lustre/osd-zfs/osd_object.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2009, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2012, 2017, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  * Lustre is a trademark of Sun Microsystems, Inc.
  31  *
  32  * lustre/osd-zfs/osd_object.c
  33  *
  34  * Author: Alex Zhuravlev <bzzz@whamcloud.com>
  35  * Author: Mike Pershin <tappro@whamcloud.com>
  36  * Author: Johann Lombardi <johann@whamcloud.com>
  37  */
  38
  39 #define DEBUG_SUBSYSTEM S_OSD
  40
  41 #include <libcfs/libcfs.h>
  42 #include <obd_support.h>
  43 #include <lustre_net.h>
  44 #include <obd.h>
  45 #include <obd_class.h>
  46 #include <lustre_disk.h>
  47 #include <lustre_fid.h>
  48
  49 #include "osd_internal.h"
  50
  51 #include <sys/dnode.h>
  52 #include <sys/dbuf.h>
  53 #include <sys/spa.h>
  54 #include <sys/stat.h>
  55 #include <sys/zap.h>
  56 #include <sys/spa_impl.h>
  57 #include <sys/zfs_znode.h>
  58 #include <sys/dmu_tx.h>
  59 #include <sys/dmu_objset.h>
  60 #include <sys/dsl_prop.h>
  61 #include <sys/sa_impl.h>
  62 #include <sys/txg.h>
  63
  64 char *osd_obj_tag = "osd_object";
  65 static int osd_object_sync_delay_us = -1;
  66
  67 static struct dt_object_operations osd_obj_ops;
  68 static struct lu_object_operations osd_lu_obj_ops;
  69 static struct dt_object_operations osd_obj_otable_it_ops;
  70
  71 static void
  72 osd_object_sa_fini(struct osd_object *obj)
  73 {
  74         if (obj->oo_sa_hdl) {
  75                 sa_handle_destroy(obj->oo_sa_hdl);
  76                 obj->oo_sa_hdl = NULL;
  77         }
  78 }
  79
  80 static int
  81 osd_object_sa_init(struct osd_object *obj, struct osd_device *o)
  82 {
  83         int rc;
  84
  85         LASSERT(obj->oo_sa_hdl == NULL);
  86         LASSERT(obj->oo_dn != NULL);
  87
  88         rc = osd_sa_handle_get(obj);
  89         if (rc)
  90                 return rc;
  91
  92         /* Cache the xattr object id, valid for the life of the object */
  93         rc = -sa_lookup(obj->oo_sa_hdl, SA_ZPL_XATTR(o), &obj->oo_xattr, 8);
  94         if (rc == -ENOENT) {
  95                 obj->oo_xattr = ZFS_NO_OBJECT;
  96                 rc = 0;
  97         } else if (rc) {
  98                 osd_object_sa_fini(obj);
  99         }
 100
 101         return rc;
 102 }
 103
 104 /*
 105  * Add object to list of dirty objects in tx handle.
 106  */
 107 void osd_object_sa_dirty_add(struct osd_object *obj, struct osd_thandle *oh)
 108 {
 109         if (!list_empty(&obj->oo_sa_linkage))
 110                 return;
 111
 112         write_lock(&obj->oo_attr_lock);
 113         if (likely(list_empty(&obj->oo_sa_linkage)))
 114                 list_add(&obj->oo_sa_linkage, &oh->ot_sa_list);
 115         write_unlock(&obj->oo_attr_lock);
 116 }
 117
 118 /*
 119  * Release spill block dbuf hold for all dirty SAs.
 120  */
 121 void osd_object_sa_dirty_rele(const struct lu_env *env, struct osd_thandle *oh)
 122 {
 123         struct osd_object *obj;
 124
 125         while (!list_empty(&oh->ot_sa_list)) {
 126                 obj = list_entry(oh->ot_sa_list.next,
 127                                  struct osd_object, oo_sa_linkage);
 128                 write_lock(&obj->oo_attr_lock);
 129                 list_del_init(&obj->oo_sa_linkage);
 130                 write_unlock(&obj->oo_attr_lock);
 131                 if (obj->oo_late_xattr) {
 132                         /*
 133                          * take oo_guard to protect oo_sa_xattr buffer
 134                          * from concurrent update by osd_xattr_set()
 135                          */
 136                         LASSERT(oh->ot_assigned != 0);
 137                         down_write(&obj->oo_guard);
 138                         if (obj->oo_late_attr_set)
 139                                 __osd_sa_attr_init(env, obj, oh);
 140                         else if (obj->oo_late_xattr)
 141                                 __osd_sa_xattr_update(env, obj, oh);
 142                         up_write(&obj->oo_guard);
 143                 }
 144                 sa_spill_rele(obj->oo_sa_hdl);
 145         }
 146 }
 147
 148 /*
 149  * Update the SA and add the object to the dirty list.
 150  */
 151 int osd_object_sa_update(struct osd_object *obj, sa_attr_type_t type,
 152                          void *buf, uint32_t buflen, struct osd_thandle *oh)
 153 {
 154         int rc;
 155
 156         LASSERT(obj->oo_sa_hdl != NULL);
 157         LASSERT(oh->ot_tx != NULL);
 158
 159         rc = -sa_update(obj->oo_sa_hdl, type, buf, buflen, oh->ot_tx);
 160         osd_object_sa_dirty_add(obj, oh);
 161
 162         return rc;
 163 }
 164
 165 /*
 166  * Bulk update the SA and add the object to the dirty list.
 167  */
 168 static int
 169 osd_object_sa_bulk_update(struct osd_object *obj, sa_bulk_attr_t *attrs,
 170                           int count, struct osd_thandle *oh)
 171 {
 172         int rc;
 173
 174         LASSERT(obj->oo_sa_hdl != NULL);
 175         LASSERT(oh->ot_tx != NULL);
 176
 177         rc = -sa_bulk_update(obj->oo_sa_hdl, attrs, count, oh->ot_tx);
 178         osd_object_sa_dirty_add(obj, oh);
 179
 180         return rc;
 181 }
 182
 183 /*
 184  * Retrieve the attributes of a DMU object
 185  */
 186 static int __osd_object_attr_get(const struct lu_env *env, struct osd_device *o,
 187                                  struct osd_object *obj, struct lu_attr *la)
 188 {
 189         struct osa_attr *osa = &osd_oti_get(env)->oti_osa;
 190         sa_bulk_attr_t  *bulk = osd_oti_get(env)->oti_attr_bulk;
 191         int              cnt = 0;
 192         int              rc;
 193         ENTRY;
 194
 195         LASSERT(obj->oo_dn != NULL);
 196
 197         la->la_valid |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE | LA_TYPE |
 198                         LA_SIZE | LA_UID | LA_GID | LA_FLAGS | LA_NLINK;
 199
 200         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(o), NULL, osa->atime, 16);
 201         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(o), NULL, osa->mtime, 16);
 202         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(o), NULL, osa->ctime, 16);
 203         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(o), NULL, &osa->mode, 8);
 204         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(o), NULL, &osa->size, 8);
 205         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(o), NULL, &osa->nlink, 8);
 206         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(o), NULL, &osa->uid, 8);
 207         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(o), NULL, &osa->gid, 8);
 208         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(o), NULL, &osa->flags, 8);
 209         LASSERT(cnt <= ARRAY_SIZE(osd_oti_get(env)->oti_attr_bulk));
 210
 211         rc = -sa_bulk_lookup(obj->oo_sa_hdl, bulk, cnt);
 212         if (rc)
 213                 GOTO(out_sa, rc);
 214
 215 #ifdef ZFS_PROJINHERIT
 216         if (o->od_projectused_dn && osa->flags & ZFS_PROJID) {
