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) {
 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_late_attr_set)
 138                                 __osd_sa_attr_init(env, obj, oh);
 139                         else if (obj->oo_late_xattr)
 140                                 __osd_sa_xattr_update(env, obj, oh);
 141                         up_write(&obj->oo_guard);
 142                 }
 143                 sa_spill_rele(obj->oo_sa_hdl);
 144         }
 145 }
 146
 147 /*
 148  * Update the SA and add the object to the dirty list.
 149  */
 150 int osd_object_sa_update(struct osd_object *obj, sa_attr_type_t type,
 151                          void *buf, uint32_t buflen, struct osd_thandle *oh)
 152 {
 153         int rc;
 154
 155         LASSERT(obj->oo_sa_hdl != NULL);
 156         LASSERT(oh->ot_tx != NULL);
 157
 158         rc = -sa_update(obj->oo_sa_hdl, type, buf, buflen, oh->ot_tx);
 159         osd_object_sa_dirty_add(obj, oh);
 160
 161         return rc;
 162 }
 163
 164 /*
 165  * Bulk update the SA and add the object to the dirty list.
 166  */
 167 static int
 168 osd_object_sa_bulk_update(struct osd_object *obj, sa_bulk_attr_t *attrs,
 169                           int count, struct osd_thandle *oh)
 170 {
 171         int rc;
 172
 173         LASSERT(obj->oo_sa_hdl != NULL);
 174         LASSERT(oh->ot_tx != NULL);
 175
 176         rc = -sa_bulk_update(obj->oo_sa_hdl, attrs, count, oh->ot_tx);
 177         osd_object_sa_dirty_add(obj, oh);
 178
 179         return rc;
 180 }
 181
 182 /*
 183  * Retrieve the attributes of a DMU object
 184  */
 185 static int __osd_object_attr_get(const struct lu_env *env, struct osd_device *o,
 186                                  struct osd_object *obj, struct lu_attr *la)
 187 {
 188         struct osa_attr *osa = &osd_oti_get(env)->oti_osa;
 189         sa_bulk_attr_t *bulk = osd_oti_get(env)->oti_attr_bulk;
 190         struct lustre_mdt_attrs *lma;
 191         struct lu_buf buf;
 192         int cnt = 0;
 193         int              rc;
 194         ENTRY;
 195
 196         LASSERT(obj->oo_dn != NULL);
 197
 198         la->la_valid |= LA_ATIME | LA_MTIME | LA_CTIME | LA_BTIME | LA_MODE |
 199                         LA_TYPE | LA_SIZE | LA_UID | LA_GID | LA_FLAGS |
 200                         LA_NLINK;
 201
 202         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(o), NULL, osa->atime, 16);
 203         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(o), NULL, osa->mtime, 16);
 204         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(o), NULL, osa->ctime, 16);
 205         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CRTIME(o), NULL, osa->btime, 16);
 206         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(o), NULL, &osa->mode, 8);
 207         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(o), NULL, &osa->size, 8);
 208         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(o), NULL, &osa->nlink, 8);
 209         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(o), NULL, &osa->uid, 8);
 210         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(o), NULL, &osa->gid, 8);
 211         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(o), NULL, &osa->flags, 8);
 212         LASSERT(cnt <= ARRAY_SIZE(osd_oti_get(env)->oti_attr_bulk));
 213
 214         rc = -sa_bulk_lookup(obj->oo_sa_hdl, bulk, cnt);
 215         if (rc)
 216                 GOTO(out_sa, rc);
 217
 218 #ifdef ZFS_PROJINHERIT
 219         if (o->od_projectused_dn && osa->flags & ZFS_PROJID) {
 220                 rc = -sa_lookup(obj->oo_sa_hdl, SA_ZPL_PROJID(o),
 221                                 &osa->projid, 8);
 222                 if (rc)
 223                         GOTO(out_sa, rc);
 224
 225                 la->la_projid = osa->projid;
 226                 la->la_valid |= LA_PROJID;
 227                 obj->oo_with_projid = 1;
 228         } else {
 229                 la->la_projid = ZFS_DEFAULT_PROJID;
 230                 la->la_valid &= ~LA_PROJID;
 231         }
 232 #else
 233         la->la_projid = 0;
 234         la->la_valid &= ~LA_PROJID;
 235 #endif
 236
 237         la->la_atime = osa->atime[0];
 238         la->la_mtime = osa->mtime[0];
 239         la->la_ctime = osa->ctime[0];
 240         la->la_btime = osa->btime[0];
 241         la->la_mode = osa->mode;
 242         la->la_uid = osa->uid;
 243         la->la_gid = osa->gid;
 244         la->la_nlink = osa->nlink;
 245         la->la_flags = attrs_zfs2fs(osa->flags);
 246         la->la_size = osa->size;
 247
 248         /* Try to get extra flags from LMA */
 249         lma = (struct lustre_mdt_attrs *)osd_oti_get(env)->oti_buf;
 250         buf.lb_buf = lma;
 251         buf.lb_len = sizeof(osd_oti_get(env)->oti_buf);
 252         down_read(&obj->oo_guard);
 253         rc = osd_xattr_get_lma(env, obj, &buf);
 254         if (!rc) {
 255                 lma->lma_incompat = le32_to_cpu(lma->lma_incompat);
 256                 obj->oo_lma_flags =
 257                         lma_to_lustre_flags(lma->lma_incompat);
 258         } else if (rc == -ENODATA ||
 259                    !(S_ISDIR(la->la_mode) &&
 260                      dt_object_exists(&obj->oo_dt))) {
 261                 rc = 0;
 262         }
 263         up_read(&obj->oo_guard);
 264
 265         if (S_ISCHR(la->la_mode) || S_ISBLK(la->la_mode)) {
 266                 rc = -sa_lookup(obj->oo_sa_hdl, SA_ZPL_RDEV(o), &osa->rdev, 8);
 267                 if (rc)
 268                         GOTO(out_sa, rc);
 269                 la->la_rdev = osa->rdev;
 270                 la->la_valid |= LA_RDEV;
 271         }
 272 out_sa:
 273
 274         RETURN(rc);
 275 }
 276
 277 int __osd_obj2dnode(objset_t *os, uint64_t oid, dnode_t **dnp)
 278 {
 279         dmu_buf_t *db;
 280         dmu_buf_impl_t *dbi;
 281         int rc;
 282
 283         rc = -dmu_bonus_hold(os, oid, osd_obj_tag, &db);
 284         if (rc)
 285                 return rc;
 286
 287         dbi = (dmu_buf_impl_t *)db;
 288         DB_DNODE_ENTER(dbi);
 289         *dnp = DB_DNODE(dbi);
 290         DB_DNODE_EXIT(dbi);
 291         LASSERT(*dnp != NULL);
 292
 293         return 0;
 294 }
 295
 296 /*
 297  * Concurrency: no concurrent access is possible that early in object
 298  * life-cycle.
 299  */
 300 struct lu_object *osd_object_alloc(const struct lu_env *env,
 301                                    const struct lu_object_header *hdr,
 302                                    struct lu_device *d)
 303 {
 304         struct osd_object *mo;
 305
 306         OBD_SLAB_ALLOC_PTR_GFP(mo, osd_object_kmem, GFP_NOFS);
 307         if (mo != NULL) {
 308                 struct lu_object *l;
 309                 struct lu_object_header *h;
 310                 struct osd_device *o = osd_dev(d);
 311
 312                 l = &mo->oo_dt.do_lu;
 313                 if (unlikely(o->od_in_init)) {
 314                         OBD_ALLOC_PTR(h);
 315                         if (!h) {
 316                                 OBD_FREE_PTR(mo);
 317                                 return NULL;
 318                         }
 319
 320                         lu_object_header_init(h);
 321                         lu_object_init(l, h, d);
 322                         lu_object_add_top(h, l);
 323                         mo->oo_header = h;
 324                 } else {
 325                         dt_object_init(&mo->oo_dt, NULL, d);
 326                         mo->oo_header = NULL;
 327                 }
 328
 329                 mo->oo_dt.do_ops = &osd_obj_ops;
 330                 l->lo_ops = &osd_lu_obj_ops;
 331                 INIT_LIST_HEAD(&mo->oo_sa_linkage);
 332                 INIT_LIST_HEAD(&mo->oo_unlinked_linkage);
 333                 init_rwsem(&mo->oo_sem);
 334                 init_rwsem(&mo->oo_guard);
 335                 rwlock_init(&mo->oo_attr_lock);
 336                 mo->oo_destroy = OSD_DESTROY_NONE;
 337                 return l;
 338         } else {
 339                 return NULL;
 340         }
 341 }
 342
 343 static void osd_obj_set_blksize(const struct lu_env *env,
 344                                 struct osd_device *osd, struct osd_object *obj)
 345 {
 346         const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
 347         dmu_tx_t *tx;
 348         dnode_t *dn = obj->oo_dn;
 349         uint32_t blksz;
 350         int rc = 0;
 351         ENTRY;
 352
 353         LASSERT(!osd_oti_get(env)->oti_in_trans);
 354
 355         tx = dmu_tx_create(osd->od_os);
 356         if (!tx) {
 357                 CERROR("%s: fail to create tx to set blksize for "DFID"\n",
 358                        osd->od_svname, PFID(fid));
 359                 RETURN_EXIT;
 360         }
 361
 362         dmu_tx_hold_bonus(tx, dn->dn_object);
 363         rc = -dmu_tx_assign(tx, TXG_WAIT);
 364         if (rc) {
 365                 dmu_tx_abort(tx);
 366                 CERROR("%s: fail to assign tx to set blksize for "DFID
 367                        ": rc = %d\n", osd->od_svname, PFID(fid), rc);
 368                 RETURN_EXIT;
 369         }
 370
 371         down_write(&obj->oo_guard);
 372         if (unlikely((1 << dn->dn_datablkshift) >= PAGE_SIZE))
 373                 GOTO(out, rc = 1);
 374
 375         blksz = dn->dn_datablksz;
 376         if (!is_power_of_2(blksz))
 377                 blksz = size_roundup_power2(blksz);
 378
 379         if (blksz > osd->od_max_blksz)
 380                 blksz = osd->od_max_blksz;
 381         else if (blksz < PAGE_SIZE)
 382                 blksz = PAGE_SIZE;
 383         rc = -dmu_object_set_blocksize(osd->od_os, dn->dn_object, blksz, 0, tx);
 384
 385         GOTO(out, rc);
 386
 387 out:
 388         up_write(&obj->oo_guard);
 389         if (rc) {
 390                 dmu_tx_abort(tx);
 391                 if (unlikely(obj->oo_dn->dn_maxblkid > 0))
 392                         rc = 1;
 393                 if (rc < 0)
 394                         CERROR("%s: fail to set blksize for "DFID": rc = %d\n",
 395                                osd->od_svname, PFID(fid), rc);
 396         } else {
 397                 dmu_tx_commit(tx);
 398                 CDEBUG(D_INODE, "%s: set blksize as %u for "DFID"\n",
 399                        osd->od_svname, blksz, PFID(fid));
 400         }
 401 }
 402
 403 /*
 404  * Concurrency: shouldn't matter.
