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 extern struct dt_body_operations osd_body_ops;
  70 static struct dt_object_operations osd_obj_otable_it_ops;
  71
  72 extern struct kmem_cache *osd_object_kmem;
  73
  74 static void
  75 osd_object_sa_fini(struct osd_object *obj)
  76 {
  77         if (obj->oo_sa_hdl) {
  78                 sa_handle_destroy(obj->oo_sa_hdl);
  79                 obj->oo_sa_hdl = NULL;
  80         }
  81 }
  82
  83 static int
  84 osd_object_sa_init(struct osd_object *obj, struct osd_device *o)
  85 {
  86         int rc;
  87
  88         LASSERT(obj->oo_sa_hdl == NULL);
  89         LASSERT(obj->oo_dn != NULL);
  90
  91         rc = osd_sa_handle_get(obj);
  92         if (rc)
  93                 return rc;
  94
  95         /* Cache the xattr object id, valid for the life of the object */
  96         rc = -sa_lookup(obj->oo_sa_hdl, SA_ZPL_XATTR(o), &obj->oo_xattr, 8);
  97         if (rc == -ENOENT) {
  98                 obj->oo_xattr = ZFS_NO_OBJECT;
  99                 rc = 0;
 100         } else if (rc) {
 101                 osd_object_sa_fini(obj);
 102         }
 103
 104         return rc;
 105 }
 106
 107 /*
 108  * Add object to list of dirty objects in tx handle.
 109  */
 110 void osd_object_sa_dirty_add(struct osd_object *obj, struct osd_thandle *oh)
 111 {
 112         if (!list_empty(&obj->oo_sa_linkage))
 113                 return;
 114
 115         write_lock(&obj->oo_attr_lock);
 116         if (likely(list_empty(&obj->oo_sa_linkage)))
 117                 list_add(&obj->oo_sa_linkage, &oh->ot_sa_list);
 118         write_unlock(&obj->oo_attr_lock);
 119 }
 120
 121 /*
 122  * Release spill block dbuf hold for all dirty SAs.
 123  */
 124 void osd_object_sa_dirty_rele(const struct lu_env *env, struct osd_thandle *oh)
 125 {
 126         struct osd_object *obj;
 127
 128         while (!list_empty(&oh->ot_sa_list)) {
 129                 obj = list_entry(oh->ot_sa_list.next,
 130                                  struct osd_object, oo_sa_linkage);
 131                 write_lock(&obj->oo_attr_lock);
 132                 list_del_init(&obj->oo_sa_linkage);
 133                 write_unlock(&obj->oo_attr_lock);
 134                 if (obj->oo_late_xattr) {
 135                         /*
 136                          * take oo_guard to protect oo_sa_xattr buffer
 137                          * from concurrent update by osd_xattr_set()
 138                          */
 139                         LASSERT(oh->ot_assigned != 0);
 140                         down_write(&obj->oo_guard);
 141                         if (obj->oo_late_attr_set)
 142                                 __osd_sa_attr_init(env, obj, oh);
 143                         else if (obj->oo_late_xattr)
 144                                 __osd_sa_xattr_update(env, obj, oh);
 145                         up_write(&obj->oo_guard);
 146                 }
 147                 sa_spill_rele(obj->oo_sa_hdl);
 148         }
 149 }
 150
 151 /*
 152  * Update the SA and add the object to the dirty list.
 153  */
 154 int osd_object_sa_update(struct osd_object *obj, sa_attr_type_t type,
 155                          void *buf, uint32_t buflen, struct osd_thandle *oh)
 156 {
 157         int rc;
 158
 159         LASSERT(obj->oo_sa_hdl != NULL);
 160         LASSERT(oh->ot_tx != NULL);
 161
 162         rc = -sa_update(obj->oo_sa_hdl, type, buf, buflen, oh->ot_tx);
 163         osd_object_sa_dirty_add(obj, oh);
 164
 165         return rc;
 166 }
 167
 168 /*
 169  * Bulk update the SA and add the object to the dirty list.
 170  */
 171 static int
 172 osd_object_sa_bulk_update(struct osd_object *obj, sa_bulk_attr_t *attrs,
 173                           int count, struct osd_thandle *oh)
 174 {
 175         int rc;
 176
 177         LASSERT(obj->oo_sa_hdl != NULL);
 178         LASSERT(oh->ot_tx != NULL);
 179
 180         rc = -sa_bulk_update(obj->oo_sa_hdl, attrs, count, oh->ot_tx);
 181         osd_object_sa_dirty_add(obj, oh);
 182
 183         return rc;
 184 }
 185
 186 /*
 187  * Retrieve the attributes of a DMU object
 188  */
 189 static int __osd_object_attr_get(const struct lu_env *env, struct osd_device *o,
 190                                  struct osd_object *obj, struct lu_attr *la)
 191 {
 192         struct osa_attr *osa = &osd_oti_get(env)->oti_osa;
 193         sa_bulk_attr_t  *bulk = osd_oti_get(env)->oti_attr_bulk;
 194         int              cnt = 0;
 195         int              rc;
 196         ENTRY;
 197
 198         LASSERT(obj->oo_dn != NULL);
 199
 200         la->la_valid |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE | LA_TYPE |
 201                         LA_SIZE | LA_UID | LA_GID | LA_FLAGS | LA_NLINK;
 202
 203         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(o), NULL, osa->atime, 16);
 204         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(o), NULL, osa->mtime, 16);
 205         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(o), NULL, osa->ctime, 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_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         CLASSERT(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         OBD_SLAB_FREE_PTR(obj, osd_object_kmem);
 701         if (unlikely(h)) {
 702                 lu_object_header_fini(h);
 703                 OBD_FREE_PTR(h);
 704         }
 705 }
 706
 707 static int
 708 osd_object_unlinked_add(struct osd_object *obj, struct osd_thandle *oh)
 709 {
 710         int rc = -EBUSY;
 711
 712         LASSERT(obj->oo_destroy == OSD_DESTROY_ASYNC);
 713
 714         /* the object is supposed to be exclusively locked by
 715          * the caller (osd_destroy()), while the transaction
 716          * (oh) is per-thread and not shared */
 717         if (likely(list_empty(&obj->oo_unlinked_linkage))) {
 718                 list_add(&obj->oo_unlinked_linkage, &oh->ot_unlinked_list);
 719                 rc = 0;
 720         }
 721
 722         return rc;
 723 }
 724
 725 /* Default to max data size covered by a level-1 indirect block */
 726 static unsigned long osd_sync_destroy_max_size =
 727         1UL << (DN_MAX_INDBLKSHIFT - SPA_BLKPTRSHIFT + SPA_MAXBLOCKSHIFT);
 728 module_param(osd_sync_destroy_max_size, ulong, 0444);
 729 MODULE_PARM_DESC(osd_sync_destroy_max_size, "Maximum object size to use synchronous destroy.");
 730
 731 static inline void
 732 osd_object_set_destroy_type(struct osd_object *obj)
 733 {
 734         /*
 735          * Lock-less OST_WRITE can race with OST_DESTROY, so set destroy type
 736          * only once and use it consistently thereafter.
