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  * 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 static int osd_object_sync_delay_us = -1;
  66
  67 static struct dt_object_operations osd_obj_ops;
  68 static struct lu_object_operations osd_lu_obj_ops;
  69 static struct dt_object_operations osd_obj_otable_it_ops;
  70
  71 static void
  72 osd_object_sa_fini(struct osd_object *obj)
  73 {
  74         if (obj->oo_sa_hdl) {
  75                 sa_handle_destroy(obj->oo_sa_hdl);
  76                 obj->oo_sa_hdl = NULL;
  77         }
  78 }
  79
  80 static int
  81 osd_object_sa_init(struct osd_object *obj, struct osd_device *o)
  82 {
  83         int rc;
  84
  85         LASSERT(obj->oo_sa_hdl == NULL);
  86         LASSERT(obj->oo_dn != NULL);
  87
  88         rc = osd_sa_handle_get(obj);
  89         if (rc)
  90                 return rc;
  91
  92         /* Cache the xattr object id, valid for the life of the object */
  93         rc = -sa_lookup(obj->oo_sa_hdl, SA_ZPL_XATTR(o), &obj->oo_xattr, 8);
  94         if (rc == -ENOENT) {
  95                 obj->oo_xattr = ZFS_NO_OBJECT;
  96                 rc = 0;
  97         } else if (rc) {
  98                 osd_object_sa_fini(obj);
  99         }
 100
 101         return rc;
 102 }
 103
 104 /*
 105  * Add object to list of dirty objects in tx handle.
 106  */
 107 void osd_object_sa_dirty_add(struct osd_object *obj, struct osd_thandle *oh)
 108 {
 109         if (!list_empty(&obj->oo_sa_linkage))
 110                 return;
 111
 112         write_lock(&obj->oo_attr_lock);
 113         if (likely(list_empty(&obj->oo_sa_linkage)))
 114                 list_add(&obj->oo_sa_linkage, &oh->ot_sa_list);
 115         write_unlock(&obj->oo_attr_lock);
 116 }
 117
 118 /*
 119  * Release spill block dbuf hold for all dirty SAs.
 120  */
 121 void osd_object_sa_dirty_rele(const struct lu_env *env, struct osd_thandle *oh)
 122 {
 123         struct osd_object *obj;
 124
 125         while (!list_empty(&oh->ot_sa_list)) {
 126                 obj = list_entry(oh->ot_sa_list.next,
 127                                  struct osd_object, oo_sa_linkage);
 128                 write_lock(&obj->oo_attr_lock);
 129                 list_del_init(&obj->oo_sa_linkage);
 130                 write_unlock(&obj->oo_attr_lock);
 131                 if (obj->oo_late_xattr) {
 132                         /*
 133                          * take oo_guard to protect oo_sa_xattr buffer
 134                          * from concurrent update by osd_xattr_set()
 135                          */
 136                         LASSERT(oh->ot_assigned != 0);
 137                         down_write(&obj->oo_guard);
 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                         up_write(&obj->oo_guard);
 143                 }
 144                 sa_spill_rele(obj->oo_sa_hdl);
 145         }
 146 }
 147
 148 /*
 149  * Update the SA and add the object to the dirty list.
 150  */
 151 int osd_object_sa_update(struct osd_object *obj, sa_attr_type_t type,
 152                          void *buf, uint32_t buflen, struct osd_thandle *oh)
 153 {
 154         int rc;
 155
 156         LASSERT(obj->oo_sa_hdl != NULL);
 157         LASSERT(oh->ot_tx != NULL);
 158
 159         rc = -sa_update(obj->oo_sa_hdl, type, buf, buflen, oh->ot_tx);
 160         osd_object_sa_dirty_add(obj, oh);
 161
 162         return rc;
 163 }
 164
 165 /*
 166  * Bulk update the SA and add the object to the dirty list.
 167  */
 168 static int
 169 osd_object_sa_bulk_update(struct osd_object *obj, sa_bulk_attr_t *attrs,
 170                           int count, struct osd_thandle *oh)
 171 {
 172         int rc;
 173
 174         LASSERT(obj->oo_sa_hdl != NULL);
 175         LASSERT(oh->ot_tx != NULL);
 176
 177         rc = -sa_bulk_update(obj->oo_sa_hdl, attrs, count, oh->ot_tx);
 178         osd_object_sa_dirty_add(obj, oh);
 179
 180         return rc;
 181 }
 182
 183 /*
 184  * Retrieve the attributes of a DMU object
 185  */
 186 static int __osd_object_attr_get(const struct lu_env *env, struct osd_device *o,
 187                                  struct osd_object *obj, struct lu_attr *la)
 188 {
 189         struct osa_attr *osa = &osd_oti_get(env)->oti_osa;
 190         sa_bulk_attr_t  *bulk = osd_oti_get(env)->oti_attr_bulk;
 191         int              cnt = 0;
 192         int              rc;
 193         ENTRY;
 194
 195         LASSERT(obj->oo_dn != NULL);
 196
 197         la->la_valid |= LA_ATIME | LA_MTIME | LA_CTIME | LA_BTIME | LA_MODE |
 198                         LA_TYPE | LA_SIZE | LA_UID | LA_GID | LA_FLAGS |
 199                         LA_NLINK;
 200
 201         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(o), NULL, osa->atime, 16);
 202         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(o), NULL, osa->mtime, 16);
 203         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(o), NULL, osa->ctime, 16);
 204         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CRTIME(o), NULL, osa->btime, 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_btime = osa->btime[0];
 240         la->la_mode = osa->mode;
 241         la->la_uid = osa->uid;
 242         la->la_gid = osa->gid;
 243         la->la_nlink = osa->nlink;
 244         la->la_flags = attrs_zfs2fs(osa->flags);
 245         la->la_size = osa->size;
 246
 247         /* Try to get extra flag from LMA. Right now, only LMAI_ORPHAN
 248          * flags is stored in LMA, and it is only for orphan directory */
 249         if (S_ISDIR(la->la_mode) && dt_object_exists(&obj->oo_dt)) {
 250                 struct osd_thread_info *info = osd_oti_get(env);
 251                 struct lustre_mdt_attrs *lma;
 252                 struct lu_buf buf;
 253
 254                 lma = (struct lustre_mdt_attrs *)info->oti_buf;
 255                 buf.lb_buf = lma;
 256                 buf.lb_len = sizeof(info->oti_buf);
 257                 rc = osd_xattr_get(env, &obj->oo_dt, &buf, XATTR_NAME_LMA);
 258                 if (rc > 0) {
 259                         rc = 0;
 260                         lma->lma_incompat = le32_to_cpu(lma->lma_incompat);
 261                         obj->oo_lma_flags =
 262                                 lma_to_lustre_flags(lma->lma_incompat);
 263
 264                 } else if (rc == -ENODATA) {
 265                         rc = 0;
 266                 }
 267         }
 268
 269         if (S_ISCHR(la->la_mode) || S_ISBLK(la->la_mode)) {
 270                 rc = -sa_lookup(obj->oo_sa_hdl, SA_ZPL_RDEV(o), &osa->rdev, 8);
 271                 if (rc)
 272                         GOTO(out_sa, rc);
 273                 la->la_rdev = osa->rdev;
 274                 la->la_valid |= LA_RDEV;
 275         }
 276 out_sa:
 277
 278         RETURN(rc);
 279 }
 280
 281 int __osd_obj2dnode(objset_t *os, uint64_t oid, dnode_t **dnp)
 282 {
 283         dmu_buf_t *db;
 284         dmu_buf_impl_t *dbi;
 285         int rc;
 286
 287         rc = -dmu_bonus_hold(os, oid, osd_obj_tag, &db);
 288         if (rc)
 289                 return rc;
 290
 291         dbi = (dmu_buf_impl_t *)db;
 292         DB_DNODE_ENTER(dbi);
 293         *dnp = DB_DNODE(dbi);
 294         DB_DNODE_EXIT(dbi);
 295         LASSERT(*dnp != NULL);
 296
 297         return 0;
 298 }
 299
 300 /*
 301  * Concurrency: no concurrent access is possible that early in object
 302  * life-cycle.
 303  */
 304 struct lu_object *osd_object_alloc(const struct lu_env *env,
 305                                    const struct lu_object_header *hdr,
 306                                    struct lu_device *d)
 307 {
 308         struct osd_object *mo;
 309
 310         OBD_SLAB_ALLOC_PTR_GFP(mo, osd_object_kmem, GFP_NOFS);
 311         if (mo != NULL) {
 312                 struct lu_object *l;
 313                 struct lu_object_header *h;
 314                 struct osd_device *o = osd_dev(d);
 315
 316                 l = &mo->oo_dt.do_lu;
 317                 if (unlikely(o->od_in_init)) {
 318                         OBD_ALLOC_PTR(h);
 319                         if (!h) {
 320                                 OBD_FREE_PTR(mo);
 321                                 return NULL;
 322                         }
 323
 324                         lu_object_header_init(h);
 325                         lu_object_init(l, h, d);
 326                         lu_object_add_top(h, l);
 327                         mo->oo_header = h;
 328                 } else {
 329                         dt_object_init(&mo->oo_dt, NULL, d);
 330                         mo->oo_header = NULL;
 331                 }
 332
 333                 mo->oo_dt.do_ops = &osd_obj_ops;
 334                 l->lo_ops = &osd_lu_obj_ops;
 335                 INIT_LIST_HEAD(&mo->oo_sa_linkage);
 336                 INIT_LIST_HEAD(&mo->oo_unlinked_linkage);
 337                 init_rwsem(&mo->oo_sem);
 338                 init_rwsem(&mo->oo_guard);
 339                 rwlock_init(&mo->oo_attr_lock);
 340                 mo->oo_destroy = OSD_DESTROY_NONE;
 341                 return l;
 342         } else {
 343                 return NULL;
 344         }
 345 }
 346
 347 static void osd_obj_set_blksize(const struct lu_env *env,
 348                                 struct osd_device *osd, struct osd_object *obj)
 349 {
 350         const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
 351         dmu_tx_t *tx;
 352         dnode_t *dn = obj->oo_dn;
 353         uint32_t blksz;
 354         int rc = 0;
 355         ENTRY;
 356
 357         LASSERT(!osd_oti_get(env)->oti_in_trans);
 358
 359         tx = dmu_tx_create(osd->od_os);
 360         if (!tx) {
 361                 CERROR("%s: fail to create tx to set blksize for "DFID"\n",
 362                        osd->od_svname, PFID(fid));
 363                 RETURN_EXIT;
 364         }
 365
 366         dmu_tx_hold_bonus(tx, dn->dn_object);
 367         rc = -dmu_tx_assign(tx, TXG_WAIT);
 368         if (rc) {
 369                 dmu_tx_abort(tx);
 370                 CERROR("%s: fail to assign tx to set blksize for "DFID
 371                        ": rc = %d\n", osd->od_svname, PFID(fid), rc);
 372                 RETURN_EXIT;
 373         }
 374
 375         down_write(&obj->oo_guard);
 376         if (unlikely((1 << dn->dn_datablkshift) >= PAGE_SIZE))
 377                 GOTO(out, rc = 1);
 378
 379         blksz = dn->dn_datablksz;
 380         if (!is_power_of_2(blksz))
 381                 blksz = size_roundup_power2(blksz);
 382
 383         if (blksz > osd->od_max_blksz)
 384                 blksz = osd->od_max_blksz;
 385         else if (blksz < PAGE_SIZE)
 386                 blksz = PAGE_SIZE;
 387         rc = -dmu_object_set_blocksize(osd->od_os, dn->dn_object, blksz, 0, tx);
 388
 389         GOTO(out, rc);
 390
 391 out:
 392         up_write(&obj->oo_guard);
 393         if (rc) {
 394                 dmu_tx_abort(tx);
 395                 if (unlikely(obj->oo_dn->dn_maxblkid > 0))
 396                         rc = 1;
 397                 if (rc < 0)
 398                         CERROR("%s: fail to set blksize for "DFID": rc = %d\n",
 399                                osd->od_svname, PFID(fid), rc);
 400         } else {
 401                 dmu_tx_commit(tx);
 402                 CDEBUG(D_INODE, "%s: set blksize as %u for "DFID"\n",
 403                        osd->od_svname, blksz, PFID(fid));
 404         }
 405 }
 406
 407 /*
 408  * Concurrency: shouldn't matter.
