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 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         LASSERT(*dnp != NULL);
 295
 296         return 0;
 297 }
 298
 299 /*
 300  * Concurrency: no concurrent access is possible that early in object
 301  * life-cycle.
 302  */
 303 struct lu_object *osd_object_alloc(const struct lu_env *env,
 304                                    const struct lu_object_header *hdr,
 305                                    struct lu_device *d)
 306 {
 307         struct osd_object *mo;
 308
 309         OBD_SLAB_ALLOC_PTR_GFP(mo, osd_object_kmem, GFP_NOFS);
 310         if (mo != NULL) {
 311                 struct lu_object *l;
 312                 struct lu_object_header *h;
 313                 struct osd_device *o = osd_dev(d);
 314
 315                 l = &mo->oo_dt.do_lu;
 316                 if (unlikely(o->od_in_init)) {
 317                         OBD_ALLOC_PTR(h);
 318                         if (!h) {
 319                                 OBD_FREE_PTR(mo);
 320                                 return NULL;
 321                         }
 322
 323                         lu_object_header_init(h);
 324                         lu_object_init(l, h, d);
 325                         lu_object_add_top(h, l);
 326                         mo->oo_header = h;
 327                 } else {
 328                         dt_object_init(&mo->oo_dt, NULL, d);
 329                         mo->oo_header = NULL;
 330                 }
 331
 332                 mo->oo_dt.do_ops = &osd_obj_ops;
 333                 l->lo_ops = &osd_lu_obj_ops;
 334                 INIT_LIST_HEAD(&mo->oo_sa_linkage);
 335                 INIT_LIST_HEAD(&mo->oo_unlinked_linkage);
 336                 init_rwsem(&mo->oo_sem);
 337                 init_rwsem(&mo->oo_guard);
 338                 rwlock_init(&mo->oo_attr_lock);
 339                 mo->oo_destroy = OSD_DESTROY_NONE;
 340                 return l;
 341         } else {
 342                 return NULL;
 343         }
 344 }
 345
 346 static void osd_obj_set_blksize(const struct lu_env *env,
 347                                 struct osd_device *osd, struct osd_object *obj)
 348 {
 349         const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
 350         dmu_tx_t *tx;
 351         dnode_t *dn = obj->oo_dn;
 352         uint32_t blksz;
 353         int rc = 0;
 354         ENTRY;
 355
 356         LASSERT(!osd_oti_get(env)->oti_in_trans);
 357
 358         tx = dmu_tx_create(osd->od_os);
 359         if (!tx) {
 360                 CERROR("%s: fail to create tx to set blksize for "DFID"\n",
 361                        osd->od_svname, PFID(fid));
 362                 RETURN_EXIT;
 363         }
 364
 365         dmu_tx_hold_bonus(tx, dn->dn_object);
 366         rc = -dmu_tx_assign(tx, TXG_WAIT);
 367         if (rc) {
 368                 dmu_tx_abort(tx);
 369                 CERROR("%s: fail to assign tx to set blksize for "DFID
 370                        ": rc = %d\n", osd->od_svname, PFID(fid), rc);
 371                 RETURN_EXIT;
 372         }
 373
 374         down_write(&obj->oo_guard);
 375         if (unlikely((1 << dn->dn_datablkshift) >= PAGE_SIZE))
 376                 GOTO(out, rc = 1);
 377
 378         blksz = dn->dn_datablksz;
 379         if (!is_power_of_2(blksz))
 380                 blksz = size_roundup_power2(blksz);
 381
 382         if (blksz > osd->od_max_blksz)
 383                 blksz = osd->od_max_blksz;
 384         else if (blksz < PAGE_SIZE)
 385                 blksz = PAGE_SIZE;
 386         rc = -dmu_object_set_blocksize(osd->od_os, dn->dn_object, blksz, 0, tx);
 387
 388         GOTO(out, rc);
 389
 390 out:
 391         up_write(&obj->oo_guard);
 392         if (rc) {
 393                 dmu_tx_abort(tx);
 394                 if (unlikely(obj->oo_dn->dn_maxblkid > 0))
 395                         rc = 1;
 396                 if (rc < 0)
 397                         CERROR("%s: fail to set blksize for "DFID": rc = %d\n",
 398                                osd->od_svname, PFID(fid), rc);
 399         } else {
 400                 dmu_tx_commit(tx);
 401                 CDEBUG(D_INODE, "%s: set blksize as %u for "DFID"\n",
 402                        osd->od_svname, blksz, PFID(fid));
 403         }
 404 }
 405
 406 /*
 407  * Concurrency: shouldn't matter.
 408  */
 409 int osd_object_init0(const struct lu_env *env, struct osd_object *obj)
 410 {
 411         struct osd_device       *osd = osd_obj2dev(obj);
 412         const struct lu_fid     *fid = lu_object_fid(&obj->oo_dt.do_lu);
 413         int                      rc = 0;
 414         ENTRY;
 415
 416         LASSERT(obj->oo_dn);
 417
 418         rc = osd_object_sa_init(obj, osd);
 419         if (rc)
 420                 RETURN(rc);
 421
 422         /* cache attrs in object */
 423         rc = __osd_object_attr_get(env, osd, obj, &obj->oo_attr);
 424         if (rc)
 425                 RETURN(rc);
 426
 427         if (likely(!fid_is_acct(fid))) {
 428                 /* no body operations for accounting objects */
 429                 obj->oo_dt.do_body_ops = &osd_body_ops;
 430
 431                 if (S_ISREG(obj->oo_attr.la_mode) &&
 432                     obj->oo_dn->dn_maxblkid == 0 &&
 433                     (1 << obj->oo_dn->dn_datablkshift) < PAGE_SIZE &&
 434                     (fid_is_idif(fid) || fid_is_norm(fid) ||
 435                      fid_is_echo(fid)) &&
 436                     osd->od_is_ost && !osd->od_dt_dev.dd_rdonly)
 437                         osd_obj_set_blksize(env, osd, obj);
 438         }
 439
 440         /*
 441          * initialize object before marking it existing
 442          */
 443         obj->oo_dt.do_lu.lo_header->loh_attr |= obj->oo_attr.la_mode & S_IFMT;
 444
 445         smp_mb();
 446         obj->oo_dt.do_lu.lo_header->loh_attr |= LOHA_EXISTS;
 447
 448         RETURN(0);
 449 }
 450
 451 static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
 452 {
 453         struct osd_thread_info  *info = osd_oti_get(env);
 454         struct lu_buf           buf;
 455         int                     rc;
 456         struct lustre_mdt_attrs *lma;
 457         const struct lu_fid *rfid = lu_object_fid(&obj->oo_dt.do_lu);
 458         ENTRY;
 459
 460         CLASSERT(sizeof(info->oti_buf) >= sizeof(*lma));
 461         lma = (struct lustre_mdt_attrs *)info->oti_buf;
 462         buf.lb_buf = lma;
 463         buf.lb_len = sizeof(info->oti_buf);
 464
 465         rc = osd_xattr_get(env, &obj->oo_dt, &buf, XATTR_NAME_LMA);
 466         if (rc > 0) {
 467                 rc = 0;
 468                 lustre_lma_swab(lma);
 469                 if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
 470                              CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT))) {
 471                         CWARN("%s: unsupported incompat LMA feature(s) %#x for "
 472                               "fid = "DFID"\n", osd_obj2dev(obj)->od_svname,
 473                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
 474                               PFID(rfid));
 475                         rc = -EOPNOTSUPP;
 476                 } else if (unlikely(!lu_fid_eq(rfid, &lma->lma_self_fid))) {
 477                         CERROR("%s: FID-in-LMA "DFID" does not match the "
 478                               "object self-fid "DFID"\n",
 479                               osd_obj2dev(obj)->od_svname,
 480                               PFID(&lma->lma_self_fid), PFID(rfid));
 481                         rc = -EREMCHG;
 482                 } else {
 483                         struct osd_device *osd = osd_obj2dev(obj);
 484
 485                         if (lma->lma_compat & LMAC_STRIPE_INFO &&
 486                             osd->od_is_ost)
 487                                 obj->oo_pfid_in_lma = 1;
 488                         if (unlikely(lma->lma_incompat & LMAI_REMOTE_PARENT) &&
 489                             osd->od_remote_parent_dir != ZFS_NO_OBJECT)
 490                                 lu_object_set_agent_entry(&obj->oo_dt.do_lu);
 491                 }
 492         } else if (rc == -ENODATA) {
 493                 /* haven't initialize LMA xattr */
 494                 rc = 0;
 495         }
 496
 497         RETURN(rc);
 498 }
 499
 500 /**
 501  * Helper function to retrieve DMU object id from fid for accounting object
 502  */
 503 static dnode_t *osd_quota_fid2dmu(const struct osd_device *osd,
 504                                   const struct lu_fid *fid)
 505 {
 506         dnode_t *dn = NULL;
 507
 508         LASSERT(fid_is_acct(fid));
 509
 510         switch (fid_oid(fid)) {
 511         case ACCT_USER_OID:
 512                 dn = osd->od_userused_dn;
 513                 break;
 514         case ACCT_GROUP_OID:
 515                 dn = osd->od_groupused_dn;
 516                 break;
 517 #ifdef ZFS_PROJINHERIT
 518         case ACCT_PROJECT_OID:
 519                 dn = osd->od_projectused_dn;
 520                 break;
 521 #endif
 522         default:
 523                 break;
 524         }
 525
 526         return dn;
 527 }
 528
 529 /*
 530  * Concurrency: no concurrent access is possible that early in object
 531  * life-cycle.