 737          */
 738         down_write(&obj->oo_guard);
 739         if (obj->oo_destroy == OSD_DESTROY_NONE) {
 740                 if (obj->oo_attr.la_size <= osd_sync_destroy_max_size)
 741                         obj->oo_destroy = OSD_DESTROY_SYNC;
 742                 else /* Larger objects are destroyed asynchronously */
 743                         obj->oo_destroy = OSD_DESTROY_ASYNC;
 744         }
 745         up_write(&obj->oo_guard);
 746 }
 747
 748 static int osd_declare_destroy(const struct lu_env *env, struct dt_object *dt,
 749                                struct thandle *th)
 750 {
 751         const struct lu_fid     *fid = lu_object_fid(&dt->do_lu);
 752         struct osd_object       *obj = osd_dt_obj(dt);
 753         struct osd_device       *osd = osd_obj2dev(obj);
 754         struct osd_thandle      *oh;
 755         dnode_t *dn;
 756         int                      rc;
 757         uint64_t                 zapid;
 758         ENTRY;
 759
 760         LASSERT(th != NULL);
 761         LASSERT(dt_object_exists(dt));
 762
 763         oh = container_of0(th, struct osd_thandle, ot_super);
 764         LASSERT(oh->ot_tx != NULL);
 765
 766         dmu_tx_mark_netfree(oh->ot_tx);
 767
 768         /* declare that we'll remove object from fid-dnode mapping */
 769         zapid = osd_get_name_n_idx(env, osd, fid, NULL, 0, &dn);
 770         osd_tx_hold_zap(oh->ot_tx, zapid, dn, FALSE, NULL);
 771
 772         osd_declare_xattrs_destroy(env, obj, oh);
 773
 774         /* one less inode */
 775         rc = osd_declare_quota(env, osd, obj->oo_attr.la_uid,
 776                                obj->oo_attr.la_gid, obj->oo_attr.la_projid,
 777                                -1, oh, NULL, OSD_QID_INODE);
 778         if (rc)
 779                 RETURN(rc);
 780
 781         /* data to be truncated */
 782         rc = osd_declare_quota(env, osd, obj->oo_attr.la_uid,
 783                                obj->oo_attr.la_gid, obj->oo_attr.la_projid,
 784                                0, oh, NULL, OSD_QID_BLK);
 785         if (rc)
 786                 RETURN(rc);
 787
 788         osd_object_set_destroy_type(obj);
 789         if (obj->oo_destroy == OSD_DESTROY_SYNC)
 790                 dmu_tx_hold_free(oh->ot_tx, obj->oo_dn->dn_object,
 791                                  0, DMU_OBJECT_END);
 792         else
 793                 osd_tx_hold_zap(oh->ot_tx, osd->od_unlinked->dn_object,
 794                                 osd->od_unlinked, TRUE, NULL);
 795
 796         /* remove agent entry (if have) from remote parent */
 797         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu))
 798                 osd_tx_hold_zap(oh->ot_tx, osd->od_remote_parent_dir,
 799                                 NULL, FALSE, NULL);
 800
 801         /* will help to find FID->ino when this object is being
 802          * added to PENDING/ */
 803         osd_idc_find_and_init(env, osd, obj);
 804
 805         RETURN(0);
 806 }
 807
 808 static int osd_destroy(const struct lu_env *env, struct dt_object *dt,
 809                        struct thandle *th)
 810 {
 811         struct osd_thread_info  *info = osd_oti_get(env);
 812         char                    *buf = info->oti_str;
 813         struct osd_object       *obj = osd_dt_obj(dt);
 814         struct osd_device       *osd = osd_obj2dev(obj);
 815         const struct lu_fid     *fid = lu_object_fid(&dt->do_lu);
 816         struct osd_thandle      *oh;
 817         int                      rc;
 818         uint64_t                 oid, zapid;
 819         dnode_t *zdn;
 820         ENTRY;
 821
 822         down_write(&obj->oo_guard);
 823
 824         if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
 825                 GOTO(out, rc = -ENOENT);
 826
 827         LASSERT(obj->oo_dn != NULL);
 828
 829         oh = container_of0(th, struct osd_thandle, ot_super);
 830         LASSERT(oh != NULL);
 831         LASSERT(oh->ot_tx != NULL);
 832
 833         /* remove obj ref from index dir (it depends) */
 834         zapid = osd_get_name_n_idx(env, osd, fid, buf,
 835                                    sizeof(info->oti_str), &zdn);
 836         rc = osd_xattrs_destroy(env, obj, oh);
 837         if (rc) {
 838                 CERROR("%s: cannot destroy xattrs for %s: rc = %d\n",
 839                        osd->od_svname, buf, rc);
 840                 GOTO(out, rc);
 841         }
 842
 843         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu)) {
 844                 rc = osd_delete_from_remote_parent(env, osd, obj, oh, true);
 845                 if (rc)
 846                         GOTO(out, rc);
 847         }
 848
 849         oid = obj->oo_dn->dn_object;
 850         if (unlikely(obj->oo_destroy == OSD_DESTROY_NONE)) {
 851                 /* this may happen if the destroy wasn't declared
 852                  * e.g. when the object is created and then destroyed
 853                  * in the same transaction - we don't need additional
 854                  * space for destroy specifically */
 855                 LASSERT(obj->oo_attr.la_size <= osd_sync_destroy_max_size);
 856                 rc = -dmu_object_free(osd->od_os, oid, oh->ot_tx);
 857                 if (rc)
 858                         CERROR("%s: failed to free %s %llu: rc = %d\n",
 859                                osd->od_svname, buf, oid, rc);
 860         } else if (obj->oo_destroy == OSD_DESTROY_SYNC) {
 861                 rc = -dmu_object_free(osd->od_os, oid, oh->ot_tx);
 862                 if (rc)
 863                         CERROR("%s: failed to free %s %llu: rc = %d\n",
 864                                osd->od_svname, buf, oid, rc);
 865         } else { /* asynchronous destroy */
 866                 char *key = info->oti_key;
 867
 868                 rc = osd_object_unlinked_add(obj, oh);
 869                 if (rc)
 870                         GOTO(out, rc);
 871
 872                 snprintf(key, sizeof(info->oti_key), "%llx", oid);
 873                 rc = osd_zap_add(osd, osd->od_unlinked->dn_object,
 874                                  osd->od_unlinked, key, 8, 1, &oid, oh->ot_tx);
 875                 if (rc)
 876                         CERROR("%s: zap_add_int() failed %s %llu: rc = %d\n",
 877                                osd->od_svname, buf, oid, rc);
 878         }
 879
 880         /* Remove the OI mapping after the destroy to handle the race with
 881          * OI scrub that may insert missed OI mapping during the interval. */
 882         rc = osd_zap_remove(osd, zapid, zdn, buf, oh->ot_tx);
 883         if (unlikely(rc == -ENOENT))
 884                 rc = 0;
 885         if (rc)
 886                 CERROR("%s: zap_remove(%s) failed: rc = %d\n",
 887                        osd->od_svname, buf, rc);
 888
 889         GOTO(out, rc);
 890
 891 out:
 892         /* not needed in the cache anymore */
 893         set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
 894         if (rc == 0)
 895                 obj->oo_destroyed = 1;
 896         up_write(&obj->oo_guard);
 897         RETURN (0);
 898 }
 899
 900 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
 901 {
 902         struct osd_object *obj = osd_obj(l);
 903         const struct lu_fid *fid = lu_object_fid(l);
 904
 905         if (obj->oo_dn) {
 906                 if (likely(!fid_is_acct(fid))) {
 907                         osd_object_sa_fini(obj);
 908                         if (obj->oo_sa_xattr) {
 909                                 nvlist_free(obj->oo_sa_xattr);
 910                                 obj->oo_sa_xattr = NULL;
 911                         }
 912                         osd_dnode_rele(obj->oo_dn);
 913                         list_del(&obj->oo_sa_linkage);
 914                 }
 915                 obj->oo_dn = NULL;
 916         }
 917 }
 918
 919 /*
 920  * Concurrency: ->loo_object_release() is called under site spin-lock.
 921  */
 922 static void osd_object_release(const struct lu_env *env,
 923                                struct lu_object *l)
 924 {
 925 }
 926
 927 /*
 928  * Concurrency: shouldn't matter.
