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         dmu_tx_mark_netfree(oh->ot_tx);
 766
 767         /* declare that we'll remove object from fid-dnode mapping */
 768         zapid = osd_get_name_n_idx(env, osd, fid, NULL, 0, &dn);
 769         osd_tx_hold_zap(oh->ot_tx, zapid, dn, FALSE, NULL);
 770
 771         osd_declare_xattrs_destroy(env, obj, oh);
 772
 773         /* one less inode */
 774         rc = osd_declare_quota(env, osd, obj->oo_attr.la_uid,
 775                                obj->oo_attr.la_gid, obj->oo_attr.la_projid,
 776                                -1, oh, NULL, OSD_QID_INODE);
 777         if (rc)
 778                 RETURN(rc);
 779
 780         /* data to be truncated */
 781         rc = osd_declare_quota(env, osd, obj->oo_attr.la_uid,
 782                                obj->oo_attr.la_gid, obj->oo_attr.la_projid,
 783                                0, oh, NULL, OSD_QID_BLK);
 784         if (rc)
 785                 RETURN(rc);
 786
 787         osd_object_set_destroy_type(obj);
 788         if (obj->oo_destroy == OSD_DESTROY_SYNC)
 789                 dmu_tx_hold_free(oh->ot_tx, obj->oo_dn->dn_object,
 790                                  0, DMU_OBJECT_END);
 791         else
 792                 osd_tx_hold_zap(oh->ot_tx, osd->od_unlinked->dn_object,
 793                                 osd->od_unlinked, TRUE, NULL);
 794
 795         /* remove agent entry (if have) from remote parent */
 796         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu))
 797                 osd_tx_hold_zap(oh->ot_tx, osd->od_remote_parent_dir,
 798                                 NULL, FALSE, NULL);
 799
 800         /* will help to find FID->ino when this object is being
 801          * added to PENDING/ */
 802         osd_idc_find_and_init(env, osd, obj);
 803
 804         RETURN(0);
 805 }
 806
 807 static int osd_destroy(const struct lu_env *env, struct dt_object *dt,
 808                        struct thandle *th)
 809 {
 810         struct osd_thread_info  *info = osd_oti_get(env);
 811         char                    *buf = info->oti_str;
 812         struct osd_object       *obj = osd_dt_obj(dt);
 813         struct osd_device       *osd = osd_obj2dev(obj);
 814         const struct lu_fid     *fid = lu_object_fid(&dt->do_lu);
 815         struct osd_thandle      *oh;
 816         int                      rc;
 817         uint64_t                 oid, zapid;
 818         dnode_t *zdn;
 819         ENTRY;
 820
 821         down_write(&obj->oo_guard);
 822
 823         if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
 824                 GOTO(out, rc = -ENOENT);
 825
 826         LASSERT(obj->oo_dn != NULL);
 827
 828         oh = container_of0(th, struct osd_thandle, ot_super);
 829         LASSERT(oh != NULL);
 830         LASSERT(oh->ot_tx != NULL);
 831
 832         /* remove obj ref from index dir (it depends) */
 833         zapid = osd_get_name_n_idx(env, osd, fid, buf,
 834                                    sizeof(info->oti_str), &zdn);
 835         rc = osd_xattrs_destroy(env, obj, oh);
 836         if (rc) {
 837                 CERROR("%s: cannot destroy xattrs for %s: rc = %d\n",
 838                        osd->od_svname, buf, rc);
 839                 GOTO(out, rc);
 840         }
 841
 842         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu)) {
 843                 rc = osd_delete_from_remote_parent(env, osd, obj, oh, true);
 844                 if (rc)
 845                         GOTO(out, rc);
 846         }
 847
 848         oid = obj->oo_dn->dn_object;
 849         if (unlikely(obj->oo_destroy == OSD_DESTROY_NONE)) {
 850                 /* this may happen if the destroy wasn't declared
 851                  * e.g. when the object is created and then destroyed
 852                  * in the same transaction - we don't need additional
 853                  * space for destroy specifically */
 854                 LASSERT(obj->oo_attr.la_size <= osd_sync_destroy_max_size);
 855                 rc = -dmu_object_free(osd->od_os, oid, oh->ot_tx);
 856                 if (rc)
 857                         CERROR("%s: failed to free %s %llu: rc = %d\n",
 858                                osd->od_svname, buf, oid, rc);
 859         } else if (obj->oo_destroy == OSD_DESTROY_SYNC) {
 860                 rc = -dmu_object_free(osd->od_os, oid, oh->ot_tx);
 861                 if (rc)
 862                         CERROR("%s: failed to free %s %llu: rc = %d\n",
 863                                osd->od_svname, buf, oid, rc);
 864         } else { /* asynchronous destroy */
 865                 char *key = info->oti_key;
 866
 867                 rc = osd_object_unlinked_add(obj, oh);
 868                 if (rc)
 869                         GOTO(out, rc);
 870
 871                 snprintf(key, sizeof(info->oti_key), "%llx", oid);
 872                 rc = osd_zap_add(osd, osd->od_unlinked->dn_object,
 873                                  osd->od_unlinked, key, 8, 1, &oid, oh->ot_tx);
 874                 if (rc)
 875                         CERROR("%s: zap_add_int() failed %s %llu: rc = %d\n",
 876                                osd->od_svname, buf, oid, rc);
 877         }
 878
 879         /* Remove the OI mapping after the destroy to handle the race with
 880          * OI scrub that may insert missed OI mapping during the interval. */
 881         rc = osd_zap_remove(osd, zapid, zdn, buf, oh->ot_tx);
 882         if (unlikely(rc == -ENOENT))
 883                 rc = 0;
 884         if (rc)
 885                 CERROR("%s: zap_remove(%s) failed: rc = %d\n",
 886                        osd->od_svname, buf, rc);
 887
 888         GOTO(out, rc);
 889
 890 out:
 891         /* not needed in the cache anymore */
 892         set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
 893         if (rc == 0)
 894                 obj->oo_destroyed = 1;
 895         up_write(&obj->oo_guard);
 896         RETURN (0);
 897 }
 898
 899 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
 900 {
 901         struct osd_object *obj = osd_obj(l);
 902         const struct lu_fid *fid = lu_object_fid(l);
 903
 904         if (obj->oo_dn) {
 905                 if (likely(!fid_is_acct(fid))) {
 906                         osd_object_sa_fini(obj);
 907                         if (obj->oo_sa_xattr) {
 908                                 nvlist_free(obj->oo_sa_xattr);
 909                                 obj->oo_sa_xattr = NULL;
 910                         }
 911                         osd_dnode_rele(obj->oo_dn);
 912                         list_del(&obj->oo_sa_linkage);
 913                 }
 914                 obj->oo_dn = NULL;
 915         }
 916 }
 917
 918 /*
 919  * Concurrency: ->loo_object_release() is called under site spin-lock.
 920  */
 921 static void osd_object_release(const struct lu_env *env,
 922                                struct lu_object *l)
 923 {
 924 }
 925
 926 /*
 927  * Concurrency: shouldn't matter.