 532  */
 533 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
 534                            const struct lu_object_conf *conf)
 535 {
 536         struct osd_object *obj = osd_obj(l);
 537         struct osd_device *osd = osd_obj2dev(obj);
 538         const struct lu_fid *fid = lu_object_fid(l);
 539         struct lustre_scrub *scrub = &osd->od_scrub;
 540         struct osd_thread_info *info = osd_oti_get(env);
 541         struct luz_direntry *zde = &info->oti_zde;
 542         struct osd_idmap_cache *idc;
 543         char *name = info->oti_str;
 544         uint64_t oid;
 545         int rc = 0;
 546         int rc1;
 547         bool remote = false;
 548         ENTRY;
 549
 550         LASSERT(osd_invariant(obj));
 551
 552         if (fid_is_otable_it(&l->lo_header->loh_fid)) {
 553                 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
 554                 l->lo_header->loh_attr |= LOHA_EXISTS;
 555
 556                 GOTO(out, rc = 0);
 557         }
 558
 559         if (conf && conf->loc_flags & LOC_F_NEW)
 560                 GOTO(out, rc = 0);
 561
 562         if (unlikely(fid_is_acct(fid))) {
 563                 obj->oo_dn = osd_quota_fid2dmu(osd, fid);
 564                 if (obj->oo_dn) {
 565                         obj->oo_dt.do_index_ops = &osd_acct_index_ops;
 566                         l->lo_header->loh_attr |= LOHA_EXISTS;
 567                 }
 568
 569                 GOTO(out, rc = 0);
 570         }
 571
 572         idc = osd_idc_find(env, osd, fid);
 573         if (idc && !idc->oic_remote && idc->oic_dnode != ZFS_NO_OBJECT) {
 574                 oid = idc->oic_dnode;
 575                 goto zget;
 576         }
 577
 578         rc = -ENOENT;
 579         if (!list_empty(&osd->od_scrub.os_inconsistent_items))
 580                 rc = osd_oii_lookup(osd, fid, &oid);
 581
 582         if (rc)
 583                 rc = osd_fid_lookup(env, osd, fid, &oid);
 584
 585         if (rc == -ENOENT) {
 586                 if (likely(!(fid_is_norm(fid) || fid_is_igif(fid)) ||
 587                            fid_is_on_ost(env, osd, fid) ||
 588                            !zfs_test_bit(osd_oi_fid2idx(osd, fid),
 589                                          scrub->os_file.sf_oi_bitmap)))
 590                         GOTO(out, rc = 0);
 591
 592                 rc = -EREMCHG;
 593                 goto trigger;
 594         }
 595
 596         if (rc)
 597                 GOTO(out, rc);
 598
 599 zget:
 600         LASSERT(obj->oo_dn == NULL);
 601
 602         rc = __osd_obj2dnode(osd->od_os, oid, &obj->oo_dn);
 603         /* EEXIST will be returned if object is being deleted in ZFS */
 604         if (rc == -EEXIST)
 605                 GOTO(out, rc = 0);
 606
 607         if (rc) {
 608                 CERROR("%s: lookup "DFID"/%#llx failed: rc = %d\n",
 609                        osd->od_svname, PFID(lu_object_fid(l)), oid, rc);
 610                 GOTO(out, rc);
 611         }
 612
 613         rc = osd_object_init0(env, obj);
 614         if (rc)
 615                 GOTO(out, rc);
 616
 617         if (unlikely(obj->oo_header))
 618                 GOTO(out, rc = 0);
 619
 620         rc = osd_check_lma(env, obj);
 621         if ((!rc && !remote) || (rc != -EREMCHG))
 622                 GOTO(out, rc);
 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 (thread_is_running(&scrub->os_thread)) {
 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         LCONSOLE_WARN("%s: trigger OI scrub by RPC for the "DFID": rc = %d\n",
 667                       osd_name(osd), PFID(fid), rc1);
 668         if (!rc) {
 669                 LASSERT(remote);
 670
 671                 lu_object_set_agent_entry(l);
 672                 if (!rc1)
 673                         osd_oii_insert(env, osd, fid, oid, false);
 674         } else {
 675                 if (!rc1)
 676                         rc = -EINPROGRESS;
 677                 else
 678                         rc = -EREMCHG;
 679         }
 680
 681         GOTO(out, rc);
 682
 683 out:
 684         RETURN(rc);
 685 }
 686
 687 /*
 688  * Concurrency: no concurrent access is possible that late in object
 689  * life-cycle.
 690  */
 691 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
 692 {
 693         struct osd_object *obj = osd_obj(l);
 694         struct lu_object_header *h = obj->oo_header;
 695
 696         LASSERT(osd_invariant(obj));
 697
 698         dt_object_fini(&obj->oo_dt);
 699         OBD_SLAB_FREE_PTR(obj, osd_object_kmem);
 700         if (unlikely(h)) {
 701                 lu_object_header_fini(h);
 702                 OBD_FREE_PTR(h);
 703         }
 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_of0(th, struct osd_thandle, ot_super);
 763         LASSERT(oh->ot_tx != NULL);
 764
 765         /* declare that we'll remove object from fid-dnode mapping */
 766         zapid = osd_get_name_n_idx(env, osd, fid, NULL, 0, &dn);
 767         osd_tx_hold_zap(oh->ot_tx, zapid, dn, FALSE, NULL);
 768
 769         osd_declare_xattrs_destroy(env, obj, oh);
 770
 771         /* one less inode */
 772         rc = osd_declare_quota(env, osd, obj->oo_attr.la_uid,
 773                                obj->oo_attr.la_gid, obj->oo_attr.la_projid,
 774                                -1, oh, NULL, OSD_QID_INODE);
 775         if (rc)
 776                 RETURN(rc);
 777
 778         /* data to be truncated */
 779         rc = osd_declare_quota(env, osd, obj->oo_attr.la_uid,
 780                                obj->oo_attr.la_gid, obj->oo_attr.la_projid,
 781                                0, oh, NULL, OSD_QID_BLK);
 782         if (rc)
 783                 RETURN(rc);
 784
 785         osd_object_set_destroy_type(obj);
 786         if (obj->oo_destroy == OSD_DESTROY_SYNC)
 787                 dmu_tx_hold_free(oh->ot_tx, obj->oo_dn->dn_object,
 788                                  0, DMU_OBJECT_END);
 789         else
 790                 osd_tx_hold_zap(oh->ot_tx, osd->od_unlinked->dn_object,
 791                                 osd->od_unlinked, TRUE, NULL);
 792
 793         /* remove agent entry (if have) from remote parent */
 794         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu))
 795                 osd_tx_hold_zap(oh->ot_tx, osd->od_remote_parent_dir,
 796                                 NULL, FALSE, NULL);
 797
 798         /* will help to find FID->ino when this object is being
 799          * added to PENDING/ */
 800         osd_idc_find_and_init(env, osd, obj);
 801
 802         RETURN(0);
 803 }
 804
 805 static int osd_destroy(const struct lu_env *env, struct dt_object *dt,
 806                        struct thandle *th)
 807 {
 808         struct osd_thread_info  *info = osd_oti_get(env);
 809         char                    *buf = info->oti_str;
 810         struct osd_object       *obj = osd_dt_obj(dt);
 811         struct osd_device       *osd = osd_obj2dev(obj);
 812         const struct lu_fid     *fid = lu_object_fid(&dt->do_lu);
 813         struct osd_thandle      *oh;
 814         int                      rc;
 815         uint64_t                 oid, zapid;
 816         dnode_t *zdn;
 817         ENTRY;
 818
 819         down_write(&obj->oo_guard);
 820
 821         if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
 822                 GOTO(out, rc = -ENOENT);
 823
 824         LASSERT(obj->oo_dn != NULL);
 825
 826         oh = container_of0(th, struct osd_thandle, ot_super);
 827         LASSERT(oh != NULL);
 828         LASSERT(oh->ot_tx != NULL);
 829
 830         /* remove obj ref from index dir (it depends) */
 831         zapid = osd_get_name_n_idx(env, osd, fid, buf,
 832                                    sizeof(info->oti_str), &zdn);
 833         rc = osd_xattrs_destroy(env, obj, oh);
 834         if (rc) {
 835                 CERROR("%s: cannot destroy xattrs for %s: rc = %d\n",
 836                        osd->od_svname, buf, rc);
 837                 GOTO(out, rc);
 838         }
 839
 840         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu)) {
 841                 rc = osd_delete_from_remote_parent(env, osd, obj, oh, true);
 842                 if (rc)
 843                         GOTO(out, rc);
 844         }
 845
 846         oid = obj->oo_dn->dn_object;
 847         if (unlikely(obj->oo_destroy == OSD_DESTROY_NONE)) {
 848                 /* this may happen if the destroy wasn't declared
 849                  * e.g. when the object is created and then destroyed
 850                  * in the same transaction - we don't need additional
 851                  * space for destroy specifically */
 852                 LASSERT(obj->oo_attr.la_size <= osd_sync_destroy_max_size);
 853                 rc = -dmu_object_free(osd->od_os, oid, oh->ot_tx);
 854                 if (rc)
 855                         CERROR("%s: failed to free %s %llu: rc = %d\n",
 856                                osd->od_svname, buf, oid, rc);
 857         } else if (obj->oo_destroy == OSD_DESTROY_SYNC) {
 858                 rc = -dmu_object_free(osd->od_os, oid, oh->ot_tx);
 859                 if (rc)
 860                         CERROR("%s: failed to free %s %llu: rc = %d\n",
 861                                osd->od_svname, buf, oid, rc);
 862         } else { /* asynchronous destroy */
 863                 char *key = info->oti_key;
 864
 865                 rc = osd_object_unlinked_add(obj, oh);
 866                 if (rc)
 867                         GOTO(out, rc);