 929  */
 930 static int osd_object_print(const struct lu_env *env, void *cookie,
 931                             lu_printer_t p, const struct lu_object *l)
 932 {
 933         struct osd_object *o = osd_obj(l);
 934
 935         return (*p)(env, cookie, LUSTRE_OSD_ZFS_NAME"-object@%p", o);
 936 }
 937
 938 static void osd_read_lock(const struct lu_env *env, struct dt_object *dt,
 939                           unsigned role)
 940 {
 941         struct osd_object *obj = osd_dt_obj(dt);
 942
 943         LASSERT(osd_invariant(obj));
 944
 945         down_read_nested(&obj->oo_sem, role);
 946 }
 947
 948 static void osd_write_lock(const struct lu_env *env, struct dt_object *dt,
 949                            unsigned role)
 950 {
 951         struct osd_object *obj = osd_dt_obj(dt);
 952
 953         LASSERT(osd_invariant(obj));
 954
 955         down_write_nested(&obj->oo_sem, role);
 956 }
 957
 958 static void osd_read_unlock(const struct lu_env *env, struct dt_object *dt)
 959 {
 960         struct osd_object *obj = osd_dt_obj(dt);
 961
 962         LASSERT(osd_invariant(obj));
 963         up_read(&obj->oo_sem);
 964 }
 965
 966 static void osd_write_unlock(const struct lu_env *env, struct dt_object *dt)
 967 {
 968         struct osd_object *obj = osd_dt_obj(dt);
 969
 970         LASSERT(osd_invariant(obj));
 971         up_write(&obj->oo_sem);
 972 }
 973
 974 static int osd_write_locked(const struct lu_env *env, struct dt_object *dt)
 975 {
 976         struct osd_object *obj = osd_dt_obj(dt);
 977         int rc = 1;
 978
 979         LASSERT(osd_invariant(obj));
 980
 981         if (down_write_trylock(&obj->oo_sem)) {
 982                 rc = 0;
 983                 up_write(&obj->oo_sem);
 984         }
 985         return rc;
 986 }
 987
 988 static int osd_attr_get(const struct lu_env *env, struct dt_object *dt,
 989                         struct lu_attr *attr)
 990 {
 991         struct osd_object       *obj = osd_dt_obj(dt);
 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         read_unlock(&obj->oo_attr_lock);
1014         if (attr->la_valid & LA_FLAGS && attr->la_flags & LUSTRE_ORPHAN_FL)
1015                 CDEBUG(D_INFO, "%s: set orphan flag on "DFID" (%llx/%x)\n",
1016                        osd_obj2dev(obj)->od_svname,
1017                        PFID(lu_object_fid(&dt->do_lu)),
1018                        attr->la_valid, obj->oo_lma_flags);
1019
1020         /* with ZFS_DEBUG zrl_add_debug() called by DB_DNODE_ENTER()
1021          * from within sa_object_size() can block on a mutex, so
1022          * we can't call sa_object_size() holding rwlock */
1023         sa_object_size(obj->oo_sa_hdl, &blksize, &blocks);
1024         /* we do not control size of indices, so always calculate
1025          * it from number of blocks reported by DMU */
1026         if (S_ISDIR(attr->la_mode))
1027                 attr->la_size = 512 * blocks;
1028         /* Block size may be not set; suggest maximal I/O transfers. */
1029         if (blksize == 0)
1030                 blksize = osd_spa_maxblocksize(
1031                         dmu_objset_spa(osd_obj2dev(obj)->od_os));
1032
1033         attr->la_blksize = blksize;
1034         attr->la_blocks = blocks;
1035         attr->la_valid |= LA_BLOCKS | LA_BLKSIZE;
1036
1037 out:
1038         up_read(&obj->oo_guard);
1039         return rc;
1040 }
1041
1042 /* Simple wrapper on top of qsd API which implement quota transfer for osd
1043  * setattr needs. As a reminder, only the root user can change ownership of
1044  * a file, that's why EDQUOT & EINPROGRESS errors are discarded */
1045 static inline int qsd_transfer(const struct lu_env *env,
1046                                struct qsd_instance *qsd,
1047                                struct lquota_trans *trans, int qtype,
1048                                __u64 orig_id, __u64 new_id, __u64 bspace,
1049                                struct lquota_id_info *qi, bool ignore_edquot)
1050 {
1051         int     rc;
1052
1053         if (unlikely(qsd == NULL))
1054                 return 0;
1055
1056         LASSERT(qtype >= 0 && qtype < LL_MAXQUOTAS);
1057         qi->lqi_type = qtype;
1058
1059         /* inode accounting */
1060         qi->lqi_is_blk = false;
1061
1062         /* one more inode for the new owner ... */
1063         qi->lqi_id.qid_uid = new_id;
1064         qi->lqi_space      = 1;
1065         rc = qsd_op_begin(env, qsd, trans, qi, NULL);
1066         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
1067                 rc = 0;
1068         if (rc)
1069                 return rc;
1070
1071         /* and one less inode for the current id */
1072         qi->lqi_id.qid_uid = orig_id;;
1073         qi->lqi_space      = -1;
1074         /* can't get EDQUOT when reducing usage */
1075         rc = qsd_op_begin(env, qsd, trans, qi, NULL);
1076         if (rc == -EINPROGRESS)
1077                 rc = 0;
1078         if (rc)
1079                 return rc;
1080
1081         /* block accounting */
1082         qi->lqi_is_blk = true;
1083
1084         /* more blocks for the new owner ... */
1085         qi->lqi_id.qid_uid = new_id;
1086         qi->lqi_space      = bspace;
1087         rc = qsd_op_begin(env, qsd, trans, qi, NULL);
1088         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
1089                 rc = 0;
1090         if (rc)
1091                 return rc;
1092
1093         /* and finally less blocks for the current owner */
1094         qi->lqi_id.qid_uid = orig_id;
1095         qi->lqi_space      = -bspace;
1096         rc = qsd_op_begin(env, qsd, trans, qi, NULL);
1097         /* can't get EDQUOT when reducing usage */
1098         if (rc == -EINPROGRESS)
1099                 rc = 0;
1100         return rc;
1101 }
1102
1103 static int osd_declare_attr_set(const struct lu_env *env,
1104                                 struct dt_object *dt,
1105                                 const struct lu_attr *attr,
1106                                 struct thandle *handle)
1107 {
1108         struct osd_thread_info  *info = osd_oti_get(env);
1109         struct osd_object       *obj = osd_dt_obj(dt);
1110         struct osd_device       *osd = osd_obj2dev(obj);
1111         dmu_tx_hold_t           *txh;
1112         struct osd_thandle      *oh;
1113         uint64_t                 bspace;
1114         uint32_t                 blksize;
1115         int                      rc = 0;
1116         bool                     found;
1117         ENTRY;
1118
1119
1120         LASSERT(handle != NULL);
1121         LASSERT(osd_invariant(obj));
1122
1123         oh = container_of0(handle, struct osd_thandle, ot_super);
1124
1125         down_read(&obj->oo_guard);
1126         if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
1127                 GOTO(out, rc = 0);
1128
1129         LASSERT(obj->oo_sa_hdl != NULL);
1130         LASSERT(oh->ot_tx != NULL);
1131         /* regular attributes are part of the bonus buffer */
1132         /* let's check whether this object is already part of
1133          * transaction.. */
1134         found = false;
1135         for (txh = list_head(&oh->ot_tx->tx_holds); txh;
1136              txh = list_next(&oh->ot_tx->tx_holds, txh)) {
1137                 if (txh->txh_dnode == NULL)
1138                         continue;
1139                 if (txh->txh_dnode->dn_object != obj->oo_dn->dn_object)
1140                         continue;
1141                 /* this object is part of the transaction already
1142                  * we don't need to declare bonus again */
1143                 found = true;
1144                 break;
1145         }
1146         if (!found)
1147                 dmu_tx_hold_bonus(oh->ot_tx, obj->oo_dn->dn_object);
1148         if (oh->ot_tx->tx_err != 0)
1149                 GOTO(out, rc = -oh->ot_tx->tx_err);
1150
1151         if (attr && attr->la_valid & LA_FLAGS) {
1152                 /* LMA is usually a part of bonus, no need to declare
1153                  * anything else */
1154         }
1155
1156         if (attr && (attr->la_valid & (LA_UID | LA_GID | LA_PROJID))) {
1157                 sa_object_size(obj->oo_sa_hdl, &blksize, &bspace);
1158                 bspace = toqb(bspace * 512);
1159
1160                 CDEBUG(D_QUOTA, "%s: enforce quota on UID %u, GID %u,"
1161                        "the quota space is %lld (%u)\n", osd->od_svname,
1162                        attr->la_uid, attr->la_gid, bspace, blksize);
1163         }
1164
1165         if (attr && attr->la_valid & LA_UID) {
1166                 /* quota enforcement for user */
1167                 if (attr->la_uid != obj->oo_attr.la_uid) {
1168                         rc = qsd_transfer(env, osd_def_qsd(osd),
1169                                           &oh->ot_quota_trans, USRQUOTA,
1170                                           obj->oo_attr.la_uid, attr->la_uid,
1171                                           bspace, &info->oti_qi, true);
1172                         if (rc)
1173                                 GOTO(out, rc);
1174                 }
1175         }
1176         if (attr && attr->la_valid & LA_GID) {
1177                 /* quota enforcement for group */
1178                 if (attr->la_gid != obj->oo_attr.la_gid) {
1179                         rc = qsd_transfer(env, osd_def_qsd(osd),
1180                                           &oh->ot_quota_trans, GRPQUOTA,
1181                                           obj->oo_attr.la_gid, attr->la_gid,
1182                                           bspace, &info->oti_qi,
1183                                           !(attr->la_flags &
1184                                                         LUSTRE_SET_SYNC_FL));
1185                         if (rc)
1186                                 GOTO(out, rc);
1187                 }
1188         }
1189 #ifdef ZFS_PROJINHERIT
1190         if (attr && attr->la_valid & LA_PROJID) {
1191                 /* quota enforcement for project */
1192                 if (attr->la_projid != obj->oo_attr.la_projid) {
1193                         if (!osd->od_projectused_dn)
1194                                 GOTO(out, rc = -EOPNOTSUPP);
1195
1196                         /* Usually, if project quota is upgradable for the
1197                          * device, then the upgrade will be done before or when
1198                          * mount the device. So when we come here, this project
1199                          * should have project ID attribute already (that is
1200                          * zero by default).  Otherwise, there was something
1201                          * wrong during the former upgrade, let's return failure
1202                          * to report that.