 928  */
 929 static int osd_object_print(const struct lu_env *env, void *cookie,
 930                             lu_printer_t p, const struct lu_object *l)
 931 {
 932         struct osd_object *o = osd_obj(l);
 933
 934         return (*p)(env, cookie, LUSTRE_OSD_ZFS_NAME"-object@%p", o);
 935 }
 936
 937 static void osd_read_lock(const struct lu_env *env, struct dt_object *dt,
 938                           unsigned role)
 939 {
 940         struct osd_object *obj = osd_dt_obj(dt);
 941
 942         LASSERT(osd_invariant(obj));
 943
 944         down_read_nested(&obj->oo_sem, role);
 945 }
 946
 947 static void osd_write_lock(const struct lu_env *env, struct dt_object *dt,
 948                            unsigned role)
 949 {
 950         struct osd_object *obj = osd_dt_obj(dt);
 951
 952         LASSERT(osd_invariant(obj));
 953
 954         down_write_nested(&obj->oo_sem, role);
 955 }
 956
 957 static void osd_read_unlock(const struct lu_env *env, struct dt_object *dt)
 958 {
 959         struct osd_object *obj = osd_dt_obj(dt);
 960
 961         LASSERT(osd_invariant(obj));
 962         up_read(&obj->oo_sem);
 963 }
 964
 965 static void osd_write_unlock(const struct lu_env *env, struct dt_object *dt)
 966 {
 967         struct osd_object *obj = osd_dt_obj(dt);
 968
 969         LASSERT(osd_invariant(obj));
 970         up_write(&obj->oo_sem);
 971 }
 972
 973 static int osd_write_locked(const struct lu_env *env, struct dt_object *dt)
 974 {
 975         struct osd_object *obj = osd_dt_obj(dt);
 976         int rc = 1;
 977
 978         LASSERT(osd_invariant(obj));
 979
 980         if (down_write_trylock(&obj->oo_sem)) {
 981                 rc = 0;
 982                 up_write(&obj->oo_sem);
 983         }
 984         return rc;
 985 }
 986
 987 static int osd_attr_get(const struct lu_env *env,
 988                         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_flags |= LUSTRE_ORPHAN_FL;
1011         read_unlock(&obj->oo_attr_lock);
1012
1013         /* with ZFS_DEBUG zrl_add_debug() called by DB_DNODE_ENTER()
1014          * from within sa_object_size() can block on a mutex, so
1015          * we can't call sa_object_size() holding rwlock */
1016         sa_object_size(obj->oo_sa_hdl, &blksize, &blocks);
1017         /* we do not control size of indices, so always calculate
1018          * it from number of blocks reported by DMU */
1019         if (S_ISDIR(attr->la_mode))
1020                 attr->la_size = 512 * blocks;
1021         /* Block size may be not set; suggest maximal I/O transfers. */
1022         if (blksize == 0)
1023                 blksize = osd_spa_maxblocksize(
1024                         dmu_objset_spa(osd_obj2dev(obj)->od_os));
1025
1026         attr->la_blksize = blksize;
1027         attr->la_blocks = blocks;
1028         attr->la_valid |= LA_BLOCKS | LA_BLKSIZE;
1029
1030 out:
1031         up_read(&obj->oo_guard);
1032         return rc;
1033 }
1034
1035 /* Simple wrapper on top of qsd API which implement quota transfer for osd
1036  * setattr needs. As a reminder, only the root user can change ownership of
1037  * a file, that's why EDQUOT & EINPROGRESS errors are discarded */
1038 static inline int qsd_transfer(const struct lu_env *env,
1039                                struct qsd_instance *qsd,
1040                                struct lquota_trans *trans, int qtype,
1041                                __u64 orig_id, __u64 new_id, __u64 bspace,
1042                                struct lquota_id_info *qi, bool ignore_edquot)
1043 {
1044         int     rc;
1045
1046         if (unlikely(qsd == NULL))
1047                 return 0;
1048
1049         LASSERT(qtype >= 0 && qtype < LL_MAXQUOTAS);
1050         qi->lqi_type = qtype;
1051
1052         /* inode accounting */
1053         qi->lqi_is_blk = false;
1054
1055         /* one more inode for the new owner ... */
1056         qi->lqi_id.qid_uid = new_id;
1057         qi->lqi_space      = 1;
1058         rc = qsd_op_begin(env, qsd, trans, qi, NULL);
1059         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
1060                 rc = 0;
1061         if (rc)
1062                 return rc;
1063
1064         /* and one less inode for the current id */
1065         qi->lqi_id.qid_uid = orig_id;;
1066         qi->lqi_space      = -1;
1067         /* can't get EDQUOT when reducing usage */
1068         rc = qsd_op_begin(env, qsd, trans, qi, NULL);
1069         if (rc == -EINPROGRESS)
1070                 rc = 0;
1071         if (rc)
1072                 return rc;
1073
1074         /* block accounting */
1075         qi->lqi_is_blk = true;
1076
1077         /* more blocks for the new owner ... */
1078         qi->lqi_id.qid_uid = new_id;
1079         qi->lqi_space      = bspace;
1080         rc = qsd_op_begin(env, qsd, trans, qi, NULL);
1081         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
1082                 rc = 0;
1083         if (rc)
1084                 return rc;
1085
1086         /* and finally less blocks for the current owner */
1087         qi->lqi_id.qid_uid = orig_id;
1088         qi->lqi_space      = -bspace;
1089         rc = qsd_op_begin(env, qsd, trans, qi, NULL);
1090         /* can't get EDQUOT when reducing usage */
1091         if (rc == -EINPROGRESS)
1092                 rc = 0;
1093         return rc;
1094 }
1095
1096 static int osd_declare_attr_set(const struct lu_env *env,
1097                                 struct dt_object *dt,
1098                                 const struct lu_attr *attr,
1099                                 struct thandle *handle)
1100 {
1101         struct osd_thread_info  *info = osd_oti_get(env);
1102         struct osd_object       *obj = osd_dt_obj(dt);
1103         struct osd_device       *osd = osd_obj2dev(obj);
1104         dmu_tx_hold_t           *txh;
1105         struct osd_thandle      *oh;
1106         uint64_t                 bspace;
1107         uint32_t                 blksize;
1108         int                      rc = 0;
1109         bool                     found;
1110         ENTRY;
1111
1112
1113         LASSERT(handle != NULL);
1114         LASSERT(osd_invariant(obj));
1115
1116         oh = container_of0(handle, struct osd_thandle, ot_super);
1117
1118         down_read(&obj->oo_guard);
1119         if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
1120                 GOTO(out, rc = 0);
1121
1122         LASSERT(obj->oo_sa_hdl != NULL);
1123         LASSERT(oh->ot_tx != NULL);
1124         /* regular attributes are part of the bonus buffer */
1125         /* let's check whether this object is already part of
1126          * transaction.. */
1127         found = false;
1128         for (txh = list_head(&oh->ot_tx->tx_holds); txh;
1129              txh = list_next(&oh->ot_tx->tx_holds, txh)) {
1130                 if (txh->txh_dnode == NULL)
1131                         continue;
1132                 if (txh->txh_dnode->dn_object != obj->oo_dn->dn_object)
1133                         continue;
1134                 /* this object is part of the transaction already
1135                  * we don't need to declare bonus again */
1136                 found = true;
1137                 break;
1138         }
1139         if (!found)
1140                 dmu_tx_hold_bonus(oh->ot_tx, obj->oo_dn->dn_object);
1141         if (oh->ot_tx->tx_err != 0)
1142                 GOTO(out, rc = -oh->ot_tx->tx_err);
1143
1144         if (attr && attr->la_valid & LA_FLAGS) {
1145                 /* LMA is usually a part of bonus, no need to declare
1146                  * anything else */
1147         }
1148
1149         if (attr && (attr->la_valid & (LA_UID | LA_GID | LA_PROJID))) {
1150                 sa_object_size(obj->oo_sa_hdl, &blksize, &bspace);
1151                 bspace = toqb(bspace * 512);
1152
1153                 CDEBUG(D_QUOTA, "%s: enforce quota on UID %u, GID %u,"
1154                        "the quota space is %lld (%u)\n", osd->od_svname,
1155                        attr->la_uid, attr->la_gid, bspace, blksize);
1156         }
1157
1158         if (attr && attr->la_valid & LA_UID) {
1159                 /* quota enforcement for user */
1160                 if (attr->la_uid != obj->oo_attr.la_uid) {