 409  */
 410 static int osd_object_init0(const struct lu_env *env, struct osd_object *obj)
 411 {
 412         struct osd_device       *osd = osd_obj2dev(obj);
 413         const struct lu_fid     *fid = lu_object_fid(&obj->oo_dt.do_lu);
 414         int                      rc = 0;
 415         ENTRY;
 416
 417         LASSERT(obj->oo_dn);
 418
 419         rc = osd_object_sa_init(obj, osd);
 420         if (rc)
 421                 RETURN(rc);
 422
 423         /* cache attrs in object */
 424         rc = __osd_object_attr_get(env, osd, obj, &obj->oo_attr);
 425         if (rc)
 426                 RETURN(rc);
 427
 428         if (likely(!fid_is_acct(fid))) {
 429                 /* no body operations for accounting objects */
 430                 obj->oo_dt.do_body_ops = &osd_body_ops;
 431
 432                 if (S_ISREG(obj->oo_attr.la_mode) &&
 433                     obj->oo_dn->dn_maxblkid == 0 &&
 434                     (1 << obj->oo_dn->dn_datablkshift) < PAGE_SIZE &&
 435                     (fid_is_idif(fid) || fid_is_norm(fid) ||
 436                      fid_is_echo(fid)) &&
 437                     osd->od_is_ost && !osd->od_dt_dev.dd_rdonly)
 438                         osd_obj_set_blksize(env, osd, obj);
 439         }
 440
 441         /*
 442          * initialize object before marking it existing
 443          */
 444         obj->oo_dt.do_lu.lo_header->loh_attr |= obj->oo_attr.la_mode & S_IFMT;
 445
 446         smp_mb();
 447         obj->oo_dt.do_lu.lo_header->loh_attr |= LOHA_EXISTS;
 448
 449         RETURN(0);
 450 }
 451
 452 static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
 453 {
 454         struct osd_thread_info  *info = osd_oti_get(env);
 455         struct lu_buf           buf;
 456         int                     rc;
 457         struct lustre_mdt_attrs *lma;
 458         const struct lu_fid *rfid = lu_object_fid(&obj->oo_dt.do_lu);
 459         ENTRY;
 460
 461         BUILD_BUG_ON(sizeof(info->oti_buf) < sizeof(*lma));
 462         lma = (struct lustre_mdt_attrs *)info->oti_buf;
 463         buf.lb_buf = lma;
 464         buf.lb_len = sizeof(info->oti_buf);
 465
 466         rc = osd_xattr_get(env, &obj->oo_dt, &buf, XATTR_NAME_LMA);
 467         if (rc > 0) {
 468                 rc = 0;
 469                 lustre_lma_swab(lma);
 470                 if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
 471                              CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT))) {
 472                         CWARN("%s: unsupported incompat LMA feature(s) %#x for "
 473                               "fid = "DFID"\n", osd_obj2dev(obj)->od_svname,
 474                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
 475                               PFID(rfid));
 476                         rc = -EOPNOTSUPP;
 477                 } else if (unlikely(!lu_fid_eq(rfid, &lma->lma_self_fid))) {
 478                         CERROR("%s: FID-in-LMA "DFID" does not match the "
 479                               "object self-fid "DFID"\n",
 480                               osd_obj2dev(obj)->od_svname,
 481                               PFID(&lma->lma_self_fid), PFID(rfid));
 482                         rc = -EREMCHG;
 483                 } else {
 484                         struct osd_device *osd = osd_obj2dev(obj);
 485
 486                         if (lma->lma_compat & LMAC_STRIPE_INFO &&
 487                             osd->od_is_ost)
 488                                 obj->oo_pfid_in_lma = 1;
 489                         if (unlikely(lma->lma_incompat & LMAI_REMOTE_PARENT) &&
 490                             osd->od_remote_parent_dir != ZFS_NO_OBJECT)
 491                                 lu_object_set_agent_entry(&obj->oo_dt.do_lu);
 492                 }
 493         } else if (rc == -ENODATA) {
 494                 /* haven't initialize LMA xattr */
 495                 rc = 0;
 496         }
 497
 498         RETURN(rc);
 499 }
 500
 501 /**
 502  * Helper function to retrieve DMU object id from fid for accounting object
 503  */
 504 static dnode_t *osd_quota_fid2dmu(const struct osd_device *osd,
 505                                   const struct lu_fid *fid)
 506 {
 507         dnode_t *dn = NULL;
 508
 509         LASSERT(fid_is_acct(fid));
 510
 511         switch (fid_oid(fid)) {
 512         case ACCT_USER_OID:
 513                 dn = osd->od_userused_dn;
 514                 break;
 515         case ACCT_GROUP_OID:
 516                 dn = osd->od_groupused_dn;
 517                 break;
 518 #ifdef ZFS_PROJINHERIT
 519         case ACCT_PROJECT_OID:
 520                 dn = osd->od_projectused_dn;
 521                 break;
 522 #endif
 523         default:
 524                 break;
 525         }
 526
 527         return dn;
 528 }
 529
 530 /*
 531  * Concurrency: no concurrent access is possible that early in object
 532  * life-cycle.
 533  */
 534 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
 535                            const struct lu_object_conf *conf)
 536 {
 537         struct osd_object *obj = osd_obj(l);
 538         struct osd_device *osd = osd_obj2dev(obj);
 539         const struct lu_fid *fid = lu_object_fid(l);
 540         struct lustre_scrub *scrub = &osd->od_scrub;
 541         struct osd_thread_info *info = osd_oti_get(env);
 542         struct luz_direntry *zde = &info->oti_zde;
 543         struct osd_idmap_cache *idc;
 544         char *name = info->oti_str;
 545         uint64_t oid;
 546         int rc = 0;
 547         int rc1;
 548         bool remote = false;
 549         ENTRY;
 550
 551         LASSERT(osd_invariant(obj));
 552
 553         if (fid_is_otable_it(&l->lo_header->loh_fid)) {
 554                 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
 555                 l->lo_header->loh_attr |= LOHA_EXISTS;
 556
 557                 GOTO(out, rc = 0);
 558         }
 559
 560         if (conf && conf->loc_flags & LOC_F_NEW)
 561                 GOTO(out, rc = 0);
 562
 563         if (unlikely(fid_is_acct(fid))) {
 564                 obj->oo_dn = osd_quota_fid2dmu(osd, fid);
 565                 if (obj->oo_dn) {
 566                         obj->oo_dt.do_index_ops = &osd_acct_index_ops;
 567                         l->lo_header->loh_attr |= LOHA_EXISTS;
 568                 }
 569
 570                 GOTO(out, rc = 0);
 571         }
 572
 573         idc = osd_idc_find(env, osd, fid);
 574         if (idc && !idc->oic_remote && idc->oic_dnode != ZFS_NO_OBJECT) {
 575                 oid = idc->oic_dnode;
 576                 goto zget;
 577         }
 578
 579         rc = -ENOENT;
 580         if (!list_empty(&osd->od_scrub.os_inconsistent_items))
 581                 rc = osd_oii_lookup(osd, fid, &oid);
 582
 583         if (rc)
 584                 rc = osd_fid_lookup(env, osd, fid, &oid);
 585
 586         if (rc == -ENOENT) {
 587                 if (likely(!(fid_is_norm(fid) || fid_is_igif(fid)) ||
 588                            fid_is_on_ost(env, osd, fid) ||
 589                            !zfs_test_bit(osd_oi_fid2idx(osd, fid),
 590                                          scrub->os_file.sf_oi_bitmap)))
 591                         GOTO(out, rc = 0);
 592
 593                 rc = -EREMCHG;
 594                 goto trigger;
 595         }
 596
 597         if (rc)
 598                 GOTO(out, rc);
 599
 600 zget:
 601         LASSERT(obj->oo_dn == NULL);
 602
 603         rc = __osd_obj2dnode(osd->od_os, oid, &obj->oo_dn);
 604         /* EEXIST will be returned if object is being deleted in ZFS */
 605         if (rc == -EEXIST)
 606                 GOTO(out, rc = 0);
 607
 608         if (rc) {
 609                 CERROR("%s: lookup "DFID"/%#llx failed: rc = %d\n",
 610                        osd->od_svname, PFID(lu_object_fid(l)), oid, rc);
 611                 GOTO(out, rc);
 612         }
 613
 614         rc = osd_object_init0(env, obj);
 615         if (rc)
 616                 GOTO(out, rc);
 617
 618         if (unlikely(obj->oo_header))
 619                 GOTO(out, rc = 0);
 620
 621         rc = osd_check_lma(env, obj);
 622         if ((!rc && !remote) || (rc != -EREMCHG))
 623                 GOTO(out, rc);
 624
 625 trigger:
 626         /* We still have chance to get the valid dnode: for the object that is
 627          * referenced by remote name entry, the object on the local MDT will be
 628          * linked under the dir /REMOTE_PARENT_DIR with its FID string as name.
