798                                  &dn, oh->ot_tx, la);
1799         if (rc)
1800                 return ERR_PTR(rc);
1801         return dn;
1802 }
1803
1804 static dnode_t *osd_mknod(const struct lu_env *env, struct osd_object *obj,
1805                           struct lu_attr *la, struct osd_thandle *oh)
1806 {
1807         dnode_t *dn;
1808         int rc;
1809
1810         if (S_ISCHR(la->la_mode) || S_ISBLK(la->la_mode))
1811                 la->la_valid |= LA_RDEV;
1812
1813         rc = __osd_object_create(env, osd_obj2dev(obj), obj,
1814                                  lu_object_fid(&obj->oo_dt.do_lu),
1815                                  &dn, oh->ot_tx, la);
1816         if (rc)
1817                 return ERR_PTR(rc);
1818         return dn;
1819 }
1820
1821 typedef dnode_t *(*osd_obj_type_f)(const struct lu_env *env,
1822                                    struct osd_object *obj,
1823                                    struct lu_attr *la,
1824                                    struct osd_thandle *oh);
1825
1826 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
1827 {
1828         osd_obj_type_f result;
1829
1830         switch (type) {
1831         case DFT_DIR:
1832                 result = osd_mkdir;
1833                 break;
1834         case DFT_INDEX:
1835                 result = osd_mkidx;
1836                 break;
1837         case DFT_REGULAR:
1838                 result = osd_mkreg;
1839                 break;
1840         case DFT_SYM:
1841                 result = osd_mksym;
1842                 break;
1843         case DFT_NODE:
1844                 result = osd_mknod;
1845                 break;
1846         default:
1847                 LBUG();
1848                 break;
1849         }
1850         return result;
1851 }
1852
1853 /*
1854  * Concurrency: @dt is write locked.
1855  */
1856 static int osd_create(const struct lu_env *env, struct dt_object *dt,
1857                       struct lu_attr *attr, struct dt_allocation_hint *hint,
1858                       struct dt_object_format *dof, struct thandle *th)
1859 {
1860         struct osd_thread_info  *info = osd_oti_get(env);
1861         struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
1862         struct zpl_direntry     *zde = &info->oti_zde.lzd_reg;
1863         const struct lu_fid     *fid = lu_object_fid(&dt->do_lu);
1864         struct osd_object       *obj = osd_dt_obj(dt);
1865         struct osd_device       *osd = osd_obj2dev(obj);
1866         char                    *buf = info->oti_str;
1867         struct osd_thandle      *oh;
1868         dnode_t *dn = NULL, *zdn = NULL;
1869         uint64_t                 zapid, parent = 0;
1870         int                      rc;
1871         __u32 compat = 0;
1872
1873         ENTRY;
1874
1875         LASSERT(!fid_is_acct(fid));
1876
1877         /* concurrent create declarations should not see
1878          * the object inconsistent (db, attr, etc).
1879          * in regular cases acquisition should be cheap */
1880         down_write(&obj->oo_guard);
1881
1882         if (unlikely(dt_object_exists(dt)))
1883                 GOTO(out, rc = -EEXIST);
1884
1885         LASSERT(osd_invariant(obj));
1886         LASSERT(dof != NULL);
1887
1888         LASSERT(th != NULL);
1889         oh = container_of(th, struct osd_thandle, ot_super);
1890
1891         LASSERT(obj->oo_dn == NULL);
1892
1893         /* to follow ZFS on-disk format we need
1894          * to initialize parent dnode properly */
1895         if (hint != NULL && hint->dah_parent != NULL &&
1896             !dt_object_remote(hint->dah_parent))
1897                 parent = osd_dt_obj(hint->dah_parent)->oo_dn->dn_object;
1898
1899         /* we may fix some attributes, better do not change the source */
1900         obj->oo_attr = *attr;
1901         obj->oo_attr.la_size = 0;
1902         obj->oo_attr.la_nlink = 0;
1903         obj->oo_attr.la_valid |= LA_SIZE | LA_NLINK | LA_TYPE;
1904
1905 #ifdef ZFS_PROJINHERIT
1906         if (osd->od_projectused_dn) {
1907                 if (!(obj->oo_attr.la_valid & LA_PROJID))
1908                         obj->oo_attr.la_projid = ZFS_DEFAULT_PROJID;
1909                 obj->oo_with_projid = 1;
1910         }
1911 #endif
1912
1913         dn = osd_create_type_f(dof->dof_type)(env, obj, &obj->oo_attr, oh);
1914         if (IS_ERR(dn)) {
1915                 rc = PTR_ERR(dn);