 868
 869                 snprintf(key, sizeof(info->oti_key), "%llx", oid);
 870                 rc = osd_zap_add(osd, osd->od_unlinked->dn_object,
 871                                  osd->od_unlinked, key, 8, 1, &oid, oh->ot_tx);
 872                 if (rc)
 873                         CERROR("%s: zap_add_int() failed %s %llu: rc = %d\n",
 874                                osd->od_svname, buf, oid, rc);
 875         }
 876
 877         /* Remove the OI mapping after the destroy to handle the race with
 878          * OI scrub that may insert missed OI mapping during the interval. */
 879         rc = osd_zap_remove(osd, zapid, zdn, buf, oh->ot_tx);
 880         if (unlikely(rc == -ENOENT))
 881                 rc = 0;
 882         if (rc)
 883                 CERROR("%s: zap_remove(%s) failed: rc = %d\n",
 884                        osd->od_svname, buf, rc);
 885
 886         GOTO(out, rc);
 887
 888 out:
 889         /* not needed in the cache anymore */
 890         set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
 891         if (rc == 0)
 892                 obj->oo_destroyed = 1;
 893         up_write(&obj->oo_guard);
 894         RETURN (0);
 895 }
 896
 897 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
 898 {
 899         struct osd_object *obj = osd_obj(l);
 900         const struct lu_fid *fid = lu_object_fid(l);
 901
 902         if (obj->oo_dn) {
 903                 if (likely(!fid_is_acct(fid))) {
 904                         osd_object_sa_fini(obj);
 905                         if (obj->oo_sa_xattr) {
 906                                 nvlist_free(obj->oo_sa_xattr);
 907                                 obj->oo_sa_xattr = NULL;
 908                         }
 909                         osd_dnode_rele(obj->oo_dn);
 910                         list_del(&obj->oo_sa_linkage);
 911                 }
 912                 obj->oo_dn = NULL;
 913         }
 914 }
 915
 916 /*
 917  * Concurrency: ->loo_object_release() is called under site spin-lock.
 918  */
 919 static void osd_object_release(const struct lu_env *env,
 920                                struct lu_object *l)
 921 {
 922 }
 923
 924 /*
 925  * Concurrency: shouldn't matter.
 926  */
 927 static int osd_object_print(const struct lu_env *env, void *cookie,
 928                             lu_printer_t p, const struct lu_object *l)
 929 {
 930         struct osd_object *o = osd_obj(l);
 931
 932         return (*p)(env, cookie, LUSTRE_OSD_ZFS_NAME"-object@%p", o);
 933 }
 934
 935 static void osd_read_lock(const struct lu_env *env, struct dt_object *dt,
 936                           unsigned role)
 937 {
 938         struct osd_object *obj = osd_dt_obj(dt);
 939
 940         LASSERT(osd_invariant(obj));
 941
 942         down_read_nested(&obj->oo_sem, role);
 943 }
 944
 945 static void osd_write_lock(const struct lu_env *env, struct dt_object *dt,
 946                            unsigned role)
 947 {
 948         struct osd_object *obj = osd_dt_obj(dt);
 949
 950         LASSERT(osd_invariant(obj));
 951
 952         down_write_nested(&obj->oo_sem, role);
 953 }
 954
 955 static void osd_read_unlock(const struct lu_env *env, struct dt_object *dt)
 956 {
 957         struct osd_object *obj = osd_dt_obj(dt);
 958
 959         LASSERT(osd_invariant(obj));
 960         up_read(&obj->oo_sem);
 961 }
 962
 963 static void osd_write_unlock(const struct lu_env *env, struct dt_object *dt)
 964 {
 965         struct osd_object *obj = osd_dt_obj(dt);
 966
 967         LASSERT(osd_invariant(obj));
 968         up_write(&obj->oo_sem);
 969 }
 970
 971 static int osd_write_locked(const struct lu_env *env, struct dt_object *dt)
 972 {
 973         struct osd_object *obj = osd_dt_obj(dt);
 974         int rc = 1;
 975
 976         LASSERT(osd_invariant(obj));
 977
 978         if (down_write_trylock(&obj->oo_sem)) {
 979                 rc = 0;
 980                 up_write(&obj->oo_sem);
 981         }
 982         return rc;
 983 }
 984
 985 static int osd_attr_get(const struct lu_env *env,
 986                         struct dt_object *dt,
 987                         struct lu_attr *attr)
 988 {
 989         struct osd_object       *obj = osd_dt_obj(dt);
 990         uint64_t                 blocks;
 991         uint32_t                 blksize;
 992         int                      rc = 0;
 993
 994         down_read(&obj->oo_guard);
 995
 996         if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
 997                 GOTO(out, rc = -ENOENT);
 998
 999         if (unlikely(fid_is_acct(lu_object_fid(&dt->do_lu))))
1000                 GOTO(out, rc = 0);
1001
1002         LASSERT(osd_invariant(obj));
1003         LASSERT(obj->oo_dn);
1004
1005         read_lock(&obj->oo_attr_lock);
1006         *attr = obj->oo_attr;
1007         if (obj->oo_lma_flags & LUSTRE_ORPHAN_FL)
1008                 attr->la_flags |= LUSTRE_ORPHAN_FL;
1009         read_unlock(&obj->oo_attr_lock);
1010
1011         /* with ZFS_DEBUG zrl_add_debug() called by DB_DNODE_ENTER()
1012          * from within sa_object_size() can block on a mutex, so
1013          * we can't call sa_object_size() holding rwlock */
1014         sa_object_size(obj->oo_sa_hdl, &blksize, &blocks);
1015         /* we do not control size of indices, so always calculate
1016          * it from number of blocks reported by DMU */
1017         if (S_ISDIR(attr->la_mode))
1018                 attr->la_size = 512 * blocks;
1019         /* Block size may be not set; suggest maximal I/O transfers. */
1020         if (blksize == 0)
1021                 blksize = osd_spa_maxblocksize(
1022                         dmu_objset_spa(osd_obj2dev(obj)->od_os));
1023
1024         attr->la_blksize = blksize;
1025         attr->la_blocks = blocks;
1026         attr->la_valid |= LA_BLOCKS | LA_BLKSIZE;
1027
1028 out:
1029         up_read(&obj->oo_guard);
1030         return rc;
1031 }
1032
1033 /* Simple wrapper on top of qsd API which implement quota transfer for osd
1034  * setattr needs. As a reminder, only the root user can change ownership of
1035  * a file, that's why EDQUOT & EINPROGRESS errors are discarded */
1036 static inline int qsd_transfer(const struct lu_env *env,
1037                                struct qsd_instance *qsd,
1038                                struct lquota_trans *trans, int qtype,
1039                                __u64 orig_id, __u64 new_id, __u64 bspace,
1040                                struct lquota_id_info *qi)
1041 {
1042         int     rc;
1043
1044         if (unlikely(qsd == NULL))
1045                 return 0;
1046
1047         LASSERT(qtype >= 0 && qtype < LL_MAXQUOTAS);
1048         qi->lqi_type = qtype;
1049
1050         /* inode accounting */
1051         qi->lqi_is_blk = false;
1052
1053         /* one more inode for the new owner ... */
1054         qi->lqi_id.qid_uid = new_id;
1055         qi->lqi_space      = 1;
1056         rc = qsd_op_begin(env, qsd, trans, qi, NULL);
1057         if (rc == -EDQUOT || rc == -EINPROGRESS)
1058                 rc = 0;
1059         if (rc)
1060                 return rc;
1061
1062         /* and one less inode for the current id */
1063         qi->lqi_id.qid_uid = orig_id;;
1064         qi->lqi_space      = -1;
1065         /* can't get EDQUOT when reducing usage */
1066         rc = qsd_op_begin(env, qsd, trans, qi, NULL);
1067         if (rc == -EINPROGRESS)
1068                 rc = 0;
1069         if (rc)
1070                 return rc;
1071
1072         /* block accounting */
1073         qi->lqi_is_blk = true;
1074
1075         /* more blocks for the new owner ... */
1076         qi->lqi_id.qid_uid = new_id;
1077         qi->lqi_space      = bspace;
1078         rc = qsd_op_begin(env, qsd, trans, qi, NULL);
1079         if (rc == -EDQUOT || rc == -EINPROGRESS)
1080                 rc = 0;
1081         if (rc)
1082                 return rc;
1083
1084         /* and finally less blocks for the current owner */
1085         qi->lqi_id.qid_uid = orig_id;
1086         qi->lqi_space      = -bspace;
1087         rc = qsd_op_begin(env, qsd, trans, qi, NULL);
1088         /* can't get EDQUOT when reducing usage */
1089         if (rc == -EINPROGRESS)
1090                 rc = 0;
1091         return rc;
1092 }
1093
1094 static int osd_declare_attr_set(const struct lu_env *env,
1095                                 struct dt_object *dt,
1096                                 const struct lu_attr *attr,
1097                                 struct thandle *handle)
1098 {
1099         struct osd_thread_info  *info = osd_oti_get(env);
1100         struct osd_object       *obj = osd_dt_obj(dt);
1101         struct osd_device       *osd = osd_obj2dev(obj);
1102         dmu_tx_hold_t           *txh;
1103         struct osd_thandle      *oh;
1104         uint64_t                 bspace;
1105         uint32_t                 blksize;
1106         int                      rc = 0;
1107         bool                     found;
1108         ENTRY;
1109
1110
1111         LASSERT(handle != NULL);
1112         LASSERT(osd_invariant(obj));
1113
1114         oh = container_of0(handle, struct osd_thandle, ot_super);
1115
1116         down_read(&obj->oo_guard);
1117         if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
1118                 GOTO(out, rc = 0);
1119
1120         LASSERT(obj->oo_sa_hdl != NULL);
1121         LASSERT(oh->ot_tx != NULL);
1122         /* regular attributes are part of the bonus buffer */