1203                          *
1204                          * Please note that, different from other attributes,
1205                          * you can NOT simply set the project ID attribute under
1206                          * such case, because adding (NOT change) project ID
1207                          * attribute needs to change the object's attribute
1208                          * layout to match zfs backend quota accounting
1209                          * requirement. */
1210                         if (unlikely(!obj->oo_with_projid))
1211                                 GOTO(out, rc = -ENXIO);
1212
1213                         rc = qsd_transfer(env, osd_def_qsd(osd),
1214                                           &oh->ot_quota_trans, PRJQUOTA,
1215                                           obj->oo_attr.la_projid,
1216                                           attr->la_projid, bspace,
1217                                           &info->oti_qi, true);
1218                         if (rc)
1219                                 GOTO(out, rc);
1220                 }
1221         }
1222 #endif
1223 out:
1224         up_read(&obj->oo_guard);
1225         RETURN(rc);
1226 }
1227
1228 /*
1229  * Set the attributes of an object
1230  *
1231  * The transaction passed to this routine must have
1232  * dmu_tx_hold_bonus(tx, oid) called and then assigned
1233  * to a transaction group.
1234  */
1235 static int osd_attr_set(const struct lu_env *env, struct dt_object *dt,
1236                         const struct lu_attr *la, struct thandle *handle)
1237 {
1238         struct osd_thread_info  *info = osd_oti_get(env);
1239         sa_bulk_attr_t          *bulk = osd_oti_get(env)->oti_attr_bulk;
1240         struct osd_object       *obj = osd_dt_obj(dt);
1241         struct osd_device       *osd = osd_obj2dev(obj);
1242         struct osd_thandle      *oh;
1243         struct osa_attr         *osa = &info->oti_osa;
1244         __u64                    valid = la->la_valid;
1245         int                      cnt;
1246         int                      rc = 0;
1247
1248         ENTRY;
1249
1250         down_read(&obj->oo_guard);
1251         if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
1252                 GOTO(out, rc = -ENOENT);
1253
1254         LASSERT(handle != NULL);
1255         LASSERT(osd_invariant(obj));
1256         LASSERT(obj->oo_sa_hdl);
1257
1258         oh = container_of0(handle, struct osd_thandle, ot_super);
1259         /* Assert that the transaction has been assigned to a
1260            transaction group. */
1261         LASSERT(oh->ot_tx->tx_txg != 0);
1262
1263         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING) && !osd->od_is_ost) {
1264                 struct zpl_direntry *zde = &info->oti_zde.lzd_reg;
1265                 char *buf = info->oti_str;
1266                 dnode_t *zdn = NULL;
1267                 uint64_t zapid;
1268
1269                 zapid = osd_get_name_n_idx(env, osd, lu_object_fid(&dt->do_lu),
1270                                            buf, sizeof(info->oti_str), &zdn);
1271                 rc = osd_zap_lookup(osd, zapid, zdn, buf, 8,
1272                                     sizeof(*zde) / 8, zde);
1273                 if (!rc) {
1274                         zde->zde_dnode -= 1;
1275                         rc = -zap_update(osd->od_os, zapid, buf, 8,
1276                                          sizeof(*zde) / 8, zde, oh->ot_tx);
1277                 }
1278                 if (rc > 0)
1279                         rc = 0;
1280                 GOTO(out, rc);
1281         }
1282
1283         /* Only allow set size for regular file */
1284         if (!S_ISREG(dt->do_lu.lo_header->loh_attr))
1285                 valid &= ~(LA_SIZE | LA_BLOCKS);
1286
1287         if (valid & LA_CTIME && la->la_ctime == obj->oo_attr.la_ctime)
1288                 valid &= ~LA_CTIME;
1289
1290         if (valid & LA_MTIME && la->la_mtime == obj->oo_attr.la_mtime)
1291                 valid &= ~LA_MTIME;
1292
1293         if (valid & LA_ATIME && la->la_atime == obj->oo_attr.la_atime)
1294                 valid &= ~LA_ATIME;
1295
1296         if (valid == 0)
1297                 GOTO(out, rc = 0);
1298
1299         if (valid & LA_FLAGS) {
1300                 struct lustre_mdt_attrs *lma;
1301                 struct lu_buf buf;
1302                 int size = 0;
1303
1304                 if (la->la_flags & LUSTRE_LMA_FL_MASKS) {
1305                         LASSERT(!obj->oo_pfid_in_lma);
1306                         CLASSERT(sizeof(info->oti_buf) >= sizeof(*lma));
1307                         lma = (struct lustre_mdt_attrs *)&info->oti_buf;
1308                         buf.lb_buf = lma;
1309                         buf.lb_len = sizeof(info->oti_buf);
1310
1311                         /* Please do NOT call osd_xattr_get() directly, that
1312                          * will cause recursive down_read() on oo_guard. */
1313                         rc = osd_xattr_get_internal(env, obj, &buf,
1314                                                     XATTR_NAME_LMA, &size);
1315                         if (!rc && unlikely(size < sizeof(*lma))) {
1316                                 rc = -EINVAL;
1317                         } else if (!rc) {
1318                                 lma->lma_incompat =
1319                                         le32_to_cpu(lma->lma_incompat);
1320                                 lma->lma_incompat |=
1321                                         lustre_to_lma_flags(la->la_flags);
1322                                 lma->lma_incompat =
1323                                         cpu_to_le32(lma->lma_incompat);
1324                                 buf.lb_buf = lma;
1325                                 buf.lb_len = sizeof(*lma);
1326                                 rc = osd_xattr_set_internal(env, obj, &buf,
1327                                                             XATTR_NAME_LMA,
1328                                                             LU_XATTR_REPLACE,
1329                                                             oh);
1330                         }
1331                         if (rc < 0) {
1332                                 CWARN("%s: failed to set LMA flags: rc = %d\n",
1333                                        osd->od_svname, rc);
1334                                 GOTO(out, rc);
1335                         }
1336                 }
1337         }
1338
1339         write_lock(&obj->oo_attr_lock);
1340         cnt = 0;
1341
1342         if (valid & LA_PROJID) {
1343 #ifdef ZFS_PROJINHERIT
1344                 if (osd->od_projectused_dn) {
1345                         LASSERT(obj->oo_with_projid);
1346
1347                         osa->projid = obj->oo_attr.la_projid = la->la_projid;
1348                         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_PROJID(osd), NULL,
1349                                          &osa->projid, 8);
1350                 } else
1351 #endif
1352                         valid &= ~LA_PROJID;
1353         }
1354
1355         if (valid & LA_ATIME) {
1356                 osa->atime[0] = obj->oo_attr.la_atime = la->la_atime;
1357                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(osd), NULL,
1358                                  osa->atime, 16);
1359         }
1360         if (valid & LA_MTIME) {
1361                 osa->mtime[0] = obj->oo_attr.la_mtime = la->la_mtime;
1362                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(osd), NULL,
1363                                  osa->mtime, 16);
1364         }
1365         if (valid & LA_CTIME) {
1366                 osa->ctime[0] = obj->oo_attr.la_ctime = la->la_ctime;
1367                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(osd), NULL,
1368                                  osa->ctime, 16);
1369         }
1370         if (valid & LA_MODE) {
1371                 /* mode is stored along with type, so read it first */
1372                 obj->oo_attr.la_mode = (obj->oo_attr.la_mode & S_IFMT) |
1373                         (la->la_mode & ~S_IFMT);
1374                 osa->mode = obj->oo_attr.la_mode;
1375                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(osd), NULL,
1376                                  &osa->mode, 8);
1377         }
1378         if (valid & LA_SIZE) {
1379                 osa->size = obj->oo_attr.la_size = la->la_size;
1380                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(osd), NULL,
1381                                  &osa->size, 8);
1382         }
1383         if (valid & LA_NLINK) {
1384                 osa->nlink = obj->oo_attr.la_nlink = la->la_nlink;
1385                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(osd), NULL,
1386                                  &osa->nlink, 8);
1387         }
1388         if (valid & LA_RDEV) {
1389                 osa->rdev = obj->oo_attr.la_rdev = la->la_rdev;
1390                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_RDEV(osd), NULL,
1391                                  &osa->rdev, 8);
1392         }
1393         if (valid & LA_FLAGS) {
1394                 osa->flags = attrs_fs2zfs(la->la_flags);
1395                 /* many flags are not supported by zfs, so ensure a good cached
1396                  * copy */
1397                 obj->oo_attr.la_flags = attrs_zfs2fs(osa->flags);
1398 #ifdef ZFS_PROJINHERIT
1399                 if (obj->oo_with_projid && osd->od_projectused_dn)
1400                         osa->flags |= ZFS_PROJID;
1401 #endif
1402                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(osd), NULL,
1403                                  &osa->flags, 8);
1404         }
1405         if (valid & LA_UID) {
1406                 osa->uid = obj->oo_attr.la_uid = la->la_uid;
1407                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(osd), NULL,
1408                                  &osa->uid, 8);
1409         }
1410         if (valid & LA_GID) {
1411                 osa->gid = obj->oo_attr.la_gid = la->la_gid;
1412                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(osd), NULL,
1413                                  &osa->gid, 8);
1414         }
1415         obj->oo_attr.la_valid |= valid;
1416         write_unlock(&obj->oo_attr_lock);
1417
1418         LASSERT(cnt <= ARRAY_SIZE(osd_oti_get(env)->oti_attr_bulk));
1419         rc = osd_object_sa_bulk_update(obj, bulk, cnt, oh);
1420
1421 out:
1422         up_read(&obj->oo_guard);
1423         RETURN(rc);
1424 }
1425
1426 /*
1427  * Object creation.