 405  */
 406 static int osd_object_init0(const struct lu_env *env, struct osd_object *obj)
 407 {
 408         struct osd_device       *osd = osd_obj2dev(obj);
 409         const struct lu_fid     *fid = lu_object_fid(&obj->oo_dt.do_lu);
 410         int                      rc = 0;
 411         ENTRY;
 412
 413         LASSERT(obj->oo_dn);
 414
 415         rc = osd_object_sa_init(obj, osd);
 416         if (rc)
 417                 RETURN(rc);
 418
 419         /* cache attrs in object */
 420         rc = __osd_object_attr_get(env, osd, obj, &obj->oo_attr);
 421         if (rc)
 422                 RETURN(rc);
 423
 424         if (likely(!fid_is_acct(fid))) {
 425                 /* no body operations for accounting objects */
 426                 obj->oo_dt.do_body_ops = &osd_body_ops;
 427
 428                 if (S_ISREG(obj->oo_attr.la_mode) &&
 429                     obj->oo_dn->dn_maxblkid == 0 &&
 430                     (1 << obj->oo_dn->dn_datablkshift) < PAGE_SIZE &&
 431                     (fid_is_idif(fid) || fid_is_norm(fid) ||
 432                      fid_is_echo(fid)) &&
 433                     osd->od_is_ost && !osd->od_dt_dev.dd_rdonly)
 434                         osd_obj_set_blksize(env, osd, obj);
 435         }
 436
 437         /*
 438          * initialize object before marking it existing
 439          */
 440         obj->oo_dt.do_lu.lo_header->loh_attr |= obj->oo_attr.la_mode & S_IFMT;
 441
 442         smp_mb();
 443         obj->oo_dt.do_lu.lo_header->loh_attr |= LOHA_EXISTS;
 444
 445         RETURN(0);
 446 }
 447
 448 static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
 449 {
 450         struct osd_thread_info  *info = osd_oti_get(env);
 451         struct lu_buf           buf;
 452         int                     rc;
 453         struct lustre_mdt_attrs *lma;
 454         const struct lu_fid *rfid = lu_object_fid(&obj->oo_dt.do_lu);
 455         ENTRY;
 456
 457         BUILD_BUG_ON(sizeof(info->oti_buf) < sizeof(*lma));
 458         lma = (struct lustre_mdt_attrs *)info->oti_buf;
 459         buf.lb_buf = lma;
 460         buf.lb_len = sizeof(info->oti_buf);
 461
 462         rc = osd_xattr_get(env, &obj->oo_dt, &buf, XATTR_NAME_LMA);
 463         if (rc > 0) {
 464                 rc = 0;
 465                 lustre_lma_swab(lma);
 466                 if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
 467                              CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT))) {
 468                         CWARN("%s: unsupported incompat LMA feature(s) %#x for "
 469                               "fid = "DFID"\n", osd_obj2dev(obj)->od_svname,
 470                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
 471                               PFID(rfid));
 472                         rc = -EOPNOTSUPP;
 473                 } else if (unlikely(!lu_fid_eq(rfid, &lma->lma_self_fid))) {
 474                         CERROR("%s: FID-in-LMA "DFID" does not match the "
 475                               "object self-fid "DFID"\n",
 476                               osd_obj2dev(obj)->od_svname,
 477                               PFID(&lma->lma_self_fid), PFID(rfid));
 478                         rc = -EREMCHG;
 479                 } else {
 480                         struct osd_device *osd = osd_obj2dev(obj);
 481
 482                         if (lma->lma_compat & LMAC_STRIPE_INFO &&
 483                             osd->od_is_ost)
 484                                 obj->oo_pfid_in_lma = 1;
 485                         if (unlikely(lma->lma_incompat & LMAI_REMOTE_PARENT) &&
 486                             osd->od_remote_parent_dir != ZFS_NO_OBJECT)
 487                                 lu_object_set_agent_entry(&obj->oo_dt.do_lu);
 488                 }
 489         } else if (rc == -ENODATA) {
 490                 /* haven't initialize LMA xattr */
 491                 rc = 0;
 492         }
 493
 494         RETURN(rc);
 495 }
 496
 497 /**
 498  * Helper function to retrieve DMU object id from fid for accounting object
 499  */
 500 static dnode_t *osd_quota_fid2dmu(const struct osd_device *osd,
 501                                   const struct lu_fid *fid)
 502 {
 503         dnode_t *dn = NULL;
 504
 505         LASSERT(fid_is_acct(fid));
 506
 507         switch (fid_oid(fid)) {
 508         case ACCT_USER_OID:
 509                 dn = osd->od_userused_dn;
 510                 break;
 511         case ACCT_GROUP_OID:
 512                 dn = osd->od_groupused_dn;
 513                 break;
 514 #ifdef ZFS_PROJINHERIT
 515         case ACCT_PROJECT_OID:
 516                 dn = osd->od_projectused_dn;
 517                 break;
 518 #endif
 519         default:
 520                 break;
 521         }
 522
 523         return dn;
 524 }
 525
 526 /*
 527  * Concurrency: no concurrent access is possible that early in object
 528  * life-cycle.
 529  */
 530 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
 531                            const struct lu_object_conf *conf)
 532 {
 533         struct osd_object *obj = osd_obj(l);
 534         struct osd_device *osd = osd_obj2dev(obj);
 535         const struct lu_fid *fid = lu_object_fid(l);
 536         struct lustre_scrub *scrub = &osd->od_scrub;
 537         struct osd_thread_info *info = osd_oti_get(env);
 538         struct luz_direntry *zde = &info->oti_zde;
 539         struct osd_idmap_cache *idc;
 540         char *name = info->oti_str;
 541         uint64_t oid;
 542         int rc = 0;
 543         int rc1;
 544         bool remote = false;
 545         ENTRY;
 546
 547         LASSERT(osd_invariant(obj));
 548
 549         if (fid_is_otable_it(&l->lo_header->loh_fid)) {
 550                 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
 551                 l->lo_header->loh_attr |= LOHA_EXISTS;
 552
 553                 GOTO(out, rc = 0);
 554         }
 555
 556         if (conf && conf->loc_flags & LOC_F_NEW)
 557                 GOTO(out, rc = 0);
 558
 559         if (unlikely(fid_is_acct(fid))) {
 560                 obj->oo_dn = osd_quota_fid2dmu(osd, fid);
 561                 if (obj->oo_dn) {
 562                         obj->oo_dt.do_index_ops = &osd_acct_index_ops;
 563                         l->lo_header->loh_attr |= LOHA_EXISTS;
 564                 }
 565
 566                 GOTO(out, rc = 0);
 567         }
 568
 569         idc = osd_idc_find(env, osd, fid);
 570         if (idc && !idc->oic_remote && idc->oic_dnode != ZFS_NO_OBJECT) {
 571                 oid = idc->oic_dnode;
 572                 goto zget;
 573         }
 574
 575         rc = -ENOENT;
 576         if (!list_empty(&osd->od_scrub.os_inconsistent_items))
 577                 rc = osd_oii_lookup(osd, fid, &oid);
 578
 579         if (rc)
 580                 rc = osd_fid_lookup(env, osd, fid, &oid);
 581
 582         if (rc == -ENOENT) {
 583                 if (likely(!(fid_is_norm(fid) || fid_is_igif(fid)) ||
 584                            fid_is_on_ost(env, osd, fid) ||
 585                            !zfs_test_bit(osd_oi_fid2idx(osd, fid),
 586                                          scrub->os_file.sf_oi_bitmap)))
 587                         GOTO(out, rc = 0);
 588
 589                 rc = -EREMCHG;
 590                 goto trigger;
 591         }
 592
 593         if (rc)
 594                 GOTO(out, rc);
 595
 596 zget:
 597         LASSERT(obj->oo_dn == NULL);
 598
 599         rc = __osd_obj2dnode(osd->od_os, oid, &obj->oo_dn);
 600         /* EEXIST will be returned if object is being deleted in ZFS */
 601         if (rc == -EEXIST)
 602                 GOTO(out, rc = 0);
 603
 604         if (rc) {
 605                 CERROR("%s: lookup "DFID"/%#llx failed: rc = %d\n",
 606                        osd->od_svname, PFID(lu_object_fid(l)), oid, rc);
 607                 GOTO(out, rc);
 608         }
 609
 610         rc = osd_object_init0(env, obj);
 611         if (rc)
 612                 GOTO(out, rc);
 613
 614         if (unlikely(obj->oo_header))
 615                 GOTO(out, rc = 0);
 616
 617         rc = osd_check_lma(env, obj);
 618         if (rc != -EREMCHG)
 619                 GOTO(out, rc);
 620
 621         osd_scrub_refresh_mapping(env, osd, fid, oid, DTO_INDEX_DELETE, true,
 622                                   NULL);
 623
 624 trigger:
 625         /* We still have chance to get the valid dnode: for the object that is
 626          * referenced by remote name entry, the object on the local MDT will be
 627          * linked under the dir /REMOTE_PARENT_DIR with its FID string as name.