 217                 rc = -sa_lookup(obj->oo_sa_hdl, SA_ZPL_PROJID(o),
 218                                 &osa->projid, 8);
 219                 if (rc)
 220                         GOTO(out_sa, rc);
 221
 222                 la->la_projid = osa->projid;
 223                 la->la_valid |= LA_PROJID;
 224                 obj->oo_with_projid = 1;
 225         } else {
 226                 la->la_projid = ZFS_DEFAULT_PROJID;
 227                 la->la_valid &= ~LA_PROJID;
 228         }
 229 #else
 230         la->la_projid = 0;
 231         la->la_valid &= ~LA_PROJID;
 232 #endif
 233
 234         la->la_atime = osa->atime[0];
 235         la->la_mtime = osa->mtime[0];
 236         la->la_ctime = osa->ctime[0];
 237         la->la_mode = osa->mode;
 238         la->la_uid = osa->uid;
 239         la->la_gid = osa->gid;
 240         la->la_nlink = osa->nlink;
 241         la->la_flags = attrs_zfs2fs(osa->flags);
 242         la->la_size = osa->size;
 243
 244         /* Try to get extra flag from LMA. Right now, only LMAI_ORPHAN
 245          * flags is stored in LMA, and it is only for orphan directory */
 246         if (S_ISDIR(la->la_mode) && dt_object_exists(&obj->oo_dt)) {
 247                 struct osd_thread_info *info = osd_oti_get(env);
 248                 struct lustre_mdt_attrs *lma;
 249                 struct lu_buf buf;
 250
 251                 lma = (struct lustre_mdt_attrs *)info->oti_buf;
 252                 buf.lb_buf = lma;
 253                 buf.lb_len = sizeof(info->oti_buf);
 254                 rc = osd_xattr_get(env, &obj->oo_dt, &buf, XATTR_NAME_LMA);
 255                 if (rc > 0) {
 256                         rc = 0;
 257                         lma->lma_incompat = le32_to_cpu(lma->lma_incompat);
 258                         obj->oo_lma_flags =
 259                                 lma_to_lustre_flags(lma->lma_incompat);
 260
 261                 } else if (rc == -ENODATA) {
 262                         rc = 0;
 263                 }
 264         }
 265
 266         if (S_ISCHR(la->la_mode) || S_ISBLK(la->la_mode)) {
 267                 rc = -sa_lookup(obj->oo_sa_hdl, SA_ZPL_RDEV(o), &osa->rdev, 8);
 268                 if (rc)
 269                         GOTO(out_sa, rc);
 270                 la->la_rdev = osa->rdev;
 271                 la->la_valid |= LA_RDEV;
 272         }
 273 out_sa:
 274
 275         RETURN(rc);
 276 }
 277
 278 int __osd_obj2dnode(objset_t *os, uint64_t oid, dnode_t **dnp)
 279 {
 280         dmu_buf_t *db;
 281         dmu_buf_impl_t *dbi;
 282         int rc;
 283
 284         rc = -dmu_bonus_hold(os, oid, osd_obj_tag, &db);
 285         if (rc)
 286                 return rc;
 287
 288         dbi = (dmu_buf_impl_t *)db;
 289         DB_DNODE_ENTER(dbi);
 290         *dnp = DB_DNODE(dbi);
 291         DB_DNODE_EXIT(dbi);
 292         LASSERT(*dnp != NULL);
 293
 294         return 0;
 295 }
 296
 297 /*
 298  * Concurrency: no concurrent access is possible that early in object
 299  * life-cycle.
 300  */
 301 struct lu_object *osd_object_alloc(const struct lu_env *env,
 302                                    const struct lu_object_header *hdr,
 303                                    struct lu_device *d)
 304 {
 305         struct osd_object *mo;
 306
 307         OBD_SLAB_ALLOC_PTR_GFP(mo, osd_object_kmem, GFP_NOFS);
 308         if (mo != NULL) {
 309                 struct lu_object *l;
 310                 struct lu_object_header *h;
 311                 struct osd_device *o = osd_dev(d);
 312
 313                 l = &mo->oo_dt.do_lu;
 314                 if (unlikely(o->od_in_init)) {
 315                         OBD_ALLOC_PTR(h);
 316                         if (!h) {
 317                                 OBD_FREE_PTR(mo);
 318                                 return NULL;
 319                         }
 320
 321                         lu_object_header_init(h);
 322                         lu_object_init(l, h, d);
 323                         lu_object_add_top(h, l);
 324                         mo->oo_header = h;
 325                 } else {
 326                         dt_object_init(&mo->oo_dt, NULL, d);
 327                         mo->oo_header = NULL;
 328                 }
 329
 330                 mo->oo_dt.do_ops = &osd_obj_ops;
 331                 l->lo_ops = &osd_lu_obj_ops;
 332                 INIT_LIST_HEAD(&mo->oo_sa_linkage);
 333                 INIT_LIST_HEAD(&mo->oo_unlinked_linkage);
 334                 init_rwsem(&mo->oo_sem);
 335                 init_rwsem(&mo->oo_guard);
 336                 rwlock_init(&mo->oo_attr_lock);
 337                 mo->oo_destroy = OSD_DESTROY_NONE;
 338                 return l;
 339         } else {
 340                 return NULL;
 341         }
 342 }
 343
 344 static void osd_obj_set_blksize(const struct lu_env *env,
 345                                 struct osd_device *osd, struct osd_object *obj)
 346 {
 347         const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
 348         dmu_tx_t *tx;
 349         dnode_t *dn = obj->oo_dn;
 350         uint32_t blksz;
 351         int rc = 0;
 352         ENTRY;
 353
 354         LASSERT(!osd_oti_get(env)->oti_in_trans);
 355
 356         tx = dmu_tx_create(osd->od_os);
 357         if (!tx) {
 358                 CERROR("%s: fail to create tx to set blksize for "DFID"\n",
 359                        osd->od_svname, PFID(fid));
 360                 RETURN_EXIT;
 361         }
 362
 363         dmu_tx_hold_bonus(tx, dn->dn_object);
 364         rc = -dmu_tx_assign(tx, TXG_WAIT);
 365         if (rc) {
 366                 dmu_tx_abort(tx);
 367                 CERROR("%s: fail to assign tx to set blksize for "DFID
 368                        ": rc = %d\n", osd->od_svname, PFID(fid), rc);
 369                 RETURN_EXIT;
 370         }
 371
 372         down_write(&obj->oo_guard);
 373         if (unlikely((1 << dn->dn_datablkshift) >= PAGE_SIZE))
 374                 GOTO(out, rc = 1);
 375
 376         blksz = dn->dn_datablksz;
 377         if (!is_power_of_2(blksz))
 378                 blksz = size_roundup_power2(blksz);
 379
 380         if (blksz > osd->od_max_blksz)
 381                 blksz = osd->od_max_blksz;
 382         else if (blksz < PAGE_SIZE)
 383                 blksz = PAGE_SIZE;
 384         rc = -dmu_object_set_blocksize(osd->od_os, dn->dn_object, blksz, 0, tx);
 385
 386         GOTO(out, rc);
 387
 388 out:
 389         up_write(&obj->oo_guard);
 390         if (rc) {
 391                 dmu_tx_abort(tx);
 392                 if (unlikely(obj->oo_dn->dn_maxblkid > 0))
 393                         rc = 1;
 394                 if (rc < 0)
 395                         CERROR("%s: fail to set blksize for "DFID": rc = %d\n",
 396                                osd->od_svname, PFID(fid), rc);
 397         } else {
 398                 dmu_tx_commit(tx);
 399                 CDEBUG(D_INODE, "%s: set blksize as %u for "DFID"\n",
 400                        osd->od_svname, blksz, PFID(fid));
 401         }
 402 }
 403
 404 /*
 405  * Concurrency: shouldn't matter.