 629          *
 630          * During the OI scrub, if we cannot find the OI mapping, we may still
 631          * have change to map the FID to local OID via lookup the dir
 632          * /REMOTE_PARENT_DIR. */
 633         if (!remote && !fid_is_on_ost(env, osd, fid)) {
 634                 osd_fid2str(name, fid, sizeof(info->oti_str));
 635                 rc = osd_zap_lookup(osd, osd->od_remote_parent_dir,
 636                                     NULL, name, 8, 3, (void *)zde);
 637                 if (!rc) {
 638                         oid = zde->lzd_reg.zde_dnode;
 639                         osd_dnode_rele(obj->oo_dn);
 640                         obj->oo_dn = NULL;
 641                         remote = true;
 642                         goto zget;
 643                 }
 644         }
 645
 646         /* The case someone triggered the OI scrub already. */
 647         if (thread_is_running(&scrub->os_thread)) {
 648                 if (!rc) {
 649                         LASSERT(remote);
 650
 651                         lu_object_set_agent_entry(l);
 652                         osd_oii_insert(env, osd, fid, oid, false);
 653                 } else {
 654                         rc = -EINPROGRESS;
 655                 }
 656
 657                 GOTO(out, rc);
 658         }
 659
 660         /* The case NOT allow to trigger OI scrub automatically. */
 661         if (osd->od_auto_scrub_interval == AS_NEVER)
 662                 GOTO(out, rc);
 663
 664         /* It is me to trigger the OI scrub. */
 665         rc1 = osd_scrub_start(env, osd, SS_CLEAR_DRYRUN |
 666                               SS_CLEAR_FAILOUT | SS_AUTO_FULL);
 667         LCONSOLE_WARN("%s: trigger OI scrub by RPC for the "DFID": rc = %d\n",
 668                       osd_name(osd), PFID(fid), rc1);
 669         if (!rc) {
 670                 LASSERT(remote);
 671
 672                 lu_object_set_agent_entry(l);
 673                 if (!rc1)
 674                         osd_oii_insert(env, osd, fid, oid, false);
 675         } else {
 676                 if (!rc1)
 677                         rc = -EINPROGRESS;
 678                 else
 679                         rc = -EREMCHG;
 680         }
 681
 682         GOTO(out, rc);
 683
 684 out:
 685         RETURN(rc);
 686 }
 687
 688 /*
 689  * Concurrency: no concurrent access is possible that late in object
 690  * life-cycle.
 691  */
 692 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
 693 {
 694         struct osd_object *obj = osd_obj(l);
 695         struct lu_object_header *h = obj->oo_header;
 696
 697         LASSERT(osd_invariant(obj));
 698
 699         dt_object_fini(&obj->oo_dt);
 700         /* obj doesn't contain an lu_object_header, so we don't need call_rcu */
 701         OBD_SLAB_FREE_PTR(obj, osd_object_kmem);
 702         if (unlikely(h)) {
 703                 lu_object_header_fini(h);
 704                 OBD_FREE_PRE(h, sizeof(*h), "kfreed");
 705                 kfree_rcu(h, loh_rcu);
 706         }
 707 }
 708
 709 static int
 710 osd_object_unlinked_add(struct osd_object *obj, struct osd_thandle *oh)
 711 {
 712         int rc = -EBUSY;
 713
 714         LASSERT(obj->oo_destroy == OSD_DESTROY_ASYNC);
 715
 716         /* the object is supposed to be exclusively locked by
 717          * the caller (osd_destroy()), while the transaction
 718          * (oh) is per-thread and not shared */
 719         if (likely(list_empty(&obj->oo_unlinked_linkage))) {
 720                 list_add(&obj->oo_unlinked_linkage, &oh->ot_unlinked_list);
 721                 rc = 0;
 722         }
 723
 724         return rc;
 725 }
 726
 727 /* Default to max data size covered by a level-1 indirect block */
 728 static unsigned long osd_sync_destroy_max_size =
 729         1UL << (DN_MAX_INDBLKSHIFT - SPA_BLKPTRSHIFT + SPA_MAXBLOCKSHIFT);
 730 module_param(osd_sync_destroy_max_size, ulong, 0444);
 731 MODULE_PARM_DESC(osd_sync_destroy_max_size, "Maximum object size to use synchronous destroy.");
 732
 733 static inline void
 734 osd_object_set_destroy_type(struct osd_object *obj)
 735 {
 736         /*
 737          * Lock-less OST_WRITE can race with OST_DESTROY, so set destroy type
 738          * only once and use it consistently thereafter.
 739          */
 740         down_write(&obj->oo_guard);
 741         if (obj->oo_destroy == OSD_DESTROY_NONE) {
 742                 if (obj->oo_attr.la_size <= osd_sync_destroy_max_size)
 743                         obj->oo_destroy = OSD_DESTROY_SYNC;
 744                 else /* Larger objects are destroyed asynchronously */
 745                         obj->oo_destroy = OSD_DESTROY_ASYNC;
 746         }
 747         up_write(&obj->oo_guard);
 748 }
 749
 750 static int osd_declare_destroy(const struct lu_env *env, struct dt_object *dt,
 751                                struct thandle *th)
 752 {
 753         const struct lu_fid     *fid = lu_object_fid(&dt->do_lu);
 754         struct osd_object       *obj = osd_dt_obj(dt);
 755         struct osd_device       *osd = osd_obj2dev(obj);
 756         struct osd_thandle      *oh;
 757         dnode_t *dn;
 758         int                      rc;
 759         uint64_t                 zapid;
 760         ENTRY;
 761
 762         LASSERT(th != NULL);
 763         LASSERT(dt_object_exists(dt));
 764
 765         oh = container_of(th, struct osd_thandle, ot_super);
 766         LASSERT(oh->ot_tx != NULL);
 767
 768         dmu_tx_mark_netfree(oh->ot_tx);
 769
 770         /* declare that we'll remove object from fid-dnode mapping */
 771         zapid = osd_get_name_n_idx(env, osd, fid, NULL, 0, &dn);
 772         osd_tx_hold_zap(oh->ot_tx, zapid, dn, FALSE, NULL);
 773
 774         osd_declare_xattrs_destroy(env, obj, oh);
 775
 776         /* one less inode */
 777         rc = osd_declare_quota(env, osd, obj->oo_attr.la_uid,
 778                                obj->oo_attr.la_gid, obj->oo_attr.la_projid,
 779                                -1, oh, NULL, OSD_QID_INODE);
 780         if (rc)
 781                 RETURN(rc);
 782
 783         /* data to be truncated */
 784         rc = osd_declare_quota(env, osd, obj->oo_attr.la_uid,
 785                                obj->oo_attr.la_gid, obj->oo_attr.la_projid,
 786                                0, oh, NULL, OSD_QID_BLK);
 787         if (rc)
 788                 RETURN(rc);
 789
 790         osd_object_set_destroy_type(obj);
 791         if (obj->oo_destroy == OSD_DESTROY_SYNC)
 792                 dmu_tx_hold_free(oh->ot_tx, obj->oo_dn->dn_object,
 793                                  0, DMU_OBJECT_END);
 794         else
 795                 osd_tx_hold_zap(oh->ot_tx, osd->od_unlinked->dn_object,
 796                                 osd->od_unlinked, TRUE, NULL);
 797
 798         /* remove agent entry (if have) from remote parent */
 799         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu))
 800                 osd_tx_hold_zap(oh->ot_tx, osd->od_remote_parent_dir,
 801                                 NULL, FALSE, NULL);
 802
 803         /* will help to find FID->ino when this object is being
 804          * added to PENDING/ */
 805         osd_idc_find_and_init(env, osd, obj);
 806
 807         RETURN(0);
 808 }
 809
 810 static int osd_destroy(const struct lu_env *env, struct dt_object *dt,
 811                        struct thandle *th)
 812 {
 813         struct osd_thread_info  *info = osd_oti_get(env);
 814         char                    *buf = info->oti_str;
 815         struct osd_object       *obj = osd_dt_obj(dt);
 816         struct osd_device       *osd = osd_obj2dev(obj);
 817         const struct lu_fid     *fid = lu_object_fid(&dt->do_lu);
 818         struct osd_thandle      *oh;
 819         int                      rc;
 820         uint64_t                 oid, zapid;
 821         dnode_t *zdn;
 822         ENTRY;
 823
 824         down_write(&obj->oo_guard);
 825
 826         if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
 827                 GOTO(out, rc = -ENOENT);
 828
 829         LASSERT(obj->oo_dn != NULL);
 830
 831         oh = container_of(th, struct osd_thandle, ot_super);
 832         LASSERT(oh != NULL);
 833         LASSERT(oh->ot_tx != NULL);
 834
 835         /* remove obj ref from index dir (it depends) */
 836         zapid = osd_get_name_n_idx(env, osd, fid, buf,
 837                                    sizeof(info->oti_str), &zdn);
 838         rc = osd_xattrs_destroy(env, obj, oh);
 839         if (rc) {
 840                 CERROR("%s: cannot destroy xattrs for %s: rc = %d\n",
 841                        osd->od_svname, buf, rc);
 842                 GOTO(out, rc);
 843         }
 844
 845         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu)) {
 846                 rc = osd_delete_from_remote_parent(env, osd, obj, oh, true);
 847                 if (rc)
 848                         GOTO(out, rc);
 849         }
 850
 851         oid = obj->oo_dn->dn_object;
 852         if (unlikely(obj->oo_destroy == OSD_DESTROY_NONE)) {
 853                 /* this may happen if the destroy wasn't declared
 854                  * e.g. when the object is created and then destroyed
 855                  * in the same transaction - we don't need additional
 856                  * space for destroy specifically */
 857                 LASSERT(obj->oo_attr.la_size <= osd_sync_destroy_max_size);
 858                 rc = -dmu_object_free(osd->od_os, oid, oh->ot_tx);
 859                 if (rc)
 860                         CERROR("%s: failed to free %s %llu: rc = %d\n",
 861                                osd->od_svname, buf, oid, rc);
 862         } else if (obj->oo_destroy == OSD_DESTROY_SYNC) {
 863                 rc = -dmu_object_free(osd->od_os, oid, oh->ot_tx);
 864                 if (rc)
 865                         CERROR("%s: failed to free %s %llu: rc = %d\n",
 866                                osd->od_svname, buf, oid, rc);
 867         } else { /* asynchronous destroy */
 868                 char *key = info->oti_key;
 869
 870                 rc = osd_object_unlinked_add(obj, oh);
 871                 if (rc)
 872                         GOTO(out, rc);
 873
 874                 snprintf(key, sizeof(info->oti_key), "%llx", oid);
 875                 rc = osd_zap_add(osd, osd->od_unlinked->dn_object,
 876                                  osd->od_unlinked, key, 8, 1, &oid, oh->ot_tx);
 877                 if (rc)
 878                         CERROR("%s: zap_add_int() failed %s %llu: rc = %d\n",
 879                                osd->od_svname, buf, oid, rc);
 880         }
 881
 882         /* Remove the OI mapping after the destroy to handle the race with
 883          * OI scrub that may insert missed OI mapping during the interval. */
 884         rc = osd_zap_remove(osd, zapid, zdn, buf, oh->ot_tx);
 885         if (unlikely(rc == -ENOENT))
 886                 rc = 0;
 887         if (rc)
 888                 CERROR("%s: zap_remove(%s) failed: rc = %d\n",
 889                        osd->od_svname, buf, rc);
 890
 891         GOTO(out, rc);
 892
 893 out:
 894         /* not needed in the cache anymore */
 895         set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
 896         if (rc == 0)
 897                 obj->oo_destroyed = 1;
 898         up_write(&obj->oo_guard);
 899         RETURN (0);
 900 }
 901
 902 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
 903 {
 904         struct osd_object *obj = osd_obj(l);
 905         const struct lu_fid *fid = lu_object_fid(l);
 906
 907         if (obj->oo_dn) {
 908                 if (likely(!fid_is_acct(fid))) {
 909                         osd_object_sa_fini(obj);
 910                         if (obj->oo_sa_xattr) {
 911                                 nvlist_free(obj->oo_sa_xattr);
 912                                 obj->oo_sa_xattr = NULL;
 913                         }
 914                         osd_dnode_rele(obj->oo_dn);
 915                         list_del(&obj->oo_sa_linkage);
 916                 }
 917                 obj->oo_dn = NULL;
 918         }
 919 }
 920
 921 /*
 922  * Concurrency: ->loo_object_release() is called under site spin-lock.