1161                         rc = qsd_transfer(env, osd->od_quota_slave,
1162                                           &oh->ot_quota_trans, USRQUOTA,
1163                                           obj->oo_attr.la_uid, attr->la_uid,
1164                                           bspace, &info->oti_qi, true);
1165                         if (rc)
1166                                 GOTO(out, rc);
1167                 }
1168         }
1169         if (attr && attr->la_valid & LA_GID) {
1170                 /* quota enforcement for group */
1171                 if (attr->la_gid != obj->oo_attr.la_gid) {
1172                         rc = qsd_transfer(env, osd->od_quota_slave,
1173                                           &oh->ot_quota_trans, GRPQUOTA,
1174                                           obj->oo_attr.la_gid, attr->la_gid,
1175                                           bspace, &info->oti_qi,
1176                                           !(attr->la_flags &
1177                                                         LUSTRE_SET_SYNC_FL));
1178                         if (rc)
1179                                 GOTO(out, rc);
1180                 }
1181         }
1182 #ifdef ZFS_PROJINHERIT
1183         if (attr && attr->la_valid & LA_PROJID) {
1184                 if (!osd->od_projectused_dn)
1185                         GOTO(out, rc = -EOPNOTSUPP);
1186
1187                 /* Usually, if project quota is upgradable for the device,
1188                  * then the upgrade will be done before or when mount the
1189                  * device. So when we come here, this project should have
1190                  * project ID attribute already (that is zero by default).
1191                  * Otherwise, there was something wrong during the former
1192                  * upgrade, let's return failure to report that.
1193                  *
1194                  * Please note that, different from other attributes, you
1195                  * can NOT simply set the project ID attribute under such
1196                  * case, because adding (NOT change) project ID attribute
1197                  * needs to change the object's attribute layout to match
1198                  * zfs backend quota accounting requirement. */
1199                 if (unlikely(!obj->oo_with_projid))
1200                         GOTO(out, rc = -ENXIO);
1201
1202                 /* quota enforcement for project */
1203                 if (attr->la_projid != obj->oo_attr.la_projid) {
1204                         rc = qsd_transfer(env, osd->od_quota_slave,
1205                                           &oh->ot_quota_trans, PRJQUOTA,
1206                                           obj->oo_attr.la_projid,
1207                                           attr->la_projid, bspace,
1208                                           &info->oti_qi, true);
1209                         if (rc)
1210                                 GOTO(out, rc);
1211                 }
1212         }
1213 #endif
1214 out:
1215         up_read(&obj->oo_guard);
1216         RETURN(rc);
1217 }
1218
1219 /*
1220  * Set the attributes of an object
1221  *
1222  * The transaction passed to this routine must have
1223  * dmu_tx_hold_bonus(tx, oid) called and then assigned
1224  * to a transaction group.
1225  */
1226 static int osd_attr_set(const struct lu_env *env, struct dt_object *dt,
1227                         const struct lu_attr *la, struct thandle *handle)
1228 {
1229         struct osd_thread_info  *info = osd_oti_get(env);
1230         sa_bulk_attr_t          *bulk = osd_oti_get(env)->oti_attr_bulk;
1231         struct osd_object       *obj = osd_dt_obj(dt);
1232         struct osd_device       *osd = osd_obj2dev(obj);
1233         struct osd_thandle      *oh;
1234         struct osa_attr         *osa = &info->oti_osa;
1235         __u64                    valid = la->la_valid;
1236         int                      cnt;
1237         int                      rc = 0;
1238
1239         ENTRY;
1240
1241         down_read(&obj->oo_guard);
1242         if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
1243                 GOTO(out, rc = -ENOENT);
1244
1245         LASSERT(handle != NULL);
1246         LASSERT(osd_invariant(obj));
1247         LASSERT(obj->oo_sa_hdl);
1248
1249         oh = container_of0(handle, struct osd_thandle, ot_super);
1250         /* Assert that the transaction has been assigned to a
1251            transaction group. */
1252         LASSERT(oh->ot_tx->tx_txg != 0);
1253
1254         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING) && !osd->od_is_ost) {
1255                 struct zpl_direntry *zde = &info->oti_zde.lzd_reg;
1256                 char *buf = info->oti_str;
1257                 dnode_t *zdn = NULL;
1258                 uint64_t zapid;
1259
1260                 zapid = osd_get_name_n_idx(env, osd, lu_object_fid(&dt->do_lu),
1261                                            buf, sizeof(info->oti_str), &zdn);
1262                 rc = osd_zap_lookup(osd, zapid, zdn, buf, 8,
1263                                     sizeof(*zde) / 8, zde);
1264                 if (!rc) {
1265                         zde->zde_dnode -= 1;
1266                         rc = -zap_update(osd->od_os, zapid, buf, 8,
1267                                          sizeof(*zde) / 8, zde, oh->ot_tx);
1268                 }
1269                 up_read(&obj->oo_guard);
1270
1271                 RETURN(rc > 0 ? 0 : rc);
1272         }
1273
1274         /* Only allow set size for regular file */
1275         if (!S_ISREG(dt->do_lu.lo_header->loh_attr))
1276                 valid &= ~(LA_SIZE | LA_BLOCKS);
1277
1278         if (valid & LA_CTIME && la->la_ctime == obj->oo_attr.la_ctime)
1279                 valid &= ~LA_CTIME;
1280
1281         if (valid & LA_MTIME && la->la_mtime == obj->oo_attr.la_mtime)
1282                 valid &= ~LA_MTIME;
1283
1284         if (valid & LA_ATIME && la->la_atime == obj->oo_attr.la_atime)
1285                 valid &= ~LA_ATIME;
1286
1287         if (valid == 0)
1288                 GOTO(out, rc = 0);
1289
1290         if (valid & LA_FLAGS) {
1291                 struct lustre_mdt_attrs *lma;
1292                 struct lu_buf buf;
1293
1294                 if (la->la_flags & LUSTRE_LMA_FL_MASKS) {
1295                         LASSERT(!obj->oo_pfid_in_lma);
1296                         CLASSERT(sizeof(info->oti_buf) >= sizeof(*lma));
1297                         lma = (struct lustre_mdt_attrs *)&info->oti_buf;
1298                         buf.lb_buf = lma;
1299                         buf.lb_len = sizeof(info->oti_buf);
1300                         rc = osd_xattr_get(env, &obj->oo_dt, &buf,
1301                                            XATTR_NAME_LMA);
1302                         if (rc > 0) {
1303                                 lma->lma_incompat =
1304                                         le32_to_cpu(lma->lma_incompat);
1305                                 lma->lma_incompat |=
1306                                         lustre_to_lma_flags(la->la_flags);
1307                                 lma->lma_incompat =
1308                                         cpu_to_le32(lma->lma_incompat);
1309                                 buf.lb_buf = lma;
1310                                 buf.lb_len = sizeof(*lma);
1311                                 rc = osd_xattr_set_internal(env, obj, &buf,
1312                                                             XATTR_NAME_LMA,
1313                                                             LU_XATTR_REPLACE,
1314                                                             oh);
1315                         }
1316                         if (rc < 0) {
1317                                 CWARN("%s: failed to set LMA flags: rc = %d\n",
1318                                        osd->od_svname, rc);
1319                                 GOTO(out, rc);
1320                         }
1321                 }
1322         }
1323
1324         write_lock(&obj->oo_attr_lock);
1325         cnt = 0;
1326
1327         if (valid & LA_PROJID) {
1328 #ifdef ZFS_PROJINHERIT
1329                 /* osd_declare_attr_set() must be called firstly.