 628          *
 629          * During the OI scrub, if we cannot find the OI mapping, we may still
 630          * have change to map the FID to local OID via lookup the dir
 631          * /REMOTE_PARENT_DIR. */
 632         if (!remote && !fid_is_on_ost(env, osd, fid)) {
 633                 osd_fid2str(name, fid, sizeof(info->oti_str));
 634                 rc = osd_zap_lookup(osd, osd->od_remote_parent_dir,
 635                                     NULL, name, 8, 3, (void *)zde);
 636                 if (!rc) {
 637                         oid = zde->lzd_reg.zde_dnode;
 638                         osd_dnode_rele(obj->oo_dn);
 639                         obj->oo_dn = NULL;
 640                         remote = true;
 641                         goto zget;
 642                 }
 643         }
 644
 645         /* The case someone triggered the OI scrub already. */
 646         if (scrub->os_running) {
 647                 if (!rc) {
 648                         LASSERT(remote);
 649
 650                         lu_object_set_agent_entry(l);
 651                         osd_oii_insert(env, osd, fid, oid, false);
 652                 } else {
 653                         rc = -EINPROGRESS;
 654                 }
 655
 656                 GOTO(out, rc);
 657         }
 658
 659         /* The case NOT allow to trigger OI scrub automatically. */
 660         if (osd->od_auto_scrub_interval == AS_NEVER)
 661                 GOTO(out, rc);
 662
 663         /* It is me to trigger the OI scrub. */
 664         rc1 = osd_scrub_start(env, osd, SS_CLEAR_DRYRUN |
 665                               SS_CLEAR_FAILOUT | SS_AUTO_FULL);
 666         CDEBUG_LIMIT(D_LFSCK | D_CONSOLE | D_WARNING,
 667                      "%s: trigger OI scrub by RPC for "DFID"/%#llx: rc = %d\n",
 668                      osd_name(osd), PFID(fid), oid, 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_free(h);
 704 }
 705
 706 static int
 707 osd_object_unlinked_add(struct osd_object *obj, struct osd_thandle *oh)
 708 {
 709         int rc = -EBUSY;
 710
 711         LASSERT(obj->oo_destroy == OSD_DESTROY_ASYNC);
 712
 713         /* the object is supposed to be exclusively locked by
 714          * the caller (osd_destroy()), while the transaction
 715          * (oh) is per-thread and not shared */
 716         if (likely(list_empty(&obj->oo_unlinked_linkage))) {
 717                 list_add(&obj->oo_unlinked_linkage, &oh->ot_unlinked_list);
 718                 rc = 0;
 719         }
 720
 721         return rc;
 722 }
 723
 724 /* Default to max data size covered by a level-1 indirect block */
 725 static unsigned long osd_sync_destroy_max_size =
 726         1UL << (DN_MAX_INDBLKSHIFT - SPA_BLKPTRSHIFT + SPA_MAXBLOCKSHIFT);
 727 module_param(osd_sync_destroy_max_size, ulong, 0444);
 728 MODULE_PARM_DESC(osd_sync_destroy_max_size, "Maximum object size to use synchronous destroy.");
 729
 730 static inline void
 731 osd_object_set_destroy_type(struct osd_object *obj)
 732 {
 733         /*
 734          * Lock-less OST_WRITE can race with OST_DESTROY, so set destroy type
 735          * only once and use it consistently thereafter.
 736          */
 737         down_write(&obj->oo_guard);
 738         if (obj->oo_destroy == OSD_DESTROY_NONE) {
 739                 if (obj->oo_attr.la_size <= osd_sync_destroy_max_size)
 740                         obj->oo_destroy = OSD_DESTROY_SYNC;
 741                 else /* Larger objects are destroyed asynchronously */
 742                         obj->oo_destroy = OSD_DESTROY_ASYNC;
 743         }
 744         up_write(&obj->oo_guard);
 745 }
 746
 747 static int osd_declare_destroy(const struct lu_env *env, struct dt_object *dt,
 748                                struct thandle *th)
 749 {
 750         const struct lu_fid     *fid = lu_object_fid(&dt->do_lu);
 751         struct osd_object       *obj = osd_dt_obj(dt);
 752         struct osd_device       *osd = osd_obj2dev(obj);
 753         struct osd_thandle      *oh;
 754         dnode_t *dn;
 755         int                      rc;
 756         uint64_t                 zapid;
 757         ENTRY;
 758
 759         LASSERT(th != NULL);
 760         LASSERT(dt_object_exists(dt));
 761
 762         oh = container_of(th, struct osd_thandle, ot_super);
 763         LASSERT(oh->ot_tx != NULL);
 764
 765         dmu_tx_mark_netfree(oh->ot_tx);
 766
 767         /* declare that we'll remove object from fid-dnode mapping */
 768         zapid = osd_get_name_n_idx(env, osd, fid, NULL, 0, &dn);
 769         osd_tx_hold_zap(oh->ot_tx, zapid, dn, FALSE, NULL);
 770
 771         osd_declare_xattrs_destroy(env, obj, oh);
 772
 773         /* one less inode */
 774         rc = osd_declare_quota(env, osd, obj->oo_attr.la_uid,
 775                                obj->oo_attr.la_gid, obj->oo_attr.la_projid,
 776                                -1, oh, NULL, OSD_QID_INODE);
 777         if (rc)
 778                 RETURN(rc);
 779
 780         /* data to be truncated */
 781         rc = osd_declare_quota(env, osd, obj->oo_attr.la_uid,
 782                                obj->oo_attr.la_gid, obj->oo_attr.la_projid,
 783                                0, oh, NULL, OSD_QID_BLK);
 784         if (rc)
 785                 RETURN(rc);
 786
 787         osd_object_set_destroy_type(obj);
 788         if (obj->oo_destroy == OSD_DESTROY_SYNC)
 789                 dmu_tx_hold_free(oh->ot_tx, obj->oo_dn->dn_object,
 790                                  0, DMU_OBJECT_END);
 791         else
 792                 osd_tx_hold_zap(oh->ot_tx, osd->od_unlinked->dn_object,
 793                                 osd->od_unlinked, TRUE, NULL);
 794
 795         /* remove agent entry (if have) from remote parent */
 796         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu))
 797                 osd_tx_hold_zap(oh->ot_tx, osd->od_remote_parent_dir,
 798                                 NULL, FALSE, NULL);
 799
 800         /* will help to find FID->ino when this object is being
 801          * added to PENDING/ */
 802         osd_idc_find_and_init(env, osd, obj);
 803
 804         RETURN(0);
 805 }
 806
 807 static int osd_destroy(const struct lu_env *env, struct dt_object *dt,
 808                        struct thandle *th)
 809 {
 810         struct osd_thread_info  *info = osd_oti_get(env);
 811         char                    *buf = info->oti_str;
 812         struct osd_object       *obj = osd_dt_obj(dt);
 813         struct osd_device       *osd = osd_obj2dev(obj);
 814         const struct lu_fid     *fid = lu_object_fid(&dt->do_lu);
 815         struct osd_thandle      *oh;
 816         int                      rc;
 817         uint64_t                 oid, zapid;
 818         dnode_t *zdn;
 819         ENTRY;
 820
 821         down_write(&obj->oo_guard);
 822
 823         if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
 824                 GOTO(out, rc = -ENOENT);
 825
 826         LASSERT(obj->oo_dn != NULL);
 827
 828         oh = container_of(th, struct osd_thandle, ot_super);
 829         LASSERT(oh != NULL);
 830         LASSERT(oh->ot_tx != NULL);
 831
 832         /* remove obj ref from index dir (it depends) */
 833         zapid = osd_get_name_n_idx(env, osd, fid, buf,
 834                                    sizeof(info->oti_str), &zdn);
 835         rc = osd_xattrs_destroy(env, obj, oh);
 836         if (rc) {
 837                 CERROR("%s: cannot destroy xattrs for %s: rc = %d\n",
 838                        osd->od_svname, buf, rc);
 839                 GOTO(out, rc);
 840         }
 841
 842         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu)) {
 843                 rc = osd_delete_from_remote_parent(env, osd, obj, oh, true);
 844                 if (rc)
 845                         GOTO(out, rc);
 846         }
 847
 848         oid = obj->oo_dn->dn_object;
 849         if (unlikely(obj->oo_destroy == OSD_DESTROY_NONE)) {
 850                 /* this may happen if the destroy wasn't declared
 851                  * e.g. when the object is created and then destroyed
 852                  * in the same transaction - we don't need additional
 853                  * space for destroy specifically */
 854                 LASSERT(obj->oo_attr.la_size <= osd_sync_destroy_max_size);
 855                 rc = -dmu_object_free(osd->od_os, oid, oh->ot_tx);
 856                 if (rc)
 857                         CERROR("%s: failed to free %s/%#llx: rc = %d\n",
 858                                osd->od_svname, buf, oid, rc);
 859         } else if (obj->oo_destroy == OSD_DESTROY_SYNC) {
 860                 rc = -dmu_object_free(osd->od_os, oid, oh->ot_tx);
 861                 if (rc)
 862                         CERROR("%s: failed to free %s/%#llx: rc = %d\n",
 863                                osd->od_svname, buf, oid, rc);
 864         } else { /* asynchronous destroy */
 865                 char *key = info->oti_key;
 866
 867                 rc = osd_object_unlinked_add(obj, oh);
 868                 if (rc)
 869                         GOTO(out, rc);
 870
 871                 snprintf(key, sizeof(info->oti_key), "%llx", oid);
 872                 rc = osd_zap_add(osd, osd->od_unlinked->dn_object,
 873                                  osd->od_unlinked, key, 8, 1, &oid, oh->ot_tx);
 874                 if (rc)
 875                         CERROR("%s: zap_add_int() failed %s/%#llx: rc = %d\n",
 876                                osd->od_svname, buf, oid, rc);
 877         }
 878
 879         /* Remove the OI mapping after the destroy to handle the race with
 880          * OI scrub that may insert missed OI mapping during the interval. */
 881         rc = osd_zap_remove(osd, zapid, zdn, buf, oh->ot_tx);
 882         if (unlikely(rc == -ENOENT))
 883                 rc = 0;
 884         if (rc)
 885                 CERROR("%s: zap_remove(%s) failed: rc = %d\n",
 886                        osd->od_svname, buf, rc);
 887
 888         GOTO(out, rc);
 889
 890 out:
 891         /* not needed in the cache anymore */
 892         set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
 893         if (rc == 0)
 894                 obj->oo_destroyed = 1;
 895         up_write(&obj->oo_guard);
 896         RETURN (0);
 897 }
 898
 899 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
 900 {
 901         struct osd_object *obj = osd_obj(l);
 902         const struct lu_fid *fid = lu_object_fid(l);
 903
 904         if (obj->oo_dn) {
 905                 if (likely(!fid_is_acct(fid))) {
 906                         osd_object_sa_fini(obj);
 907                         if (obj->oo_sa_xattr) {
 908                                 nvlist_free(obj->oo_sa_xattr);
 909                                 obj->oo_sa_xattr = NULL;
 910                         }
 911                         osd_dnode_rele(obj->oo_dn);
 912                         list_del(&obj->oo_sa_linkage);
 913                 }
 914                 obj->oo_dn = NULL;
 915         }
 916 }
 917
 918 /*
 919  * Concurrency: ->loo_object_release() is called under site spin-lock.
 920  */
 921 static void osd_object_release(const struct lu_env *env,
 922                                struct lu_object *l)
 923 {
 924 }
 925
 926 /*
 927  * Concurrency: shouldn't matter.