 406  */
 407 static int osd_object_init0(const struct lu_env *env, struct osd_object *obj)
 408 {
 409         struct osd_device       *osd = osd_obj2dev(obj);
 410         const struct lu_fid     *fid = lu_object_fid(&obj->oo_dt.do_lu);
 411         int                      rc = 0;
 412         ENTRY;
 413
 414         LASSERT(obj->oo_dn);
 415
 416         rc = osd_object_sa_init(obj, osd);
 417         if (rc)
 418                 RETURN(rc);
 419
 420         /* cache attrs in object */
 421         rc = __osd_object_attr_get(env, osd, obj, &obj->oo_attr);
 422         if (rc)
 423                 RETURN(rc);
 424
 425         if (likely(!fid_is_acct(fid))) {
 426                 /* no body operations for accounting objects */
 427                 obj->oo_dt.do_body_ops = &osd_body_ops;
 428
 429                 if (S_ISREG(obj->oo_attr.la_mode) &&
 430                     obj->oo_dn->dn_maxblkid == 0 &&
 431                     (1 << obj->oo_dn->dn_datablkshift) < PAGE_SIZE &&
 432                     (fid_is_idif(fid) || fid_is_norm(fid) ||
 433                      fid_is_echo(fid)) &&
 434                     osd->od_is_ost && !osd->od_dt_dev.dd_rdonly)
 435                         osd_obj_set_blksize(env, osd, obj);
 436         }
 437
 438         /*
 439          * initialize object before marking it existing
 440          */
 441         obj->oo_dt.do_lu.lo_header->loh_attr |= obj->oo_attr.la_mode & S_IFMT;
 442
 443         smp_mb();
 444         obj->oo_dt.do_lu.lo_header->loh_attr |= LOHA_EXISTS;
 445
 446         RETURN(0);
 447 }
 448
 449 static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
 450 {
 451         struct osd_thread_info  *info = osd_oti_get(env);
 452         struct lu_buf           buf;
 453         int                     rc;
 454         struct lustre_mdt_attrs *lma;
 455         const struct lu_fid *rfid = lu_object_fid(&obj->oo_dt.do_lu);
 456         ENTRY;
 457
 458         BUILD_BUG_ON(sizeof(info->oti_buf) < sizeof(*lma));
 459         lma = (struct lustre_mdt_attrs *)info->oti_buf;
 460         buf.lb_buf = lma;
 461         buf.lb_len = sizeof(info->oti_buf);
 462
 463         rc = osd_xattr_get(env, &obj->oo_dt, &buf, XATTR_NAME_LMA);
 464         if (rc > 0) {
 465                 rc = 0;
 466                 lustre_lma_swab(lma);
 467                 if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
 468                              CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT))) {
 469                         CWARN("%s: unsupported incompat LMA feature(s) %#x for "
 470                               "fid = "DFID"\n", osd_obj2dev(obj)->od_svname,
 471                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
 472                               PFID(rfid));
 473                         rc = -EOPNOTSUPP;
 474                 } else if (unlikely(!lu_fid_eq(rfid, &lma->lma_self_fid))) {
 475                         CERROR("%s: FID-in-LMA "DFID" does not match the "
 476                               "object self-fid "DFID"\n",
 477                               osd_obj2dev(obj)->od_svname,
 478                               PFID(&lma->lma_self_fid), PFID(rfid));
 479                         rc = -EREMCHG;
 480                 } else {
 481                         struct osd_device *osd = osd_obj2dev(obj);
 482
 483                         if (lma->lma_compat & LMAC_STRIPE_INFO &&
 484                             osd->od_is_ost)
 485                                 obj->oo_pfid_in_lma = 1;
 486                         if (unlikely(lma->lma_incompat & LMAI_REMOTE_PARENT) &&
 487                             osd->od_remote_parent_dir != ZFS_NO_OBJECT)
 488                                 lu_object_set_agent_entry(&obj->oo_dt.do_lu);
 489                 }
 490         } else if (rc == -ENODATA) {
 491                 /* haven't initialize LMA xattr */
 492                 rc = 0;
 493         }
 494
 495         RETURN(rc);
 496 }
 497
 498 /**
 499  * Helper function to retrieve DMU object id from fid for accounting object
 500  */
 501 static dnode_t *osd_quota_fid2dmu(const struct osd_device *osd,
 502                                   const struct lu_fid *fid)
 503 {
 504         dnode_t *dn = NULL;
 505
 506         LASSERT(fid_is_acct(fid));
 507
 508         switch (fid_oid(fid)) {
 509         case ACCT_USER_OID:
 510                 dn = osd->od_userused_dn;
 511                 break;
 512         case ACCT_GROUP_OID:
 513                 dn = osd->od_groupused_dn;
 514                 break;
 515 #ifdef ZFS_PROJINHERIT
 516         case ACCT_PROJECT_OID:
 517                 dn = osd->od_projectused_dn;
 518                 break;
 519 #endif
 520         default:
 521                 break;
 522         }
 523
 524         return dn;
 525 }
 526
 527 /*
 528  * Concurrency: no concurrent access is possible that early in object
 529  * life-cycle.
 530  */
 531 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
 532                            const struct lu_object_conf *conf)
 533 {
 534         struct osd_object *obj = osd_obj(l);
 535         struct osd_device *osd = osd_obj2dev(obj);
 536         const struct lu_fid *fid = lu_object_fid(l);
 537         struct lustre_scrub *scrub = &osd->od_scrub;
 538         struct osd_thread_info *info = osd_oti_get(env);
 539         struct luz_direntry *zde = &info->oti_zde;
 540         struct osd_idmap_cache *idc;
 541         char *name = info->oti_str;
 542         uint64_t oid;
 543         int rc = 0;
 544         int rc1;
 545         bool remote = false;
 546         ENTRY;
 547
 548         LASSERT(osd_invariant(obj));
 549
 550         if (fid_is_otable_it(&l->lo_header->loh_fid)) {
 551                 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
 552                 l->lo_header->loh_attr |= LOHA_EXISTS;
 553
 554                 GOTO(out, rc = 0);
 555         }
 556
 557         if (conf && conf->loc_flags & LOC_F_NEW)
 558                 GOTO(out, rc = 0);
 559
 560         if (unlikely(fid_is_acct(fid))) {
 561                 obj->oo_dn = osd_quota_fid2dmu(osd, fid);
 562                 if (obj->oo_dn) {
 563                         obj->oo_dt.do_index_ops = &osd_acct_index_ops;
 564                         l->lo_header->loh_attr |= LOHA_EXISTS;
 565                 }
 566
 567                 GOTO(out, rc = 0);
 568         }
 569
 570         idc = osd_idc_find(env, osd, fid);
 571         if (idc && !idc->oic_remote && idc->oic_dnode != ZFS_NO_OBJECT) {
 572                 oid = idc->oic_dnode;
 573                 goto zget;
 574         }
 575
 576         rc = -ENOENT;
 577         if (!list_empty(&osd->od_scrub.os_inconsistent_items))
 578                 rc = osd_oii_lookup(osd, fid, &oid);
 579
 580         if (rc)
 581                 rc = osd_fid_lookup(env, osd, fid, &oid);
 582
 583         if (rc == -ENOENT) {
 584                 if (likely(!(fid_is_norm(fid) || fid_is_igif(fid)) ||
 585                            fid_is_on_ost(env, osd, fid) ||
 586                            !zfs_test_bit(osd_oi_fid2idx(osd, fid),
 587                                          scrub->os_file.sf_oi_bitmap)))
 588                         GOTO(out, rc = 0);
 589
 590                 rc = -EREMCHG;
 591                 goto trigger;
 592         }
 593
 594         if (rc)
 595                 GOTO(out, rc);
 596
 597 zget:
 598         LASSERT(obj->oo_dn == NULL);
 599
 600         rc = __osd_obj2dnode(osd->od_os, oid, &obj->oo_dn);
 601         /* EEXIST will be returned if object is being deleted in ZFS */
 602         if (rc == -EEXIST)
 603                 GOTO(out, rc = 0);
 604
 605         if (rc) {
 606                 CERROR("%s: lookup "DFID"/%#llx failed: rc = %d\n",
 607                        osd->od_svname, PFID(lu_object_fid(l)), oid, rc);
 608                 GOTO(out, rc);
 609         }
 610
 611         rc = osd_object_init0(env, obj);
 612         if (rc)
 613                 GOTO(out, rc);
 614
 615         if (unlikely(obj->oo_header))
 616                 GOTO(out, rc = 0);
 617
 618         rc = osd_check_lma(env, obj);
 619         if ((!rc && !remote) || (rc != -EREMCHG))
 620                 GOTO(out, rc);
 621
 622 trigger:
 623         /* We still have chance to get the valid dnode: for the object that is
 624          * referenced by remote name entry, the object on the local MDT will be
 625          * linked under the dir /REMOTE_PARENT_DIR with its FID string as name.