1330                  * If osd::od_projectused_dn is not set, then we
1331                  * can not arrive at here. */
1332                 LASSERT(osd->od_projectused_dn);
1333                 LASSERT(obj->oo_with_projid);
1334
1335                 osa->projid = obj->oo_attr.la_projid = la->la_projid;
1336                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_PROJID(osd), NULL,
1337                                  &osa->projid, 8);
1338 #else
1339                 valid &= ~LA_PROJID;
1340 #endif
1341         }
1342
1343         if (valid & LA_ATIME) {
1344                 osa->atime[0] = obj->oo_attr.la_atime = la->la_atime;
1345                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(osd), NULL,
1346                                  osa->atime, 16);
1347         }
1348         if (valid & LA_MTIME) {
1349                 osa->mtime[0] = obj->oo_attr.la_mtime = la->la_mtime;
1350                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(osd), NULL,
1351                                  osa->mtime, 16);
1352         }
1353         if (valid & LA_CTIME) {
1354                 osa->ctime[0] = obj->oo_attr.la_ctime = la->la_ctime;
1355                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(osd), NULL,
1356                                  osa->ctime, 16);
1357         }
1358         if (valid & LA_MODE) {
1359                 /* mode is stored along with type, so read it first */
1360                 obj->oo_attr.la_mode = (obj->oo_attr.la_mode & S_IFMT) |
1361                         (la->la_mode & ~S_IFMT);
1362                 osa->mode = obj->oo_attr.la_mode;
1363                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(osd), NULL,
1364                                  &osa->mode, 8);
1365         }
1366         if (valid & LA_SIZE) {
1367                 osa->size = obj->oo_attr.la_size = la->la_size;
1368                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(osd), NULL,
1369                                  &osa->size, 8);
1370         }
1371         if (valid & LA_NLINK) {
1372                 osa->nlink = obj->oo_attr.la_nlink = la->la_nlink;
1373                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(osd), NULL,
1374                                  &osa->nlink, 8);
1375         }
1376         if (valid & LA_RDEV) {
1377                 osa->rdev = obj->oo_attr.la_rdev = la->la_rdev;
1378                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_RDEV(osd), NULL,
1379                                  &osa->rdev, 8);
1380         }
1381         if (valid & LA_FLAGS) {
1382                 osa->flags = attrs_fs2zfs(la->la_flags);
1383                 /* many flags are not supported by zfs, so ensure a good cached
1384                  * copy */
1385                 obj->oo_attr.la_flags = attrs_zfs2fs(osa->flags);
1386 #ifdef ZFS_PROJINHERIT
1387                 if (obj->oo_with_projid)
1388                         osa->flags |= ZFS_PROJID;
1389 #endif
1390                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(osd), NULL,
1391                                  &osa->flags, 8);
1392         }
1393         if (valid & LA_UID) {
1394                 osa->uid = obj->oo_attr.la_uid = la->la_uid;
1395                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(osd), NULL,
1396                                  &osa->uid, 8);
1397         }
1398         if (valid & LA_GID) {
1399                 osa->gid = obj->oo_attr.la_gid = la->la_gid;
1400                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(osd), NULL,
1401                                  &osa->gid, 8);
1402         }
1403         obj->oo_attr.la_valid |= valid;
1404         write_unlock(&obj->oo_attr_lock);
1405
1406         LASSERT(cnt <= ARRAY_SIZE(osd_oti_get(env)->oti_attr_bulk));
1407         rc = osd_object_sa_bulk_update(obj, bulk, cnt, oh);
1408
1409 out:
1410         up_read(&obj->oo_guard);
1411         RETURN(rc);
1412 }
1413
1414 /*
1415  * Object creation.
1416  *
1417  * XXX temporary solution.
1418  */
1419
1420 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
1421                         struct dt_object *parent, struct dt_object *child,
1422                         umode_t child_mode)
1423 {
1424         LASSERT(ah);
1425
1426         ah->dah_parent = parent;
1427         ah->dah_mode = child_mode;
1428
1429         if (parent != NULL && !dt_object_remote(parent)) {
1430                 /* will help to find FID->ino at dt_insert("..") */
1431                 struct osd_object *pobj = osd_dt_obj(parent);
1432
1433                 osd_idc_find_and_init(env, osd_obj2dev(pobj), pobj);
1434         }
1435 }
1436
1437 static int osd_declare_create(const struct lu_env *env, struct dt_object *dt,
1438                               struct lu_attr *attr,
1439                               struct dt_allocation_hint *hint,
1440                               struct dt_object_format *dof,
1441                               struct thandle *handle)
1442 {
1443         const struct lu_fid     *fid = lu_object_fid(&dt->do_lu);
1444         struct osd_object       *obj = osd_dt_obj(dt);
1445         struct osd_device       *osd = osd_obj2dev(obj);
1446         struct osd_thandle      *oh;
1447         uint64_t                 zapid;
1448         dnode_t                 *dn;
1449         int                      rc, dnode_size;
1450         ENTRY;
1451
1452         LASSERT(dof);
1453
1454         switch (dof->dof_type) {
1455                 case DFT_REGULAR:
1456                 case DFT_SYM:
1457                 case DFT_NODE:
1458                         if (obj->oo_dt.do_body_ops == NULL)
1459                                 obj->oo_dt.do_body_ops = &osd_body_ops;
1460                         break;
1461                 default:
1462                         break;
1463         }
1464
1465         LASSERT(handle != NULL);
1466         oh = container_of0(handle, struct osd_thandle, ot_super);
1467         LASSERT(oh->ot_tx != NULL);
1468
1469         /* this is the minimum set of EAs on every Lustre object */
1470         obj->oo_ea_in_bonus = OSD_BASE_EA_IN_BONUS;
1471         /* reserve 32 bytes for extra stuff like ACLs */
1472         dnode_size = size_roundup_power2(obj->oo_ea_in_bonus + 32);
1473
1474         switch (dof->dof_type) {
1475                 case DFT_DIR:
1476                         dt->do_index_ops = &osd_dir_ops;
1477                 case DFT_INDEX:
1478                         /* for zap create */
1479                         dmu_tx_hold_zap(oh->ot_tx, DMU_NEW_OBJECT, FALSE, NULL);
1480                         dmu_tx_hold_sa_create(oh->ot_tx, dnode_size);
1481                         break;
1482                 case DFT_REGULAR:
1483                 case DFT_SYM:
1484                 case DFT_NODE:
1485                         /* first, we'll create new object */
1486                         dmu_tx_hold_sa_create(oh->ot_tx, dnode_size);
1487                         break;
1488
1489                 default:
1490                         LBUG();
1491                         break;
1492         }
1493
1494         /* and we'll add it to some mapping */
1495         zapid = osd_get_name_n_idx(env, osd, fid, NULL, 0, &dn);
1496         osd_tx_hold_zap(oh->ot_tx, zapid, dn, TRUE, NULL);
1497
1498         /* will help to find FID->ino mapping at dt_insert() */
1499         osd_idc_find_and_init(env, osd, obj);
1500
1501         rc = osd_declare_quota(env, osd, attr->la_uid, attr->la_gid,
1502                                attr->la_projid, 1, oh, NULL, OSD_QID_INODE);
1503
1504         RETURN(rc);
1505 }
1506
1507 int __osd_attr_init(const struct lu_env *env, struct osd_device *osd,