 928  */
 929 static int osd_object_print(const struct lu_env *env, void *cookie,
 930                             lu_printer_t p, const struct lu_object *l)
 931 {
 932         struct osd_object *o = osd_obj(l);
 933
 934         return (*p)(env, cookie, LUSTRE_OSD_ZFS_NAME"-object@%p", o);
 935 }
 936
 937 static void osd_read_lock(const struct lu_env *env, struct dt_object *dt,
 938                           unsigned role)
 939 {
 940         struct osd_object *obj = osd_dt_obj(dt);
 941
 942         LASSERT(osd_invariant(obj));
 943
 944         down_read_nested(&obj->oo_sem, role);
 945 }
 946
 947 static void osd_write_lock(const struct lu_env *env, struct dt_object *dt,
 948                            unsigned role)
 949 {
 950         struct osd_object *obj = osd_dt_obj(dt);
 951
 952         LASSERT(osd_invariant(obj));
 953
 954         down_write_nested(&obj->oo_sem, role);
 955 }
 956
 957 static void osd_read_unlock(const struct lu_env *env, struct dt_object *dt)
 958 {
 959         struct osd_object *obj = osd_dt_obj(dt);
 960
 961         LASSERT(osd_invariant(obj));
 962         up_read(&obj->oo_sem);
 963 }
 964
 965 static void osd_write_unlock(const struct lu_env *env, struct dt_object *dt)
 966 {
 967         struct osd_object *obj = osd_dt_obj(dt);
 968
 969         LASSERT(osd_invariant(obj));
 970         up_write(&obj->oo_sem);
 971 }
 972
 973 static int osd_write_locked(const struct lu_env *env, struct dt_object *dt)
 974 {
 975         struct osd_object *obj = osd_dt_obj(dt);
 976         int rc = 1;
 977
 978         LASSERT(osd_invariant(obj));
 979
 980         if (down_write_trylock(&obj->oo_sem)) {
 981                 rc = 0;
 982                 up_write(&obj->oo_sem);
 983         }
 984         return rc;
 985 }
 986
 987 static int osd_attr_get(const struct lu_env *env, struct dt_object *dt,
 988                         struct lu_attr *attr)
 989 {
 990         struct osd_object *obj = osd_dt_obj(dt);
 991         struct osd_device *osd = osd_obj2dev(obj);
 992         uint64_t blocks;
 993         uint32_t blksize;
 994         int rc = 0;
 995
 996         down_read(&obj->oo_guard);
 997
 998         if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
 999                 GOTO(out, rc = -ENOENT);
1000
1001         if (unlikely(fid_is_acct(lu_object_fid(&dt->do_lu))))
1002                 GOTO(out, rc = 0);
1003
1004         LASSERT(osd_invariant(obj));
1005         LASSERT(obj->oo_dn);
1006
1007         read_lock(&obj->oo_attr_lock);
1008         *attr = obj->oo_attr;
1009         if (obj->oo_lma_flags & LUSTRE_ORPHAN_FL) {
1010                 attr->la_valid |= LA_FLAGS;
1011                 attr->la_flags |= LUSTRE_ORPHAN_FL;
1012         }
1013         if (obj->oo_lma_flags & LUSTRE_ENCRYPT_FL) {
1014                 attr->la_valid |= LA_FLAGS;
1015                 attr->la_flags |= LUSTRE_ENCRYPT_FL;
1016         }
1017         read_unlock(&obj->oo_attr_lock);
1018         if (attr->la_valid & LA_FLAGS && attr->la_flags & LUSTRE_ORPHAN_FL)
1019                 CDEBUG(D_INFO, "%s: set orphan flag on "DFID" (%#llx/%#x)\n",
1020                        osd_obj2dev(obj)->od_svname,
1021                        PFID(lu_object_fid(&dt->do_lu)),
1022                        attr->la_valid, obj->oo_lma_flags);
1023
1024         /* with ZFS_DEBUG zrl_add_debug() called by DB_DNODE_ENTER()
1025          * from within sa_object_size() can block on a mutex, so
1026          * we can't call sa_object_size() holding rwlock */
1027         sa_object_size(obj->oo_sa_hdl, &blksize, &blocks);
1028         /* we do not control size of indices, so always calculate
1029          * it from number of blocks reported by DMU */
1030         if (S_ISDIR(attr->la_mode)) {
1031                 attr->la_size = 512 * blocks;
1032                 rc = -zap_count(osd->od_os, obj->oo_dn->dn_object,
1033                                 &attr->la_dirent_count);
1034         }
1035         /* Block size may be not set; suggest maximal I/O transfers. */
1036         if (blksize == 0)
1037                 blksize = osd_spa_maxblocksize(
1038                         dmu_objset_spa(osd_obj2dev(obj)->od_os));
1039
1040         attr->la_blksize = blksize;
1041         attr->la_blocks = blocks;
1042         attr->la_valid |= LA_BLOCKS | LA_BLKSIZE;
1043
1044 out:
1045         up_read(&obj->oo_guard);
1046         return rc;
1047 }
1048
1049 /* Simple wrapper on top of qsd API which implement quota transfer for osd
1050  * setattr needs. As a reminder, only the root user can change ownership of
1051  * a file, that's why EDQUOT & EINPROGRESS errors are discarded */
1052 static inline int qsd_transfer(const struct lu_env *env,
1053                                struct qsd_instance *qsd,
1054                                struct lquota_trans *trans, int qtype,
1055                                __u64 orig_id, __u64 new_id, __u64 bspace,
1056                                struct lquota_id_info *qi, bool ignore_edquot)
1057 {
1058         int     rc;
1059
1060         if (unlikely(qsd == NULL))
1061                 return 0;
1062
1063         LASSERT(qtype >= 0 && qtype < LL_MAXQUOTAS);
1064         qi->lqi_type = qtype;
1065
1066         /* inode accounting */
1067         qi->lqi_is_blk = false;
1068
1069         /* one more inode for the new owner ... */
1070         qi->lqi_id.qid_uid = new_id;
1071         qi->lqi_space      = 1;
1072         rc = qsd_op_begin(env, qsd, trans, qi, NULL);
1073         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
1074                 rc = 0;
1075         if (rc)
1076                 return rc;
1077
1078         /* and one less inode for the current id */
1079         qi->lqi_id.qid_uid = orig_id;;
1080         qi->lqi_space      = -1;
1081         /* can't get EDQUOT when reducing usage */
1082         rc = qsd_op_begin(env, qsd, trans, qi, NULL);
1083         if (rc == -EINPROGRESS)
1084                 rc = 0;
1085         if (rc)
1086                 return rc;
1087
1088         /* block accounting */
1089         qi->lqi_is_blk = true;
1090
1091         /* more blocks for the new owner ... */
1092         qi->lqi_id.qid_uid = new_id;
1093         qi->lqi_space      = bspace;
1094         rc = qsd_op_begin(env, qsd, trans, qi, NULL);
1095         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
1096                 rc = 0;
1097         if (rc)
1098                 return rc;
1099
1100         /* and finally less blocks for the current owner */
1101         qi->lqi_id.qid_uid = orig_id;
1102         qi->lqi_space      = -bspace;
1103         rc = qsd_op_begin(env, qsd, trans, qi, NULL);
1104         /* can't get EDQUOT when reducing usage */
1105         if (rc == -EINPROGRESS)
1106                 rc = 0;
1107         return rc;
1108 }
1109
1110 static int osd_declare_attr_set(const struct lu_env *env,
1111                                 struct dt_object *dt,
1112                                 const struct lu_attr *attr,
1113                                 struct thandle *handle)
1114 {
1115         struct osd_thread_info  *info = osd_oti_get(env);
1116         struct osd_object       *obj = osd_dt_obj(dt);
1117         struct osd_device       *osd = osd_obj2dev(obj);
1118         dmu_tx_hold_t           *txh;
1119         struct osd_thandle      *oh;
1120         uint64_t                 bspace;
1121         uint32_t                 blksize;
1122         int                      rc = 0;
1123         bool                     found;
1124         ENTRY;
1125
1126
1127         LASSERT(handle != NULL);
1128         LASSERT(osd_invariant(obj));
1129
1130         oh = container_of(handle, struct osd_thandle, ot_super);
1131
1132         down_read(&obj->oo_guard);
1133         if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
1134                 GOTO(out_sem, rc = 0);
1135
1136         LASSERT(obj->oo_sa_hdl != NULL);
1137         LASSERT(oh->ot_tx != NULL);
1138         /* regular attributes are part of the bonus buffer */
1139         /* let's check whether this object is already part of
1140          * transaction.. */
1141         found = false;
1142         for (txh = list_head(&oh->ot_tx->tx_holds); txh;
1143              txh = list_next(&oh->ot_tx->tx_holds, txh)) {
1144                 if (txh->txh_dnode == NULL)
1145                         continue;
1146                 if (txh->txh_dnode->dn_object != obj->oo_dn->dn_object)
1147                         continue;
1148                 /* this object is part of the transaction already
1149                  * we don't need to declare bonus again */
1150                 found = true;
1151                 break;
1152         }
1153         if (!found)
1154                 dmu_tx_hold_bonus(oh->ot_tx, obj->oo_dn->dn_object);
1155         if (oh->ot_tx->tx_err != 0)
1156                 GOTO(out_sem, rc = -oh->ot_tx->tx_err);
1157
1158         if (attr && attr->la_valid & LA_FLAGS) {
1159                 /* LMA is usually a part of bonus, no need to declare
1160                  * anything else */
1161         }
1162
1163         if (attr && (attr->la_valid & (LA_UID | LA_GID | LA_PROJID))) {
1164                 sa_object_size(obj->oo_sa_hdl, &blksize, &bspace);
1165                 bspace = toqb(bspace * 512);
1166
1167                 CDEBUG(D_QUOTA,
1168                        "%s: enforce quota on UID %u, GID %u, the quota space is %lld (%u)\n",
1169                        osd->od_svname,
1170                        attr->la_uid, attr->la_gid, bspace, blksize);
1171         }
1172         /* to preserve locking order - qsd_transfer() may need to flush
1173          * currently running transaction when we're out of quota. */
1174         up_read(&obj->oo_guard);
1175
1176         if (attr && attr->la_valid & LA_UID) {
1177                 /* quota enforcement for user */
1178                 if (attr->la_uid != obj->oo_attr.la_uid) {
1179                         rc = qsd_transfer(env, osd_def_qsd(osd),
1180                                           &oh->ot_quota_trans, USRQUOTA,
1181                                           obj->oo_attr.la_uid, attr->la_uid,
1182                                           bspace, &info->oti_qi, true);
1183                         if (rc)
1184                                 GOTO(out, rc);
1185                 }
1186         }
1187         if (attr && attr->la_valid & LA_GID) {
1188                 /* quota enforcement for group */
1189                 if (attr->la_gid != obj->oo_attr.la_gid) {
1190                         rc = qsd_transfer(env, osd_def_qsd(osd),
1191                                           &oh->ot_quota_trans, GRPQUOTA,
1192                                           obj->oo_attr.la_gid, attr->la_gid,
1193                                           bspace, &info->oti_qi,
1194                                           !(attr->la_flags &
1195                                                         LUSTRE_SET_SYNC_FL));
1196                         if (rc)
1197                                 GOTO(out, rc);
1198                 }
1199         }
1200 #ifdef ZFS_PROJINHERIT
1201         if (attr && attr->la_valid & LA_PROJID) {
1202                 /* quota enforcement for project */
1203                 if (attr->la_projid != obj->oo_attr.la_projid) {
1204                         if (!osd->od_projectused_dn)
1205                                 GOTO(out, rc = -EOPNOTSUPP);
1206
1207                         /* Usually, if project quota is upgradable for the
1208                          * device, then the upgrade will be done before or when
1209                          * mount the device. So when we come here, this project
1210                          * should have project ID attribute already (that is
1211                          * zero by default).  Otherwise, there was something
1212                          * wrong during the former upgrade, let's return failure
1213                          * to report that.