 626          *
 627          * During the OI scrub, if we cannot find the OI mapping, we may still
 628          * have change to map the FID to local OID via lookup the dir
 629          * /REMOTE_PARENT_DIR. */
 630         if (!remote && !fid_is_on_ost(env, osd, fid)) {
 631                 osd_fid2str(name, fid, sizeof(info->oti_str));
 632                 rc = osd_zap_lookup(osd, osd->od_remote_parent_dir,
 633                                     NULL, name, 8, 3, (void *)zde);
 634                 if (!rc) {
 635                         oid = zde->lzd_reg.zde_dnode;
 636                         osd_dnode_rele(obj->oo_dn);
 637                         obj->oo_dn = NULL;
 638                         remote = true;
 639                         goto zget;
 640                 }
 641         }
 642
 643         /* The case someone triggered the OI scrub already. */
 644         if (thread_is_running(&scrub->os_thread)) {
 645                 if (!rc) {
 646                         LASSERT(remote);
 647
 648                         lu_object_set_agent_entry(l);
 649                         osd_oii_insert(env, osd, fid, oid, false);
 650                 } else {
 651                         rc = -EINPROGRESS;
 652                 }
 653
 654                 GOTO(out, rc);
 655         }
 656
 657         /* The case NOT allow to trigger OI scrub automatically. */
 658         if (osd->od_auto_scrub_interval == AS_NEVER)
 659                 GOTO(out, rc);
 660
 661         /* It is me to trigger the OI scrub. */
 662         rc1 = osd_scrub_start(env, osd, SS_CLEAR_DRYRUN |
 663                               SS_CLEAR_FAILOUT | SS_AUTO_FULL);
 664         LCONSOLE_WARN("%s: trigger OI scrub by RPC for the "DFID": rc = %d\n",
 665                       osd_name(osd), PFID(fid), rc1);
 666         if (!rc) {
 667                 LASSERT(remote);
 668
 669                 lu_object_set_agent_entry(l);
 670                 if (!rc1)
 671                         osd_oii_insert(env, osd, fid, oid, false);
 672         } else {
 673                 if (!rc1)
 674                         rc = -EINPROGRESS;
 675                 else
 676                         rc = -EREMCHG;
 677         }
 678
 679         GOTO(out, rc);
 680
 681 out:
 682         RETURN(rc);
 683 }
 684
 685 /*
 686  * Concurrency: no concurrent access is possible that late in object
 687  * life-cycle.
 688  */
 689 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
 690 {
 691         struct osd_object *obj = osd_obj(l);
 692         struct lu_object_header *h = obj->oo_header;
 693
 694         LASSERT(osd_invariant(obj));
 695
 696         dt_object_fini(&obj->oo_dt);
 697         OBD_SLAB_FREE_PTR(obj, osd_object_kmem);
 698         if (unlikely(h)) {
 699                 lu_object_header_fini(h);
 700                 OBD_FREE_PTR(h);
 701         }
 702 }
 703
 704 static int
 705 osd_object_unlinked_add(struct osd_object *obj, struct osd_thandle *oh)
 706 {
 707         int rc = -EBUSY;
 708
 709         LASSERT(obj->oo_destroy == OSD_DESTROY_ASYNC);
 710
 711         /* the object is supposed to be exclusively locked by
 712          * the caller (osd_destroy()), while the transaction
 713          * (oh) is per-thread and not shared */
 714         if (likely(list_empty(&obj->oo_unlinked_linkage))) {
 715                 list_add(&obj->oo_unlinked_linkage, &oh->ot_unlinked_list);
 716                 rc = 0;
 717         }
 718
 719         return rc;
 720 }
 721
 722 /* Default to max data size covered by a level-1 indirect block */
 723 static unsigned long osd_sync_destroy_max_size =
 724         1UL << (DN_MAX_INDBLKSHIFT - SPA_BLKPTRSHIFT + SPA_MAXBLOCKSHIFT);
 725 module_param(osd_sync_destroy_max_size, ulong, 0444);
 726 MODULE_PARM_DESC(osd_sync_destroy_max_size, "Maximum object size to use synchronous destroy.");
 727
 728 static inline void
 729 osd_object_set_destroy_type(struct osd_object *obj)
 730 {
 731         /*
 732          * Lock-less OST_WRITE can race with OST_DESTROY, so set destroy type
 733          * only once and use it consistently thereafter.
 734          */
 735         down_write(&obj->oo_guard);
 736         if (obj->oo_destroy == OSD_DESTROY_NONE) {
 737                 if (obj->oo_attr.la_size <= osd_sync_destroy_max_size)
 738                         obj->oo_destroy = OSD_DESTROY_SYNC;
 739                 else /* Larger objects are destroyed asynchronously */
 740                         obj->oo_destroy = OSD_DESTROY_ASYNC;
 741         }
 742         up_write(&obj->oo_guard);
 743 }
 744
 745 static int osd_declare_destroy(const struct lu_env *env, struct dt_object *dt,
 746                                struct thandle *th)
 747 {
 748         const struct lu_fid     *fid = lu_object_fid(&dt->do_lu);
 749         struct osd_object       *obj = osd_dt_obj(dt);
 750         struct osd_device       *osd = osd_obj2dev(obj);
 751         struct osd_thandle      *oh;
 752         dnode_t *dn;
 753         int                      rc;
 754         uint64_t                 zapid;
 755         ENTRY;
 756
 757         LASSERT(th != NULL);
 758         LASSERT(dt_object_exists(dt));
 759
 760         oh = container_of0(th, struct osd_thandle, ot_super);
 761         LASSERT(oh->ot_tx != NULL);
 762
 763         dmu_tx_mark_netfree(oh->ot_tx);
 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, struct dt_object *dt,
 986                         struct lu_attr *attr)
 987 {
 988         struct osd_object       *obj = osd_dt_obj(dt);
 989         uint64_t                 blocks;
 990         uint32_t                 blksize;
 991         int                      rc = 0;
 992
 993         down_read(&obj->oo_guard);
 994
 995         if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
 996                 GOTO(out, rc = -ENOENT);
 997
 998         if (unlikely(fid_is_acct(lu_object_fid(&dt->do_lu))))
 999                 GOTO(out, rc = 0);
1000
1001         LASSERT(osd_invariant(obj));
1002         LASSERT(obj->oo_dn);
1003
1004         read_lock(&obj->oo_attr_lock);
1005         *attr = obj->oo_attr;
1006         if (obj->oo_lma_flags & LUSTRE_ORPHAN_FL) {
1007                 attr->la_valid |= LA_FLAGS;
1008                 attr->la_flags |= LUSTRE_ORPHAN_FL;
1009         }
1010         read_unlock(&obj->oo_attr_lock);
1011         if (attr->la_valid & LA_FLAGS && attr->la_flags & LUSTRE_ORPHAN_FL)
1012                 CDEBUG(D_INFO, "%s: set orphan flag on "DFID" (%llx/%x)\n",
1013                        osd_obj2dev(obj)->od_svname,
1014                        PFID(lu_object_fid(&dt->do_lu)),
1015                        attr->la_valid, obj->oo_lma_flags);
1016
1017         /* with ZFS_DEBUG zrl_add_debug() called by DB_DNODE_ENTER()
1018          * from within sa_object_size() can block on a mutex, so
1019          * we can't call sa_object_size() holding rwlock */
1020         sa_object_size(obj->oo_sa_hdl, &blksize, &blocks);
1021         /* we do not control size of indices, so always calculate
1022          * it from number of blocks reported by DMU */
1023         if (S_ISDIR(attr->la_mode))
1024                 attr->la_size = 512 * blocks;
1025         /* Block size may be not set; suggest maximal I/O transfers. */
1026         if (blksize == 0)
1027                 blksize = osd_spa_maxblocksize(
1028                         dmu_objset_spa(osd_obj2dev(obj)->od_os));
1029
1030         attr->la_blksize = blksize;
1031         attr->la_blocks = blocks;
1032         attr->la_valid |= LA_BLOCKS | LA_BLKSIZE;
1033
1034 out:
1035         up_read(&obj->oo_guard);
1036         return rc;
1037 }
1038
1039 /* Simple wrapper on top of qsd API which implement quota transfer for osd
1040  * setattr needs. As a reminder, only the root user can change ownership of
1041  * a file, that's why EDQUOT & EINPROGRESS errors are discarded */
1042 static inline int qsd_transfer(const struct lu_env *env,
1043                                struct qsd_instance *qsd,
1044                                struct lquota_trans *trans, int qtype,
1045                                __u64 orig_id, __u64 new_id, __u64 bspace,
1046                                struct lquota_id_info *qi, bool ignore_edquot)