1428  *
1429  * XXX temporary solution.
1430  */
1431
1432 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
1433                         struct dt_object *parent, struct dt_object *child,
1434                         umode_t child_mode)
1435 {
1436         LASSERT(ah);
1437
1438         ah->dah_parent = parent;
1439         ah->dah_mode = child_mode;
1440
1441         if (parent != NULL && !dt_object_remote(parent)) {
1442                 /* will help to find FID->ino at dt_insert("..") */
1443                 struct osd_object *pobj = osd_dt_obj(parent);
1444
1445                 osd_idc_find_and_init(env, osd_obj2dev(pobj), pobj);
1446         }
1447 }
1448
1449 static int osd_declare_create(const struct lu_env *env, struct dt_object *dt,
1450                               struct lu_attr *attr,
1451                               struct dt_allocation_hint *hint,
1452                               struct dt_object_format *dof,
1453                               struct thandle *handle)
1454 {
1455         const struct lu_fid     *fid = lu_object_fid(&dt->do_lu);
1456         struct osd_object       *obj = osd_dt_obj(dt);
1457         struct osd_device       *osd = osd_obj2dev(obj);
1458         struct osd_thandle      *oh;
1459         uint64_t                 zapid;
1460         dnode_t                 *dn;
1461         int                      rc, dnode_size;
1462         ENTRY;
1463
1464         LASSERT(dof);
1465
1466         switch (dof->dof_type) {
1467                 case DFT_REGULAR:
1468                 case DFT_SYM:
1469                 case DFT_NODE:
1470                         if (obj->oo_dt.do_body_ops == NULL)
1471                                 obj->oo_dt.do_body_ops = &osd_body_ops;
1472                         break;
1473                 default:
1474                         break;
1475         }
1476
1477         LASSERT(handle != NULL);
1478         oh = container_of0(handle, struct osd_thandle, ot_super);
1479         LASSERT(oh->ot_tx != NULL);
1480
1481         /* this is the minimum set of EAs on every Lustre object */
1482         obj->oo_ea_in_bonus = OSD_BASE_EA_IN_BONUS;
1483         /* reserve 32 bytes for extra stuff like ACLs */
1484         dnode_size = size_roundup_power2(obj->oo_ea_in_bonus + 32);
1485
1486         switch (dof->dof_type) {
1487                 case DFT_DIR:
1488                         dt->do_index_ops = &osd_dir_ops;
1489                 case DFT_INDEX:
1490                         /* for zap create */
1491                         dmu_tx_hold_zap(oh->ot_tx, DMU_NEW_OBJECT, FALSE, NULL);
1492                         dmu_tx_hold_sa_create(oh->ot_tx, dnode_size);
1493                         break;
1494                 case DFT_REGULAR:
1495                 case DFT_SYM:
1496                 case DFT_NODE:
1497                         /* first, we'll create new object */
1498                         dmu_tx_hold_sa_create(oh->ot_tx, dnode_size);
1499                         break;
1500
1501                 default:
1502                         LBUG();
1503                         break;
1504         }
1505
1506         /* and we'll add it to some mapping */
1507         zapid = osd_get_name_n_idx(env, osd, fid, NULL, 0, &dn);
1508         osd_tx_hold_zap(oh->ot_tx, zapid, dn, TRUE, NULL);
1509
1510         /* will help to find FID->ino mapping at dt_insert() */
1511         osd_idc_find_and_init(env, osd, obj);
1512
1513         rc = osd_declare_quota(env, osd, attr->la_uid, attr->la_gid,
1514                                attr->la_projid, 1, oh, NULL, OSD_QID_INODE);
1515
1516         RETURN(rc);
1517 }
1518
1519 int __osd_attr_init(const struct lu_env *env, struct osd_device *osd,
1520                     struct osd_object *obj, sa_handle_t *sa_hdl, dmu_tx_t *tx,
1521                     struct lu_attr *la, uint64_t parent,
1522                     nvlist_t *xattr)
1523 {
1524         sa_bulk_attr_t *bulk = osd_oti_get(env)->oti_attr_bulk;
1525         struct osa_attr *osa = &osd_oti_get(env)->oti_osa;
1526         uint64_t gen;
1527         uint64_t crtime[2];
1528         inode_timespec_t now;
1529         int cnt;
1530         int rc;
1531         char *dxattr = NULL;
1532         size_t sa_size;
1533
1534
1535         LASSERT(sa_hdl);
1536
1537         gen = dmu_tx_get_txg(tx);
1538         gethrestime(&now);
1539         ZFS_TIME_ENCODE(&now, crtime);
1540
1541         osa->atime[0] = la->la_atime;
1542         osa->ctime[0] = la->la_ctime;
1543         osa->mtime[0] = la->la_mtime;
1544         osa->mode = la->la_mode;
1545         osa->uid = la->la_uid;
1546         osa->gid = la->la_gid;
1547         osa->rdev = la->la_rdev;
1548         osa->nlink = la->la_nlink;
1549         if (la->la_valid & LA_FLAGS)
1550                 osa->flags = attrs_fs2zfs(la->la_flags);
1551         else
1552                 osa->flags = 0;
1553         osa->size  = la->la_size;
1554 #ifdef ZFS_PROJINHERIT
1555         if (osd->od_projectused_dn) {
1556                 if (la->la_valid & LA_PROJID)
1557                         osa->projid = la->la_projid;
1558                 else
1559                         osa->projid = ZFS_DEFAULT_PROJID;
1560                 osa->flags |= ZFS_PROJID;
1561                 if (obj)
1562                         obj->oo_with_projid = 1;
1563         } else {
1564                 osa->flags &= ~ZFS_PROJID;
1565         }
1566 #endif
1567
1568         /*
1569          * we need to create all SA below upon object create.
1570          *
1571          * XXX The attribute order matters since the accounting callback relies
1572          * on static offsets (i.e. SA_*_OFFSET, see zfs_space_delta_cb()) to
1573          * look up the UID/GID/PROJID attributes. Moreover, the callback does
1574          * not seem to support the spill block.
1575          * We define attributes in the same order as SA_*_OFFSET in order to
1576          * work around the problem. See ORI-610.