 923  */
 924 static void osd_object_release(const struct lu_env *env,
 925                                struct lu_object *l)
 926 {
 927 }
 928
 929 /*
 930  * Concurrency: shouldn't matter.
 931  */
 932 static int osd_object_print(const struct lu_env *env, void *cookie,
 933                             lu_printer_t p, const struct lu_object *l)
 934 {
 935         struct osd_object *o = osd_obj(l);
 936
 937         return (*p)(env, cookie, LUSTRE_OSD_ZFS_NAME"-object@%p", o);
 938 }
 939
 940 static void osd_read_lock(const struct lu_env *env, struct dt_object *dt,
 941                           unsigned role)
 942 {
 943         struct osd_object *obj = osd_dt_obj(dt);
 944
 945         LASSERT(osd_invariant(obj));
 946
 947         down_read_nested(&obj->oo_sem, role);
 948 }
 949
 950 static void osd_write_lock(const struct lu_env *env, struct dt_object *dt,
 951                            unsigned role)
 952 {
 953         struct osd_object *obj = osd_dt_obj(dt);
 954
 955         LASSERT(osd_invariant(obj));
 956
 957         down_write_nested(&obj->oo_sem, role);
 958 }
 959
 960 static void osd_read_unlock(const struct lu_env *env, struct dt_object *dt)
 961 {
 962         struct osd_object *obj = osd_dt_obj(dt);
 963
 964         LASSERT(osd_invariant(obj));
 965         up_read(&obj->oo_sem);
 966 }
 967
 968 static void osd_write_unlock(const struct lu_env *env, struct dt_object *dt)
 969 {
 970         struct osd_object *obj = osd_dt_obj(dt);
 971
 972         LASSERT(osd_invariant(obj));
 973         up_write(&obj->oo_sem);
 974 }
 975
 976 static int osd_write_locked(const struct lu_env *env, struct dt_object *dt)
 977 {
 978         struct osd_object *obj = osd_dt_obj(dt);
 979         int rc = 1;
 980
 981         LASSERT(osd_invariant(obj));
 982
 983         if (down_write_trylock(&obj->oo_sem)) {
 984                 rc = 0;
 985                 up_write(&obj->oo_sem);
 986         }
 987         return rc;
 988 }
 989
 990 static int osd_attr_get(const struct lu_env *env, struct dt_object *dt,
 991                         struct lu_attr *attr)
 992 {
 993         struct osd_object       *obj = osd_dt_obj(dt);
 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         read_unlock(&obj->oo_attr_lock);
1016         if (attr->la_valid & LA_FLAGS && attr->la_flags & LUSTRE_ORPHAN_FL)
1017                 CDEBUG(D_INFO, "%s: set orphan flag on "DFID" (%llx/%x)\n",
1018                        osd_obj2dev(obj)->od_svname,
1019                        PFID(lu_object_fid(&dt->do_lu)),
1020                        attr->la_valid, obj->oo_lma_flags);
1021
1022         /* with ZFS_DEBUG zrl_add_debug() called by DB_DNODE_ENTER()
1023          * from within sa_object_size() can block on a mutex, so
1024          * we can't call sa_object_size() holding rwlock */
1025         sa_object_size(obj->oo_sa_hdl, &blksize, &blocks);
1026         /* we do not control size of indices, so always calculate
1027          * it from number of blocks reported by DMU */
1028         if (S_ISDIR(attr->la_mode))
1029                 attr->la_size = 512 * blocks;
1030         /* Block size may be not set; suggest maximal I/O transfers. */
1031         if (blksize == 0)
1032                 blksize = osd_spa_maxblocksize(
1033                         dmu_objset_spa(osd_obj2dev(obj)->od_os));
1034
1035         attr->la_blksize = blksize;
1036         attr->la_blocks = blocks;
1037         attr->la_valid |= LA_BLOCKS | LA_BLKSIZE;
1038
1039 out:
1040         up_read(&obj->oo_guard);
1041         return rc;
1042 }
1043
1044 /* Simple wrapper on top of qsd API which implement quota transfer for osd
1045  * setattr needs. As a reminder, only the root user can change ownership of
1046  * a file, that's why EDQUOT & EINPROGRESS errors are discarded */
1047 static inline int qsd_transfer(const struct lu_env *env,
1048                                struct qsd_instance *qsd,
1049                                struct lquota_trans *trans, int qtype,
1050                                __u64 orig_id, __u64 new_id, __u64 bspace,
1051                                struct lquota_id_info *qi, bool ignore_edquot)
1052 {
1053         int     rc;
1054
1055         if (unlikely(qsd == NULL))
1056                 return 0;
1057
1058         LASSERT(qtype >= 0 && qtype < LL_MAXQUOTAS);
1059         qi->lqi_type = qtype;
1060
1061         /* inode accounting */
1062         qi->lqi_is_blk = false;
1063
1064         /* one more inode for the new owner ... */
1065         qi->lqi_id.qid_uid = new_id;
1066         qi->lqi_space      = 1;
1067         rc = qsd_op_begin(env, qsd, trans, qi, NULL);
1068         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
1069                 rc = 0;
1070         if (rc)
1071                 return rc;
1072
1073         /* and one less inode for the current id */
1074         qi->lqi_id.qid_uid = orig_id;;
1075         qi->lqi_space      = -1;
1076         /* can't get EDQUOT when reducing usage */
1077         rc = qsd_op_begin(env, qsd, trans, qi, NULL);
1078         if (rc == -EINPROGRESS)
1079                 rc = 0;
1080         if (rc)
1081                 return rc;
1082
1083         /* block accounting */
1084         qi->lqi_is_blk = true;
1085
1086         /* more blocks for the new owner ... */
1087         qi->lqi_id.qid_uid = new_id;
1088         qi->lqi_space      = bspace;
1089         rc = qsd_op_begin(env, qsd, trans, qi, NULL);
1090         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
1091                 rc = 0;
1092         if (rc)
1093                 return rc;
1094
1095         /* and finally less blocks for the current owner */
1096         qi->lqi_id.qid_uid = orig_id;
1097         qi->lqi_space      = -bspace;
1098         rc = qsd_op_begin(env, qsd, trans, qi, NULL);
1099         /* can't get EDQUOT when reducing usage */
1100         if (rc == -EINPROGRESS)
1101                 rc = 0;
1102         return rc;
1103 }
1104
1105 static int osd_declare_attr_set(const struct lu_env *env,
1106                                 struct dt_object *dt,
1107                                 const struct lu_attr *attr,
1108                                 struct thandle *handle)
1109 {
1110         struct osd_thread_info  *info = osd_oti_get(env);
1111         struct osd_object       *obj = osd_dt_obj(dt);
1112         struct osd_device       *osd = osd_obj2dev(obj);
1113         dmu_tx_hold_t           *txh;
1114         struct osd_thandle      *oh;
1115         uint64_t                 bspace;
1116         uint32_t                 blksize;
1117         int                      rc = 0;
1118         bool                     found;
1119         ENTRY;
1120
1121
1122         LASSERT(handle != NULL);
1123         LASSERT(osd_invariant(obj));
1124
1125         oh = container_of(handle, struct osd_thandle, ot_super);
1126
1127         down_read(&obj->oo_guard);
1128         if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
1129                 GOTO(out, rc = 0);
1130
1131         LASSERT(obj->oo_sa_hdl != NULL);
1132         LASSERT(oh->ot_tx != NULL);
1133         /* regular attributes are part of the bonus buffer */
1134         /* let's check whether this object is already part of
1135          * transaction.. */
1136         found = false;
1137         for (txh = list_head(&oh->ot_tx->tx_holds); txh;
1138              txh = list_next(&oh->ot_tx->tx_holds, txh)) {
1139                 if (txh->txh_dnode == NULL)
1140                         continue;
1141                 if (txh->txh_dnode->dn_object != obj->oo_dn->dn_object)
1142                         continue;
1143                 /* this object is part of the transaction already
1144                  * we don't need to declare bonus again */
1145                 found = true;
1146                 break;
1147         }
1148         if (!found)
1149                 dmu_tx_hold_bonus(oh->ot_tx, obj->oo_dn->dn_object);
1150         if (oh->ot_tx->tx_err != 0)
1151                 GOTO(out, rc = -oh->ot_tx->tx_err);
1152
1153         if (attr && attr->la_valid & LA_FLAGS) {
1154                 /* LMA is usually a part of bonus, no need to declare
1155                  * anything else */
1156         }
1157
1158         if (attr && (attr->la_valid & (LA_UID | LA_GID | LA_PROJID))) {
1159                 sa_object_size(obj->oo_sa_hdl, &blksize, &bspace);
1160                 bspace = toqb(bspace * 512);
1161
1162                 CDEBUG(D_QUOTA,
1163                        "%s: enforce quota on UID %u, GID %u, the quota space is %lld (%u)\n",
1164                        osd->od_svname,
1165                        attr->la_uid, attr->la_gid, bspace, blksize);
1166         }
1167
1168         if (attr && attr->la_valid & LA_UID) {
1169                 /* quota enforcement for user */
1170                 if (attr->la_uid != obj->oo_attr.la_uid) {
1171                         rc = qsd_transfer(env, osd_def_qsd(osd),
1172                                           &oh->ot_quota_trans, USRQUOTA,
1173                                           obj->oo_attr.la_uid, attr->la_uid,
1174                                           bspace, &info->oti_qi, true);
1175                         if (rc)
1176                                 GOTO(out, rc);
1177                 }
1178         }
1179         if (attr && attr->la_valid & LA_GID) {
1180                 /* quota enforcement for group */
1181                 if (attr->la_gid != obj->oo_attr.la_gid) {
1182                         rc = qsd_transfer(env, osd_def_qsd(osd),
1183                                           &oh->ot_quota_trans, GRPQUOTA,
1184                                           obj->oo_attr.la_gid, attr->la_gid,
1185                                           bspace, &info->oti_qi,
1186                                           !(attr->la_flags &
1187                                                         LUSTRE_SET_SYNC_FL));
1188                         if (rc)
1189                                 GOTO(out, rc);
1190                 }
1191         }
1192 #ifdef ZFS_PROJINHERIT
1193         if (attr && attr->la_valid & LA_PROJID) {
1194                 /* quota enforcement for project */
1195                 if (attr->la_projid != obj->oo_attr.la_projid) {
1196                         if (!osd->od_projectused_dn)
1197                                 GOTO(out, rc = -EOPNOTSUPP);
1198
1199                         /* Usually, if project quota is upgradable for the
1200                          * device, then the upgrade will be done before or when
1201                          * mount the device. So when we come here, this project
1202                          * should have project ID attribute already (that is
1203                          * zero by default).  Otherwise, there was something
1204                          * wrong during the former upgrade, let's return failure
1205                          * to report that.