1916                 dn = NULL;
1917                 GOTO(out, rc);
1918         }
1919
1920         zde->zde_pad = 0;
1921         zde->zde_dnode = dn->dn_object;
1922         zde->zde_type = S_DT(attr->la_mode & S_IFMT);
1923
1924         zapid = osd_get_name_n_idx(env, osd, fid, buf,
1925                                    sizeof(info->oti_str), &zdn);
1926         if (CFS_FAIL_CHECK(OBD_FAIL_OSD_NO_OI_ENTRY) ||
1927             (osd->od_is_ost && OBD_FAIL_CHECK(OBD_FAIL_OSD_COMPAT_NO_ENTRY)))
1928                 goto skip_add;
1929
1930         if (osd->od_is_ost && OBD_FAIL_CHECK(OBD_FAIL_OSD_COMPAT_INVALID_ENTRY))
1931                 zde->zde_dnode++;
1932
1933         rc = osd_zap_add(osd, zapid, zdn, buf, 8, 1, zde, oh->ot_tx);
1934         if (rc)
1935                 GOTO(out, rc);
1936
1937 skip_add:
1938         obj->oo_dn = dn;
1939         /* Now add in all of the "SA" attributes */
1940         rc = osd_sa_handle_get(obj);
1941         if (rc)
1942                 GOTO(out, rc);
1943
1944         rc = -nvlist_alloc(&obj->oo_sa_xattr, NV_UNIQUE_NAME, KM_SLEEP);
1945         if (rc)
1946                 GOTO(out, rc);
1947
1948         /* initialize LMA */
1949         if (fid_is_idif(fid) || (fid_is_norm(fid) && osd->od_is_ost))
1950                 compat |= LMAC_FID_ON_OST;
1951         lustre_lma_init(lma, fid, compat, 0);
1952         lustre_lma_swab(lma);
1953         rc = -nvlist_add_byte_array(obj->oo_sa_xattr, XATTR_NAME_LMA,
1954                                     (uchar_t *)lma, sizeof(*lma));
1955         if (rc)
1956                 GOTO(out, rc);
1957
1958         /* configure new osd object */
1959         obj->oo_parent = parent != 0 ? parent : zapid;
1960         obj->oo_late_attr_set = 1;
1961         rc = __osd_sa_xattr_schedule_update(env, obj, oh);
1962         if (rc)
1963                 GOTO(out, rc);
1964
1965         /* XXX: oo_lma_flags */
1966         obj->oo_dt.do_lu.lo_header->loh_attr |= obj->oo_attr.la_mode & S_IFMT;
1967         if (likely(!fid_is_acct(lu_object_fid(&obj->oo_dt.do_lu))))
1968                 /* no body operations for accounting objects */
1969                 obj->oo_dt.do_body_ops = &osd_body_ops;
1970
1971         osd_idc_find_and_init(env, osd, obj);
1972
1973 out:
1974         if (unlikely(rc && dn)) {
1975                 dmu_object_free(osd->od_os, dn->dn_object, oh->ot_tx);
1976                 osd_dnode_rele(dn);
1977                 obj->oo_dn = NULL;
1978         } else if (!rc) {
1979                 obj->oo_dt.do_lu.lo_header->loh_attr |= LOHA_EXISTS;
1980         }
1981         up_write(&obj->oo_guard);
1982         RETURN(rc);
1983 }
1984
1985 static int osd_declare_ref_add(const struct lu_env *env, struct dt_object *dt,
1986                                struct thandle *th)
1987 {
1988         osd_idc_find_and_init(env, osd_dev(dt->do_lu.lo_dev), osd_dt_obj(dt));
1989         return osd_declare_attr_set(env, dt, NULL, th);
1990 }
1991
1992 /*
1993  * Concurrency: @dt is write locked.
1994  */
1995 static int osd_ref_add(const struct lu_env *env, struct dt_object *dt,
1996                        struct thandle *handle)
1997 {
1998         struct osd_object       *obj = osd_dt_obj(dt);
1999         struct osd_thandle      *oh;
2000         struct osd_device       *osd = osd_obj2dev(obj);
2001         uint64_t                 nlink;
2002         int rc;
2003
2004         ENTRY;
2005
2006         down_read(&obj->oo_guard);
2007         if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
2008                 GOTO(out, rc = -ENOENT);
2009
2010         LASSERT(osd_invariant(obj));
2011         LASSERT(obj->oo_sa_hdl != NULL);
2012
2013         oh = container_of(handle, struct osd_thandle, ot_super);
2014
2015         write_lock(&obj->oo_attr_lock);
2016         nlink = ++obj->oo_attr.la_nlink;
2017         write_unlock(&obj->oo_attr_lock);
2018
2019         rc = osd_object_sa_update(obj, SA_ZPL_LINKS(osd), &nlink, 8, oh);
2020
2021 out:
2022         up_read(&obj->oo_guard);
2023         RETURN(rc);
2024 }
2025
2026 static int osd_declare_ref_del(const struct lu_env *env, struct dt_object *dt,
2027                                struct thandle *handle)
2028 {
2029         osd_idc_find_and_init(env, osd_dev(dt->do_lu.lo_dev), osd_dt_obj(dt));
2030         return osd_declare_attr_set(env, dt, NULL, handle);
2031 }
2032
2033 /*
2034  * Concurrency: @dt is write locked.