1577          */
1578         cnt = 0;
1579         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(osd), NULL, &osa->mode, 8);
1580         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(osd), NULL, &osa->size, 8);
1581         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GEN(osd), NULL, &gen, 8);
1582         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(osd), NULL, &osa->uid, 8);
1583         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(osd), NULL, &osa->gid, 8);
1584         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_PARENT(osd), NULL, &parent, 8);
1585         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(osd), NULL, &osa->flags, 8);
1586         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(osd), NULL, osa->atime, 16);
1587         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(osd), NULL, osa->mtime, 16);
1588         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(osd), NULL, osa->ctime, 16);
1589         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CRTIME(osd), NULL, crtime, 16);
1590         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(osd), NULL, &osa->nlink, 8);
1591 #ifdef ZFS_PROJINHERIT
1592         if (osd->od_projectused_dn)
1593                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_PROJID(osd), NULL,
1594                                  &osa->projid, 8);
1595 #endif
1596         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_RDEV(osd), NULL, &osa->rdev, 8);
1597         LASSERT(cnt <= ARRAY_SIZE(osd_oti_get(env)->oti_attr_bulk));
1598
1599         if (xattr) {
1600                 rc = -nvlist_size(xattr, &sa_size, NV_ENCODE_XDR);
1601                 LASSERT(rc == 0);
1602
1603                 dxattr = osd_zio_buf_alloc(sa_size);
1604                 LASSERT(dxattr);
1605
1606                 rc = -nvlist_pack(xattr, &dxattr, &sa_size,
1607                                 NV_ENCODE_XDR, KM_SLEEP);
1608                 LASSERT(rc == 0);
1609
1610                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_DXATTR(osd),
1611                                 NULL, dxattr, sa_size);
1612         }
1613
1614         rc = -sa_replace_all_by_template(sa_hdl, bulk, cnt, tx);
1615         if (dxattr)
1616                 osd_zio_buf_free(dxattr, sa_size);
1617
1618         return rc;
1619 }
1620
1621 int osd_find_new_dnode(const struct lu_env *env, dmu_tx_t *tx,
1622                        uint64_t oid, dnode_t **dnp)
1623 {
1624         dmu_tx_hold_t *txh;
1625         int rc = 0;
1626
1627         /* take dnode_t from tx to save on dnode#->dnode_t lookup */
1628         for (txh = list_tail(&tx->tx_holds); txh;
1629              txh = list_prev(&tx->tx_holds, txh)) {
1630                 dnode_t *dn = txh->txh_dnode;
1631                 dmu_buf_impl_t *db;
1632
1633                 if (dn == NULL)
1634                         continue;
1635                 if (dn->dn_object != oid)
1636                         continue;
1637                 db = dn->dn_bonus;
1638                 if (db == NULL) {
1639                         rw_enter(&dn->dn_struct_rwlock, RW_WRITER);
1640                         if (dn->dn_bonus == NULL)
1641                                 dbuf_create_bonus(dn);
1642                         rw_exit(&dn->dn_struct_rwlock);
1643                 }
1644                 db = dn->dn_bonus;
1645                 LASSERT(db);
1646                 LASSERT(dn->dn_handle);
1647                 DB_DNODE_ENTER(db);
1648                 if (zfs_refcount_add(&db->db_holds, osd_obj_tag) == 1) {
1649                         zfs_refcount_add(&dn->dn_holds, osd_obj_tag);
1650                         atomic_inc_32(&dn->dn_dbufs_count);
1651                 }
1652                 *dnp = dn;
1653                 DB_DNODE_EXIT(db);
1654                 dbuf_read(db, NULL, DB_RF_MUST_SUCCEED | DB_RF_NOPREFETCH);
1655                 break;
1656         }
1657
1658         if (unlikely(*dnp == NULL))
1659                 rc = __osd_obj2dnode(tx->tx_objset, oid, dnp);
1660
1661         return rc;
1662 }
1663
1664 #ifdef HAVE_DMU_OBJECT_ALLOC_DNSIZE
1665 int osd_find_dnsize(struct osd_device *osd, int ea_in_bonus)
1666 {
1667         int dnsize;
1668
1669         if (osd->od_dnsize == ZFS_DNSIZE_AUTO) {
1670                 dnsize = DNODE_MIN_SIZE;
1671                 do {
1672                         if (DN_BONUS_SIZE(dnsize) >= ea_in_bonus + 32)
1673                                 break;
1674                         dnsize <<= 1;
1675                 } while (dnsize < DNODE_MAX_SIZE);
1676                 if (dnsize > DNODE_MAX_SIZE)
1677                         dnsize = DNODE_MAX_SIZE;
1678         } else if (osd->od_dnsize == ZFS_DNSIZE_1K) {
1679                 dnsize = 1024;
1680         } else if (osd->od_dnsize == ZFS_DNSIZE_2K) {
1681                 dnsize = 2048;
1682         } else if (osd->od_dnsize == ZFS_DNSIZE_4K) {
1683                 dnsize = 4096;
1684         } else if (osd->od_dnsize == ZFS_DNSIZE_8K) {
1685                 dnsize = 8192;
1686         } else if (osd->od_dnsize == ZFS_DNSIZE_16K) {
1687                 dnsize = 16384;
1688         } else {
1689                 dnsize = DNODE_MIN_SIZE;
1690         }
1691         return dnsize;
1692 }
1693 #endif
1694
1695 /*
1696  * The transaction passed to this routine must have
1697  * dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT) called and then assigned
1698  * to a transaction group.
1699  */
1700 int __osd_object_create(const struct lu_env *env, struct osd_device *osd,
1701                         struct osd_object *obj, const struct lu_fid *fid,
1702                         dnode_t **dnp, dmu_tx_t *tx, struct lu_attr *la)
1703 {
1704         dmu_object_type_t type = DMU_OT_PLAIN_FILE_CONTENTS;
1705         uint64_t oid;
1706         int size;
1707
1708         /* Use DMU_OTN_UINT8_METADATA for local objects so their data blocks
1709          * would get an additional ditto copy */
1710         if (unlikely(S_ISREG(la->la_mode) &&
1711                      fid_seq_is_local_file(fid_seq(fid))))
1712                 type = DMU_OTN_UINT8_METADATA;
1713
1714         /* Create a new DMU object using the default dnode size. */
1715         if (obj)
1716                 size = obj->oo_ea_in_bonus;
1717         else
1718                 size = OSD_BASE_EA_IN_BONUS;
1719         oid = osd_dmu_object_alloc(osd->od_os, type, 0,
1720                                    osd_find_dnsize(osd, size), tx);
1721
1722         LASSERT(la->la_valid & LA_MODE);
1723         la->la_size = 0;
1724         la->la_nlink = 1;
1725
1726         return osd_find_new_dnode(env, tx, oid, dnp);
1727 }
1728
1729 /*
1730  * The transaction passed to this routine must have
1731  * dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, ...) called and then assigned
1732  * to a transaction group.
1733  *
1734  * Using ZAP_FLAG_HASH64 will force the ZAP to always be a FAT ZAP.
1735  * This is fine for directories today, because storing the FID in the dirent
1736  * will also require a FAT ZAP.  If there is a new type of micro ZAP created
1737  * then we might need to re-evaluate the use of this flag and instead do
1738  * a conversion from the different internal ZAP hash formats being used. */
1739 int __osd_zap_create(const struct lu_env *env, struct osd_device *osd,
1740                      dnode_t **dnp, dmu_tx_t *tx, struct lu_attr *la,
1741                      unsigned dnsize, zap_flags_t flags)
1742 {
1743         uint64_t oid;
1744
1745         /* Assert that the transaction has been assigned to a
1746            transaction group. */
1747         LASSERT(tx->tx_txg != 0);
1748         *dnp = NULL;
1749
1750         oid = osd_zap_create_flags(osd->od_os, 0, flags | ZAP_FLAG_HASH64,
1751                                    DMU_OT_DIRECTORY_CONTENTS,
1752                                    14, /* == ZFS fzap_default_blockshift */
1753                                    DN_MAX_INDBLKSHIFT, /* indirect blockshift */
1754                                    dnsize, tx);
1755
1756         la->la_size = 2;
1757         la->la_nlink = 1;
1758
1759         return osd_find_new_dnode(env, tx, oid, dnp);
1760 }
1761
1762 static dnode_t *osd_mkidx(const struct lu_env *env, struct osd_object *obj,
1763                           struct lu_attr *la, struct osd_thandle *oh)
1764 {
1765         struct osd_device *osd = osd_obj2dev(obj);
1766         dnode_t *dn;
1767         int rc;
1768
1769         /* Index file should be created as regular file in order not to confuse
1770          * ZPL which could interpret them as directory.