1214                          *
1215                          * Please note that, different from other attributes,
1216                          * you can NOT simply set the project ID attribute under
1217                          * such case, because adding (NOT change) project ID
1218                          * attribute needs to change the object's attribute
1219                          * layout to match zfs backend quota accounting
1220                          * requirement. */
1221                         if (unlikely(!obj->oo_with_projid))
1222                                 GOTO(out, rc = -ENXIO);
1223
1224                         rc = qsd_transfer(env, osd_def_qsd(osd),
1225                                           &oh->ot_quota_trans, PRJQUOTA,
1226                                           obj->oo_attr.la_projid,
1227                                           attr->la_projid, bspace,
1228                                           &info->oti_qi, true);
1229                         if (rc)
1230                                 GOTO(out, rc);
1231                 }
1232         }
1233 #endif
1234 out:
1235         RETURN(rc);
1236 out_sem:
1237         up_read(&obj->oo_guard);
1238         RETURN(rc);
1239 }
1240
1241 /*
1242  * Set the attributes of an object
1243  *
1244  * The transaction passed to this routine must have
1245  * dmu_tx_hold_bonus(tx, oid) called and then assigned
1246  * to a transaction group.
1247  */
1248 static int osd_attr_set(const struct lu_env *env, struct dt_object *dt,
1249                         const struct lu_attr *la, struct thandle *handle)
1250 {
1251         struct osd_thread_info  *info = osd_oti_get(env);
1252         sa_bulk_attr_t          *bulk = osd_oti_get(env)->oti_attr_bulk;
1253         struct osd_object       *obj = osd_dt_obj(dt);
1254         struct osd_device       *osd = osd_obj2dev(obj);
1255         struct osd_thandle      *oh;
1256         struct osa_attr         *osa = &info->oti_osa;
1257         __u64                    valid = la->la_valid;
1258         int                      cnt;
1259         int                      rc = 0;
1260
1261         ENTRY;
1262
1263         down_read(&obj->oo_guard);
1264         if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
1265                 GOTO(out, rc = -ENOENT);
1266
1267         LASSERT(handle != NULL);
1268         LASSERT(osd_invariant(obj));
1269         LASSERT(obj->oo_sa_hdl);
1270
1271         oh = container_of(handle, struct osd_thandle, ot_super);
1272         /* Assert that the transaction has been assigned to a
1273            transaction group. */
1274         LASSERT(oh->ot_tx->tx_txg != 0);
1275
1276         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING) && !osd->od_is_ost) {
1277                 struct zpl_direntry *zde = &info->oti_zde.lzd_reg;
1278                 char *buf = info->oti_str;
1279                 dnode_t *zdn = NULL;
1280                 uint64_t zapid;
1281
1282                 zapid = osd_get_name_n_idx(env, osd, lu_object_fid(&dt->do_lu),
1283                                            buf, sizeof(info->oti_str), &zdn);
1284                 rc = osd_zap_lookup(osd, zapid, zdn, buf, 8,
1285                                     sizeof(*zde) / 8, zde);
1286                 if (!rc) {
1287                         zde->zde_dnode -= 1;
1288                         rc = -zap_update(osd->od_os, zapid, buf, 8,
1289                                          sizeof(*zde) / 8, zde, oh->ot_tx);
1290                 }
1291                 if (rc > 0)
1292                         rc = 0;
1293                 GOTO(out, rc);
1294         }
1295
1296         /* Only allow set size for regular file */
1297         if (!S_ISREG(dt->do_lu.lo_header->loh_attr))
1298                 valid &= ~(LA_SIZE | LA_BLOCKS);
1299
1300         if (valid & LA_CTIME && la->la_ctime == obj->oo_attr.la_ctime)
1301                 valid &= ~LA_CTIME;
1302
1303         if (valid & LA_MTIME && la->la_mtime == obj->oo_attr.la_mtime)
1304                 valid &= ~LA_MTIME;
1305
1306         if (valid & LA_ATIME && la->la_atime == obj->oo_attr.la_atime)
1307                 valid &= ~LA_ATIME;
1308
1309         if (valid == 0)
1310                 GOTO(out, rc = 0);
1311
1312         if (valid & LA_FLAGS) {
1313                 struct lustre_mdt_attrs *lma;
1314                 struct lu_buf buf;
1315                 int size = 0;
1316
1317                 if (la->la_flags & LUSTRE_LMA_FL_MASKS) {
1318                         LASSERT(!obj->oo_pfid_in_lma);
1319                         BUILD_BUG_ON(sizeof(info->oti_buf) < sizeof(*lma));
1320                         lma = (struct lustre_mdt_attrs *)&info->oti_buf;
1321                         buf.lb_buf = lma;
1322                         buf.lb_len = sizeof(info->oti_buf);
1323
1324                         /* Please do NOT call osd_xattr_get() directly, that
1325                          * will cause recursive down_read() on oo_guard. */
1326                         rc = osd_xattr_get_internal(env, obj, &buf,
1327                                                     XATTR_NAME_LMA, &size);
1328                         if (!rc && unlikely(size < sizeof(*lma))) {
1329                                 rc = -EINVAL;
1330                         } else if (!rc) {
1331                                 lma->lma_incompat =
1332                                         le32_to_cpu(lma->lma_incompat);
1333                                 lma->lma_incompat |=
1334                                         lustre_to_lma_flags(la->la_flags);
1335                                 lma->lma_incompat =
1336                                         cpu_to_le32(lma->lma_incompat);
1337                                 buf.lb_buf = lma;
1338                                 buf.lb_len = sizeof(*lma);
1339                                 rc = osd_xattr_set_internal(env, obj, &buf,
1340                                                             XATTR_NAME_LMA,
1341                                                             LU_XATTR_REPLACE,
1342                                                             oh);
1343                         }
1344                         if (rc < 0) {
1345                                 CWARN("%s: failed to set LMA flags: rc = %d\n",
1346                                        osd->od_svname, rc);
1347                                 GOTO(out, rc);
1348                         } else {
1349                                 obj->oo_lma_flags =
1350                                         la->la_flags & LUSTRE_LMA_FL_MASKS;
1351                         }
1352                 }
1353         }
1354
1355         write_lock(&obj->oo_attr_lock);
1356         cnt = 0;
1357
1358         if (valid & LA_PROJID) {
1359 #ifdef ZFS_PROJINHERIT
1360                 if (osd->od_projectused_dn) {
1361                         LASSERT(obj->oo_with_projid);
1362
1363                         osa->projid = obj->oo_attr.la_projid = la->la_projid;
1364                         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_PROJID(osd), NULL,
1365                                          &osa->projid, 8);
1366                 } else
1367 #endif
1368                         valid &= ~LA_PROJID;
1369         }
1370
1371         if (valid & LA_ATIME) {
1372                 osa->atime[0] = obj->oo_attr.la_atime = la->la_atime;
1373                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(osd), NULL,
1374                                  osa->atime, 16);
1375         }
1376         if (valid & LA_MTIME) {
1377                 osa->mtime[0] = obj->oo_attr.la_mtime = la->la_mtime;
1378                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(osd), NULL,
1379                                  osa->mtime, 16);
1380         }
1381         if (valid & LA_CTIME) {
1382                 osa->ctime[0] = obj->oo_attr.la_ctime = la->la_ctime;
1383                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(osd), NULL,
1384                                  osa->ctime, 16);
1385         }
1386         if (valid & LA_MODE) {
1387                 /* mode is stored along with type, so read it first */
1388                 obj->oo_attr.la_mode = (obj->oo_attr.la_mode & S_IFMT) |
1389                         (la->la_mode & ~S_IFMT);
1390                 osa->mode = obj->oo_attr.la_mode;
1391                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(osd), NULL,
1392                                  &osa->mode, 8);
1393         }
1394         if (valid & LA_SIZE) {
1395                 osa->size = obj->oo_attr.la_size = la->la_size;
1396                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(osd), NULL,
1397                                  &osa->size, 8);
1398         }
1399         if (valid & LA_NLINK) {
1400                 osa->nlink = obj->oo_attr.la_nlink = la->la_nlink;
1401                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(osd), NULL,
1402                                  &osa->nlink, 8);
1403         }
1404         if (valid & LA_RDEV) {
1405                 osa->rdev = obj->oo_attr.la_rdev = la->la_rdev;
1406                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_RDEV(osd), NULL,
1407                                  &osa->rdev, 8);
1408         }
1409         if (valid & LA_FLAGS) {
1410                 osa->flags = attrs_fs2zfs(la->la_flags);
1411                 /* many flags are not supported by zfs, so ensure a good cached
1412                  * copy */
1413                 obj->oo_attr.la_flags = attrs_zfs2fs(osa->flags);
1414 #ifdef ZFS_PROJINHERIT
1415                 if (obj->oo_with_projid && osd->od_projectused_dn)
1416                         osa->flags |= ZFS_PROJID;
1417 #endif
1418                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(osd), NULL,
1419                                  &osa->flags, 8);
1420         }
1421         if (valid & LA_UID) {
1422                 osa->uid = obj->oo_attr.la_uid = la->la_uid;
1423                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(osd), NULL,
1424                                  &osa->uid, 8);
1425         }
1426         if (valid & LA_GID) {
1427                 osa->gid = obj->oo_attr.la_gid = la->la_gid;
1428                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(osd), NULL,
1429                                  &osa->gid, 8);
1430         }
1431         obj->oo_attr.la_valid |= valid;
1432         write_unlock(&obj->oo_attr_lock);
1433
1434         LASSERT(cnt <= ARRAY_SIZE(osd_oti_get(env)->oti_attr_bulk));
1435         rc = osd_object_sa_bulk_update(obj, bulk, cnt, oh);
1436
1437 out:
1438         up_read(&obj->oo_guard);
1439         RETURN(rc);
1440 }
1441
1442 /*
1443  * Object creation.
1444  *
1445  * XXX temporary solution.