2035  */
2036 static int osd_ref_del(const struct lu_env *env, struct dt_object *dt,
2037                        struct thandle *handle)
2038 {
2039         struct osd_object       *obj = osd_dt_obj(dt);
2040         struct osd_thandle      *oh;
2041         struct osd_device       *osd = osd_obj2dev(obj);
2042         uint64_t                 nlink;
2043         int                      rc;
2044
2045         ENTRY;
2046
2047         down_read(&obj->oo_guard);
2048
2049         if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
2050                 GOTO(out, rc = -ENOENT);
2051
2052         LASSERT(osd_invariant(obj));
2053         LASSERT(obj->oo_sa_hdl != NULL);
2054
2055         oh = container_of(handle, struct osd_thandle, ot_super);
2056         LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
2057
2058         write_lock(&obj->oo_attr_lock);
2059         nlink = --obj->oo_attr.la_nlink;
2060         write_unlock(&obj->oo_attr_lock);
2061
2062         rc = osd_object_sa_update(obj, SA_ZPL_LINKS(osd), &nlink, 8, oh);
2063
2064 out:
2065         up_read(&obj->oo_guard);
2066         RETURN(rc);
2067 }
2068
2069 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
2070                            __u64 start, __u64 end)
2071 {
2072         struct osd_device *osd = osd_obj2dev(osd_dt_obj(dt));
2073         uint64_t txg = 0;
2074         ENTRY;
2075
2076         if (osd->od_dt_dev.dd_rdonly)
2077                 RETURN(0);
2078
2079         txg = osd_db_dirty_txg(osd_dt_obj(dt)->oo_dn->dn_dbuf);
2080         if (txg) {
2081                 /* the object is dirty or being synced */
2082                 if (osd_object_sync_delay_us < 0)
2083                         txg_wait_synced(dmu_objset_pool(osd->od_os), txg);
2084                 else
2085                         udelay(osd_object_sync_delay_us);
2086         }
2087
2088         RETURN(0);
2089 }
2090
2091 static int osd_invalidate(const struct lu_env *env, struct dt_object *dt)
2092 {
2093         return 0;
2094 }
2095
2096 static bool osd_check_stale(struct dt_object *dt)
2097 {
2098         return false;
2099 }
2100
2101 static const struct dt_object_operations osd_obj_ops = {
2102         .do_read_lock           = osd_read_lock,
2103         .do_write_lock          = osd_write_lock,
2104         .do_read_unlock         = osd_read_unlock,
2105         .do_write_unlock        = osd_write_unlock,
2106         .do_write_locked        = osd_write_locked,
2107         .do_attr_get            = osd_attr_get,
2108         .do_declare_attr_set    = osd_declare_attr_set,
2109         .do_attr_set            = osd_attr_set,
2110         .do_ah_init             = osd_ah_init,
2111         .do_declare_create      = osd_declare_create,
2112         .do_create              = osd_create,
2113         .do_declare_destroy     = osd_declare_destroy,
2114         .do_destroy             = osd_destroy,
2115         .do_index_try           = osd_index_try,
2116         .do_declare_ref_add     = osd_declare_ref_add,
2117         .do_ref_add             = osd_ref_add,
2118         .do_declare_ref_del     = osd_declare_ref_del,
2119         .do_ref_del             = osd_ref_del,
2120         .do_xattr_get           = osd_xattr_get,
2121         .do_declare_xattr_set   = osd_declare_xattr_set,
2122         .do_xattr_set           = osd_xattr_set,
2123         .do_declare_xattr_del   = osd_declare_xattr_del,
2124         .do_xattr_del           = osd_xattr_del,
2125         .do_xattr_list          = osd_xattr_list,
2126         .do_object_sync         = osd_object_sync,
2127         .do_invalidate          = osd_invalidate,
2128         .do_check_stale         = osd_check_stale,
2129 };
2130
2131 static const struct lu_object_operations osd_lu_obj_ops = {
2132         .loo_object_init        = osd_object_init,
2133         .loo_object_delete      = osd_object_delete,
2134         .loo_object_release     = osd_object_release,
2135         .loo_object_free        = osd_object_free,
2136         .loo_object_print       = osd_object_print,
2137         .loo_object_invariant   = osd_object_invariant,
2138 };
2139
2140 static int osd_otable_it_attr_get(const struct lu_env *env,
2141                                 struct dt_object *dt,
2142                                 struct lu_attr *attr)
2143 {
2144         attr->la_valid = 0;
2145         return 0;
2146 }
2147
2148 static const struct dt_object_operations osd_obj_otable_it_ops = {
2149         .do_attr_get            = osd_otable_it_attr_get,
2150         .do_index_try           = osd_index_try,
2151 };
2152
2153 module_param(osd_object_sync_delay_us, int, 0644);
2154 MODULE_PARM_DESC(osd_object_sync_delay_us,
2155                  "If zero or larger delay N usec instead of doing object sync");