1206                          *
1207                          * Please note that, different from other attributes,
1208                          * you can NOT simply set the project ID attribute under
1209                          * such case, because adding (NOT change) project ID
1210                          * attribute needs to change the object's attribute
1211                          * layout to match zfs backend quota accounting
1212                          * requirement. */
1213                         if (unlikely(!obj->oo_with_projid))
1214                                 GOTO(out, rc = -ENXIO);
1215
1216                         rc = qsd_transfer(env, osd_def_qsd(osd),
1217                                           &oh->ot_quota_trans, PRJQUOTA,
1218                                           obj->oo_attr.la_projid,
1219                                           attr->la_projid, bspace,
1220                                           &info->oti_qi, true);
1221                         if (rc)
1222                                 GOTO(out, rc);
1223                 }
1224         }
1225 #endif
1226 out:
1227         up_read(&obj->oo_guard);
1228         RETURN(rc);
1229 }
1230
1231 /*
1232  * Set the attributes of an object
1233  *
1234  * The transaction passed to this routine must have
1235  * dmu_tx_hold_bonus(tx, oid) called and then assigned
1236  * to a transaction group.
1237  */
1238 static int osd_attr_set(const struct lu_env *env, struct dt_object *dt,
1239                         const struct lu_attr *la, struct thandle *handle)
1240 {
1241         struct osd_thread_info  *info = osd_oti_get(env);
1242         sa_bulk_attr_t          *bulk = osd_oti_get(env)->oti_attr_bulk;
1243         struct osd_object       *obj = osd_dt_obj(dt);
1244         struct osd_device       *osd = osd_obj2dev(obj);
1245         struct osd_thandle      *oh;
1246         struct osa_attr         *osa = &info->oti_osa;
1247         __u64                    valid = la->la_valid;
1248         int                      cnt;
1249         int                      rc = 0;
1250
1251         ENTRY;
1252
1253         down_read(&obj->oo_guard);
1254         if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
1255                 GOTO(out, rc = -ENOENT);
1256
1257         LASSERT(handle != NULL);
1258         LASSERT(osd_invariant(obj));
1259         LASSERT(obj->oo_sa_hdl);
1260
1261         oh = container_of(handle, struct osd_thandle, ot_super);
1262         /* Assert that the transaction has been assigned to a
1263            transaction group. */
1264         LASSERT(oh->ot_tx->tx_txg != 0);
1265
1266         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING) && !osd->od_is_ost) {
1267                 struct zpl_direntry *zde = &info->oti_zde.lzd_reg;
1268                 char *buf = info->oti_str;
1269                 dnode_t *zdn = NULL;
1270                 uint64_t zapid;
1271
1272                 zapid = osd_get_name_n_idx(env, osd, lu_object_fid(&dt->do_lu),
1273                                            buf, sizeof(info->oti_str), &zdn);
1274                 rc = osd_zap_lookup(osd, zapid, zdn, buf, 8,
1275                                     sizeof(*zde) / 8, zde);
1276                 if (!rc) {
1277                         zde->zde_dnode -= 1;
1278                         rc = -zap_update(osd->od_os, zapid, buf, 8,
1279                                          sizeof(*zde) / 8, zde, oh->ot_tx);
1280                 }
1281                 if (rc > 0)
1282                         rc = 0;
1283                 GOTO(out, rc);
1284         }
1285
1286         /* Only allow set size for regular file */
1287         if (!S_ISREG(dt->do_lu.lo_header->loh_attr))
1288                 valid &= ~(LA_SIZE | LA_BLOCKS);
1289
1290         if (valid & LA_CTIME && la->la_ctime == obj->oo_attr.la_ctime)
1291                 valid &= ~LA_CTIME;
1292
1293         if (valid & LA_MTIME && la->la_mtime == obj->oo_attr.la_mtime)
1294                 valid &= ~LA_MTIME;
1295
1296         if (valid & LA_ATIME && la->la_atime == obj->oo_attr.la_atime)
1297                 valid &= ~LA_ATIME;
1298
1299         if (valid == 0)
1300                 GOTO(out, rc = 0);
1301
1302         if (valid & LA_FLAGS) {
1303                 struct lustre_mdt_attrs *lma;
1304                 struct lu_buf buf;
1305                 int size = 0;
1306
1307                 if (la->la_flags & LUSTRE_LMA_FL_MASKS) {
1308                         LASSERT(!obj->oo_pfid_in_lma);
1309                         BUILD_BUG_ON(sizeof(info->oti_buf) < sizeof(*lma));
1310                         lma = (struct lustre_mdt_attrs *)&info->oti_buf;
1311                         buf.lb_buf = lma;
1312                         buf.lb_len = sizeof(info->oti_buf);
1313
1314                         /* Please do NOT call osd_xattr_get() directly, that
1315                          * will cause recursive down_read() on oo_guard. */
1316                         rc = osd_xattr_get_internal(env, obj, &buf,
1317                                                     XATTR_NAME_LMA, &size);
1318                         if (!rc && unlikely(size < sizeof(*lma))) {
1319                                 rc = -EINVAL;
1320                         } else if (!rc) {
1321                                 lma->lma_incompat =
1322                                         le32_to_cpu(lma->lma_incompat);
1323                                 lma->lma_incompat |=
1324                                         lustre_to_lma_flags(la->la_flags);
1325                                 lma->lma_incompat =
1326                                         cpu_to_le32(lma->lma_incompat);
1327                                 buf.lb_buf = lma;
1328                                 buf.lb_len = sizeof(*lma);
1329                                 rc = osd_xattr_set_internal(env, obj, &buf,
1330                                                             XATTR_NAME_LMA,
1331                                                             LU_XATTR_REPLACE,
1332                                                             oh);
1333                         }
1334                         if (rc < 0) {
1335                                 CWARN("%s: failed to set LMA flags: rc = %d\n",
1336                                        osd->od_svname, rc);
1337                                 GOTO(out, rc);
1338                         }
1339                 }
1340         }
1341
1342         write_lock(&obj->oo_attr_lock);
1343         cnt = 0;
1344
1345         if (valid & LA_PROJID) {
1346 #ifdef ZFS_PROJINHERIT
1347                 if (osd->od_projectused_dn) {
1348                         LASSERT(obj->oo_with_projid);
1349
1350                         osa->projid = obj->oo_attr.la_projid = la->la_projid;
1351                         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_PROJID(osd), NULL,
1352                                          &osa->projid, 8);
1353                 } else
1354 #endif
1355                         valid &= ~LA_PROJID;
1356         }
1357
1358         if (valid & LA_ATIME) {
1359                 osa->atime[0] = obj->oo_attr.la_atime = la->la_atime;
1360                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(osd), NULL,
1361                                  osa->atime, 16);
1362         }
1363         if (valid & LA_MTIME) {
1364                 osa->mtime[0] = obj->oo_attr.la_mtime = la->la_mtime;
1365                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(osd), NULL,
1366                                  osa->mtime, 16);
1367         }
1368         if (valid & LA_CTIME) {
1369                 osa->ctime[0] = obj->oo_attr.la_ctime = la->la_ctime;
1370                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(osd), NULL,
1371                                  osa->ctime, 16);
1372         }
1373         if (valid & LA_MODE) {
1374                 /* mode is stored along with type, so read it first */
1375                 obj->oo_attr.la_mode = (obj->oo_attr.la_mode & S_IFMT) |
1376                         (la->la_mode & ~S_IFMT);
1377                 osa->mode = obj->oo_attr.la_mode;
1378                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(osd), NULL,
1379                                  &osa->mode, 8);
1380         }
1381         if (valid & LA_SIZE) {
1382                 osa->size = obj->oo_attr.la_size = la->la_size;
1383                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(osd), NULL,
1384                                  &osa->size, 8);
1385         }
1386         if (valid & LA_NLINK) {
1387                 osa->nlink = obj->oo_attr.la_nlink = la->la_nlink;
1388                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(osd), NULL,
1389                                  &osa->nlink, 8);
1390         }
1391         if (valid & LA_RDEV) {
1392                 osa->rdev = obj->oo_attr.la_rdev = la->la_rdev;
1393                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_RDEV(osd), NULL,
1394                                  &osa->rdev, 8);
1395         }
1396         if (valid & LA_FLAGS) {
1397                 osa->flags = attrs_fs2zfs(la->la_flags);
1398                 /* many flags are not supported by zfs, so ensure a good cached
1399                  * copy */
1400                 obj->oo_attr.la_flags = attrs_zfs2fs(osa->flags);
1401 #ifdef ZFS_PROJINHERIT
1402                 if (obj->oo_with_projid && osd->od_projectused_dn)
1403                         osa->flags |= ZFS_PROJID;
1404 #endif
1405                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(osd), NULL,
1406                                  &osa->flags, 8);
1407         }
1408         if (valid & LA_UID) {
1409                 osa->uid = obj->oo_attr.la_uid = la->la_uid;
1410                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(osd), NULL,
1411                                  &osa->uid, 8);
1412         }
1413         if (valid & LA_GID) {
1414                 osa->gid = obj->oo_attr.la_gid = la->la_gid;
1415                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(osd), NULL,
1416                                  &osa->gid, 8);
1417         }
1418         obj->oo_attr.la_valid |= valid;
1419         write_unlock(&obj->oo_attr_lock);
1420
1421         LASSERT(cnt <= ARRAY_SIZE(osd_oti_get(env)->oti_attr_bulk));
1422         rc = osd_object_sa_bulk_update(obj, bulk, cnt, oh);
1423
1424 out:
1425         up_read(&obj->oo_guard);
1426         RETURN(rc);
1427 }
1428
1429 /*
1430  * Object creation.
1431  *
1432  * XXX temporary solution.