1446  */
1447
1448 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
1449                         struct dt_object *parent, struct dt_object *child,
1450                         umode_t child_mode)
1451 {
1452         LASSERT(ah);
1453
1454         ah->dah_parent = parent;
1455         ah->dah_mode = child_mode;
1456
1457         if (parent != NULL && !dt_object_remote(parent)) {
1458                 /* will help to find FID->ino at dt_insert("..") */
1459                 struct osd_object *pobj = osd_dt_obj(parent);
1460
1461                 osd_idc_find_and_init(env, osd_obj2dev(pobj), pobj);
1462         }
1463 }
1464
1465 static int osd_declare_create(const struct lu_env *env, struct dt_object *dt,
1466                               struct lu_attr *attr,
1467                               struct dt_allocation_hint *hint,
1468                               struct dt_object_format *dof,
1469                               struct thandle *handle)
1470 {
1471         const struct lu_fid     *fid = lu_object_fid(&dt->do_lu);
1472         struct osd_object       *obj = osd_dt_obj(dt);
1473         struct osd_device       *osd = osd_obj2dev(obj);
1474         struct osd_thandle      *oh;
1475         uint64_t                 zapid;
1476         dnode_t                 *dn;
1477         int                      rc, dnode_size;
1478         ENTRY;
1479
1480         LASSERT(dof);
1481
1482         switch (dof->dof_type) {
1483                 case DFT_REGULAR:
1484                 case DFT_SYM:
1485                 case DFT_NODE:
1486                         if (obj->oo_dt.do_body_ops == NULL)
1487                                 obj->oo_dt.do_body_ops = &osd_body_ops;
1488                         break;
1489                 default:
1490                         break;
1491         }
1492
1493         LASSERT(handle != NULL);
1494         oh = container_of(handle, struct osd_thandle, ot_super);
1495         LASSERT(oh->ot_tx != NULL);
1496
1497         /* this is the minimum set of EAs on every Lustre object */
1498         obj->oo_ea_in_bonus = OSD_BASE_EA_IN_BONUS;
1499         /* reserve 32 bytes for extra stuff like ACLs */
1500         dnode_size = size_roundup_power2(obj->oo_ea_in_bonus + 32);
1501
1502         switch (dof->dof_type) {
1503                 case DFT_DIR:
1504                         dt->do_index_ops = &osd_dir_ops;
1505                         /* fallthrough */
1506                 case DFT_INDEX:
1507                         /* for zap create */
1508                         dmu_tx_hold_zap(oh->ot_tx, DMU_NEW_OBJECT, FALSE, NULL);
1509                         dmu_tx_hold_sa_create(oh->ot_tx, dnode_size);
1510                         break;
1511                 case DFT_REGULAR:
1512                 case DFT_SYM:
1513                 case DFT_NODE:
1514                         /* first, we'll create new object */
1515                         dmu_tx_hold_sa_create(oh->ot_tx, dnode_size);
1516                         break;
1517
1518                 default:
1519                         LBUG();
1520                         break;
1521         }
1522
1523         /* and we'll add it to some mapping */
1524         zapid = osd_get_name_n_idx(env, osd, fid, NULL, 0, &dn);
1525         osd_tx_hold_zap(oh->ot_tx, zapid, dn, TRUE, NULL);
1526
1527         /* will help to find FID->ino mapping at dt_insert() */
1528         osd_idc_find_and_init(env, osd, obj);
1529
1530         rc = osd_declare_quota(env, osd, attr->la_uid, attr->la_gid,
1531                                attr->la_projid, 1, oh, NULL, OSD_QID_INODE);
1532
1533         RETURN(rc);
1534 }
1535
1536 int __osd_attr_init(const struct lu_env *env, struct osd_device *osd,
1537                     struct osd_object *obj, sa_handle_t *sa_hdl, dmu_tx_t *tx,
1538                     struct lu_attr *la, uint64_t parent,
1539                     nvlist_t *xattr)
1540 {
1541         sa_bulk_attr_t *bulk = osd_oti_get(env)->oti_attr_bulk;
1542         struct osa_attr *osa = &osd_oti_get(env)->oti_osa;
1543         uint64_t gen;
1544         inode_timespec_t now;
1545         int cnt;
1546         int rc;
1547         char *dxattr = NULL;
1548         size_t sa_size;
1549
1550
1551         LASSERT(sa_hdl);
1552
1553         gen = dmu_tx_get_txg(tx);
1554         gethrestime(&now);
1555         ZFS_TIME_ENCODE(&now, osa->btime);
1556
1557         osa->atime[0] = la->la_atime;
1558         osa->ctime[0] = la->la_ctime;
1559         osa->mtime[0] = la->la_mtime;
1560         osa->mode = la->la_mode;
1561         osa->uid = la->la_uid;
1562         osa->gid = la->la_gid;
1563         osa->rdev = la->la_rdev;
1564         osa->nlink = la->la_nlink;
1565         if (la->la_valid & LA_FLAGS)
1566                 osa->flags = attrs_fs2zfs(la->la_flags);
1567         else
1568                 osa->flags = 0;
1569         osa->size  = la->la_size;
1570 #ifdef ZFS_PROJINHERIT
1571         if (osd->od_projectused_dn) {
1572                 if (la->la_valid & LA_PROJID)
1573                         osa->projid = la->la_projid;
1574                 else
1575                         osa->projid = ZFS_DEFAULT_PROJID;
1576                 osa->flags |= ZFS_PROJID;
1577                 if (obj)
1578                         obj->oo_with_projid = 1;
1579         } else {
1580                 osa->flags &= ~ZFS_PROJID;
1581         }
1582 #endif
1583
1584         /*
1585          * we need to create all SA below upon object create.
1586          *
1587          * XXX The attribute order matters since the accounting callback relies
1588          * on static offsets (i.e. SA_*_OFFSET, see zfs_space_delta_cb()) to
1589          * look up the UID/GID/PROJID attributes. Moreover, the callback does
1590          * not seem to support the spill block.
1591          * We define attributes in the same order as SA_*_OFFSET in order to
1592          * work around the problem. See ORI-610.
1593          */
1594         cnt = 0;
1595         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(osd), NULL, &osa->mode, 8);
1596         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(osd), NULL, &osa->size, 8);
1597         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GEN(osd), NULL, &gen, 8);
1598         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(osd), NULL, &osa->uid, 8);
1599         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(osd), NULL, &osa->gid, 8);
1600         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_PARENT(osd), NULL, &parent, 8);
1601         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(osd), NULL, &osa->flags, 8);
1602         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(osd), NULL, osa->atime, 16);
1603         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(osd), NULL, osa->mtime, 16);
1604         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(osd), NULL, osa->ctime, 16);
1605         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CRTIME(osd), NULL, osa->btime, 16);
1606         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(osd), NULL, &osa->nlink, 8);
1607 #ifdef ZFS_PROJINHERIT
1608         if (osd->od_projectused_dn)
1609                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_PROJID(osd), NULL,
1610                                  &osa->projid, 8);
1611 #endif
1612         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_RDEV(osd), NULL, &osa->rdev, 8);
1613         LASSERT(cnt <= ARRAY_SIZE(osd_oti_get(env)->oti_attr_bulk));
1614
1615         if (xattr) {
1616                 rc = -nvlist_size(xattr, &sa_size, NV_ENCODE_XDR);
1617                 LASSERT(rc == 0);
1618
1619                 dxattr = osd_zio_buf_alloc(sa_size);
1620                 LASSERT(dxattr);
1621
1622                 rc = -nvlist_pack(xattr, &dxattr, &sa_size,
1623                                 NV_ENCODE_XDR, KM_SLEEP);
1624                 LASSERT(rc == 0);
1625
1626                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_DXATTR(osd),
1627                                 NULL, dxattr, sa_size);
1628         }
1629
1630         rc = -sa_replace_all_by_template(sa_hdl, bulk, cnt, tx);
1631         if (dxattr)
1632                 osd_zio_buf_free(dxattr, sa_size);
1633
1634         return rc;
1635 }
1636
1637 int osd_find_new_dnode(const struct lu_env *env, dmu_tx_t *tx,
1638                        uint64_t oid, dnode_t **dnp)
1639 {
1640         dmu_tx_hold_t *txh;
1641         int rc = 0;
1642
1643         /* take dnode_t from tx to save on dnode#->dnode_t lookup */
1644         for (txh = list_tail(&tx->tx_holds); txh;
1645              txh = list_prev(&tx->tx_holds, txh)) {
1646                 dnode_t *dn = txh->txh_dnode;
1647                 dmu_buf_impl_t *db;
1648
1649                 if (dn == NULL)
1650                         continue;
1651                 if (dn->dn_object != oid)
1652                         continue;
1653                 db = dn->dn_bonus;
1654                 if (db == NULL) {
1655                         rw_enter(&dn->dn_struct_rwlock, RW_WRITER);
1656                         if (dn->dn_bonus == NULL)
1657                                 dbuf_create_bonus(dn);
1658                         rw_exit(&dn->dn_struct_rwlock);
1659                 }
1660                 db = dn->dn_bonus;
1661                 LASSERT(db);
1662                 LASSERT(dn->dn_handle);
1663                 DB_DNODE_ENTER(db);
1664                 if (zfs_refcount_add(&db->db_holds, osd_obj_tag) == 1) {
1665                         zfs_refcount_add(&dn->dn_holds, osd_obj_tag);
1666                         atomic_inc_32(&dn->dn_dbufs_count);
1667                 }
1668                 *dnp = dn;
1669                 DB_DNODE_EXIT(db);
1670                 dbuf_read(db, NULL, DB_RF_MUST_SUCCEED | DB_RF_NOPREFETCH);
1671                 break;
1672         }
1673
1674         if (unlikely(*dnp == NULL))
1675                 rc = __osd_obj2dnode(tx->tx_objset, oid, dnp);
1676
1677         return rc;
1678 }
1679
1680 #ifdef HAVE_DMU_OBJECT_ALLOC_DNSIZE
1681 int osd_find_dnsize(struct osd_device *osd, int ea_in_bonus)
1682 {
1683         int dnsize;
1684
1685         if (osd->od_dnsize == ZFS_DNSIZE_AUTO) {
1686                 dnsize = DNODE_MIN_SIZE;
1687                 do {
1688                         if (DN_BONUS_SIZE(dnsize) >= ea_in_bonus + 32)
1689                                 break;
1690                         dnsize <<= 1;
1691                 } while (dnsize < DNODE_MAX_SIZE);
1692                 if (dnsize > DNODE_MAX_SIZE)
1693                         dnsize = DNODE_MAX_SIZE;
1694         } else if (osd->od_dnsize == ZFS_DNSIZE_1K) {
1695                 dnsize = 1024;
1696         } else if (osd->od_dnsize == ZFS_DNSIZE_2K) {
1697                 dnsize = 2048;
1698         } else if (osd->od_dnsize == ZFS_DNSIZE_4K) {
1699                 dnsize = 4096;
1700         } else if (osd->od_dnsize == ZFS_DNSIZE_8K) {
1701                 dnsize = 8192;
1702         } else if (osd->od_dnsize == ZFS_DNSIZE_16K) {
1703                 dnsize = 16384;
1704         } else {
1705                 dnsize = DNODE_MIN_SIZE;
1706         }
1707         return dnsize;
1708 }
1709 #endif
1710
1711 /*
1712  * The transaction passed to this routine must have
1713  * dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT) called and then assigned
1714  * to a transaction group.