1047 {
1048         int     rc;
1049
1050         if (unlikely(qsd == NULL))
1051                 return 0;
1052
1053         LASSERT(qtype >= 0 && qtype < LL_MAXQUOTAS);
1054         qi->lqi_type = qtype;
1055
1056         /* inode accounting */
1057         qi->lqi_is_blk = false;
1058
1059         /* one more inode for the new owner ... */
1060         qi->lqi_id.qid_uid = new_id;
1061         qi->lqi_space      = 1;
1062         rc = qsd_op_begin(env, qsd, trans, qi, NULL);
1063         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
1064                 rc = 0;
1065         if (rc)
1066                 return rc;
1067
1068         /* and one less inode for the current id */
1069         qi->lqi_id.qid_uid = orig_id;;
1070         qi->lqi_space      = -1;
1071         /* can't get EDQUOT when reducing usage */
1072         rc = qsd_op_begin(env, qsd, trans, qi, NULL);
1073         if (rc == -EINPROGRESS)
1074                 rc = 0;
1075         if (rc)
1076                 return rc;
1077
1078         /* block accounting */
1079         qi->lqi_is_blk = true;
1080
1081         /* more blocks for the new owner ... */
1082         qi->lqi_id.qid_uid = new_id;
1083         qi->lqi_space      = bspace;
1084         rc = qsd_op_begin(env, qsd, trans, qi, NULL);
1085         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
1086                 rc = 0;
1087         if (rc)
1088                 return rc;
1089
1090         /* and finally less blocks for the current owner */
1091         qi->lqi_id.qid_uid = orig_id;
1092         qi->lqi_space      = -bspace;
1093         rc = qsd_op_begin(env, qsd, trans, qi, NULL);
1094         /* can't get EDQUOT when reducing usage */
1095         if (rc == -EINPROGRESS)
1096                 rc = 0;
1097         return rc;
1098 }
1099
1100 static int osd_declare_attr_set(const struct lu_env *env,
1101                                 struct dt_object *dt,
1102                                 const struct lu_attr *attr,
1103                                 struct thandle *handle)
1104 {
1105         struct osd_thread_info  *info = osd_oti_get(env);
1106         struct osd_object       *obj = osd_dt_obj(dt);
1107         struct osd_device       *osd = osd_obj2dev(obj);
1108         dmu_tx_hold_t           *txh;
1109         struct osd_thandle      *oh;
1110         uint64_t                 bspace;
1111         uint32_t                 blksize;
1112         int                      rc = 0;
1113         bool                     found;
1114         ENTRY;
1115
1116
1117         LASSERT(handle != NULL);
1118         LASSERT(osd_invariant(obj));
1119
1120         oh = container_of0(handle, struct osd_thandle, ot_super);
1121
1122         down_read(&obj->oo_guard);
1123         if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
1124                 GOTO(out, rc = 0);
1125
1126         LASSERT(obj->oo_sa_hdl != NULL);
1127         LASSERT(oh->ot_tx != NULL);
1128         /* regular attributes are part of the bonus buffer */
1129         /* let's check whether this object is already part of
1130          * transaction.. */
1131         found = false;
1132         for (txh = list_head(&oh->ot_tx->tx_holds); txh;
1133              txh = list_next(&oh->ot_tx->tx_holds, txh)) {
1134                 if (txh->txh_dnode == NULL)
1135                         continue;
1136                 if (txh->txh_dnode->dn_object != obj->oo_dn->dn_object)
1137                         continue;
1138                 /* this object is part of the transaction already
1139                  * we don't need to declare bonus again */
1140                 found = true;
1141                 break;
1142         }
1143         if (!found)
1144                 dmu_tx_hold_bonus(oh->ot_tx, obj->oo_dn->dn_object);
1145         if (oh->ot_tx->tx_err != 0)
1146                 GOTO(out, rc = -oh->ot_tx->tx_err);
1147
1148         if (attr && attr->la_valid & LA_FLAGS) {
1149                 /* LMA is usually a part of bonus, no need to declare
1150                  * anything else */
1151         }
1152
1153         if (attr && (attr->la_valid & (LA_UID | LA_GID | LA_PROJID))) {
1154                 sa_object_size(obj->oo_sa_hdl, &blksize, &bspace);
1155                 bspace = toqb(bspace * 512);
1156
1157                 CDEBUG(D_QUOTA,
1158                        "%s: enforce quota on UID %u, GID %u, the quota space is %lld (%u)\n",
1159                        osd->od_svname,
1160                        attr->la_uid, attr->la_gid, bspace, blksize);
1161         }
1162
1163         if (attr && attr->la_valid & LA_UID) {
1164                 /* quota enforcement for user */
1165                 if (attr->la_uid != obj->oo_attr.la_uid) {
1166                         rc = qsd_transfer(env, osd_def_qsd(osd),
1167                                           &oh->ot_quota_trans, USRQUOTA,
1168                                           obj->oo_attr.la_uid, attr->la_uid,
1169                                           bspace, &info->oti_qi, true);
1170                         if (rc)
1171                                 GOTO(out, rc);
1172                 }
1173         }
1174         if (attr && attr->la_valid & LA_GID) {
1175                 /* quota enforcement for group */
1176                 if (attr->la_gid != obj->oo_attr.la_gid) {
1177                         rc = qsd_transfer(env, osd_def_qsd(osd),
1178                                           &oh->ot_quota_trans, GRPQUOTA,
1179                                           obj->oo_attr.la_gid, attr->la_gid,
1180                                           bspace, &info->oti_qi,
1181                                           !(attr->la_flags &
1182                                                         LUSTRE_SET_SYNC_FL));
1183                         if (rc)
1184                                 GOTO(out, rc);
1185                 }
1186         }
1187 #ifdef ZFS_PROJINHERIT
1188         if (attr && attr->la_valid & LA_PROJID) {
1189                 /* quota enforcement for project */
1190                 if (attr->la_projid != obj->oo_attr.la_projid) {
1191                         if (!osd->od_projectused_dn)
1192                                 GOTO(out, rc = -EOPNOTSUPP);
1193
1194                         /* Usually, if project quota is upgradable for the
1195                          * device, then the upgrade will be done before or when
1196                          * mount the device. So when we come here, this project
1197                          * should have project ID attribute already (that is
1198                          * zero by default).  Otherwise, there was something
1199                          * wrong during the former upgrade, let's return failure
1200                          * to report that.
1201                          *
1202                          * Please note that, different from other attributes,
1203                          * you can NOT simply set the project ID attribute under
1204                          * such case, because adding (NOT change) project ID
1205                          * attribute needs to change the object's attribute
1206                          * layout to match zfs backend quota accounting
1207                          * requirement. */
1208                         if (unlikely(!obj->oo_with_projid))
1209                                 GOTO(out, rc = -ENXIO);
1210
1211                         rc = qsd_transfer(env, osd_def_qsd(osd),
1212                                           &oh->ot_quota_trans, PRJQUOTA,
1213                                           obj->oo_attr.la_projid,
1214                                           attr->la_projid, bspace,
1215                                           &info->oti_qi, true);
1216                         if (rc)
1217                                 GOTO(out, rc);
1218                 }
1219         }
1220 #endif
1221 out:
1222         up_read(&obj->oo_guard);
1223         RETURN(rc);
1224 }
1225
1226 /*
1227  * Set the attributes of an object
1228  *
1229  * The transaction passed to this routine must have
1230  * dmu_tx_hold_bonus(tx, oid) called and then assigned
1231  * to a transaction group.