1433  */
1434
1435 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
1436                         struct dt_object *parent, struct dt_object *child,
1437                         umode_t child_mode)
1438 {
1439         LASSERT(ah);
1440
1441         ah->dah_parent = parent;
1442         ah->dah_mode = child_mode;
1443
1444         if (parent != NULL && !dt_object_remote(parent)) {
1445                 /* will help to find FID->ino at dt_insert("..") */
1446                 struct osd_object *pobj = osd_dt_obj(parent);
1447
1448                 osd_idc_find_and_init(env, osd_obj2dev(pobj), pobj);
1449         }
1450 }
1451
1452 static int osd_declare_create(const struct lu_env *env, struct dt_object *dt,
1453                               struct lu_attr *attr,
1454                               struct dt_allocation_hint *hint,
1455                               struct dt_object_format *dof,
1456                               struct thandle *handle)
1457 {
1458         const struct lu_fid     *fid = lu_object_fid(&dt->do_lu);
1459         struct osd_object       *obj = osd_dt_obj(dt);
1460         struct osd_device       *osd = osd_obj2dev(obj);
1461         struct osd_thandle      *oh;
1462         uint64_t                 zapid;
1463         dnode_t                 *dn;
1464         int                      rc, dnode_size;
1465         ENTRY;
1466
1467         LASSERT(dof);
1468
1469         switch (dof->dof_type) {
1470                 case DFT_REGULAR:
1471                 case DFT_SYM:
1472                 case DFT_NODE:
1473                         if (obj->oo_dt.do_body_ops == NULL)
1474                                 obj->oo_dt.do_body_ops = &osd_body_ops;
1475                         break;
1476                 default:
1477                         break;
1478         }
1479
1480         LASSERT(handle != NULL);
1481         oh = container_of(handle, struct osd_thandle, ot_super);
1482         LASSERT(oh->ot_tx != NULL);
1483
1484         /* this is the minimum set of EAs on every Lustre object */
1485         obj->oo_ea_in_bonus = OSD_BASE_EA_IN_BONUS;
1486         /* reserve 32 bytes for extra stuff like ACLs */
1487         dnode_size = size_roundup_power2(obj->oo_ea_in_bonus + 32);
1488
1489         switch (dof->dof_type) {
1490                 case DFT_DIR:
1491                         dt->do_index_ops = &osd_dir_ops;
1492                         /* fallthrough */
1493                 case DFT_INDEX:
1494                         /* for zap create */
1495                         dmu_tx_hold_zap(oh->ot_tx, DMU_NEW_OBJECT, FALSE, NULL);
1496                         dmu_tx_hold_sa_create(oh->ot_tx, dnode_size);
1497                         break;
1498                 case DFT_REGULAR:
1499                 case DFT_SYM:
1500                 case DFT_NODE:
1501                         /* first, we'll create new object */
1502                         dmu_tx_hold_sa_create(oh->ot_tx, dnode_size);
1503                         break;
1504
1505                 default:
1506                         LBUG();
1507                         break;
1508         }
1509
1510         /* and we'll add it to some mapping */
1511         zapid = osd_get_name_n_idx(env, osd, fid, NULL, 0, &dn);
1512         osd_tx_hold_zap(oh->ot_tx, zapid, dn, TRUE, NULL);
1513
1514         /* will help to find FID->ino mapping at dt_insert() */
1515         osd_idc_find_and_init(env, osd, obj);
1516
1517         rc = osd_declare_quota(env, osd, attr->la_uid, attr->la_gid,
1518                                attr->la_projid, 1, oh, NULL, OSD_QID_INODE);
1519
1520         RETURN(rc);
1521 }
1522
1523 int __osd_attr_init(const struct lu_env *env, struct osd_device *osd,
1524                     struct osd_object *obj, sa_handle_t *sa_hdl, dmu_tx_t *tx,
1525                     struct lu_attr *la, uint64_t parent,
1526                     nvlist_t *xattr)
1527 {
1528         sa_bulk_attr_t *bulk = osd_oti_get(env)->oti_attr_bulk;
1529         struct osa_attr *osa = &osd_oti_get(env)->oti_osa;
1530         uint64_t gen;
1531         inode_timespec_t now;
1532         int cnt;
1533         int rc;
1534         char *dxattr = NULL;
1535         size_t sa_size;
1536
1537
1538         LASSERT(sa_hdl);
1539
1540         gen = dmu_tx_get_txg(tx);
1541         gethrestime(&now);
1542         ZFS_TIME_ENCODE(&now, osa->btime);
1543
1544         osa->atime[0] = la->la_atime;
1545         osa->ctime[0] = la->la_ctime;
1546         osa->mtime[0] = la->la_mtime;
1547         osa->mode = la->la_mode;
1548         osa->uid = la->la_uid;
1549         osa->gid = la->la_gid;
1550         osa->rdev = la->la_rdev;
1551         osa->nlink = la->la_nlink;
1552         if (la->la_valid & LA_FLAGS)
1553                 osa->flags = attrs_fs2zfs(la->la_flags);
1554         else
1555                 osa->flags = 0;
1556         osa->size  = la->la_size;
1557 #ifdef ZFS_PROJINHERIT
1558         if (osd->od_projectused_dn) {
1559                 if (la->la_valid & LA_PROJID)
1560                         osa->projid = la->la_projid;
1561                 else
1562                         osa->projid = ZFS_DEFAULT_PROJID;
1563                 osa->flags |= ZFS_PROJID;
1564                 if (obj)
1565                         obj->oo_with_projid = 1;
1566         } else {
1567                 osa->flags &= ~ZFS_PROJID;
1568         }
1569 #endif
1570
1571         /*
1572          * we need to create all SA below upon object create.
1573          *
1574          * XXX The attribute order matters since the accounting callback relies
1575          * on static offsets (i.e. SA_*_OFFSET, see zfs_space_delta_cb()) to
1576          * look up the UID/GID/PROJID attributes. Moreover, the callback does
1577          * not seem to support the spill block.
1578          * We define attributes in the same order as SA_*_OFFSET in order to
1579          * work around the problem. See ORI-610.
1580          */
1581         cnt = 0;
1582         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(osd), NULL, &osa->mode, 8);
1583         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(osd), NULL, &osa->size, 8);
1584         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GEN(osd), NULL, &gen, 8);
1585         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(osd), NULL, &osa->uid, 8);
1586         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(osd), NULL, &osa->gid, 8);
1587         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_PARENT(osd), NULL, &parent, 8);
1588         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(osd), NULL, &osa->flags, 8);
1589         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(osd), NULL, osa->atime, 16);
1590         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(osd), NULL, osa->mtime, 16);
1591         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(osd), NULL, osa->ctime, 16);
1592         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CRTIME(osd), NULL, osa->btime, 16);
1593         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(osd), NULL, &osa->nlink, 8);
1594 #ifdef ZFS_PROJINHERIT
1595         if (osd->od_projectused_dn)
1596                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_PROJID(osd), NULL,
1597                                  &osa->projid, 8);
1598 #endif
1599         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_RDEV(osd), NULL, &osa->rdev, 8);
1600         LASSERT(cnt <= ARRAY_SIZE(osd_oti_get(env)->oti_attr_bulk));
1601
1602         if (xattr) {
1603                 rc = -nvlist_size(xattr, &sa_size, NV_ENCODE_XDR);
1604                 LASSERT(rc == 0);
1605
1606                 dxattr = osd_zio_buf_alloc(sa_size);
1607                 LASSERT(dxattr);
1608
1609                 rc = -nvlist_pack(xattr, &dxattr, &sa_size,
1610                                 NV_ENCODE_XDR, KM_SLEEP);
1611                 LASSERT(rc == 0);
1612
1613                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_DXATTR(osd),
1614                                 NULL, dxattr, sa_size);
1615         }
1616
1617         rc = -sa_replace_all_by_template(sa_hdl, bulk, cnt, tx);
1618         if (dxattr)
1619                 osd_zio_buf_free(dxattr, sa_size);
1620
1621         return rc;
1622 }
1623
1624 int osd_find_new_dnode(const struct lu_env *env, dmu_tx_t *tx,
1625                        uint64_t oid, dnode_t **dnp)
1626 {
1627         dmu_tx_hold_t *txh;
1628         int rc = 0;
1629
1630         /* take dnode_t from tx to save on dnode#->dnode_t lookup */
1631         for (txh = list_tail(&tx->tx_holds); txh;
1632              txh = list_prev(&tx->tx_holds, txh)) {
1633                 dnode_t *dn = txh->txh_dnode;
1634                 dmu_buf_impl_t *db;
1635
1636                 if (dn == NULL)
1637                         continue;
1638                 if (dn->dn_object != oid)
1639                         continue;
1640                 db = dn->dn_bonus;
1641                 if (db == NULL) {
1642                         rw_enter(&dn->dn_struct_rwlock, RW_WRITER);
1643                         if (dn->dn_bonus == NULL)
1644                                 dbuf_create_bonus(dn);
1645                         rw_exit(&dn->dn_struct_rwlock);
1646                 }
1647                 db = dn->dn_bonus;
1648                 LASSERT(db);
1649                 LASSERT(dn->dn_handle);
1650                 DB_DNODE_ENTER(db);
1651                 if (zfs_refcount_add(&db->db_holds, osd_obj_tag) == 1) {
1652                         zfs_refcount_add(&dn->dn_holds, osd_obj_tag);
1653                         atomic_inc_32(&dn->dn_dbufs_count);
1654                 }
1655                 *dnp = dn;
1656                 DB_DNODE_EXIT(db);
1657                 dbuf_read(db, NULL, DB_RF_MUST_SUCCEED | DB_RF_NOPREFETCH);
1658                 break;
1659         }
1660
1661         if (unlikely(*dnp == NULL))
1662                 rc = __osd_obj2dnode(tx->tx_objset, oid, dnp);
1663
1664         return rc;
1665 }
1666
1667 #ifdef HAVE_DMU_OBJECT_ALLOC_DNSIZE
1668 int osd_find_dnsize(struct osd_device *osd, int ea_in_bonus)
1669 {
1670         int dnsize;
1671
1672         if (osd->od_dnsize == ZFS_DNSIZE_AUTO) {
1673                 dnsize = DNODE_MIN_SIZE;
1674                 do {
1675                         if (DN_BONUS_SIZE(dnsize) >= ea_in_bonus + 32)
1676                                 break;
1677                         dnsize <<= 1;
1678                 } while (dnsize < DNODE_MAX_SIZE);
1679                 if (dnsize > DNODE_MAX_SIZE)
1680                         dnsize = DNODE_MAX_SIZE;
1681         } else if (osd->od_dnsize == ZFS_DNSIZE_1K) {
1682                 dnsize = 1024;
1683         } else if (osd->od_dnsize == ZFS_DNSIZE_2K) {
1684                 dnsize = 2048;
1685         } else if (osd->od_dnsize == ZFS_DNSIZE_4K) {
1686                 dnsize = 4096;
1687         } else if (osd->od_dnsize == ZFS_DNSIZE_8K) {
1688                 dnsize = 8192;
1689         } else if (osd->od_dnsize == ZFS_DNSIZE_16K) {
1690                 dnsize = 16384;
1691         } else {
1692                 dnsize = DNODE_MIN_SIZE;
1693         }
1694         return dnsize;
1695 }
1696 #endif
1697
1698 /*
1699  * The transaction passed to this routine must have
1700  * dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT) called and then assigned
1701  * to a transaction group.