1715  */
1716 int __osd_object_create(const struct lu_env *env, struct osd_device *osd,
1717                         struct osd_object *obj, const struct lu_fid *fid,
1718                         dnode_t **dnp, dmu_tx_t *tx, struct lu_attr *la)
1719 {
1720         dmu_object_type_t type = DMU_OT_PLAIN_FILE_CONTENTS;
1721         uint64_t oid;
1722         int size;
1723
1724         /* Use DMU_OTN_UINT8_METADATA for local objects so their data blocks
1725          * would get an additional ditto copy */
1726         if (unlikely(S_ISREG(la->la_mode) &&
1727                      fid_seq_is_local_file(fid_seq(fid))))
1728                 type = DMU_OTN_UINT8_METADATA;
1729
1730         /* Create a new DMU object using the default dnode size. */
1731         if (obj)
1732                 size = obj->oo_ea_in_bonus;
1733         else
1734                 size = OSD_BASE_EA_IN_BONUS;
1735         oid = osd_dmu_object_alloc(osd->od_os, type, 0,
1736                                    osd_find_dnsize(osd, size), tx);
1737
1738         LASSERT(la->la_valid & LA_MODE);
1739         la->la_size = 0;
1740         la->la_nlink = 1;
1741
1742         return osd_find_new_dnode(env, tx, oid, dnp);
1743 }
1744
1745 /*
1746  * The transaction passed to this routine must have
1747  * dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, ...) called and then assigned
1748  * to a transaction group.
1749  *
1750  * Using ZAP_FLAG_HASH64 will force the ZAP to always be a FAT ZAP.
1751  * This is fine for directories today, because storing the FID in the dirent
1752  * will also require a FAT ZAP.  If there is a new type of micro ZAP created
1753  * then we might need to re-evaluate the use of this flag and instead do
1754  * a conversion from the different internal ZAP hash formats being used. */
1755 int __osd_zap_create(const struct lu_env *env, struct osd_device *osd,
1756                      dnode_t **dnp, dmu_tx_t *tx, struct lu_attr *la,
1757                      unsigned dnsize, zap_flags_t flags)
1758 {
1759         uint64_t oid;
1760
1761         /* Assert that the transaction has been assigned to a
1762            transaction group. */
1763         LASSERT(tx->tx_txg != 0);
1764         *dnp = NULL;
1765
1766         oid = osd_zap_create_flags(osd->od_os, 0, flags | ZAP_FLAG_HASH64,
1767                                    DMU_OT_DIRECTORY_CONTENTS,
1768                                    14, /* == ZFS fzap_default_blockshift */
1769                                    DN_MAX_INDBLKSHIFT, /* indirect blockshift */
1770                                    dnsize, tx);
1771
1772         la->la_size = 2;
1773         la->la_nlink = 1;
1774
1775         return osd_find_new_dnode(env, tx, oid, dnp);
1776 }
1777
1778 static dnode_t *osd_mkidx(const struct lu_env *env, struct osd_object *obj,
1779                           struct lu_attr *la, struct osd_thandle *oh)
1780 {
1781         struct osd_device *osd = osd_obj2dev(obj);
1782         dnode_t *dn;
1783         int rc;
1784
1785         /* Index file should be created as regular file in order not to confuse
1786          * ZPL which could interpret them as directory.
1787          * We set ZAP_FLAG_UINT64_KEY to let ZFS know than we are going to use
1788          * binary keys */
1789         LASSERT(S_ISREG(la->la_mode));
1790         rc = __osd_zap_create(env, osd, &dn, oh->ot_tx, la,
1791                 osd_find_dnsize(osd, obj->oo_ea_in_bonus), ZAP_FLAG_UINT64_KEY);
1792         if (rc)
1793                 return ERR_PTR(rc);
1794         return dn;
1795 }
1796
1797 static dnode_t *osd_mkdir(const struct lu_env *env, struct osd_object *obj,
1798                           struct lu_attr *la, struct osd_thandle *oh)
1799 {
1800         struct osd_device *osd = osd_obj2dev(obj);
1801         dnode_t *dn;
1802         int rc;
1803
1804         LASSERT(S_ISDIR(la->la_mode));
1805         rc = __osd_zap_create(env, osd, &dn, oh->ot_tx, la,
1806                               osd_find_dnsize(osd, obj->oo_ea_in_bonus), 0);
1807         if (rc)
1808                 return ERR_PTR(rc);
1809         return dn;
1810 }
1811
1812 static dnode_t *osd_mkreg(const struct lu_env *env, struct osd_object *obj,
1813                           struct lu_attr *la, struct osd_thandle *oh)
1814 {
1815         const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
1816         struct osd_device *osd = osd_obj2dev(obj);
1817         dnode_t *dn;
1818         int rc;
1819
1820         LASSERT(S_ISREG(la->la_mode));
1821         rc = __osd_object_create(env, osd, obj, fid, &dn, oh->ot_tx, la);
1822         if (rc)
1823                 return ERR_PTR(rc);
1824
1825         if ((fid_is_idif(fid) || fid_is_norm(fid) || fid_is_echo(fid))) {
1826                 /* The minimum block size must be at least page size otherwise
1827                  * it will break the assumption in tgt_thread_big_cache where
1828                  * the array size is PTLRPC_MAX_BRW_PAGES. It will also affect
1829                  * RDMA due to subpage transfer size */
1830                 rc = -dmu_object_set_blocksize(osd->od_os, dn->dn_object,
1831                                                PAGE_SIZE, 0, oh->ot_tx);
1832                 if (unlikely(rc)) {
1833                         CERROR("%s: can't change blocksize: %d\n",
1834                                osd->od_svname, rc);
1835                         return ERR_PTR(rc);
1836                 }
1837         } else if ((fid_is_llog(fid))) {
1838                 rc = -dmu_object_set_blocksize(osd->od_os, dn->dn_object,
1839                                                LLOG_MIN_CHUNK_SIZE, 0, oh->ot_tx);
1840                 if (unlikely(rc)) {
1841                         CERROR("%s: can't change blocksize: %d\n",
1842                                osd->od_svname, rc);
1843                         return ERR_PTR(rc);
1844                 }
1845         }
1846
1847         return dn;
1848 }
1849
1850 static dnode_t *osd_mksym(const struct lu_env *env, struct osd_object *obj,
1851                           struct lu_attr *la, struct osd_thandle *oh)
1852 {
1853         dnode_t *dn;
1854         int rc;
1855
1856         LASSERT(S_ISLNK(la->la_mode));
1857         rc = __osd_object_create(env, osd_obj2dev(obj), obj,
1858                                  lu_object_fid(&obj->oo_dt.do_lu),
1859                                  &dn, oh->ot_tx, la);
1860         if (rc)
1861                 return ERR_PTR(rc);
1862         return dn;
1863 }
1864
1865 static dnode_t *osd_mknod(const struct lu_env *env, struct osd_object *obj,
1866                           struct lu_attr *la, struct osd_thandle *oh)
1867 {
1868         dnode_t *dn;
1869         int rc;
1870
1871         if (S_ISCHR(la->la_mode) || S_ISBLK(la->la_mode))
1872                 la->la_valid |= LA_RDEV;
1873
1874         rc = __osd_object_create(env, osd_obj2dev(obj), obj,
1875                                  lu_object_fid(&obj->oo_dt.do_lu),
1876                                  &dn, oh->ot_tx, la);
1877         if (rc)
1878                 return ERR_PTR(rc);
1879         return dn;
1880 }
1881
1882 typedef dnode_t *(*osd_obj_type_f)(const struct lu_env *env,
1883                                    struct osd_object *obj,
1884                                    struct lu_attr *la,
1885                                    struct osd_thandle *oh);
1886
1887 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
1888 {
1889         osd_obj_type_f result;
1890
1891         switch (type) {
1892         case DFT_DIR:
1893                 result = osd_mkdir;
1894                 break;
1895         case DFT_INDEX:
1896                 result = osd_mkidx;
1897                 break;
1898         case DFT_REGULAR:
1899                 result = osd_mkreg;
1900                 break;
1901         case DFT_SYM:
1902                 result = osd_mksym;
1903                 break;
1904         case DFT_NODE:
1905                 result = osd_mknod;
1906                 break;
1907         default:
1908                 LBUG();
1909                 break;
1910         }
1911         return result;
1912 }
1913
1914 /*
1915  * Concurrency: @dt is write locked.
1916  */
1917 static int osd_create(const struct lu_env *env, struct dt_object *dt,
1918                       struct lu_attr *attr, struct dt_allocation_hint *hint,
1919                       struct dt_object_format *dof, struct thandle *th)
1920 {
1921         struct osd_thread_info  *info = osd_oti_get(env);
1922         struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
1923         struct zpl_direntry     *zde = &info->oti_zde.lzd_reg;
1924         const struct lu_fid     *fid = lu_object_fid(&dt->do_lu);
1925         struct osd_object       *obj = osd_dt_obj(dt);
1926         struct osd_device       *osd = osd_obj2dev(obj);
1927         char                    *buf = info->oti_str;
1928         struct osd_thandle      *oh;
1929         dnode_t *dn = NULL, *zdn = NULL;
1930         uint64_t                 zapid, parent = 0;
1931         int                      rc;
1932         __u32 compat = 0;
1933
1934         ENTRY;
1935
1936         LASSERT(!fid_is_acct(fid));
1937
1938         /* concurrent create declarations should not see
1939          * the object inconsistent (db, attr, etc).