1232  */
1233 static int osd_attr_set(const struct lu_env *env, struct dt_object *dt,
1234                         const struct lu_attr *la, struct thandle *handle)
1235 {
1236         struct osd_thread_info  *info = osd_oti_get(env);
1237         sa_bulk_attr_t          *bulk = osd_oti_get(env)->oti_attr_bulk;
1238         struct osd_object       *obj = osd_dt_obj(dt);
1239         struct osd_device       *osd = osd_obj2dev(obj);
1240         struct osd_thandle      *oh;
1241         struct osa_attr         *osa = &info->oti_osa;
1242         __u64                    valid = la->la_valid;
1243         int                      cnt;
1244         int                      rc = 0;
1245
1246         ENTRY;
1247
1248         down_read(&obj->oo_guard);
1249         if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
1250                 GOTO(out, rc = -ENOENT);
1251
1252         LASSERT(handle != NULL);
1253         LASSERT(osd_invariant(obj));
1254         LASSERT(obj->oo_sa_hdl);
1255
1256         oh = container_of0(handle, struct osd_thandle, ot_super);
1257         /* Assert that the transaction has been assigned to a
1258            transaction group. */
1259         LASSERT(oh->ot_tx->tx_txg != 0);
1260
1261         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING) && !osd->od_is_ost) {
1262                 struct zpl_direntry *zde = &info->oti_zde.lzd_reg;
1263                 char *buf = info->oti_str;
1264                 dnode_t *zdn = NULL;
1265                 uint64_t zapid;
1266
1267                 zapid = osd_get_name_n_idx(env, osd, lu_object_fid(&dt->do_lu),
1268                                            buf, sizeof(info->oti_str), &zdn);
1269                 rc = osd_zap_lookup(osd, zapid, zdn, buf, 8,
1270                                     sizeof(*zde) / 8, zde);
1271                 if (!rc) {
1272                         zde->zde_dnode -= 1;
1273                         rc = -zap_update(osd->od_os, zapid, buf, 8,
1274                                          sizeof(*zde) / 8, zde, oh->ot_tx);
1275                 }
1276                 if (rc > 0)
1277                         rc = 0;
1278                 GOTO(out, rc);
1279         }
1280
1281         /* Only allow set size for regular file */
1282         if (!S_ISREG(dt->do_lu.lo_header->loh_attr))
1283                 valid &= ~(LA_SIZE | LA_BLOCKS);
1284
1285         if (valid & LA_CTIME && la->la_ctime == obj->oo_attr.la_ctime)
1286                 valid &= ~LA_CTIME;
1287
1288         if (valid & LA_MTIME && la->la_mtime == obj->oo_attr.la_mtime)
1289                 valid &= ~LA_MTIME;
1290
1291         if (valid & LA_ATIME && la->la_atime == obj->oo_attr.la_atime)
1292                 valid &= ~LA_ATIME;
1293
1294         if (valid == 0)
1295                 GOTO(out, rc = 0);
1296
1297         if (valid & LA_FLAGS) {
1298                 struct lustre_mdt_attrs *lma;
1299                 struct lu_buf buf;
1300                 int size = 0;
1301
1302                 if (la->la_flags & LUSTRE_LMA_FL_MASKS) {
1303                         LASSERT(!obj->oo_pfid_in_lma);
1304                         BUILD_BUG_ON(sizeof(info->oti_buf) < sizeof(*lma));
1305                         lma = (struct lustre_mdt_attrs *)&info->oti_buf;
1306                         buf.lb_buf = lma;
1307                         buf.lb_len = sizeof(info->oti_buf);
1308
1309                         /* Please do NOT call osd_xattr_get() directly, that
1310                          * will cause recursive down_read() on oo_guard. */
1311                         rc = osd_xattr_get_internal(env, obj, &buf,
1312                                                     XATTR_NAME_LMA, &size);
1313                         if (!rc && unlikely(size < sizeof(*lma))) {
1314                                 rc = -EINVAL;
1315                         } else if (!rc) {
1316                                 lma->lma_incompat =
1317                                         le32_to_cpu(lma->lma_incompat);
1318                                 lma->lma_incompat |=
1319                                         lustre_to_lma_flags(la->la_flags);
1320                                 lma->lma_incompat =
1321                                         cpu_to_le32(lma->lma_incompat);
1322                                 buf.lb_buf = lma;
1323                                 buf.lb_len = sizeof(*lma);
1324                                 rc = osd_xattr_set_internal(env, obj, &buf,
1325                                                             XATTR_NAME_LMA,
1326                                                             LU_XATTR_REPLACE,
1327                                                             oh);
1328                         }
1329                         if (rc < 0) {
1330                                 CWARN("%s: failed to set LMA flags: rc = %d\n",
1331                                        osd->od_svname, rc);
1332                                 GOTO(out, rc);
1333                         }
1334                 }
1335         }
1336
1337         write_lock(&obj->oo_attr_lock);
1338         cnt = 0;
1339
1340         if (valid & LA_PROJID) {
1341 #ifdef ZFS_PROJINHERIT
1342                 if (osd->od_projectused_dn) {
1343                         LASSERT(obj->oo_with_projid);
1344
1345                         osa->projid = obj->oo_attr.la_projid = la->la_projid;
1346                         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_PROJID(osd), NULL,
1347                                          &osa->projid, 8);
1348                 } else
1349 #endif
1350                         valid &= ~LA_PROJID;
1351         }
1352
1353         if (valid & LA_ATIME) {
1354                 osa->atime[0] = obj->oo_attr.la_atime = la->la_atime;
1355                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(osd), NULL,
1356                                  osa->atime, 16);
1357         }
1358         if (valid & LA_MTIME) {
1359                 osa->mtime[0] = obj->oo_attr.la_mtime = la->la_mtime;
1360                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(osd), NULL,
1361                                  osa->mtime, 16);
1362         }
1363         if (valid & LA_CTIME) {
1364                 osa->ctime[0] = obj->oo_attr.la_ctime = la->la_ctime;
1365                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(osd), NULL,
1366                                  osa->ctime, 16);
1367         }
1368         if (valid & LA_MODE) {
1369                 /* mode is stored along with type, so read it first */
1370                 obj->oo_attr.la_mode = (obj->oo_attr.la_mode & S_IFMT) |
1371                         (la->la_mode & ~S_IFMT);
1372                 osa->mode = obj->oo_attr.la_mode;
1373                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(osd), NULL,
1374                                  &osa->mode, 8);
1375         }
1376         if (valid & LA_SIZE) {
1377                 osa->size = obj->oo_attr.la_size = la->la_size;
1378                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(osd), NULL,
1379                                  &osa->size, 8);
1380         }
1381         if (valid & LA_NLINK) {
1382                 osa->nlink = obj->oo_attr.la_nlink = la->la_nlink;
1383                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(osd), NULL,
1384                                  &osa->nlink, 8);
1385         }
1386         if (valid & LA_RDEV) {
1387                 osa->rdev = obj->oo_attr.la_rdev = la->la_rdev;
1388                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_RDEV(osd), NULL,
1389                                  &osa->rdev, 8);
1390         }
1391         if (valid & LA_FLAGS) {
1392                 osa->flags = attrs_fs2zfs(la->la_flags);
1393                 /* many flags are not supported by zfs, so ensure a good cached
1394                  * copy */
1395                 obj->oo_attr.la_flags = attrs_zfs2fs(osa->flags);
1396 #ifdef ZFS_PROJINHERIT
1397                 if (obj->oo_with_projid && osd->od_projectused_dn)
1398                         osa->flags |= ZFS_PROJID;
1399 #endif
1400                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(osd), NULL,
1401                                  &osa->flags, 8);
1402         }
1403         if (valid & LA_UID) {
1404                 osa->uid = obj->oo_attr.la_uid = la->la_uid;
1405                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(osd), NULL,
1406                                  &osa->uid, 8);
1407         }
1408         if (valid & LA_GID) {
1409                 osa->gid = obj->oo_attr.la_gid = la->la_gid;
1410                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(osd), NULL,
1411                                  &osa->gid, 8);
1412         }
1413         obj->oo_attr.la_valid |= valid;
1414         write_unlock(&obj->oo_attr_lock);
1415
1416         LASSERT(cnt <= ARRAY_SIZE(osd_oti_get(env)->oti_attr_bulk));
1417         rc = osd_object_sa_bulk_update(obj, bulk, cnt, oh);
1418
1419 out:
1420         up_read(&obj->oo_guard);
1421         RETURN(rc);
1422 }
1423
1424 /*
1425  * Object creation.
1426  *
1427  * XXX temporary solution.