1771          * We set ZAP_FLAG_UINT64_KEY to let ZFS know than we are going to use
1772          * binary keys */
1773         LASSERT(S_ISREG(la->la_mode));
1774         rc = __osd_zap_create(env, osd, &dn, oh->ot_tx, la,
1775                 osd_find_dnsize(osd, obj->oo_ea_in_bonus), ZAP_FLAG_UINT64_KEY);
1776         if (rc)
1777                 return ERR_PTR(rc);
1778         return dn;
1779 }
1780
1781 static dnode_t *osd_mkdir(const struct lu_env *env, struct osd_object *obj,
1782                           struct lu_attr *la, struct osd_thandle *oh)
1783 {
1784         struct osd_device *osd = osd_obj2dev(obj);
1785         dnode_t *dn;
1786         int rc;
1787
1788         LASSERT(S_ISDIR(la->la_mode));
1789         rc = __osd_zap_create(env, osd, &dn, oh->ot_tx, la,
1790                               osd_find_dnsize(osd, obj->oo_ea_in_bonus), 0);
1791         if (rc)
1792                 return ERR_PTR(rc);
1793         return dn;
1794 }
1795
1796 static dnode_t *osd_mkreg(const struct lu_env *env, struct osd_object *obj,
1797                           struct lu_attr *la, struct osd_thandle *oh)
1798 {
1799         const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
1800         struct osd_device *osd = osd_obj2dev(obj);
1801         dnode_t *dn;
1802         int rc;
1803
1804         LASSERT(S_ISREG(la->la_mode));
1805         rc = __osd_object_create(env, osd, obj, fid, &dn, oh->ot_tx, la);
1806         if (rc)
1807                 return ERR_PTR(rc);
1808
1809         if ((fid_is_idif(fid) || fid_is_norm(fid) || fid_is_echo(fid))) {
1810                 /* The minimum block size must be at least page size otherwise
1811                  * it will break the assumption in tgt_thread_big_cache where
1812                  * the array size is PTLRPC_MAX_BRW_PAGES. It will also affect
1813                  * RDMA due to subpage transfer size */
1814                 rc = -dmu_object_set_blocksize(osd->od_os, dn->dn_object,
1815                                                PAGE_SIZE, 0, oh->ot_tx);
1816                 if (unlikely(rc)) {
1817                         CERROR("%s: can't change blocksize: %d\n",
1818                                osd->od_svname, rc);
1819                         return ERR_PTR(rc);
1820                 }
1821         }
1822
1823         return dn;
1824 }
1825
1826 static dnode_t *osd_mksym(const struct lu_env *env, struct osd_object *obj,
1827                           struct lu_attr *la, struct osd_thandle *oh)
1828 {
1829         dnode_t *dn;
1830         int rc;
1831
1832         LASSERT(S_ISLNK(la->la_mode));
1833         rc = __osd_object_create(env, osd_obj2dev(obj), obj,
1834                                  lu_object_fid(&obj->oo_dt.do_lu),
1835                                  &dn, oh->ot_tx, la);
1836         if (rc)
1837                 return ERR_PTR(rc);
1838         return dn;
1839 }
1840
1841 static dnode_t *osd_mknod(const struct lu_env *env, struct osd_object *obj,
1842                           struct lu_attr *la, struct osd_thandle *oh)
1843 {
1844         dnode_t *dn;
1845         int rc;
1846
1847         if (S_ISCHR(la->la_mode) || S_ISBLK(la->la_mode))
1848                 la->la_valid |= LA_RDEV;
1849
1850         rc = __osd_object_create(env, osd_obj2dev(obj), obj,
1851                                  lu_object_fid(&obj->oo_dt.do_lu),
1852                                  &dn, oh->ot_tx, la);
1853         if (rc)
1854                 return ERR_PTR(rc);
1855         return dn;
1856 }
1857
1858 typedef dnode_t *(*osd_obj_type_f)(const struct lu_env *env,
1859                                    struct osd_object *obj,
1860                                    struct lu_attr *la,
1861                                    struct osd_thandle *oh);
1862
1863 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
1864 {
1865         osd_obj_type_f result;
1866
1867         switch (type) {
1868         case DFT_DIR:
1869                 result = osd_mkdir;
1870                 break;
1871         case DFT_INDEX:
1872                 result = osd_mkidx;
1873                 break;
1874         case DFT_REGULAR:
1875                 result = osd_mkreg;
1876                 break;
1877         case DFT_SYM:
1878                 result = osd_mksym;
1879                 break;
1880         case DFT_NODE:
1881                 result = osd_mknod;
1882                 break;
1883         default:
1884                 LBUG();
1885                 break;
1886         }
1887         return result;
1888 }
1889
1890 /*
1891  * Concurrency: @dt is write locked.
1892  */
1893 static int osd_create(const struct lu_env *env, struct dt_object *dt,
1894                       struct lu_attr *attr, struct dt_allocation_hint *hint,
1895                       struct dt_object_format *dof, struct thandle *th)
1896 {
1897         struct osd_thread_info  *info = osd_oti_get(env);
1898         struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
1899         struct zpl_direntry     *zde = &info->oti_zde.lzd_reg;
1900         const struct lu_fid     *fid = lu_object_fid(&dt->do_lu);
1901         struct osd_object       *obj = osd_dt_obj(dt);
1902         struct osd_device       *osd = osd_obj2dev(obj);
1903         char                    *buf = info->oti_str;
1904         struct osd_thandle      *oh;
1905         dnode_t *dn = NULL, *zdn = NULL;
1906         uint64_t                 zapid, parent = 0;
1907         int                      rc;
1908         __u32 compat = 0;
1909
1910         ENTRY;
1911
1912         LASSERT(!fid_is_acct(fid));
1913
1914         /* concurrent create declarations should not see
1915          * the object inconsistent (db, attr, etc).
1916          * in regular cases acquisition should be cheap */
1917         down_write(&obj->oo_guard);
1918
1919         if (unlikely(dt_object_exists(dt)))
1920                 GOTO(out, rc = -EEXIST);
1921
1922         LASSERT(osd_invariant(obj));
1923         LASSERT(dof != NULL);
1924
1925         LASSERT(th != NULL);
1926         oh = container_of0(th, struct osd_thandle, ot_super);
1927
1928         LASSERT(obj->oo_dn == NULL);
1929
1930         /* to follow ZFS on-disk format we need
1931          * to initialize parent dnode properly */
1932         if (hint != NULL && hint->dah_parent != NULL &&
1933             !dt_object_remote(hint->dah_parent))
1934                 parent = osd_dt_obj(hint->dah_parent)->oo_dn->dn_object;
1935
1936         /* we may fix some attributes, better do not change the source */
1937         obj->oo_attr = *attr;
1938         obj->oo_attr.la_valid |= LA_SIZE | LA_NLINK | LA_TYPE;
1939
1940 #ifdef ZFS_PROJINHERIT
1941         if (osd->od_projectused_dn) {
1942                 if (!(obj->oo_attr.la_valid & LA_PROJID))
1943                         obj->oo_attr.la_projid = ZFS_DEFAULT_PROJID;
1944                 obj->oo_with_projid = 1;
1945         }
1946 #endif
1947
1948         dn = osd_create_type_f(dof->dof_type)(env, obj, &obj->oo_attr, oh);
1949         if (IS_ERR(dn)) {
1950                 rc = PTR_ERR(dn);
1951                 dn = NULL;
1952                 GOTO(out, rc);
1953         }
1954
1955         zde->zde_pad = 0;
1956         zde->zde_dnode = dn->dn_object;
1957         zde->zde_type = IFTODT(attr->la_mode & S_IFMT);
1958
1959         zapid = osd_get_name_n_idx(env, osd, fid, buf,
1960                                    sizeof(info->oti_str), &zdn);
1961         if (CFS_FAIL_CHECK(OBD_FAIL_OSD_NO_OI_ENTRY) ||
1962             (osd->od_is_ost && OBD_FAIL_CHECK(OBD_FAIL_OSD_COMPAT_NO_ENTRY)))
1963                 goto skip_add;
1964
1965         if (osd->od_is_ost && OBD_FAIL_CHECK(OBD_FAIL_OSD_COMPAT_INVALID_ENTRY))
1966                 zde->zde_dnode++;
1967
1968         rc = osd_zap_add(osd, zapid, zdn, buf, 8, 1, zde, oh->ot_tx);
1969         if (rc)
1970                 GOTO(out, rc);
1971
1972 skip_add:
1973         obj->oo_dn = dn;
1974         /* Now add in all of the "SA" attributes */
1975         rc = osd_sa_handle_get(obj);
1976         if (rc)
1977                 GOTO(out, rc);
1978
1979         rc = -nvlist_alloc(&obj->oo_sa_xattr, NV_UNIQUE_NAME, KM_SLEEP);
1980         if (rc)
1981                 GOTO(out, rc);
1982
1983         /* initialize LMA */
1984         if (fid_is_idif(fid) || (fid_is_norm(fid) && osd->od_is_ost))
1985                 compat |= LMAC_FID_ON_OST;
1986         lustre_lma_init(lma, fid, compat, 0);
1987         lustre_lma_swab(lma);
1988         rc = -nvlist_add_byte_array(obj->oo_sa_xattr, XATTR_NAME_LMA,
1989                                     (uchar_t *)lma, sizeof(*lma));
1990         if (rc)
1991                 GOTO(out, rc);
1992
1993         /* configure new osd object */
1994         obj->oo_parent = parent != 0 ? parent : zapid;
1995         obj->oo_late_attr_set = 1;
1996         rc = __osd_sa_xattr_schedule_update(env, obj, oh);
1997         if (rc)
1998                 GOTO(out, rc);
1999
2000         /* XXX: oo_lma_flags */
2001         obj->oo_dt.do_lu.lo_header->loh_attr |= obj->oo_attr.la_mode & S_IFMT;
2002         if (likely(!fid_is_acct(lu_object_fid(&obj->oo_dt.do_lu))))
2003                 /* no body operations for accounting objects */
2004                 obj->oo_dt.do_body_ops = &osd_body_ops;
2005
2006         osd_idc_find_and_init(env, osd, obj);
2007
2008 out:
2009         if (unlikely(rc && dn)) {
2010                 dmu_object_free(osd->od_os, dn->dn_object, oh->ot_tx);
2011                 osd_dnode_rele(dn);
2012                 obj->oo_dn = NULL;
2013         } else if (!rc) {
2014                 obj->oo_dt.do_lu.lo_header->loh_attr |= LOHA_EXISTS;
2015         }
2016         up_write(&obj->oo_guard);
2017         RETURN(rc);
2018 }
2019
2020 static int osd_declare_ref_add(const struct lu_env *env, struct dt_object *dt,
2021                                struct thandle *th)
2022 {
2023         osd_idc_find_and_init(env, osd_dev(dt->do_lu.lo_dev), osd_dt_obj(dt));
2024         return osd_declare_attr_set(env, dt, NULL, th);
2025 }
2026
2027 /*
2028  * Concurrency: @dt is write locked.