1123         /* let's check whether this object is already part of
1124          * transaction.. */
1125         found = false;
1126         for (txh = list_head(&oh->ot_tx->tx_holds); txh;
1127              txh = list_next(&oh->ot_tx->tx_holds, txh)) {
1128                 if (txh->txh_dnode == NULL)
1129                         continue;
1130                 if (txh->txh_dnode->dn_object != obj->oo_dn->dn_object)
1131                         continue;
1132                 /* this object is part of the transaction already
1133                  * we don't need to declare bonus again */
1134                 found = true;
1135                 break;
1136         }
1137         if (!found)
1138                 dmu_tx_hold_bonus(oh->ot_tx, obj->oo_dn->dn_object);
1139         if (oh->ot_tx->tx_err != 0)
1140                 GOTO(out, rc = -oh->ot_tx->tx_err);
1141
1142         if (attr && attr->la_valid & LA_FLAGS) {
1143                 /* LMA is usually a part of bonus, no need to declare
1144                  * anything else */
1145         }
1146
1147         if (attr && (attr->la_valid & (LA_UID | LA_GID | LA_PROJID))) {
1148                 sa_object_size(obj->oo_sa_hdl, &blksize, &bspace);
1149                 bspace = toqb(bspace * blksize);
1150         }
1151
1152         if (attr && attr->la_valid & LA_UID) {
1153                 /* quota enforcement for user */
1154                 if (attr->la_uid != obj->oo_attr.la_uid) {
1155                         rc = qsd_transfer(env, osd->od_quota_slave,
1156                                           &oh->ot_quota_trans, USRQUOTA,
1157                                           obj->oo_attr.la_uid, attr->la_uid,
1158                                           bspace, &info->oti_qi);
1159                         if (rc)
1160                                 GOTO(out, rc);
1161                 }
1162         }
1163         if (attr && attr->la_valid & LA_GID) {
1164                 /* quota enforcement for group */
1165                 if (attr->la_gid != obj->oo_attr.la_gid) {
1166                         rc = qsd_transfer(env, osd->od_quota_slave,
1167                                           &oh->ot_quota_trans, GRPQUOTA,
1168                                           obj->oo_attr.la_gid, attr->la_gid,
1169                                           bspace, &info->oti_qi);
1170                         if (rc)
1171                                 GOTO(out, rc);
1172                 }
1173         }
1174 #ifdef ZFS_PROJINHERIT
1175         if (attr && attr->la_valid & LA_PROJID) {
1176                 if (!osd->od_projectused_dn)
1177                         GOTO(out, rc = -EOPNOTSUPP);
1178
1179                 /* Usually, if project quota is upgradable for the device,
1180                  * then the upgrade will be done before or when mount the
1181                  * device. So when we come here, this project should have
1182                  * project ID attribute already (that is zero by default).
1183                  * Otherwise, there was something wrong during the former
1184                  * upgrade, let's return failure to report that.
1185                  *
1186                  * Please note that, different from other attributes, you
1187                  * can NOT simply set the project ID attribute under such
1188                  * case, because adding (NOT change) project ID attribute
1189                  * needs to change the object's attribute layout to match
1190                  * zfs backend quota accounting requirement. */
1191                 if (unlikely(!obj->oo_with_projid))
1192                         GOTO(out, rc = -ENXIO);
1193
1194                 /* quota enforcement for project */
1195                 if (attr->la_projid != obj->oo_attr.la_projid) {
1196                         rc = qsd_transfer(env, osd->od_quota_slave,
1197                                           &oh->ot_quota_trans, PRJQUOTA,
1198                                           obj->oo_attr.la_projid,
1199                                           attr->la_projid, bspace,
1200                                           &info->oti_qi);
1201                         if (rc)
1202                                 GOTO(out, rc);
1203                 }
1204         }
1205 #endif
1206 out:
1207         up_read(&obj->oo_guard);
1208         RETURN(rc);
1209 }
1210
1211 /*
1212  * Set the attributes of an object
1213  *
1214  * The transaction passed to this routine must have
1215  * dmu_tx_hold_bonus(tx, oid) called and then assigned
1216  * to a transaction group.
1217  */
1218 static int osd_attr_set(const struct lu_env *env, struct dt_object *dt,
1219                         const struct lu_attr *la, struct thandle *handle)
1220 {
1221         struct osd_thread_info  *info = osd_oti_get(env);
1222         sa_bulk_attr_t          *bulk = osd_oti_get(env)->oti_attr_bulk;
1223         struct osd_object       *obj = osd_dt_obj(dt);
1224         struct osd_device       *osd = osd_obj2dev(obj);
1225         struct osd_thandle      *oh;
1226         struct osa_attr         *osa = &info->oti_osa;
1227         __u64                    valid = la->la_valid;
1228         int                      cnt;
1229         int                      rc = 0;
1230
1231         ENTRY;
1232
1233         down_read(&obj->oo_guard);
1234         if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
1235                 GOTO(out, rc = -ENOENT);
1236
1237         LASSERT(handle != NULL);
1238         LASSERT(osd_invariant(obj));
1239         LASSERT(obj->oo_sa_hdl);
1240
1241         oh = container_of0(handle, struct osd_thandle, ot_super);
1242         /* Assert that the transaction has been assigned to a
1243            transaction group. */
1244         LASSERT(oh->ot_tx->tx_txg != 0);
1245
1246         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING) && !osd->od_is_ost) {
1247                 struct zpl_direntry *zde = &info->oti_zde.lzd_reg;
1248                 char *buf = info->oti_str;
1249                 dnode_t *zdn = NULL;
1250                 uint64_t zapid;
1251
1252                 zapid = osd_get_name_n_idx(env, osd, lu_object_fid(&dt->do_lu),
1253                                            buf, sizeof(info->oti_str), &zdn);
1254                 rc = osd_zap_lookup(osd, zapid, zdn, buf, 8,
1255                                     sizeof(*zde) / 8, zde);
1256                 if (!rc) {
1257                         zde->zde_dnode -= 1;
1258                         rc = -zap_update(osd->od_os, zapid, buf, 8,
1259                                          sizeof(*zde) / 8, zde, oh->ot_tx);
1260                 }
1261                 up_read(&obj->oo_guard);
1262
1263                 RETURN(rc > 0 ? 0 : rc);
1264         }
1265
1266         /* Only allow set size for regular file */
1267         if (!S_ISREG(dt->do_lu.lo_header->loh_attr))
1268                 valid &= ~(LA_SIZE | LA_BLOCKS);
1269
1270         if (valid & LA_CTIME && la->la_ctime == obj->oo_attr.la_ctime)
1271                 valid &= ~LA_CTIME;
1272
1273         if (valid & LA_MTIME && la->la_mtime == obj->oo_attr.la_mtime)
1274                 valid &= ~LA_MTIME;
1275
1276         if (valid & LA_ATIME && la->la_atime == obj->oo_attr.la_atime)
1277                 valid &= ~LA_ATIME;
1278
1279         if (valid == 0)
1280                 GOTO(out, rc = 0);
1281
1282         if (valid & LA_FLAGS) {
1283                 struct lustre_mdt_attrs *lma;
1284                 struct lu_buf buf;
1285
1286                 if (la->la_flags & LUSTRE_LMA_FL_MASKS) {
1287                         LASSERT(!obj->oo_pfid_in_lma);
1288                         CLASSERT(sizeof(info->oti_buf) >= sizeof(*lma));
1289                         lma = (struct lustre_mdt_attrs *)&info->oti_buf;
1290                         buf.lb_buf = lma;
1291                         buf.lb_len = sizeof(info->oti_buf);
1292                         rc = osd_xattr_get(env, &obj->oo_dt, &buf,
1293                                            XATTR_NAME_LMA);
1294                         if (rc > 0) {
1295                                 lma->lma_incompat =
1296                                         le32_to_cpu(lma->lma_incompat);
1297                                 lma->lma_incompat |=
1298                                         lustre_to_lma_flags(la->la_flags);
1299                                 lma->lma_incompat =
1300                                         cpu_to_le32(lma->lma_incompat);
1301                                 buf.lb_buf = lma;
1302                                 buf.lb_len = sizeof(*lma);
1303                                 rc = osd_xattr_set_internal(env, obj, &buf,
1304                                                             XATTR_NAME_LMA,
1305                                                             LU_XATTR_REPLACE,
1306                                                             oh);
1307                         }
1308                         if (rc < 0) {
1309                                 CWARN("%s: failed to set LMA flags: rc = %d\n",
1310                                        osd->od_svname, rc);
1311                                 GOTO(out, rc);
1312                         }
1313                 }
1314         }
1315
1316         write_lock(&obj->oo_attr_lock);
1317         cnt = 0;
1318
1319         if (valid & LA_PROJID) {
1320 #ifdef ZFS_PROJINHERIT
1321                 /* osd_declare_attr_set() must be called firstly.