1940          * in regular cases acquisition should be cheap */
1941         down_write(&obj->oo_guard);
1942
1943         if (unlikely(dt_object_exists(dt)))
1944                 GOTO(out, rc = -EEXIST);
1945
1946         LASSERT(osd_invariant(obj));
1947         LASSERT(dof != NULL);
1948
1949         LASSERT(th != NULL);
1950         oh = container_of(th, struct osd_thandle, ot_super);
1951
1952         LASSERT(obj->oo_dn == NULL);
1953
1954         /* to follow ZFS on-disk format we need
1955          * to initialize parent dnode properly */
1956         if (hint != NULL && hint->dah_parent != NULL &&
1957             !dt_object_remote(hint->dah_parent))
1958                 parent = osd_dt_obj(hint->dah_parent)->oo_dn->dn_object;
1959
1960         /* we may fix some attributes, better do not change the source */
1961         obj->oo_attr = *attr;
1962         obj->oo_attr.la_size = 0;
1963         obj->oo_attr.la_nlink = 0;
1964         obj->oo_attr.la_valid |= LA_SIZE | LA_NLINK | LA_TYPE;
1965
1966 #ifdef ZFS_PROJINHERIT
1967         if (osd->od_projectused_dn) {
1968                 if (!(obj->oo_attr.la_valid & LA_PROJID))
1969                         obj->oo_attr.la_projid = ZFS_DEFAULT_PROJID;
1970                 obj->oo_with_projid = 1;
1971         }
1972 #endif
1973
1974         dn = osd_create_type_f(dof->dof_type)(env, obj, &obj->oo_attr, oh);
1975         if (IS_ERR(dn)) {
1976                 rc = PTR_ERR(dn);
1977                 dn = NULL;
1978                 GOTO(out, rc);
1979         }
1980
1981         zde->zde_pad = 0;
1982         zde->zde_dnode = dn->dn_object;
1983         zde->zde_type = S_DT(attr->la_mode & S_IFMT);
1984
1985         zapid = osd_get_name_n_idx(env, osd, fid, buf,
1986                                    sizeof(info->oti_str), &zdn);
1987         if (CFS_FAIL_CHECK(OBD_FAIL_OSD_NO_OI_ENTRY) ||
1988             (osd->od_is_ost && OBD_FAIL_CHECK(OBD_FAIL_OSD_COMPAT_NO_ENTRY)))
1989                 goto skip_add;
1990
1991         if (osd->od_is_ost && OBD_FAIL_CHECK(OBD_FAIL_OSD_COMPAT_INVALID_ENTRY))
1992                 zde->zde_dnode++;
1993
1994         rc = osd_zap_add(osd, zapid, zdn, buf, 8, 1, zde, oh->ot_tx);
1995         if (rc)
1996                 GOTO(out, rc);
1997
1998 skip_add:
1999         obj->oo_dn = dn;
2000         /* Now add in all of the "SA" attributes */
2001         rc = osd_sa_handle_get(obj);
2002         if (rc)
2003                 GOTO(out, rc);
2004
2005         rc = -nvlist_alloc(&obj->oo_sa_xattr, NV_UNIQUE_NAME, KM_SLEEP);
2006         if (rc)
2007                 GOTO(out, rc);
2008
2009         /* initialize LMA */
2010         if (fid_is_idif(fid) || (fid_is_norm(fid) && osd->od_is_ost))
2011                 compat |= LMAC_FID_ON_OST;
2012         lustre_lma_init(lma, fid, compat, 0);
2013         lustre_lma_swab(lma);
2014         rc = -nvlist_add_byte_array(obj->oo_sa_xattr, XATTR_NAME_LMA,
2015                                     (uchar_t *)lma, sizeof(*lma));
2016         if (rc)
2017                 GOTO(out, rc);
2018
2019         /* configure new osd object */
2020         obj->oo_parent = parent != 0 ? parent : zapid;
2021         obj->oo_late_attr_set = 1;
2022         rc = __osd_sa_xattr_schedule_update(env, obj, oh);
2023         if (rc)
2024                 GOTO(out, rc);
2025
2026         /* XXX: oo_lma_flags */
2027         obj->oo_dt.do_lu.lo_header->loh_attr |= obj->oo_attr.la_mode & S_IFMT;
2028         if (likely(!fid_is_acct(lu_object_fid(&obj->oo_dt.do_lu))))
2029                 /* no body operations for accounting objects */
2030                 obj->oo_dt.do_body_ops = &osd_body_ops;
2031
2032         osd_idc_find_and_init(env, osd, obj);
2033
2034 out:
2035         if (unlikely(rc && dn)) {
2036                 dmu_object_free(osd->od_os, dn->dn_object, oh->ot_tx);
2037                 osd_dnode_rele(dn);
2038                 obj->oo_dn = NULL;
2039         } else if (!rc) {
2040                 obj->oo_dt.do_lu.lo_header->loh_attr |= LOHA_EXISTS;
2041         }
2042         up_write(&obj->oo_guard);
2043         RETURN(rc);
2044 }
2045
2046 static int osd_declare_ref_add(const struct lu_env *env, struct dt_object *dt,
2047                                struct thandle *th)
2048 {
2049         osd_idc_find_and_init(env, osd_dev(dt->do_lu.lo_dev), osd_dt_obj(dt));
2050         return osd_declare_attr_set(env, dt, NULL, th);
2051 }
2052
2053 /*
2054  * Concurrency: @dt is write locked.
2055  */
2056 static int osd_ref_add(const struct lu_env *env, struct dt_object *dt,
2057                        struct thandle *handle)
2058 {
2059         struct osd_object       *obj = osd_dt_obj(dt);
2060         struct osd_thandle      *oh;
2061         struct osd_device       *osd = osd_obj2dev(obj);
2062         uint64_t                 nlink;
2063         int rc;
2064
2065         ENTRY;
2066
2067         down_read(&obj->oo_guard);
2068         if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
2069                 GOTO(out, rc = -ENOENT);
2070
2071         LASSERT(osd_invariant(obj));
2072         LASSERT(obj->oo_sa_hdl != NULL);
2073
2074         oh = container_of(handle, struct osd_thandle, ot_super);
2075
2076         write_lock(&obj->oo_attr_lock);
2077         nlink = ++obj->oo_attr.la_nlink;
2078         write_unlock(&obj->oo_attr_lock);
2079
2080         rc = osd_object_sa_update(obj, SA_ZPL_LINKS(osd), &nlink, 8, oh);
2081
2082 out:
2083         up_read(&obj->oo_guard);
2084         RETURN(rc);
2085 }
2086
2087 static int osd_declare_ref_del(const struct lu_env *env, struct dt_object *dt,
2088                                struct thandle *handle)
2089 {
2090         osd_idc_find_and_init(env, osd_dev(dt->do_lu.lo_dev), osd_dt_obj(dt));
2091         return osd_declare_attr_set(env, dt, NULL, handle);
2092 }
2093
2094 /*
2095  * Concurrency: @dt is write locked.
2096  */
2097 static int osd_ref_del(const struct lu_env *env, struct dt_object *dt,
2098                        struct thandle *handle)
2099 {
2100         struct osd_object       *obj = osd_dt_obj(dt);
2101         struct osd_thandle      *oh;
2102         struct osd_device       *osd = osd_obj2dev(obj);
2103         uint64_t                 nlink;
2104         int                      rc;
2105
2106         ENTRY;
2107
2108         down_read(&obj->oo_guard);
2109
2110         if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
2111                 GOTO(out, rc = -ENOENT);
2112
2113         LASSERT(osd_invariant(obj));
2114         LASSERT(obj->oo_sa_hdl != NULL);
2115
2116         oh = container_of(handle, struct osd_thandle, ot_super);
2117         LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
2118
2119         write_lock(&obj->oo_attr_lock);
2120         nlink = --obj->oo_attr.la_nlink;
2121         write_unlock(&obj->oo_attr_lock);
2122
2123         rc = osd_object_sa_update(obj, SA_ZPL_LINKS(osd), &nlink, 8, oh);
2124
2125 out:
2126         up_read(&obj->oo_guard);
2127         RETURN(rc);
2128 }
2129
2130 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
2131                            __u64 start, __u64 end)
2132 {
2133         struct osd_device *osd = osd_obj2dev(osd_dt_obj(dt));
2134         uint64_t txg = 0;
2135         ENTRY;
2136
2137         if (osd->od_dt_dev.dd_rdonly)
2138                 RETURN(0);
2139
2140         txg = osd_db_dirty_txg(osd_dt_obj(dt)->oo_dn->dn_dbuf);
2141         if (txg) {
2142                 /* the object is dirty or being synced */
2143                 if (osd_object_sync_delay_us < 0)
2144                         txg_wait_synced(dmu_objset_pool(osd->od_os), txg);
2145                 else
2146                         udelay(osd_object_sync_delay_us);
2147         }
2148
2149         RETURN(0);
2150 }
2151
2152 static int osd_invalidate(const struct lu_env *env, struct dt_object *dt)
2153 {
2154         return 0;
2155 }
2156
2157 static bool osd_check_stale(struct dt_object *dt)
2158 {
2159         return false;
2160 }
2161
2162 static const struct dt_object_operations osd_obj_ops = {
2163         .do_read_lock           = osd_read_lock,
2164         .do_write_lock          = osd_write_lock,
2165         .do_read_unlock         = osd_read_unlock,
2166         .do_write_unlock        = osd_write_unlock,
2167         .do_write_locked        = osd_write_locked,
2168         .do_attr_get            = osd_attr_get,
2169         .do_declare_attr_set    = osd_declare_attr_set,
2170         .do_attr_set            = osd_attr_set,
2171         .do_ah_init             = osd_ah_init,
2172         .do_declare_create      = osd_declare_create,
2173         .do_create              = osd_create,
2174         .do_declare_destroy     = osd_declare_destroy,
2175         .do_destroy             = osd_destroy,
2176         .do_index_try           = osd_index_try,
2177         .do_declare_ref_add     = osd_declare_ref_add,
2178         .do_ref_add             = osd_ref_add,
2179         .do_declare_ref_del     = osd_declare_ref_del,
2180         .do_ref_del             = osd_ref_del,
2181         .do_xattr_get           = osd_xattr_get,
2182         .do_declare_xattr_set   = osd_declare_xattr_set,
2183         .do_xattr_set           = osd_xattr_set,
2184         .do_declare_xattr_del   = osd_declare_xattr_del,
2185         .do_xattr_del           = osd_xattr_del,
2186         .do_xattr_list          = osd_xattr_list,
2187         .do_object_sync         = osd_object_sync,
2188         .do_invalidate          = osd_invalidate,
2189         .do_check_stale         = osd_check_stale,
2190 };
2191
2192 static const struct lu_object_operations osd_lu_obj_ops = {
2193         .loo_object_init        = osd_object_init,
2194         .loo_object_delete      = osd_object_delete,
2195         .loo_object_release     = osd_object_release,
2196         .loo_object_free        = osd_object_free,
2197         .loo_object_print       = osd_object_print,
2198         .loo_object_invariant   = osd_object_invariant,
2199 };
2200
2201 static int osd_otable_it_attr_get(const struct lu_env *env,
2202                                 struct dt_object *dt,
2203                                 struct lu_attr *attr)
2204 {
2205         attr->la_valid = 0;
2206         return 0;
2207 }
2208
2209 static const struct dt_object_operations osd_obj_otable_it_ops = {
2210         .do_attr_get            = osd_otable_it_attr_get,
2211         .do_index_try           = osd_index_try,
2212 };
2213
2214 module_param(osd_object_sync_delay_us, int, 0644);
2215 MODULE_PARM_DESC(osd_object_sync_delay_us,
2216                  "If zero or larger delay N usec instead of doing object sync");