1702  */
1703 int __osd_object_create(const struct lu_env *env, struct osd_device *osd,
1704                         struct osd_object *obj, const struct lu_fid *fid,
1705                         dnode_t **dnp, dmu_tx_t *tx, struct lu_attr *la)
1706 {
1707         dmu_object_type_t type = DMU_OT_PLAIN_FILE_CONTENTS;
1708         uint64_t oid;
1709         int size;
1710
1711         /* Use DMU_OTN_UINT8_METADATA for local objects so their data blocks
1712          * would get an additional ditto copy */
1713         if (unlikely(S_ISREG(la->la_mode) &&
1714                      fid_seq_is_local_file(fid_seq(fid))))
1715                 type = DMU_OTN_UINT8_METADATA;
1716
1717         /* Create a new DMU object using the default dnode size. */
1718         if (obj)
1719                 size = obj->oo_ea_in_bonus;
1720         else
1721                 size = OSD_BASE_EA_IN_BONUS;
1722         oid = osd_dmu_object_alloc(osd->od_os, type, 0,
1723                                    osd_find_dnsize(osd, size), tx);
1724
1725         LASSERT(la->la_valid & LA_MODE);
1726         la->la_size = 0;
1727         la->la_nlink = 1;
1728
1729         return osd_find_new_dnode(env, tx, oid, dnp);
1730 }
1731
1732 /*
1733  * The transaction passed to this routine must have
1734  * dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, ...) called and then assigned
1735  * to a transaction group.
1736  *
1737  * Using ZAP_FLAG_HASH64 will force the ZAP to always be a FAT ZAP.
1738  * This is fine for directories today, because storing the FID in the dirent
1739  * will also require a FAT ZAP.  If there is a new type of micro ZAP created
1740  * then we might need to re-evaluate the use of this flag and instead do
1741  * a conversion from the different internal ZAP hash formats being used. */
1742 int __osd_zap_create(const struct lu_env *env, struct osd_device *osd,
1743                      dnode_t **dnp, dmu_tx_t *tx, struct lu_attr *la,
1744                      unsigned dnsize, zap_flags_t flags)
1745 {
1746         uint64_t oid;
1747
1748         /* Assert that the transaction has been assigned to a
1749            transaction group. */
1750         LASSERT(tx->tx_txg != 0);
1751         *dnp = NULL;
1752
1753         oid = osd_zap_create_flags(osd->od_os, 0, flags | ZAP_FLAG_HASH64,
1754                                    DMU_OT_DIRECTORY_CONTENTS,
1755                                    14, /* == ZFS fzap_default_blockshift */
1756                                    DN_MAX_INDBLKSHIFT, /* indirect blockshift */
1757                                    dnsize, tx);
1758
1759         la->la_size = 2;
1760         la->la_nlink = 1;
1761
1762         return osd_find_new_dnode(env, tx, oid, dnp);
1763 }
1764
1765 static dnode_t *osd_mkidx(const struct lu_env *env, struct osd_object *obj,
1766                           struct lu_attr *la, struct osd_thandle *oh)
1767 {
1768         struct osd_device *osd = osd_obj2dev(obj);
1769         dnode_t *dn;
1770         int rc;
1771
1772         /* Index file should be created as regular file in order not to confuse
1773          * ZPL which could interpret them as directory.
1774          * We set ZAP_FLAG_UINT64_KEY to let ZFS know than we are going to use
1775          * binary keys */
1776         LASSERT(S_ISREG(la->la_mode));
1777         rc = __osd_zap_create(env, osd, &dn, oh->ot_tx, la,
1778                 osd_find_dnsize(osd, obj->oo_ea_in_bonus), ZAP_FLAG_UINT64_KEY);
1779         if (rc)
1780                 return ERR_PTR(rc);
1781         return dn;
1782 }
1783
1784 static dnode_t *osd_mkdir(const struct lu_env *env, struct osd_object *obj,
1785                           struct lu_attr *la, struct osd_thandle *oh)
1786 {
1787         struct osd_device *osd = osd_obj2dev(obj);
1788         dnode_t *dn;
1789         int rc;
1790
1791         LASSERT(S_ISDIR(la->la_mode));
1792         rc = __osd_zap_create(env, osd, &dn, oh->ot_tx, la,
1793                               osd_find_dnsize(osd, obj->oo_ea_in_bonus), 0);
1794         if (rc)
1795                 return ERR_PTR(rc);
1796         return dn;
1797 }
1798
1799 static dnode_t *osd_mkreg(const struct lu_env *env, struct osd_object *obj,
1800                           struct lu_attr *la, struct osd_thandle *oh)
1801 {
1802         const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
1803         struct osd_device *osd = osd_obj2dev(obj);
1804         dnode_t *dn;
1805         int rc;
1806
1807         LASSERT(S_ISREG(la->la_mode));
1808         rc = __osd_object_create(env, osd, obj, fid, &dn, oh->ot_tx, la);
1809         if (rc)
1810                 return ERR_PTR(rc);
1811
1812         if ((fid_is_idif(fid) || fid_is_norm(fid) || fid_is_echo(fid))) {
1813                 /* The minimum block size must be at least page size otherwise
1814                  * it will break the assumption in tgt_thread_big_cache where
1815                  * the array size is PTLRPC_MAX_BRW_PAGES. It will also affect
1816                  * RDMA due to subpage transfer size */
1817                 rc = -dmu_object_set_blocksize(osd->od_os, dn->dn_object,
1818                                                PAGE_SIZE, 0, oh->ot_tx);
1819                 if (unlikely(rc)) {
1820                         CERROR("%s: can't change blocksize: %d\n",
1821                                osd->od_svname, rc);
1822                         return ERR_PTR(rc);
1823                 }
1824         } else if ((fid_is_llog(fid))) {
1825                 rc = -dmu_object_set_blocksize(osd->od_os, dn->dn_object,
1826                                                LLOG_MIN_CHUNK_SIZE, 0, oh->ot_tx);
1827                 if (unlikely(rc)) {
1828                         CERROR("%s: can't change blocksize: %d\n",
1829                                osd->od_svname, rc);
1830                         return ERR_PTR(rc);
1831                 }
1832         }
1833
1834         return dn;
1835 }
1836
1837 static dnode_t *osd_mksym(const struct lu_env *env, struct osd_object *obj,
1838                           struct lu_attr *la, struct osd_thandle *oh)
1839 {
1840         dnode_t *dn;
1841         int rc;
1842
1843         LASSERT(S_ISLNK(la->la_mode));
1844         rc = __osd_object_create(env, osd_obj2dev(obj), obj,
1845                                  lu_object_fid(&obj->oo_dt.do_lu),
1846                                  &dn, oh->ot_tx, la);
1847         if (rc)
1848                 return ERR_PTR(rc);
1849         return dn;
1850 }
1851
1852 static dnode_t *osd_mknod(const struct lu_env *env, struct osd_object *obj,
1853                           struct lu_attr *la, struct osd_thandle *oh)
1854 {
1855         dnode_t *dn;
1856         int rc;
1857
1858         if (S_ISCHR(la->la_mode) || S_ISBLK(la->la_mode))
1859                 la->la_valid |= LA_RDEV;
1860
1861         rc = __osd_object_create(env, osd_obj2dev(obj), obj,
1862                                  lu_object_fid(&obj->oo_dt.do_lu),
1863                                  &dn, oh->ot_tx, la);
1864         if (rc)
1865                 return ERR_PTR(rc);
1866         return dn;
1867 }
1868
1869 typedef dnode_t *(*osd_obj_type_f)(const struct lu_env *env,
1870                                    struct osd_object *obj,
1871                                    struct lu_attr *la,
1872                                    struct osd_thandle *oh);
1873
1874 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
1875 {
1876         osd_obj_type_f result;
1877
1878         switch (type) {
1879         case DFT_DIR:
1880                 result = osd_mkdir;
1881                 break;
1882         case DFT_INDEX:
1883                 result = osd_mkidx;
1884                 break;
1885         case DFT_REGULAR:
1886                 result = osd_mkreg;
1887                 break;
1888         case DFT_SYM:
1889                 result = osd_mksym;
1890                 break;
1891         case DFT_NODE:
1892                 result = osd_mknod;
1893                 break;
1894         default:
1895                 LBUG();
1896                 break;
1897         }
1898         return result;
1899 }
1900
1901 /*
1902  * Concurrency: @dt is write locked.
1903  */
1904 static int osd_create(const struct lu_env *env, struct dt_object *dt,
1905                       struct lu_attr *attr, struct dt_allocation_hint *hint,
1906                       struct dt_object_format *dof, struct thandle *th)
1907 {
1908         struct osd_thread_info  *info = osd_oti_get(env);
1909         struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
1910         struct zpl_direntry     *zde = &info->oti_zde.lzd_reg;
1911         const struct lu_fid     *fid = lu_object_fid(&dt->do_lu);
1912         struct osd_object       *obj = osd_dt_obj(dt);
1913         struct osd_device       *osd = osd_obj2dev(obj);
1914         char                    *buf = info->oti_str;
1915         struct osd_thandle      *oh;
1916         dnode_t *dn = NULL, *zdn = NULL;
1917         uint64_t                 zapid, parent = 0;
1918         int                      rc;
1919         __u32 compat = 0;
1920
1921         ENTRY;
1922
1923         LASSERT(!fid_is_acct(fid));
1924
1925         /* concurrent create declarations should not see
1926          * the object inconsistent (db, attr, etc).