1322                  * If osd::od_projectused_dn is not set, then we
1323                  * can not arrive at here. */
1324                 LASSERT(osd->od_projectused_dn);
1325                 LASSERT(obj->oo_with_projid);
1326
1327                 osa->projid = obj->oo_attr.la_projid = la->la_projid;
1328                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_PROJID(osd), NULL,
1329                                  &osa->projid, 8);
1330 #else
1331                 valid &= ~LA_PROJID;
1332 #endif
1333         }
1334
1335         if (valid & LA_ATIME) {
1336                 osa->atime[0] = obj->oo_attr.la_atime = la->la_atime;
1337                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(osd), NULL,
1338                                  osa->atime, 16);
1339         }
1340         if (valid & LA_MTIME) {
1341                 osa->mtime[0] = obj->oo_attr.la_mtime = la->la_mtime;
1342                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(osd), NULL,
1343                                  osa->mtime, 16);
1344         }
1345         if (valid & LA_CTIME) {
1346                 osa->ctime[0] = obj->oo_attr.la_ctime = la->la_ctime;
1347                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(osd), NULL,
1348                                  osa->ctime, 16);
1349         }
1350         if (valid & LA_MODE) {
1351                 /* mode is stored along with type, so read it first */
1352                 obj->oo_attr.la_mode = (obj->oo_attr.la_mode & S_IFMT) |
1353                         (la->la_mode & ~S_IFMT);
1354                 osa->mode = obj->oo_attr.la_mode;
1355                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(osd), NULL,
1356                                  &osa->mode, 8);
1357         }
1358         if (valid & LA_SIZE) {
1359                 osa->size = obj->oo_attr.la_size = la->la_size;
1360                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(osd), NULL,
1361                                  &osa->size, 8);
1362         }
1363         if (valid & LA_NLINK) {
1364                 osa->nlink = obj->oo_attr.la_nlink = la->la_nlink;
1365                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(osd), NULL,
1366                                  &osa->nlink, 8);
1367         }
1368         if (valid & LA_RDEV) {
1369                 osa->rdev = obj->oo_attr.la_rdev = la->la_rdev;
1370                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_RDEV(osd), NULL,
1371                                  &osa->rdev, 8);
1372         }
1373         if (valid & LA_FLAGS) {
1374                 osa->flags = attrs_fs2zfs(la->la_flags);
1375                 /* many flags are not supported by zfs, so ensure a good cached
1376                  * copy */
1377                 obj->oo_attr.la_flags = attrs_zfs2fs(osa->flags);
1378 #ifdef ZFS_PROJINHERIT
1379                 if (obj->oo_with_projid)
1380                         osa->flags |= ZFS_PROJID;
1381 #endif
1382                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(osd), NULL,
1383                                  &osa->flags, 8);
1384         }
1385         if (valid & LA_UID) {
1386                 osa->uid = obj->oo_attr.la_uid = la->la_uid;
1387                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(osd), NULL,
1388                                  &osa->uid, 8);
1389         }
1390         if (valid & LA_GID) {
1391                 osa->gid = obj->oo_attr.la_gid = la->la_gid;
1392                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(osd), NULL,
1393                                  &osa->gid, 8);
1394         }
1395         obj->oo_attr.la_valid |= valid;
1396         write_unlock(&obj->oo_attr_lock);
1397
1398         LASSERT(cnt <= ARRAY_SIZE(osd_oti_get(env)->oti_attr_bulk));
1399         rc = osd_object_sa_bulk_update(obj, bulk, cnt, oh);
1400
1401 out:
1402         up_read(&obj->oo_guard);
1403         RETURN(rc);
1404 }
1405
1406 /*
1407  * Object creation.
1408  *
1409  * XXX temporary solution.
1410  */
1411
1412 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
1413                         struct dt_object *parent, struct dt_object *child,
1414                         umode_t child_mode)
1415 {
1416         LASSERT(ah);
1417
1418         ah->dah_parent = parent;
1419         ah->dah_mode = child_mode;
1420
1421         if (parent != NULL && !dt_object_remote(parent)) {
1422                 /* will help to find FID->ino at dt_insert("..") */
1423                 struct osd_object *pobj = osd_dt_obj(parent);
1424
1425                 osd_idc_find_and_init(env, osd_obj2dev(pobj), pobj);
1426         }
1427 }
1428
1429 static int osd_declare_create(const struct lu_env *env, struct dt_object *dt,
1430                               struct lu_attr *attr,
1431                               struct dt_allocation_hint *hint,
1432                               struct dt_object_format *dof,
1433                               struct thandle *handle)
1434 {
1435         const struct lu_fid     *fid = lu_object_fid(&dt->do_lu);
1436         struct osd_object       *obj = osd_dt_obj(dt);
1437         struct osd_device       *osd = osd_obj2dev(obj);
1438         struct osd_thandle      *oh;
1439         uint64_t                 zapid;
1440         dnode_t                 *dn;
1441         int                      rc, dnode_size;
1442         ENTRY;
1443
1444         LASSERT(dof);
1445
1446         switch (dof->dof_type) {
1447                 case DFT_REGULAR:
1448                 case DFT_SYM:
1449                 case DFT_NODE:
1450                         if (obj->oo_dt.do_body_ops == NULL)
1451                                 obj->oo_dt.do_body_ops = &osd_body_ops;
1452                         break;
1453                 default:
1454                         break;
1455         }
1456
1457         LASSERT(handle != NULL);
1458         oh = container_of0(handle, struct osd_thandle, ot_super);
1459         LASSERT(oh->ot_tx != NULL);
1460
1461         /* this is the minimum set of EAs on every Lustre object */
1462         obj->oo_ea_in_bonus = OSD_BASE_EA_IN_BONUS;
1463         /* reserve 32 bytes for extra stuff like ACLs */
1464         dnode_size = size_roundup_power2(obj->oo_ea_in_bonus + 32);
1465
1466         switch (dof->dof_type) {
1467                 case DFT_DIR:
1468                         dt->do_index_ops = &osd_dir_ops;
1469                 case DFT_INDEX:
1470                         /* for zap create */
1471                         dmu_tx_hold_zap(oh->ot_tx, DMU_NEW_OBJECT, FALSE, NULL);
1472                         dmu_tx_hold_sa_create(oh->ot_tx, dnode_size);
1473                         break;
1474                 case DFT_REGULAR:
1475                 case DFT_SYM:
1476                 case DFT_NODE:
1477                         /* first, we'll create new object */
1478                         dmu_tx_hold_sa_create(oh->ot_tx, dnode_size);
1479                         break;
1480
1481                 default:
1482                         LBUG();
1483                         break;
1484         }
1485
1486         /* and we'll add it to some mapping */
1487         zapid = osd_get_name_n_idx(env, osd, fid, NULL, 0, &dn);
1488         osd_tx_hold_zap(oh->ot_tx, zapid, dn, TRUE, NULL);
1489
1490         /* will help to find FID->ino mapping at dt_insert() */
1491         osd_idc_find_and_init(env, osd, obj);
1492
1493         rc = osd_declare_quota(env, osd, attr->la_uid, attr->la_gid,
1494                                attr->la_projid, 1, oh, NULL, OSD_QID_INODE);
1495
1496         RETURN(rc);
1497 }
1498
1499 int __osd_attr_init(const struct lu_env *env, struct osd_device *osd,
1500                     struct osd_object *obj, sa_handle_t *sa_hdl, dmu_tx_t *tx,
1501                     struct lu_attr *la, uint64_t parent,
1502                     nvlist_t *xattr)
1503 {
1504         sa_bulk_attr_t  *bulk = osd_oti_get(env)->oti_attr_bulk;
1505         struct osa_attr *osa = &osd_oti_get(env)->oti_osa;
1506         uint64_t         gen;
1507         uint64_t         crtime[2];
1508         timestruc_t      now;
1509         int              cnt;
1510         int              rc;
1511         char *dxattr = NULL;
1512         size_t sa_size;
1513
1514
1515         LASSERT(sa_hdl);
1516
1517         gen = dmu_tx_get_txg(tx);
1518         gethrestime(&now);
1519         ZFS_TIME_ENCODE(&now, crtime);
1520
1521         osa->atime[0] = la->la_atime;
1522         osa->ctime[0] = la->la_ctime;
1523         osa->mtime[0] = la->la_mtime;
1524         osa->mode = la->la_mode;
1525         osa->uid = la->la_uid;
1526         osa->gid = la->la_gid;
1527         osa->rdev = la->la_rdev;
1528         osa->nlink = la->la_nlink;
1529         if (la->la_valid & LA_FLAGS)
1530                 osa->flags = attrs_fs2zfs(la->la_flags);
1531         else
1532                 osa->flags = 0;
1533         osa->size  = la->la_size;
1534 #ifdef ZFS_PROJINHERIT
1535         if (osd->od_projectused_dn) {
1536                 if (la->la_valid & LA_PROJID)
1537                         osa->projid = la->la_projid;
1538                 else
1539                         osa->projid = ZFS_DEFAULT_PROJID;
1540                 osa->flags |= ZFS_PROJID;
1541                 if (obj)
1542                         obj->oo_with_projid = 1;
1543         } else {
1544                 osa->flags &= ~ZFS_PROJID;
1545         }
1546 #endif
1547
1548         /*
1549          * we need to create all SA below upon object create.