1508                     struct osd_object *obj, sa_handle_t *sa_hdl, dmu_tx_t *tx,
1509                     struct lu_attr *la, uint64_t parent,
1510                     nvlist_t *xattr)
1511 {
1512         sa_bulk_attr_t  *bulk = osd_oti_get(env)->oti_attr_bulk;
1513         struct osa_attr *osa = &osd_oti_get(env)->oti_osa;
1514         uint64_t         gen;
1515         uint64_t         crtime[2];
1516         timestruc_t      now;
1517         int              cnt;
1518         int              rc;
1519         char *dxattr = NULL;
1520         size_t sa_size;
1521
1522
1523         LASSERT(sa_hdl);
1524
1525         gen = dmu_tx_get_txg(tx);
1526         gethrestime(&now);
1527         ZFS_TIME_ENCODE(&now, crtime);
1528
1529         osa->atime[0] = la->la_atime;
1530         osa->ctime[0] = la->la_ctime;
1531         osa->mtime[0] = la->la_mtime;
1532         osa->mode = la->la_mode;
1533         osa->uid = la->la_uid;
1534         osa->gid = la->la_gid;
1535         osa->rdev = la->la_rdev;
1536         osa->nlink = la->la_nlink;
1537         if (la->la_valid & LA_FLAGS)
1538                 osa->flags = attrs_fs2zfs(la->la_flags);
1539         else
1540                 osa->flags = 0;
1541         osa->size  = la->la_size;
1542 #ifdef ZFS_PROJINHERIT
1543         if (osd->od_projectused_dn) {
1544                 if (la->la_valid & LA_PROJID)
1545                         osa->projid = la->la_projid;
1546                 else
1547                         osa->projid = ZFS_DEFAULT_PROJID;
1548                 osa->flags |= ZFS_PROJID;
1549                 if (obj)
1550                         obj->oo_with_projid = 1;
1551         } else {
1552                 osa->flags &= ~ZFS_PROJID;
1553         }
1554 #endif
1555
1556         /*
1557          * we need to create all SA below upon object create.
1558          *
1559          * XXX The attribute order matters since the accounting callback relies
1560          * on static offsets (i.e. SA_*_OFFSET, see zfs_space_delta_cb()) to
1561          * look up the UID/GID/PROJID attributes. Moreover, the callback does
1562          * not seem to support the spill block.
1563          * We define attributes in the same order as SA_*_OFFSET in order to
1564          * work around the problem. See ORI-610.
1565          */
1566         cnt = 0;
1567         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(osd), NULL, &osa->mode, 8);
1568         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(osd), NULL, &osa->size, 8);
1569         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GEN(osd), NULL, &gen, 8);
1570         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(osd), NULL, &osa->uid, 8);
1571         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(osd), NULL, &osa->gid, 8);
1572         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_PARENT(osd), NULL, &parent, 8);
1573         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(osd), NULL, &osa->flags, 8);
1574         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(osd), NULL, osa->atime, 16);
1575         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(osd), NULL, osa->mtime, 16);
1576         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(osd), NULL, osa->ctime, 16);
1577         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CRTIME(osd), NULL, crtime, 16);
1578         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(osd), NULL, &osa->nlink, 8);
1579 #ifdef ZFS_PROJINHERIT
1580         if (osd->od_projectused_dn)
1581                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_PROJID(osd), NULL,
1582                                  &osa->projid, 8);
1583 #endif
1584         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_RDEV(osd), NULL, &osa->rdev, 8);
1585         LASSERT(cnt <= ARRAY_SIZE(osd_oti_get(env)->oti_attr_bulk));
1586
1587         if (xattr) {
1588                 rc = -nvlist_size(xattr, &sa_size, NV_ENCODE_XDR);
1589                 LASSERT(rc == 0);
1590
1591                 dxattr = osd_zio_buf_alloc(sa_size);
1592                 LASSERT(dxattr);
1593
1594                 rc = -nvlist_pack(xattr, &dxattr, &sa_size,
1595                                 NV_ENCODE_XDR, KM_SLEEP);
1596                 LASSERT(rc == 0);
1597
1598                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_DXATTR(osd),
1599                                 NULL, dxattr, sa_size);
1600         }
1601
1602         rc = -sa_replace_all_by_template(sa_hdl, bulk, cnt, tx);
1603         if (dxattr)
1604                 osd_zio_buf_free(dxattr, sa_size);
1605
1606         return rc;
1607 }
1608
1609 int osd_find_new_dnode(const struct lu_env *env, dmu_tx_t *tx,
1610                        uint64_t oid, dnode_t **dnp)
1611 {
1612         dmu_tx_hold_t *txh;
1613         int rc = 0;
1614
1615         /* take dnode_t from tx to save on dnode#->dnode_t lookup */
1616         for (txh = list_tail(&tx->tx_holds); txh;
1617              txh = list_prev(&tx->tx_holds, txh)) {
1618                 dnode_t *dn = txh->txh_dnode;
1619                 dmu_buf_impl_t *db;
1620
1621                 if (dn == NULL)
1622                         continue;
1623                 if (dn->dn_object != oid)
1624                         continue;
1625                 db = dn->dn_bonus;
1626                 if (db == NULL) {
1627                         rw_enter(&dn->dn_struct_rwlock, RW_WRITER);
1628                         if (dn->dn_bonus == NULL)
1629                                 dbuf_create_bonus(dn);
1630                         rw_exit(&dn->dn_struct_rwlock);
1631                 }
1632                 db = dn->dn_bonus;
1633                 LASSERT(db);
1634                 LASSERT(dn->dn_handle);
1635                 DB_DNODE_ENTER(db);
1636                 if (refcount_add(&db->db_holds, osd_obj_tag) == 1) {
1637                         refcount_add(&dn->dn_holds, tag);
1638                         atomic_inc_32(&dn->dn_dbufs_count);
1639                 }
1640                 *dnp = dn;
1641                 dbuf_read(db, NULL, DB_RF_MUST_SUCCEED | DB_RF_NOPREFETCH);
1642                 break;
1643         }
1644
1645         if (unlikely(*dnp == NULL))
1646                 rc = __osd_obj2dnode(tx->tx_objset, oid, dnp);
1647
1648         return rc;
1649 }
1650
1651 #ifdef HAVE_DMU_OBJECT_ALLOC_DNSIZE
1652 int osd_find_dnsize(struct osd_device *osd, int ea_in_bonus)
1653 {
1654         int dnsize;
1655
1656         if (osd->od_dnsize == ZFS_DNSIZE_AUTO) {
1657                 dnsize = DNODE_MIN_SIZE;
1658                 do {
1659                         if (DN_BONUS_SIZE(dnsize) >= ea_in_bonus + 32)
1660                                 break;
1661                         dnsize <<= 1;
1662                 } while (dnsize < DNODE_MAX_SIZE);
1663                 if (dnsize > DNODE_MAX_SIZE)
1664                         dnsize = DNODE_MAX_SIZE;
1665         } else if (osd->od_dnsize == ZFS_DNSIZE_1K) {
1666                 dnsize = 1024;
1667         } else if (osd->od_dnsize == ZFS_DNSIZE_2K) {
1668                 dnsize = 2048;
1669         } else if (osd->od_dnsize == ZFS_DNSIZE_4K) {
1670                 dnsize = 4096;
1671         } else if (osd->od_dnsize == ZFS_DNSIZE_8K) {
1672                 dnsize = 8192;
1673         } else if (osd->od_dnsize == ZFS_DNSIZE_16K) {
1674                 dnsize = 16384;
1675         } else {
1676                 dnsize = DNODE_MIN_SIZE;
1677         }
1678         return dnsize;
1679 }
1680 #endif
1681
1682 /*
1683  * The transaction passed to this routine must have
1684  * dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT) called and then assigned
1685  * to a transaction group.