1927          * in regular cases acquisition should be cheap */
1928         down_write(&obj->oo_guard);
1929
1930         if (unlikely(dt_object_exists(dt)))
1931                 GOTO(out, rc = -EEXIST);
1932
1933         LASSERT(osd_invariant(obj));
1934         LASSERT(dof != NULL);
1935
1936         LASSERT(th != NULL);
1937         oh = container_of(th, struct osd_thandle, ot_super);
1938
1939         LASSERT(obj->oo_dn == NULL);
1940
1941         /* to follow ZFS on-disk format we need
1942          * to initialize parent dnode properly */
1943         if (hint != NULL && hint->dah_parent != NULL &&
1944             !dt_object_remote(hint->dah_parent))
1945                 parent = osd_dt_obj(hint->dah_parent)->oo_dn->dn_object;
1946
1947         /* we may fix some attributes, better do not change the source */
1948         obj->oo_attr = *attr;
1949         obj->oo_attr.la_valid |= LA_SIZE | LA_NLINK | LA_TYPE;
1950
1951 #ifdef ZFS_PROJINHERIT
1952         if (osd->od_projectused_dn) {
1953                 if (!(obj->oo_attr.la_valid & LA_PROJID))
1954                         obj->oo_attr.la_projid = ZFS_DEFAULT_PROJID;
1955                 obj->oo_with_projid = 1;
1956         }
1957 #endif
1958
1959         dn = osd_create_type_f(dof->dof_type)(env, obj, &obj->oo_attr, oh);
1960         if (IS_ERR(dn)) {
1961                 rc = PTR_ERR(dn);
1962                 dn = NULL;
1963                 GOTO(out, rc);
1964         }
1965
1966         zde->zde_pad = 0;
1967         zde->zde_dnode = dn->dn_object;
1968         zde->zde_type = IFTODT(attr->la_mode & S_IFMT);
1969
1970         zapid = osd_get_name_n_idx(env, osd, fid, buf,
1971                                    sizeof(info->oti_str), &zdn);
1972         if (CFS_FAIL_CHECK(OBD_FAIL_OSD_NO_OI_ENTRY) ||
1973             (osd->od_is_ost && OBD_FAIL_CHECK(OBD_FAIL_OSD_COMPAT_NO_ENTRY)))
1974                 goto skip_add;
1975
1976         if (osd->od_is_ost && OBD_FAIL_CHECK(OBD_FAIL_OSD_COMPAT_INVALID_ENTRY))
1977                 zde->zde_dnode++;
1978
1979         rc = osd_zap_add(osd, zapid, zdn, buf, 8, 1, zde, oh->ot_tx);
1980         if (rc)
1981                 GOTO(out, rc);
1982
1983 skip_add:
1984         obj->oo_dn = dn;
1985         /* Now add in all of the "SA" attributes */
1986         rc = osd_sa_handle_get(obj);
1987         if (rc)
1988                 GOTO(out, rc);
1989
1990         rc = -nvlist_alloc(&obj->oo_sa_xattr, NV_UNIQUE_NAME, KM_SLEEP);
1991         if (rc)
1992                 GOTO(out, rc);
1993
1994         /* initialize LMA */
1995         if (fid_is_idif(fid) || (fid_is_norm(fid) && osd->od_is_ost))
1996                 compat |= LMAC_FID_ON_OST;
1997         lustre_lma_init(lma, fid, compat, 0);
1998         lustre_lma_swab(lma);
1999         rc = -nvlist_add_byte_array(obj->oo_sa_xattr, XATTR_NAME_LMA,
2000                                     (uchar_t *)lma, sizeof(*lma));
2001         if (rc)
2002                 GOTO(out, rc);
2003
2004         /* configure new osd object */
2005         obj->oo_parent = parent != 0 ? parent : zapid;
2006         obj->oo_late_attr_set = 1;
2007         rc = __osd_sa_xattr_schedule_update(env, obj, oh);
2008         if (rc)
2009                 GOTO(out, rc);
2010
2011         /* XXX: oo_lma_flags */
2012         obj->oo_dt.do_lu.lo_header->loh_attr |= obj->oo_attr.la_mode & S_IFMT;
2013         if (likely(!fid_is_acct(lu_object_fid(&obj->oo_dt.do_lu))))
2014                 /* no body operations for accounting objects */
2015                 obj->oo_dt.do_body_ops = &osd_body_ops;
2016
2017         osd_idc_find_and_init(env, osd, obj);
2018
2019 out:
2020         if (unlikely(rc && dn)) {
2021                 dmu_object_free(osd->od_os, dn->dn_object, oh->ot_tx);
2022                 osd_dnode_rele(dn);
2023                 obj->oo_dn = NULL;
2024         } else if (!rc) {
2025                 obj->oo_dt.do_lu.lo_header->loh_attr |= LOHA_EXISTS;
2026         }
2027         up_write(&obj->oo_guard);
2028         RETURN(rc);
2029 }
2030
2031 static int osd_declare_ref_add(const struct lu_env *env, struct dt_object *dt,
2032                                struct thandle *th)
2033 {
2034         osd_idc_find_and_init(env, osd_dev(dt->do_lu.lo_dev), osd_dt_obj(dt));
2035         return osd_declare_attr_set(env, dt, NULL, th);
2036 }
2037
2038 /*
2039  * Concurrency: @dt is write locked.
2040  */
2041 static int osd_ref_add(const struct lu_env *env, struct dt_object *dt,
2042                        struct thandle *handle)
2043 {
2044         struct osd_object       *obj = osd_dt_obj(dt);
2045         struct osd_thandle      *oh;
2046         struct osd_device       *osd = osd_obj2dev(obj);
2047         uint64_t                 nlink;
2048         int rc;
2049
2050         ENTRY;
2051
2052         down_read(&obj->oo_guard);
2053         if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
2054                 GOTO(out, rc = -ENOENT);
2055
2056         LASSERT(osd_invariant(obj));
2057         LASSERT(obj->oo_sa_hdl != NULL);
2058
2059         oh = container_of(handle, struct osd_thandle, ot_super);
2060
2061         write_lock(&obj->oo_attr_lock);
2062         nlink = ++obj->oo_attr.la_nlink;
2063         write_unlock(&obj->oo_attr_lock);
2064
2065         rc = osd_object_sa_update(obj, SA_ZPL_LINKS(osd), &nlink, 8, oh);
2066
2067 out:
2068         up_read(&obj->oo_guard);
2069         RETURN(rc);
2070 }
2071
2072 static int osd_declare_ref_del(const struct lu_env *env, struct dt_object *dt,
2073                                struct thandle *handle)
2074 {
2075         osd_idc_find_and_init(env, osd_dev(dt->do_lu.lo_dev), osd_dt_obj(dt));
2076         return osd_declare_attr_set(env, dt, NULL, handle);
2077 }
2078
2079 /*
2080  * Concurrency: @dt is write locked.
2081  */
2082 static int osd_ref_del(const struct lu_env *env, struct dt_object *dt,
2083                        struct thandle *handle)
2084 {
2085         struct osd_object       *obj = osd_dt_obj(dt);
2086         struct osd_thandle      *oh;
2087         struct osd_device       *osd = osd_obj2dev(obj);
2088         uint64_t                 nlink;
2089         int                      rc;
2090
2091         ENTRY;
2092
2093         down_read(&obj->oo_guard);
2094
2095         if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
2096                 GOTO(out, rc = -ENOENT);
2097
2098         LASSERT(osd_invariant(obj));
2099         LASSERT(obj->oo_sa_hdl != NULL);
2100
2101         oh = container_of(handle, struct osd_thandle, ot_super);
2102         LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
2103
2104         write_lock(&obj->oo_attr_lock);
2105         nlink = --obj->oo_attr.la_nlink;
2106         write_unlock(&obj->oo_attr_lock);
2107
2108         rc = osd_object_sa_update(obj, SA_ZPL_LINKS(osd), &nlink, 8, oh);
2109
2110 out:
2111         up_read(&obj->oo_guard);
2112         RETURN(rc);
2113 }
2114
2115 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
2116                            __u64 start, __u64 end)
2117 {
2118         struct osd_device *osd = osd_obj2dev(osd_dt_obj(dt));
2119         struct dmu_buf_impl *db = osd_dt_obj(dt)->oo_dn->dn_dbuf;
2120         uint64_t txg = 0;
2121         ENTRY;
2122
2123         if (osd->od_dt_dev.dd_rdonly)
2124                 RETURN(0);
2125
2126         mutex_enter(&db->db_mtx);
2127         if (db->db_last_dirty)
2128                 txg = db->db_last_dirty->dr_txg;
2129         mutex_exit(&db->db_mtx);
2130
2131         if (txg) {
2132                 /* the object is dirty or being synced */
2133                 if (osd_object_sync_delay_us < 0)
2134                         txg_wait_synced(dmu_objset_pool(osd->od_os), txg);
2135                 else
2136                         udelay(osd_object_sync_delay_us);
2137         }
2138
2139         RETURN(0);
2140 }
2141
2142 static int osd_invalidate(const struct lu_env *env, struct dt_object *dt)
2143 {
2144         return 0;
2145 }
2146
2147 static struct dt_object_operations osd_obj_ops = {
2148         .do_read_lock           = osd_read_lock,
2149         .do_write_lock          = osd_write_lock,
2150         .do_read_unlock         = osd_read_unlock,
2151         .do_write_unlock        = osd_write_unlock,
2152         .do_write_locked        = osd_write_locked,
2153         .do_attr_get            = osd_attr_get,
2154         .do_declare_attr_set    = osd_declare_attr_set,
2155         .do_attr_set            = osd_attr_set,
2156         .do_ah_init             = osd_ah_init,
2157         .do_declare_create      = osd_declare_create,
2158         .do_create              = osd_create,
2159         .do_declare_destroy     = osd_declare_destroy,
2160         .do_destroy             = osd_destroy,
2161         .do_index_try           = osd_index_try,
2162         .do_declare_ref_add     = osd_declare_ref_add,
2163         .do_ref_add             = osd_ref_add,
2164         .do_declare_ref_del     = osd_declare_ref_del,
2165         .do_ref_del             = osd_ref_del,
2166         .do_xattr_get           = osd_xattr_get,
2167         .do_declare_xattr_set   = osd_declare_xattr_set,
2168         .do_xattr_set           = osd_xattr_set,
2169         .do_declare_xattr_del   = osd_declare_xattr_del,
2170         .do_xattr_del           = osd_xattr_del,
2171         .do_xattr_list          = osd_xattr_list,
2172         .do_object_sync         = osd_object_sync,
2173         .do_invalidate          = osd_invalidate,
2174 };
2175
2176 static struct lu_object_operations osd_lu_obj_ops = {
2177         .loo_object_init        = osd_object_init,
2178         .loo_object_delete      = osd_object_delete,
2179         .loo_object_release     = osd_object_release,
2180         .loo_object_free        = osd_object_free,
2181         .loo_object_print       = osd_object_print,
2182         .loo_object_invariant   = osd_object_invariant,
2183 };
2184
2185 static int osd_otable_it_attr_get(const struct lu_env *env,
2186                                 struct dt_object *dt,
2187                                 struct lu_attr *attr)
2188 {
2189         attr->la_valid = 0;
2190         return 0;
2191 }
2192
2193 static struct dt_object_operations osd_obj_otable_it_ops = {
2194         .do_attr_get            = osd_otable_it_attr_get,
2195         .do_index_try           = osd_index_try,
2196 };
2197
2198 module_param(osd_object_sync_delay_us, int, 0644);
2199 MODULE_PARM_DESC(osd_object_sync_delay_us,
2200                  "If zero or larger delay N usec instead of doing object sync");