1550          *
1551          * XXX The attribute order matters since the accounting callback relies
1552          * on static offsets (i.e. SA_*_OFFSET, see zfs_space_delta_cb()) to
1553          * look up the UID/GID/PROJID attributes. Moreover, the callback does
1554          * not seem to support the spill block.
1555          * We define attributes in the same order as SA_*_OFFSET in order to
1556          * work around the problem. See ORI-610.
1557          */
1558         cnt = 0;
1559         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(osd), NULL, &osa->mode, 8);
1560         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(osd), NULL, &osa->size, 8);
1561         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GEN(osd), NULL, &gen, 8);
1562         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(osd), NULL, &osa->uid, 8);
1563         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(osd), NULL, &osa->gid, 8);
1564         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_PARENT(osd), NULL, &parent, 8);
1565         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(osd), NULL, &osa->flags, 8);
1566         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(osd), NULL, osa->atime, 16);
1567         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(osd), NULL, osa->mtime, 16);
1568         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(osd), NULL, osa->ctime, 16);
1569         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CRTIME(osd), NULL, crtime, 16);
1570         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(osd), NULL, &osa->nlink, 8);
1571 #ifdef ZFS_PROJINHERIT
1572         if (osd->od_projectused_dn)
1573                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_PROJID(osd), NULL,
1574                                  &osa->projid, 8);
1575 #endif
1576         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_RDEV(osd), NULL, &osa->rdev, 8);
1577         LASSERT(cnt <= ARRAY_SIZE(osd_oti_get(env)->oti_attr_bulk));
1578
1579         if (xattr) {
1580                 rc = -nvlist_size(xattr, &sa_size, NV_ENCODE_XDR);
1581                 LASSERT(rc == 0);
1582
1583                 dxattr = osd_zio_buf_alloc(sa_size);
1584                 LASSERT(dxattr);
1585
1586                 rc = -nvlist_pack(xattr, &dxattr, &sa_size,
1587                                 NV_ENCODE_XDR, KM_SLEEP);
1588                 LASSERT(rc == 0);
1589
1590                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_DXATTR(osd),
1591                                 NULL, dxattr, sa_size);
1592         }
1593
1594         rc = -sa_replace_all_by_template(sa_hdl, bulk, cnt, tx);
1595         if (dxattr)
1596                 osd_zio_buf_free(dxattr, sa_size);
1597
1598         return rc;
1599 }
1600
1601 int osd_find_new_dnode(const struct lu_env *env, dmu_tx_t *tx,
1602                        uint64_t oid, dnode_t **dnp)
1603 {
1604         dmu_tx_hold_t *txh;
1605         int rc = 0;
1606
1607         /* take dnode_t from tx to save on dnode#->dnode_t lookup */
1608         for (txh = list_tail(&tx->tx_holds); txh;
1609              txh = list_prev(&tx->tx_holds, txh)) {
1610                 dnode_t *dn = txh->txh_dnode;
1611                 dmu_buf_impl_t *db;
1612
1613                 if (dn == NULL)
1614                         continue;
1615                 if (dn->dn_object != oid)
1616                         continue;
1617                 db = dn->dn_bonus;
1618                 if (db == NULL) {
1619                         rw_enter(&dn->dn_struct_rwlock, RW_WRITER);
1620                         if (dn->dn_bonus == NULL)
1621                                 dbuf_create_bonus(dn);
1622                         rw_exit(&dn->dn_struct_rwlock);
1623                 }
1624                 db = dn->dn_bonus;
1625                 LASSERT(db);
1626                 LASSERT(dn->dn_handle);
1627                 DB_DNODE_ENTER(db);
1628                 if (refcount_add(&db->db_holds, osd_obj_tag) == 1) {
1629                         refcount_add(&dn->dn_holds, tag);
1630                         atomic_inc_32(&dn->dn_dbufs_count);
1631                 }
1632                 *dnp = dn;
1633                 dbuf_read(db, NULL, DB_RF_MUST_SUCCEED | DB_RF_NOPREFETCH);
1634                 break;
1635         }
1636
1637         if (unlikely(*dnp == NULL))
1638                 rc = __osd_obj2dnode(tx->tx_objset, oid, dnp);
1639
1640         return rc;
1641 }
1642
1643 #ifdef HAVE_DMU_OBJECT_ALLOC_DNSIZE
1644 int osd_find_dnsize(struct osd_device *osd, int ea_in_bonus)
1645 {
1646         int dnsize;
1647
1648         if (osd->od_dnsize == ZFS_DNSIZE_AUTO) {
1649                 dnsize = DNODE_MIN_SIZE;
1650                 do {
1651                         if (DN_BONUS_SIZE(dnsize) >= ea_in_bonus + 32)
1652                                 break;
1653                         dnsize <<= 1;
1654                 } while (dnsize < DNODE_MAX_SIZE);
1655                 if (dnsize > DNODE_MAX_SIZE)
1656                         dnsize = DNODE_MAX_SIZE;
1657         } else if (osd->od_dnsize == ZFS_DNSIZE_1K) {
1658                 dnsize = 1024;
1659         } else if (osd->od_dnsize == ZFS_DNSIZE_2K) {
1660                 dnsize = 2048;
1661         } else if (osd->od_dnsize == ZFS_DNSIZE_4K) {
1662                 dnsize = 4096;
1663         } else if (osd->od_dnsize == ZFS_DNSIZE_8K) {
1664                 dnsize = 8192;
1665         } else if (osd->od_dnsize == ZFS_DNSIZE_16K) {
1666                 dnsize = 16384;
1667         } else {
1668                 dnsize = DNODE_MIN_SIZE;
1669         }
1670         return dnsize;
1671 }
1672 #endif
1673
1674 /*
1675  * The transaction passed to this routine must have
1676  * dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT) called and then assigned
1677  * to a transaction group.
1678  */
1679 int __osd_object_create(const struct lu_env *env, struct osd_device *osd,
1680                         struct osd_object *obj, const struct lu_fid *fid,
1681                         dnode_t **dnp, dmu_tx_t *tx, struct lu_attr *la)
1682 {
1683         dmu_object_type_t type = DMU_OT_PLAIN_FILE_CONTENTS;
1684         uint64_t oid;
1685         int size;
1686
1687         /* Use DMU_OTN_UINT8_METADATA for local objects so their data blocks
1688          * would get an additional ditto copy */
1689         if (unlikely(S_ISREG(la->la_mode) &&
1690                      fid_seq_is_local_file(fid_seq(fid))))
1691                 type = DMU_OTN_UINT8_METADATA;
1692
1693         /* Create a new DMU object using the default dnode size. */
1694         if (obj)
1695                 size = obj->oo_ea_in_bonus;
1696         else
1697                 size = OSD_BASE_EA_IN_BONUS;
1698         oid = osd_dmu_object_alloc(osd->od_os, type, 0,
1699                                    osd_find_dnsize(osd, size), tx);
1700
1701         LASSERT(la->la_valid & LA_MODE);
1702         la->la_size = 0;
1703         la->la_nlink = 1;
1704
1705         return osd_find_new_dnode(env, tx, oid, dnp);
1706 }
1707
1708 /*
1709  * The transaction passed to this routine must have
1710  * dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, ...) called and then assigned
1711  * to a transaction group.
1712  *
1713  * Using ZAP_FLAG_HASH64 will force the ZAP to always be a FAT ZAP.