1428  */
1429
1430 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
1431                         struct dt_object *parent, struct dt_object *child,
1432                         umode_t child_mode)
1433 {
1434         LASSERT(ah);
1435
1436         ah->dah_parent = parent;
1437         ah->dah_mode = child_mode;
1438
1439         if (parent != NULL && !dt_object_remote(parent)) {
1440                 /* will help to find FID->ino at dt_insert("..") */
1441                 struct osd_object *pobj = osd_dt_obj(parent);
1442
1443                 osd_idc_find_and_init(env, osd_obj2dev(pobj), pobj);
1444         }
1445 }
1446
1447 static int osd_declare_create(const struct lu_env *env, struct dt_object *dt,
1448                               struct lu_attr *attr,
1449                               struct dt_allocation_hint *hint,
1450                               struct dt_object_format *dof,
1451                               struct thandle *handle)
1452 {
1453         const struct lu_fid     *fid = lu_object_fid(&dt->do_lu);
1454         struct osd_object       *obj = osd_dt_obj(dt);
1455         struct osd_device       *osd = osd_obj2dev(obj);
1456         struct osd_thandle      *oh;
1457         uint64_t                 zapid;
1458         dnode_t                 *dn;
1459         int                      rc, dnode_size;
1460         ENTRY;
1461
1462         LASSERT(dof);
1463
1464         switch (dof->dof_type) {
1465                 case DFT_REGULAR:
1466                 case DFT_SYM:
1467                 case DFT_NODE:
1468                         if (obj->oo_dt.do_body_ops == NULL)
1469                                 obj->oo_dt.do_body_ops = &osd_body_ops;
1470                         break;
1471                 default:
1472                         break;
1473         }
1474
1475         LASSERT(handle != NULL);
1476         oh = container_of0(handle, struct osd_thandle, ot_super);
1477         LASSERT(oh->ot_tx != NULL);
1478
1479         /* this is the minimum set of EAs on every Lustre object */
1480         obj->oo_ea_in_bonus = OSD_BASE_EA_IN_BONUS;
1481         /* reserve 32 bytes for extra stuff like ACLs */
1482         dnode_size = size_roundup_power2(obj->oo_ea_in_bonus + 32);
1483
1484         switch (dof->dof_type) {
1485                 case DFT_DIR:
1486                         dt->do_index_ops = &osd_dir_ops;
1487                         /* fallthrough */
1488                 case DFT_INDEX:
1489                         /* for zap create */
1490                         dmu_tx_hold_zap(oh->ot_tx, DMU_NEW_OBJECT, FALSE, NULL);
1491                         dmu_tx_hold_sa_create(oh->ot_tx, dnode_size);
1492                         break;
1493                 case DFT_REGULAR:
1494                 case DFT_SYM:
1495                 case DFT_NODE:
1496                         /* first, we'll create new object */
1497                         dmu_tx_hold_sa_create(oh->ot_tx, dnode_size);
1498                         break;
1499
1500                 default:
1501                         LBUG();
1502                         break;
1503         }
1504
1505         /* and we'll add it to some mapping */
1506         zapid = osd_get_name_n_idx(env, osd, fid, NULL, 0, &dn);
1507         osd_tx_hold_zap(oh->ot_tx, zapid, dn, TRUE, NULL);
1508
1509         /* will help to find FID->ino mapping at dt_insert() */
1510         osd_idc_find_and_init(env, osd, obj);
1511
1512         rc = osd_declare_quota(env, osd, attr->la_uid, attr->la_gid,
1513                                attr->la_projid, 1, oh, NULL, OSD_QID_INODE);
1514
1515         RETURN(rc);
1516 }
1517
1518 int __osd_attr_init(const struct lu_env *env, struct osd_device *osd,
1519                     struct osd_object *obj, sa_handle_t *sa_hdl, dmu_tx_t *tx,
1520                     struct lu_attr *la, uint64_t parent,
1521                     nvlist_t *xattr)
1522 {
1523         sa_bulk_attr_t *bulk = osd_oti_get(env)->oti_attr_bulk;
1524         struct osa_attr *osa = &osd_oti_get(env)->oti_osa;
1525         uint64_t gen;
1526         uint64_t crtime[2];
1527         inode_timespec_t now;
1528         int cnt;
1529         int rc;
1530         char *dxattr = NULL;
1531         size_t sa_size;
1532
1533
1534         LASSERT(sa_hdl);
1535
1536         gen = dmu_tx_get_txg(tx);
1537         gethrestime(&now);
1538         ZFS_TIME_ENCODE(&now, crtime);
1539
1540         osa->atime[0] = la->la_atime;
1541         osa->ctime[0] = la->la_ctime;
1542         osa->mtime[0] = la->la_mtime;
1543         osa->mode = la->la_mode;
1544         osa->uid = la->la_uid;
1545         osa->gid = la->la_gid;
1546         osa->rdev = la->la_rdev;
1547         osa->nlink = la->la_nlink;
1548         if (la->la_valid & LA_FLAGS)
1549                 osa->flags = attrs_fs2zfs(la->la_flags);
1550         else
1551                 osa->flags = 0;
1552         osa->size  = la->la_size;
1553 #ifdef ZFS_PROJINHERIT
1554         if (osd->od_projectused_dn) {
1555                 if (la->la_valid & LA_PROJID)
1556                         osa->projid = la->la_projid;
1557                 else
1558                         osa->projid = ZFS_DEFAULT_PROJID;
1559                 osa->flags |= ZFS_PROJID;
1560                 if (obj)
1561                         obj->oo_with_projid = 1;
1562         } else {
1563                 osa->flags &= ~ZFS_PROJID;
1564         }
1565 #endif
1566
1567         /*
1568          * we need to create all SA below upon object create.
1569          *
1570          * XXX The attribute order matters since the accounting callback relies
1571          * on static offsets (i.e. SA_*_OFFSET, see zfs_space_delta_cb()) to
1572          * look up the UID/GID/PROJID attributes. Moreover, the callback does
1573          * not seem to support the spill block.
1574          * We define attributes in the same order as SA_*_OFFSET in order to
1575          * work around the problem. See ORI-610.