1686  */
1687 int __osd_object_create(const struct lu_env *env, struct osd_device *osd,
1688                         struct osd_object *obj, const struct lu_fid *fid,
1689                         dnode_t **dnp, dmu_tx_t *tx, struct lu_attr *la)
1690 {
1691         dmu_object_type_t type = DMU_OT_PLAIN_FILE_CONTENTS;
1692         uint64_t oid;
1693         int size;
1694
1695         /* Use DMU_OTN_UINT8_METADATA for local objects so their data blocks
1696          * would get an additional ditto copy */
1697         if (unlikely(S_ISREG(la->la_mode) &&
1698                      fid_seq_is_local_file(fid_seq(fid))))
1699                 type = DMU_OTN_UINT8_METADATA;
1700
1701         /* Create a new DMU object using the default dnode size. */
1702         if (obj)
1703                 size = obj->oo_ea_in_bonus;
1704         else
1705                 size = OSD_BASE_EA_IN_BONUS;
1706         oid = osd_dmu_object_alloc(osd->od_os, type, 0,
1707                                    osd_find_dnsize(osd, size), tx);
1708
1709         LASSERT(la->la_valid & LA_MODE);
1710         la->la_size = 0;
1711         la->la_nlink = 1;
1712
1713         return osd_find_new_dnode(env, tx, oid, dnp);
1714 }
1715
1716 /*
1717  * The transaction passed to this routine must have
1718  * dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, ...) called and then assigned
1719  * to a transaction group.
1720  *
1721  * Using ZAP_FLAG_HASH64 will force the ZAP to always be a FAT ZAP.
1722  * This is fine for directories today, because storing the FID in the dirent
1723  * will also require a FAT ZAP.  If there is a new type of micro ZAP created
1724  * then we might need to re-evaluate the use of this flag and instead do
1725  * a conversion from the different internal ZAP hash formats being used. */
1726 int __osd_zap_create(const struct lu_env *env, struct osd_device *osd,
1727                      dnode_t **dnp, dmu_tx_t *tx, struct lu_attr *la,
1728                      unsigned dnsize, zap_flags_t flags)
1729 {
1730         uint64_t oid;
1731
1732         /* Assert that the transaction has been assigned to a
1733            transaction group. */
1734         LASSERT(tx->tx_txg != 0);
1735         *dnp = NULL;
1736
1737         oid = osd_zap_create_flags(osd->od_os, 0, flags | ZAP_FLAG_HASH64,
1738                                    DMU_OT_DIRECTORY_CONTENTS,
1739                                    14, /* == ZFS fzap_default_blockshift */
1740                                    DN_MAX_INDBLKSHIFT, /* indirect blockshift */
1741                                    dnsize, tx);
1742
1743         la->la_size = 2;
1744         la->la_nlink = 1;
1745
1746         return osd_find_new_dnode(env, tx, oid, dnp);
1747 }
1748
1749 static dnode_t *osd_mkidx(const struct lu_env *env, struct osd_object *obj,
1750                           struct lu_attr *la, struct osd_thandle *oh)
1751 {
1752         struct osd_device *osd = osd_obj2dev(obj);
1753         dnode_t *dn;
1754         int rc;
1755
1756         /* Index file should be created as regular file in order not to confuse
1757          * ZPL which could interpret them as directory.
1758          * We set ZAP_FLAG_UINT64_KEY to let ZFS know than we are going to use
1759          * binary keys */
1760         LASSERT(S_ISREG(la->la_mode));
1761         rc = __osd_zap_create(env, osd, &dn, oh->ot_tx, la,
1762                 osd_find_dnsize(osd, obj->oo_ea_in_bonus), ZAP_FLAG_UINT64_KEY);
1763         if (rc)
1764                 return ERR_PTR(rc);
1765         return dn;
1766 }
1767
1768 static dnode_t *osd_mkdir(const struct lu_env *env, struct osd_object *obj,
1769                           struct lu_attr *la, struct osd_thandle *oh)
1770 {
1771         struct osd_device *osd = osd_obj2dev(obj);
1772         dnode_t *dn;
1773         int rc;
1774
1775         LASSERT(S_ISDIR(la->la_mode));
1776         rc = __osd_zap_create(env, osd, &dn, oh->ot_tx, la,
1777                               osd_find_dnsize(osd, obj->oo_ea_in_bonus), 0);
1778         if (rc)
1779                 return ERR_PTR(rc);
1780         return dn;
1781 }
1782
1783 static dnode_t *osd_mkreg(const struct lu_env *env, struct osd_object *obj,
1784                           struct lu_attr *la, struct osd_thandle *oh)
1785 {
1786         const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
1787         struct osd_device *osd = osd_obj2dev(obj);
1788         dnode_t *dn;
1789         int rc;
1790
1791         LASSERT(S_ISREG(la->la_mode));
1792         rc = __osd_object_create(env, osd, obj, fid, &dn, oh->ot_tx, la);
1793         if (rc)
1794                 return ERR_PTR(rc);
1795
1796         if ((fid_is_idif(fid) || fid_is_norm(fid) || fid_is_echo(fid))) {
1797                 /* The minimum block size must be at least page size otherwise
1798                  * it will break the assumption in tgt_thread_big_cache where
1799                  * the array size is PTLRPC_MAX_BRW_PAGES. It will also affect
1800                  * RDMA due to subpage transfer size */
1801                 rc = -dmu_object_set_blocksize(osd->od_os, dn->dn_object,
1802                                                PAGE_SIZE, 0, oh->ot_tx);
1803                 if (unlikely(rc)) {
1804                         CERROR("%s: can't change blocksize: %d\n",
1805                                osd->od_svname, rc);
1806                         return ERR_PTR(rc);
1807                 }
1808         }
1809
1810         return dn;
1811 }
1812
1813 static dnode_t *osd_mksym(const struct lu_env *env, struct osd_object *obj,
1814                           struct lu_attr *la, struct osd_thandle *oh)
1815 {
1816         dnode_t *dn;
1817         int rc;
1818
1819         LASSERT(S_ISLNK(la->la_mode));
1820         rc = __osd_object_create(env, osd_obj2dev(obj), obj,
1821                                  lu_object_fid(&obj->oo_dt.do_lu),
1822                                  &dn, oh->ot_tx, la);
1823         if (rc)
1824                 return ERR_PTR(rc);
1825         return dn;
1826 }
1827
1828 static dnode_t *osd_mknod(const struct lu_env *env, struct osd_object *obj,
1829                           struct lu_attr *la, struct osd_thandle *oh)
1830 {
1831         dnode_t *dn;
1832         int rc;
1833
1834         if (S_ISCHR(la->la_mode) || S_ISBLK(la->la_mode))
1835                 la->la_valid |= LA_RDEV;
1836
1837         rc = __osd_object_create(env, osd_obj2dev(obj), obj,
1838                                  lu_object_fid(&obj->oo_dt.do_lu),
1839                                  &dn, oh->ot_tx, la);
1840         if (rc)
1841                 return ERR_PTR(rc);
1842         return dn;
1843 }
1844
1845 typedef dnode_t *(*osd_obj_type_f)(const struct lu_env *env,
1846                                    struct osd_object *obj,
1847                                    struct lu_attr *la,
1848                                    struct osd_thandle *oh);
1849
1850 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
1851 {
1852         osd_obj_type_f result;
1853
1854         switch (type) {
1855         case DFT_DIR:
1856                 result = osd_mkdir;
1857                 break;
1858         case DFT_INDEX:
1859                 result = osd_mkidx;
1860                 break;
1861         case DFT_REGULAR:
1862                 result = osd_mkreg;
1863                 break;
1864         case DFT_SYM:
1865                 result = osd_mksym;
1866                 break;
1867         case DFT_NODE:
1868                 result = osd_mknod;
1869                 break;
1870         default:
1871                 LBUG();
1872                 break;
1873         }
1874         return result;
1875 }
1876
1877 /*
1878  * Concurrency: @dt is write locked.