1714  * This is fine for directories today, because storing the FID in the dirent
1715  * will also require a FAT ZAP.  If there is a new type of micro ZAP created
1716  * then we might need to re-evaluate the use of this flag and instead do
1717  * a conversion from the different internal ZAP hash formats being used. */
1718 int __osd_zap_create(const struct lu_env *env, struct osd_device *osd,
1719                      dnode_t **dnp, dmu_tx_t *tx, struct lu_attr *la,
1720                      unsigned dnsize, zap_flags_t flags)
1721 {
1722         uint64_t oid;
1723
1724         /* Assert that the transaction has been assigned to a
1725            transaction group. */
1726         LASSERT(tx->tx_txg != 0);
1727         *dnp = NULL;
1728
1729         oid = osd_zap_create_flags(osd->od_os, 0, flags | ZAP_FLAG_HASH64,
1730                                    DMU_OT_DIRECTORY_CONTENTS,
1731                                    14, /* == ZFS fzap_default_blockshift */
1732                                    DN_MAX_INDBLKSHIFT, /* indirect blockshift */
1733                                    dnsize, tx);
1734
1735         la->la_size = 2;
1736         la->la_nlink = 1;
1737
1738         return osd_find_new_dnode(env, tx, oid, dnp);
1739 }
1740
1741 static dnode_t *osd_mkidx(const struct lu_env *env, struct osd_object *obj,
1742                           struct lu_attr *la, struct osd_thandle *oh)
1743 {
1744         struct osd_device *osd = osd_obj2dev(obj);
1745         dnode_t *dn;
1746         int rc;
1747
1748         /* Index file should be created as regular file in order not to confuse
1749          * ZPL which could interpret them as directory.
1750          * We set ZAP_FLAG_UINT64_KEY to let ZFS know than we are going to use
1751          * binary keys */
1752         LASSERT(S_ISREG(la->la_mode));
1753         rc = __osd_zap_create(env, osd, &dn, oh->ot_tx, la,
1754                 osd_find_dnsize(osd, obj->oo_ea_in_bonus), ZAP_FLAG_UINT64_KEY);
1755         if (rc)
1756                 return ERR_PTR(rc);
1757         return dn;
1758 }
1759
1760 static dnode_t *osd_mkdir(const struct lu_env *env, struct osd_object *obj,
1761                           struct lu_attr *la, struct osd_thandle *oh)
1762 {
1763         struct osd_device *osd = osd_obj2dev(obj);
1764         dnode_t *dn;
1765         int rc;
1766
1767         LASSERT(S_ISDIR(la->la_mode));
1768         rc = __osd_zap_create(env, osd, &dn, oh->ot_tx, la,
1769                               osd_find_dnsize(osd, obj->oo_ea_in_bonus), 0);
1770         if (rc)
1771                 return ERR_PTR(rc);
1772         return dn;
1773 }
1774
1775 static dnode_t *osd_mkreg(const struct lu_env *env, struct osd_object *obj,
1776                           struct lu_attr *la, struct osd_thandle *oh)
1777 {
1778         const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
1779         struct osd_device *osd = osd_obj2dev(obj);
1780         dnode_t *dn;
1781         int rc;
1782
1783         LASSERT(S_ISREG(la->la_mode));
1784         rc = __osd_object_create(env, osd, obj, fid, &dn, oh->ot_tx, la);
1785         if (rc)
1786                 return ERR_PTR(rc);
1787
1788         if ((fid_is_idif(fid) || fid_is_norm(fid) || fid_is_echo(fid))) {
1789                 /* The minimum block size must be at least page size otherwise
1790                  * it will break the assumption in tgt_thread_big_cache where
1791                  * the array size is PTLRPC_MAX_BRW_PAGES. It will also affect
1792                  * RDMA due to subpage transfer size */
1793                 rc = -dmu_object_set_blocksize(osd->od_os, dn->dn_object,
1794                                                PAGE_SIZE, 0, oh->ot_tx);
1795                 if (unlikely(rc)) {
1796                         CERROR("%s: can't change blocksize: %d\n",
1797                                osd->od_svname, rc);
1798                         return ERR_PTR(rc);
1799                 }
1800         }
1801
1802         return dn;
1803 }
1804
1805 static dnode_t *osd_mksym(const struct lu_env *env, struct osd_object *obj,
1806                           struct lu_attr *la, struct osd_thandle *oh)
1807 {
1808         dnode_t *dn;
1809         int rc;
1810
1811         LASSERT(S_ISLNK(la->la_mode));
1812         rc = __osd_object_create(env, osd_obj2dev(obj), obj,
1813                                  lu_object_fid(&obj->oo_dt.do_lu),
1814                                  &dn, oh->ot_tx, la);
1815         if (rc)
1816                 return ERR_PTR(rc);
1817         return dn;
1818 }
1819
1820 static dnode_t *osd_mknod(const struct lu_env *env, struct osd_object *obj,
1821                           struct lu_attr *la, struct osd_thandle *oh)
1822 {
1823         dnode_t *dn;
1824         int rc;
1825
1826         if (S_ISCHR(la->la_mode) || S_ISBLK(la->la_mode))
1827                 la->la_valid |= LA_RDEV;
1828
1829         rc = __osd_object_create(env, osd_obj2dev(obj), obj,
1830                                  lu_object_fid(&obj->oo_dt.do_lu),
1831                                  &dn, oh->ot_tx, la);
1832         if (rc)
1833                 return ERR_PTR(rc);
1834         return dn;
1835 }
1836
1837 typedef dnode_t *(*osd_obj_type_f)(const struct lu_env *env,
1838                                    struct osd_object *obj,
1839                                    struct lu_attr *la,
1840                                    struct osd_thandle *oh);
1841
1842 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
1843 {
1844         osd_obj_type_f result;
1845
1846         switch (type) {
1847         case DFT_DIR:
1848                 result = osd_mkdir;
1849                 break;
1850         case DFT_INDEX:
1851                 result = osd_mkidx;
1852                 break;
1853         case DFT_REGULAR:
1854                 result = osd_mkreg;
1855                 break;
1856         case DFT_SYM:
1857                 result = osd_mksym;
1858                 break;
1859         case DFT_NODE:
1860                 result = osd_mknod;
1861                 break;
1862         default:
1863                 LBUG();
1864                 break;
1865         }
1866         return result;
1867 }
1868
1869 /*
1870  * Concurrency: @dt is write locked.
1871  */
1872 static int osd_create(const struct lu_env *env, struct dt_object *dt,
1873                       struct lu_attr *attr, struct dt_allocation_hint *hint,
1874                       struct dt_object_format *dof, struct thandle *th)
1875 {
1876         struct osd_thread_info  *info = osd_oti_get(env);
1877         struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
1878         struct zpl_direntry     *zde = &info->oti_zde.lzd_reg;
1879         const struct lu_fid     *fid = lu_object_fid(&dt->do_lu);
1880         struct osd_object       *obj = osd_dt_obj(dt);
1881         struct osd_device       *osd = osd_obj2dev(obj);
1882         char                    *buf = info->oti_str;
1883         struct osd_thandle      *oh;
1884         dnode_t *dn = NULL, *zdn = NULL;
1885         uint64_t                 zapid, parent = 0;
1886         int                      rc;
1887         __u32 compat = 0;
1888
1889         ENTRY;
1890
1891         LASSERT(!fid_is_acct(fid));
1892
1893         /* concurrent create declarations should not see
1894          * the object inconsistent (db, attr, etc).