2029  */
2030 static int osd_ref_add(const struct lu_env *env, struct dt_object *dt,
2031                        struct thandle *handle)
2032 {
2033         struct osd_object       *obj = osd_dt_obj(dt);
2034         struct osd_thandle      *oh;
2035         struct osd_device       *osd = osd_obj2dev(obj);
2036         uint64_t                 nlink;
2037         int rc;
2038
2039         ENTRY;
2040
2041         down_read(&obj->oo_guard);
2042         if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
2043                 GOTO(out, rc = -ENOENT);
2044
2045         LASSERT(osd_invariant(obj));
2046         LASSERT(obj->oo_sa_hdl != NULL);
2047
2048         oh = container_of0(handle, struct osd_thandle, ot_super);
2049
2050         write_lock(&obj->oo_attr_lock);
2051         nlink = ++obj->oo_attr.la_nlink;
2052         write_unlock(&obj->oo_attr_lock);
2053
2054         rc = osd_object_sa_update(obj, SA_ZPL_LINKS(osd), &nlink, 8, oh);
2055
2056 out:
2057         up_read(&obj->oo_guard);
2058         RETURN(rc);
2059 }
2060
2061 static int osd_declare_ref_del(const struct lu_env *env, struct dt_object *dt,
2062                                struct thandle *handle)
2063 {
2064         osd_idc_find_and_init(env, osd_dev(dt->do_lu.lo_dev), osd_dt_obj(dt));
2065         return osd_declare_attr_set(env, dt, NULL, handle);
2066 }
2067
2068 /*
2069  * Concurrency: @dt is write locked.
2070  */
2071 static int osd_ref_del(const struct lu_env *env, struct dt_object *dt,
2072                        struct thandle *handle)
2073 {
2074         struct osd_object       *obj = osd_dt_obj(dt);
2075         struct osd_thandle      *oh;
2076         struct osd_device       *osd = osd_obj2dev(obj);
2077         uint64_t                 nlink;
2078         int                      rc;
2079
2080         ENTRY;
2081
2082         down_read(&obj->oo_guard);
2083
2084         if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
2085                 GOTO(out, rc = -ENOENT);
2086
2087         LASSERT(osd_invariant(obj));
2088         LASSERT(obj->oo_sa_hdl != NULL);
2089
2090         oh = container_of0(handle, struct osd_thandle, ot_super);
2091         LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
2092
2093         write_lock(&obj->oo_attr_lock);
2094         nlink = --obj->oo_attr.la_nlink;
2095         write_unlock(&obj->oo_attr_lock);
2096
2097         rc = osd_object_sa_update(obj, SA_ZPL_LINKS(osd), &nlink, 8, oh);
2098
2099 out:
2100         up_read(&obj->oo_guard);
2101         RETURN(rc);
2102 }
2103
2104 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
2105                            __u64 start, __u64 end)
2106 {
2107         struct osd_device *osd = osd_obj2dev(osd_dt_obj(dt));
2108         struct dmu_buf_impl *db = osd_dt_obj(dt)->oo_dn->dn_dbuf;
2109         uint64_t txg = 0;
2110         ENTRY;
2111
2112         if (osd->od_dt_dev.dd_rdonly)
2113                 RETURN(0);
2114
2115         mutex_enter(&db->db_mtx);
2116         if (db->db_last_dirty)
2117                 txg = db->db_last_dirty->dr_txg;
2118         mutex_exit(&db->db_mtx);
2119
2120         if (txg) {
2121                 /* the object is dirty or being synced */
2122                 if (osd_object_sync_delay_us < 0)
2123                         txg_wait_synced(dmu_objset_pool(osd->od_os), txg);
2124                 else
2125                         udelay(osd_object_sync_delay_us);
2126         }
2127
2128         RETURN(0);
2129 }
2130
2131 static int osd_invalidate(const struct lu_env *env, struct dt_object *dt)
2132 {
2133         return 0;
2134 }
2135
2136 static struct dt_object_operations osd_obj_ops = {
2137         .do_read_lock           = osd_read_lock,
2138         .do_write_lock          = osd_write_lock,
2139         .do_read_unlock         = osd_read_unlock,
2140         .do_write_unlock        = osd_write_unlock,
2141         .do_write_locked        = osd_write_locked,
2142         .do_attr_get            = osd_attr_get,
2143         .do_declare_attr_set    = osd_declare_attr_set,
2144         .do_attr_set            = osd_attr_set,
2145         .do_ah_init             = osd_ah_init,
2146         .do_declare_create      = osd_declare_create,
2147         .do_create              = osd_create,
2148         .do_declare_destroy     = osd_declare_destroy,
2149         .do_destroy             = osd_destroy,
2150         .do_index_try           = osd_index_try,
2151         .do_declare_ref_add     = osd_declare_ref_add,
2152         .do_ref_add             = osd_ref_add,
2153         .do_declare_ref_del     = osd_declare_ref_del,
2154         .do_ref_del             = osd_ref_del,
2155         .do_xattr_get           = osd_xattr_get,
2156         .do_declare_xattr_set   = osd_declare_xattr_set,
2157         .do_xattr_set           = osd_xattr_set,
2158         .do_declare_xattr_del   = osd_declare_xattr_del,
2159         .do_xattr_del           = osd_xattr_del,
2160         .do_xattr_list          = osd_xattr_list,
2161         .do_object_sync         = osd_object_sync,
2162         .do_invalidate          = osd_invalidate,
2163 };
2164
2165 static struct lu_object_operations osd_lu_obj_ops = {
2166         .loo_object_init        = osd_object_init,
2167         .loo_object_delete      = osd_object_delete,
2168         .loo_object_release     = osd_object_release,
2169         .loo_object_free        = osd_object_free,
2170         .loo_object_print       = osd_object_print,
2171         .loo_object_invariant   = osd_object_invariant,
2172 };
2173
2174 static int osd_otable_it_attr_get(const struct lu_env *env,
2175                                 struct dt_object *dt,
2176                                 struct lu_attr *attr)
2177 {
2178         attr->la_valid = 0;
2179         return 0;
2180 }
2181
2182 static struct dt_object_operations osd_obj_otable_it_ops = {
2183         .do_attr_get            = osd_otable_it_attr_get,
2184         .do_index_try           = osd_index_try,
2185 };
2186
2187 module_param(osd_object_sync_delay_us, int, 0644);
2188 MODULE_PARM_DESC(osd_object_sync_delay_us,
2189                  "If zero or larger delay N usec instead of doing object sync");