1879  */
1880 static int osd_create(const struct lu_env *env, struct dt_object *dt,
1881                       struct lu_attr *attr, struct dt_allocation_hint *hint,
1882                       struct dt_object_format *dof, struct thandle *th)
1883 {
1884         struct osd_thread_info  *info = osd_oti_get(env);
1885         struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
1886         struct zpl_direntry     *zde = &info->oti_zde.lzd_reg;
1887         const struct lu_fid     *fid = lu_object_fid(&dt->do_lu);
1888         struct osd_object       *obj = osd_dt_obj(dt);
1889         struct osd_device       *osd = osd_obj2dev(obj);
1890         char                    *buf = info->oti_str;
1891         struct osd_thandle      *oh;
1892         dnode_t *dn = NULL, *zdn = NULL;
1893         uint64_t                 zapid, parent = 0;
1894         int                      rc;
1895         __u32 compat = 0;
1896
1897         ENTRY;
1898
1899         LASSERT(!fid_is_acct(fid));
1900
1901         /* concurrent create declarations should not see
1902          * the object inconsistent (db, attr, etc).
1903          * in regular cases acquisition should be cheap */
1904         down_write(&obj->oo_guard);
1905
1906         if (unlikely(dt_object_exists(dt)))
1907                 GOTO(out, rc = -EEXIST);
1908
1909         LASSERT(osd_invariant(obj));
1910         LASSERT(dof != NULL);
1911
1912         LASSERT(th != NULL);
1913         oh = container_of0(th, struct osd_thandle, ot_super);
1914
1915         LASSERT(obj->oo_dn == NULL);
1916
1917         /* to follow ZFS on-disk format we need
1918          * to initialize parent dnode properly */
1919         if (hint != NULL && hint->dah_parent != NULL &&
1920             !dt_object_remote(hint->dah_parent))
1921                 parent = osd_dt_obj(hint->dah_parent)->oo_dn->dn_object;
1922
1923         /* we may fix some attributes, better do not change the source */
1924         obj->oo_attr = *attr;
1925         obj->oo_attr.la_valid |= LA_SIZE | LA_NLINK | LA_TYPE;
1926
1927 #ifdef ZFS_PROJINHERIT
1928         if (osd->od_projectused_dn) {
1929                 if (!(obj->oo_attr.la_valid & LA_PROJID))
1930                         obj->oo_attr.la_projid = ZFS_DEFAULT_PROJID;
1931                 obj->oo_with_projid = 1;
1932         }
1933 #endif
1934
1935         dn = osd_create_type_f(dof->dof_type)(env, obj, &obj->oo_attr, oh);
1936         if (IS_ERR(dn)) {
1937                 rc = PTR_ERR(dn);
1938                 dn = NULL;
1939                 GOTO(out, rc);
1940         }
1941
1942         zde->zde_pad = 0;
1943         zde->zde_dnode = dn->dn_object;
1944         zde->zde_type = IFTODT(attr->la_mode & S_IFMT);
1945
1946         zapid = osd_get_name_n_idx(env, osd, fid, buf,
1947                                    sizeof(info->oti_str), &zdn);
1948         if (CFS_FAIL_CHECK(OBD_FAIL_OSD_NO_OI_ENTRY) ||
1949             (osd->od_is_ost && OBD_FAIL_CHECK(OBD_FAIL_OSD_COMPAT_NO_ENTRY)))
1950                 goto skip_add;
1951
1952         if (osd->od_is_ost && OBD_FAIL_CHECK(OBD_FAIL_OSD_COMPAT_INVALID_ENTRY))
1953                 zde->zde_dnode++;
1954
1955         rc = osd_zap_add(osd, zapid, zdn, buf, 8, 1, zde, oh->ot_tx);
1956         if (rc)
1957                 GOTO(out, rc);
1958
1959 skip_add:
1960         obj->oo_dn = dn;
1961         /* Now add in all of the "SA" attributes */
1962         rc = osd_sa_handle_get(obj);
1963         if (rc)
1964                 GOTO(out, rc);
1965
1966         rc = -nvlist_alloc(&obj->oo_sa_xattr, NV_UNIQUE_NAME, KM_SLEEP);
1967         if (rc)
1968                 GOTO(out, rc);
1969
1970         /* initialize LMA */
1971         if (fid_is_idif(fid) || (fid_is_norm(fid) && osd->od_is_ost))
1972                 compat |= LMAC_FID_ON_OST;
1973         lustre_lma_init(lma, fid, compat, 0);
1974         lustre_lma_swab(lma);
1975         rc = -nvlist_add_byte_array(obj->oo_sa_xattr, XATTR_NAME_LMA,
1976                                     (uchar_t *)lma, sizeof(*lma));
1977         if (rc)
1978                 GOTO(out, rc);
1979
1980         /* configure new osd object */
1981         obj->oo_parent = parent != 0 ? parent : zapid;
1982         obj->oo_late_attr_set = 1;
1983         rc = __osd_sa_xattr_schedule_update(env, obj, oh);
1984         if (rc)
1985                 GOTO(out, rc);
1986
1987         /* XXX: oo_lma_flags */
1988         obj->oo_dt.do_lu.lo_header->loh_attr |= obj->oo_attr.la_mode & S_IFMT;
1989         if (likely(!fid_is_acct(lu_object_fid(&obj->oo_dt.do_lu))))
1990                 /* no body operations for accounting objects */
1991                 obj->oo_dt.do_body_ops = &osd_body_ops;
1992
1993         osd_idc_find_and_init(env, osd, obj);
1994
1995 out:
1996         if (unlikely(rc && dn)) {
1997                 dmu_object_free(osd->od_os, dn->dn_object, oh->ot_tx);
1998                 osd_dnode_rele(dn);
1999                 obj->oo_dn = NULL;
2000         } else if (!rc) {
2001                 obj->oo_dt.do_lu.lo_header->loh_attr |= LOHA_EXISTS;
2002         }
2003         up_write(&obj->oo_guard);
2004         RETURN(rc);
2005 }
2006
2007 static int osd_declare_ref_add(const struct lu_env *env, struct dt_object *dt,
2008                                struct thandle *th)
2009 {
2010         osd_idc_find_and_init(env, osd_dev(dt->do_lu.lo_dev), osd_dt_obj(dt));
2011         return osd_declare_attr_set(env, dt, NULL, th);
2012 }
2013
2014 /*
2015  * Concurrency: @dt is write locked.