1895          * in regular cases acquisition should be cheap */
1896         down_write(&obj->oo_guard);
1897
1898         if (unlikely(dt_object_exists(dt)))
1899                 GOTO(out, rc = -EEXIST);
1900
1901         LASSERT(osd_invariant(obj));
1902         LASSERT(dof != NULL);
1903
1904         LASSERT(th != NULL);
1905         oh = container_of0(th, struct osd_thandle, ot_super);
1906
1907         LASSERT(obj->oo_dn == NULL);
1908
1909         /* to follow ZFS on-disk format we need
1910          * to initialize parent dnode properly */
1911         if (hint != NULL && hint->dah_parent != NULL &&
1912             !dt_object_remote(hint->dah_parent))
1913                 parent = osd_dt_obj(hint->dah_parent)->oo_dn->dn_object;
1914
1915         /* we may fix some attributes, better do not change the source */
1916         obj->oo_attr = *attr;
1917         obj->oo_attr.la_valid |= LA_SIZE | LA_NLINK | LA_TYPE;
1918
1919 #ifdef ZFS_PROJINHERIT
1920         if (osd->od_projectused_dn) {
1921                 if (!(obj->oo_attr.la_valid & LA_PROJID))
1922                         obj->oo_attr.la_projid = ZFS_DEFAULT_PROJID;
1923                 obj->oo_with_projid = 1;
1924         }
1925 #endif
1926
1927         dn = osd_create_type_f(dof->dof_type)(env, obj, &obj->oo_attr, oh);
1928         if (IS_ERR(dn)) {
1929                 rc = PTR_ERR(dn);
1930                 dn = NULL;
1931                 GOTO(out, rc);
1932         }
1933
1934         zde->zde_pad = 0;
1935         zde->zde_dnode = dn->dn_object;
1936         zde->zde_type = IFTODT(attr->la_mode & S_IFMT);
1937
1938         zapid = osd_get_name_n_idx(env, osd, fid, buf,
1939                                    sizeof(info->oti_str), &zdn);
1940         if (CFS_FAIL_CHECK(OBD_FAIL_OSD_NO_OI_ENTRY) ||
1941             (osd->od_is_ost && OBD_FAIL_CHECK(OBD_FAIL_OSD_COMPAT_NO_ENTRY)))
1942                 goto skip_add;
1943
1944         if (osd->od_is_ost && OBD_FAIL_CHECK(OBD_FAIL_OSD_COMPAT_INVALID_ENTRY))
1945                 zde->zde_dnode++;
1946
1947         rc = osd_zap_add(osd, zapid, zdn, buf, 8, 1, zde, oh->ot_tx);
1948         if (rc)
1949                 GOTO(out, rc);
1950
1951 skip_add:
1952         obj->oo_dn = dn;
1953         /* Now add in all of the "SA" attributes */
1954         rc = osd_sa_handle_get(obj);
1955         if (rc)
1956                 GOTO(out, rc);
1957
1958         rc = -nvlist_alloc(&obj->oo_sa_xattr, NV_UNIQUE_NAME, KM_SLEEP);
1959         if (rc)
1960                 GOTO(out, rc);
1961
1962         /* initialize LMA */
1963         if (fid_is_idif(fid) || (fid_is_norm(fid) && osd->od_is_ost))
1964                 compat |= LMAC_FID_ON_OST;
1965         lustre_lma_init(lma, fid, compat, 0);
1966         lustre_lma_swab(lma);
1967         rc = -nvlist_add_byte_array(obj->oo_sa_xattr, XATTR_NAME_LMA,
1968                                     (uchar_t *)lma, sizeof(*lma));
1969         if (rc)
1970                 GOTO(out, rc);
1971
1972         /* configure new osd object */
1973         obj->oo_parent = parent != 0 ? parent : zapid;
1974         obj->oo_late_attr_set = 1;
1975         rc = __osd_sa_xattr_schedule_update(env, obj, oh);
1976         if (rc)
1977                 GOTO(out, rc);
1978
1979         /* XXX: oo_lma_flags */
1980         obj->oo_dt.do_lu.lo_header->loh_attr |= obj->oo_attr.la_mode & S_IFMT;
1981         if (likely(!fid_is_acct(lu_object_fid(&obj->oo_dt.do_lu))))
1982                 /* no body operations for accounting objects */
1983                 obj->oo_dt.do_body_ops = &osd_body_ops;
1984
1985         osd_idc_find_and_init(env, osd, obj);
1986
1987 out:
1988         if (unlikely(rc && dn)) {
1989                 dmu_object_free(osd->od_os, dn->dn_object, oh->ot_tx);
1990                 osd_dnode_rele(dn);
1991                 obj->oo_dn = NULL;
1992         } else if (!rc) {
1993                 obj->oo_dt.do_lu.lo_header->loh_attr |= LOHA_EXISTS;
1994         }
1995         up_write(&obj->oo_guard);
1996         RETURN(rc);
1997 }
1998
1999 static int osd_declare_ref_add(const struct lu_env *env, struct dt_object *dt,
2000                                struct thandle *th)
2001 {
2002         osd_idc_find_and_init(env, osd_dev(dt->do_lu.lo_dev), osd_dt_obj(dt));
2003         return osd_declare_attr_set(env, dt, NULL, th);
2004 }
2005
2006 /*
2007  * Concurrency: @dt is write locked.
2008  */
2009 static int osd_ref_add(const struct lu_env *env, struct dt_object *dt,
2010                        struct thandle *handle)
2011 {
2012         struct osd_object       *obj = osd_dt_obj(dt);
2013         struct osd_thandle      *oh;
2014         struct osd_device       *osd = osd_obj2dev(obj);
2015         uint64_t                 nlink;
2016         int rc;
2017
2018         ENTRY;
2019
2020         down_read(&obj->oo_guard);
2021         if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
2022                 GOTO(out, rc = -ENOENT);
2023
2024         LASSERT(osd_invariant(obj));
2025         LASSERT(obj->oo_sa_hdl != NULL);
2026
2027         oh = container_of0(handle, struct osd_thandle, ot_super);
2028
2029         write_lock(&obj->oo_attr_lock);
2030         nlink = ++obj->oo_attr.la_nlink;
2031         write_unlock(&obj->oo_attr_lock);
2032
2033         rc = osd_object_sa_update(obj, SA_ZPL_LINKS(osd), &nlink, 8, oh);
2034
2035 out:
2036         up_read(&obj->oo_guard);
2037         RETURN(rc);
2038 }
2039
2040 static int osd_declare_ref_del(const struct lu_env *env, struct dt_object *dt,
2041                                struct thandle *handle)
2042 {
2043         osd_idc_find_and_init(env, osd_dev(dt->do_lu.lo_dev), osd_dt_obj(dt));
2044         return osd_declare_attr_set(env, dt, NULL, handle);
2045 }
2046
2047 /*
2048  * Concurrency: @dt is write locked.
2049  */
2050 static int osd_ref_del(const struct lu_env *env, struct dt_object *dt,
2051                        struct thandle *handle)
2052 {
2053         struct osd_object       *obj = osd_dt_obj(dt);
2054         struct osd_thandle      *oh;
2055         struct osd_device       *osd = osd_obj2dev(obj);
2056         uint64_t                 nlink;
2057         int                      rc;
2058
2059         ENTRY;
2060
2061         down_read(&obj->oo_guard);
2062
2063         if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
2064                 GOTO(out, rc = -ENOENT);
2065
2066         LASSERT(osd_invariant(obj));
2067         LASSERT(obj->oo_sa_hdl != NULL);
2068
2069         oh = container_of0(handle, struct osd_thandle, ot_super);
2070         LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
2071
2072         write_lock(&obj->oo_attr_lock);
2073         nlink = --obj->oo_attr.la_nlink;
2074         write_unlock(&obj->oo_attr_lock);
2075
2076         rc = osd_object_sa_update(obj, SA_ZPL_LINKS(osd), &nlink, 8, oh);
2077
2078 out:
2079         up_read(&obj->oo_guard);
2080         RETURN(rc);
2081 }
2082
2083 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
2084                            __u64 start, __u64 end)
2085 {
2086         struct osd_device *osd = osd_obj2dev(osd_dt_obj(dt));
2087         ENTRY;
2088
2089         /* XXX: no other option than syncing the whole filesystem until we
2090          * support ZIL.  If the object tracked the txg that it was last
2091          * modified in, it could pass that txg here instead of "0".  Maybe
2092          * the changes are already committed, so no wait is needed at all? */
2093         if (!osd->od_dt_dev.dd_rdonly) {
2094                 if (osd_object_sync_delay_us < 0)
2095                         txg_wait_synced(dmu_objset_pool(osd->od_os), 0ULL);
2096                 else
2097                         udelay(osd_object_sync_delay_us);
2098         }
2099
2100         RETURN(0);
2101 }
2102
2103 static int osd_invalidate(const struct lu_env *env, struct dt_object *dt)
2104 {
2105         return 0;
2106 }
2107
2108 static struct dt_object_operations osd_obj_ops = {
2109         .do_read_lock           = osd_read_lock,
2110         .do_write_lock          = osd_write_lock,
2111         .do_read_unlock         = osd_read_unlock,
2112         .do_write_unlock        = osd_write_unlock,
2113         .do_write_locked        = osd_write_locked,
2114         .do_attr_get            = osd_attr_get,
2115         .do_declare_attr_set    = osd_declare_attr_set,
2116         .do_attr_set            = osd_attr_set,
2117         .do_ah_init             = osd_ah_init,
2118         .do_declare_create      = osd_declare_create,
2119         .do_create              = osd_create,
2120         .do_declare_destroy     = osd_declare_destroy,
2121         .do_destroy             = osd_destroy,
2122         .do_index_try           = osd_index_try,
2123         .do_declare_ref_add     = osd_declare_ref_add,
2124         .do_ref_add             = osd_ref_add,
2125         .do_declare_ref_del     = osd_declare_ref_del,
2126         .do_ref_del             = osd_ref_del,
2127         .do_xattr_get           = osd_xattr_get,
2128         .do_declare_xattr_set   = osd_declare_xattr_set,
2129         .do_xattr_set           = osd_xattr_set,
2130         .do_declare_xattr_del   = osd_declare_xattr_del,
2131         .do_xattr_del           = osd_xattr_del,
2132         .do_xattr_list          = osd_xattr_list,
2133         .do_object_sync         = osd_object_sync,
2134         .do_invalidate          = osd_invalidate,
2135 };
2136
2137 static struct lu_object_operations osd_lu_obj_ops = {
2138         .loo_object_init        = osd_object_init,
2139         .loo_object_delete      = osd_object_delete,
2140         .loo_object_release     = osd_object_release,
2141         .loo_object_free        = osd_object_free,
2142         .loo_object_print       = osd_object_print,
2143         .loo_object_invariant   = osd_object_invariant,
2144 };
2145
2146 static int osd_otable_it_attr_get(const struct lu_env *env,
2147                                 struct dt_object *dt,
2148                                 struct lu_attr *attr)
2149 {
2150         attr->la_valid = 0;
2151         return 0;
2152 }
2153
2154 static struct dt_object_operations osd_obj_otable_it_ops = {
2155         .do_attr_get            = osd_otable_it_attr_get,
2156         .do_index_try           = osd_index_try,
2157 };
2158
2159 module_param(osd_object_sync_delay_us, int, 0644);
2160 MODULE_PARM_DESC(osd_object_sync_delay_us,
2161                  "If zero or larger delay N usec instead of doing object sync");