2016  */
2017 static int osd_ref_add(const struct lu_env *env, struct dt_object *dt,
2018                        struct thandle *handle)
2019 {
2020         struct osd_object       *obj = osd_dt_obj(dt);
2021         struct osd_thandle      *oh;
2022         struct osd_device       *osd = osd_obj2dev(obj);
2023         uint64_t                 nlink;
2024         int rc;
2025
2026         ENTRY;
2027
2028         down_read(&obj->oo_guard);
2029         if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
2030                 GOTO(out, rc = -ENOENT);
2031
2032         LASSERT(osd_invariant(obj));
2033         LASSERT(obj->oo_sa_hdl != NULL);
2034
2035         oh = container_of0(handle, struct osd_thandle, ot_super);
2036
2037         write_lock(&obj->oo_attr_lock);
2038         nlink = ++obj->oo_attr.la_nlink;
2039         write_unlock(&obj->oo_attr_lock);
2040
2041         rc = osd_object_sa_update(obj, SA_ZPL_LINKS(osd), &nlink, 8, oh);
2042
2043 out:
2044         up_read(&obj->oo_guard);
2045         RETURN(rc);
2046 }
2047
2048 static int osd_declare_ref_del(const struct lu_env *env, struct dt_object *dt,
2049                                struct thandle *handle)
2050 {
2051         osd_idc_find_and_init(env, osd_dev(dt->do_lu.lo_dev), osd_dt_obj(dt));
2052         return osd_declare_attr_set(env, dt, NULL, handle);
2053 }
2054
2055 /*
2056  * Concurrency: @dt is write locked.
2057  */
2058 static int osd_ref_del(const struct lu_env *env, struct dt_object *dt,
2059                        struct thandle *handle)
2060 {
2061         struct osd_object       *obj = osd_dt_obj(dt);
2062         struct osd_thandle      *oh;
2063         struct osd_device       *osd = osd_obj2dev(obj);
2064         uint64_t                 nlink;
2065         int                      rc;
2066
2067         ENTRY;
2068
2069         down_read(&obj->oo_guard);
2070
2071         if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
2072                 GOTO(out, rc = -ENOENT);
2073
2074         LASSERT(osd_invariant(obj));
2075         LASSERT(obj->oo_sa_hdl != NULL);
2076
2077         oh = container_of0(handle, struct osd_thandle, ot_super);
2078         LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
2079
2080         write_lock(&obj->oo_attr_lock);
2081         nlink = --obj->oo_attr.la_nlink;
2082         write_unlock(&obj->oo_attr_lock);
2083
2084         rc = osd_object_sa_update(obj, SA_ZPL_LINKS(osd), &nlink, 8, oh);
2085
2086 out:
2087         up_read(&obj->oo_guard);
2088         RETURN(rc);
2089 }
2090
2091 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
2092                            __u64 start, __u64 end)
2093 {
2094         struct osd_device *osd = osd_obj2dev(osd_dt_obj(dt));
2095         ENTRY;
2096
2097         /* XXX: no other option than syncing the whole filesystem until we
2098          * support ZIL.  If the object tracked the txg that it was last
2099          * modified in, it could pass that txg here instead of "0".  Maybe
2100          * the changes are already committed, so no wait is needed at all? */
2101         if (!osd->od_dt_dev.dd_rdonly) {
2102                 if (osd_object_sync_delay_us < 0)
2103                         txg_wait_synced(dmu_objset_pool(osd->od_os), 0ULL);
2104                 else
2105                         udelay(osd_object_sync_delay_us);
2106         }
2107
2108         RETURN(0);
2109 }
2110
2111 static int osd_invalidate(const struct lu_env *env, struct dt_object *dt)
2112 {
2113         return 0;
2114 }
2115
2116 static struct dt_object_operations osd_obj_ops = {
2117         .do_read_lock           = osd_read_lock,
2118         .do_write_lock          = osd_write_lock,
2119         .do_read_unlock         = osd_read_unlock,
2120         .do_write_unlock        = osd_write_unlock,
2121         .do_write_locked        = osd_write_locked,
2122         .do_attr_get            = osd_attr_get,
2123         .do_declare_attr_set    = osd_declare_attr_set,
2124         .do_attr_set            = osd_attr_set,
2125         .do_ah_init             = osd_ah_init,
2126         .do_declare_create      = osd_declare_create,
2127         .do_create              = osd_create,
2128         .do_declare_destroy     = osd_declare_destroy,
2129         .do_destroy             = osd_destroy,
2130         .do_index_try           = osd_index_try,
2131         .do_declare_ref_add     = osd_declare_ref_add,
2132         .do_ref_add             = osd_ref_add,
2133         .do_declare_ref_del     = osd_declare_ref_del,
2134         .do_ref_del             = osd_ref_del,
2135         .do_xattr_get           = osd_xattr_get,
2136         .do_declare_xattr_set   = osd_declare_xattr_set,
2137         .do_xattr_set           = osd_xattr_set,
2138         .do_declare_xattr_del   = osd_declare_xattr_del,
2139         .do_xattr_del           = osd_xattr_del,
2140         .do_xattr_list          = osd_xattr_list,
2141         .do_object_sync         = osd_object_sync,
2142         .do_invalidate          = osd_invalidate,
2143 };
2144
2145 static struct lu_object_operations osd_lu_obj_ops = {
2146         .loo_object_init        = osd_object_init,
2147         .loo_object_delete      = osd_object_delete,
2148         .loo_object_release     = osd_object_release,
2149         .loo_object_free        = osd_object_free,
2150         .loo_object_print       = osd_object_print,
2151         .loo_object_invariant   = osd_object_invariant,
2152 };
2153
2154 static int osd_otable_it_attr_get(const struct lu_env *env,
2155                                 struct dt_object *dt,
2156                                 struct lu_attr *attr)
2157 {
2158         attr->la_valid = 0;
2159         return 0;
2160 }
2161
2162 static struct dt_object_operations osd_obj_otable_it_ops = {
2163         .do_attr_get            = osd_otable_it_attr_get,
2164         .do_index_try           = osd_index_try,
2165 };
2166
2167 module_param(osd_object_sync_delay_us, int, 0644);
2168 MODULE_PARM_DESC(osd_object_sync_delay_us,
2169                  "If zero or larger delay N usec instead of doing object sync");