1576          */
1577         cnt = 0;
1578         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(osd), NULL, &osa->mode, 8);
1579         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(osd), NULL, &osa->size, 8);
1580         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GEN(osd), NULL, &gen, 8);
1581         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(osd), NULL, &osa->uid, 8);
1582         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(osd), NULL, &osa->gid, 8);
1583         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_PARENT(osd), NULL, &parent, 8);
1584         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(osd), NULL, &osa->flags, 8);
1585         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(osd), NULL, osa->atime, 16);
1586         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(osd), NULL, osa->mtime, 16);
1587         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(osd), NULL, osa->ctime, 16);
1588         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CRTIME(osd), NULL, crtime, 16);
1589         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(osd), NULL, &osa->nlink, 8);
1590 #ifdef ZFS_PROJINHERIT
1591         if (osd->od_projectused_dn)
1592                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_PROJID(osd), NULL,
1593                                  &osa->projid, 8);
1594 #endif
1595         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_RDEV(osd), NULL, &osa->rdev, 8);
1596         LASSERT(cnt <= ARRAY_SIZE(osd_oti_get(env)->oti_attr_bulk));
1597
1598         if (xattr) {
1599                 rc = -nvlist_size(xattr, &sa_size, NV_ENCODE_XDR);
1600                 LASSERT(rc == 0);
1601
1602                 dxattr = osd_zio_buf_alloc(sa_size);
1603                 LASSERT(dxattr);
1604
1605                 rc = -nvlist_pack(xattr, &dxattr, &sa_size,
1606                                 NV_ENCODE_XDR, KM_SLEEP);
1607                 LASSERT(rc == 0);
1608
1609                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_DXATTR(osd),
1610                                 NULL, dxattr, sa_size);
1611         }
1612
1613         rc = -sa_replace_all_by_template(sa_hdl, bulk, cnt, tx);
1614         if (dxattr)
1615                 osd_zio_buf_free(dxattr, sa_size);
1616
1617         return rc;
1618 }
1619
1620 int osd_find_new_dnode(const struct lu_env *env, dmu_tx_t *tx,
1621                        uint64_t oid, dnode_t **dnp)
1622 {
1623         dmu_tx_hold_t *txh;
1624         int rc = 0;
1625
1626         /* take dnode_t from tx to save on dnode#->dnode_t lookup */
1627         for (txh = list_tail(&tx->tx_holds); txh;
1628              txh = list_prev(&tx->tx_holds, txh)) {
1629                 dnode_t *dn = txh->txh_dnode;
1630                 dmu_buf_impl_t *db;
1631
1632                 if (dn == NULL)
1633                         continue;
1634                 if (dn->dn_object != oid)
1635                         continue;
1636                 db = dn->dn_bonus;
1637                 if (db == NULL) {
1638                         rw_enter(&dn->dn_struct_rwlock, RW_WRITER);
1639                         if (dn->dn_bonus == NULL)
1640                                 dbuf_create_bonus(dn);
1641                         rw_exit(&dn->dn_struct_rwlock);
1642                 }
1643                 db = dn->dn_bonus;
1644                 LASSERT(db);
1645                 LASSERT(dn->dn_handle);
1646                 DB_DNODE_ENTER(db);
1647                 if (zfs_refcount_add(&db->db_holds, osd_obj_tag) == 1) {
1648                         zfs_refcount_add(&dn->dn_holds, osd_obj_tag);
1649                         atomic_inc_32(&dn->dn_dbufs_count);
1650                 }
1651                 *dnp = dn;
1652                 DB_DNODE_EXIT(db);
1653                 dbuf_read(db, NULL, DB_RF_MUST_SUCCEED | DB_RF_NOPREFETCH);
1654                 break;
1655         }
1656
1657         if (unlikely(*dnp == NULL))
1658                 rc = __osd_obj2dnode(tx->tx_objset, oid, dnp);
1659
1660         return rc;
1661 }
1662
1663 #ifdef HAVE_DMU_OBJECT_ALLOC_DNSIZE
1664 int osd_find_dnsize(struct osd_device *osd, int ea_in_bonus)
1665 {
1666         int dnsize;
1667
1668         if (osd->od_dnsize == ZFS_DNSIZE_AUTO) {
1669                 dnsize = DNODE_MIN_SIZE;
1670                 do {
1671                         if (DN_BONUS_SIZE(dnsize) >= ea_in_bonus + 32)
1672                                 break;
1673                         dnsize <<= 1;
1674                 } while (dnsize < DNODE_MAX_SIZE);
1675                 if (dnsize > DNODE_MAX_SIZE)
1676                         dnsize = DNODE_MAX_SIZE;
1677         } else if (osd->od_dnsize == ZFS_DNSIZE_1K) {
1678                 dnsize = 1024;
1679         } else if (osd->od_dnsize == ZFS_DNSIZE_2K) {
1680                 dnsize = 2048;
1681         } else if (osd->od_dnsize == ZFS_DNSIZE_4K) {
1682                 dnsize = 4096;
1683         } else if (osd->od_dnsize == ZFS_DNSIZE_8K) {
1684                 dnsize = 8192;
1685         } else if (osd->od_dnsize == ZFS_DNSIZE_16K) {
1686                 dnsize = 16384;
1687         } else {
1688                 dnsize = DNODE_MIN_SIZE;
1689         }
1690         return dnsize;
1691 }
1692 #endif
1693
1694 /*
1695  * The transaction passed to this routine must have
1696  * dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT) called and then assigned
1697  * to a transaction group.
1698  */
1699 int __osd_object_create(const struct lu_env *env, struct osd_device *osd,
1700                         struct osd_object *obj, const struct lu_fid *fid,
1701                         dnode_t **dnp, dmu_tx_t *tx, struct lu_attr *la)
1702 {
1703         dmu_object_type_t type = DMU_OT_PLAIN_FILE_CONTENTS;
1704         uint64_t oid;
1705         int size;
1706
1707         /* Use DMU_OTN_UINT8_METADATA for local objects so their data blocks
1708          * would get an additional ditto copy */
1709         if (unlikely(S_ISREG(la->la_mode) &&
1710                      fid_seq_is_local_file(fid_seq(fid))))
1711                 type = DMU_OTN_UINT8_METADATA;
1712
1713         /* Create a new DMU object using the default dnode size. */
1714         if (obj)
1715                 size = obj->oo_ea_in_bonus;
1716         else
1717                 size = OSD_BASE_EA_IN_BONUS;
1718         oid = osd_dmu_object_alloc(osd->od_os, type, 0,
1719                                    osd_find_dnsize(osd, size), tx);
1720
1721         LASSERT(la->la_valid & LA_MODE);
1722         la->la_size = 0;
1723         la->la_nlink = 1;
1724
1725         return osd_find_new_dnode(env, tx, oid, dnp);
1726 }
1727
1728 /*
1729  * The transaction passed to this routine must have
1730  * dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, ...) called and then assigned
1731  * to a transaction group.
1732  *
1733  * Using ZAP_FLAG_HASH64 will force the ZAP to always be a FAT ZAP.
1734  * This is fine for directories today, because storing the FID in the dirent
1735  * will also require a FAT ZAP.  If there is a new type of micro ZAP created
1736  * then we might need to re-evaluate the use of this flag and instead do
1737  * a conversion from the different internal ZAP hash formats being used. */
1738 int __osd_zap_create(const struct lu_env *env, struct osd_device *osd,
1739                      dnode_t **dnp, dmu_tx_t *tx, struct lu_attr *la,
1740                      unsigned dnsize, zap_flags_t flags)
1741 {
1742         uint64_t oid;
1743
1744         /* Assert that the transaction has been assigned to a
1745            transaction group. */
1746         LASSERT(tx->tx_txg != 0);
1747         *dnp = NULL;
1748
1749         oid = osd_zap_create_flags(osd->od_os, 0, flags | ZAP_FLAG_HASH64,
1750                                    DMU_OT_DIRECTORY_CONTENTS,
1751                                    14, /* == ZFS fzap_default_blockshift */
1752                                    DN_MAX_INDBLKSHIFT, /* indirect blockshift */
1753                                    dnsize, tx);
1754
1755         la->la_size = 2;
1756         la->la_nlink = 1;
1757
1758         return osd_find_new_dnode(env, tx, oid, dnp);
1759 }
1760
1761 static dnode_t *osd_mkidx(const struct lu_env *env, struct osd_object *obj,
1762                           struct lu_attr *la, struct osd_thandle *oh)
1763 {
1764         struct osd_device *osd = osd_obj2dev(obj);
1765         dnode_t *dn;
1766         int rc;
1767
1768         /* Index file should be created as regular file in order not to confuse
1769          * ZPL which could interpret them as directory.
1770          * We set ZAP_FLAG_UINT64_KEY to let ZFS know than we are going to use
1771          * binary keys */
1772         LASSERT(S_ISREG(la->la_mode));
1773         rc = __osd_zap_create(env, osd, &dn, oh->ot_tx, la,
1774                 osd_find_dnsize(osd, obj->oo_ea_in_bonus), ZAP_FLAG_UINT64_KEY);
1775         if (rc)
1776                 return ERR_PTR(rc);
1777         return dn;
1778 }
1779
1780 static dnode_t *osd_mkdir(const struct lu_env *env, struct osd_object *obj,
1781                           struct lu_attr *la, struct osd_thandle *oh)
1782 {
1783         struct osd_device *osd = osd_obj2dev(obj);
1784         dnode_t *dn;
1785         int rc;
1786
1787         LASSERT(S_ISDIR(la->la_mode));
1788         rc = __osd_zap_create(env, osd, &dn, oh->ot_tx, la,
1789                               osd_find_dnsize(osd, obj->oo_ea_in_bonus), 0);
1790         if (rc)
1791                 return ERR_PTR(rc);
1792         return dn;
1793 }
1794
1795 static dnode_t *osd_mkreg(const struct lu_env *env, struct osd_object *obj,
1796                           struct lu_attr *la, struct osd_thandle *oh)
1797 {
1798         const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
1799         struct osd_device *osd = osd_obj2dev(obj);
1800         dnode_t *dn;
1801         int rc;
1802
1803         LASSERT(S_ISREG(la->la_mode));
1804         rc = __osd_object_create(env, osd, obj, fid, &dn, oh->ot_tx, la);
1805         if (rc)
1806                 return ERR_PTR(rc);
1807
1808         if ((fid_is_idif(fid) || fid_is_norm(fid) || fid_is_echo(fid))) {
1809                 /* The minimum block size must be at least page size otherwise
1810                  * it will break the assumption in tgt_thread_big_cache where
1811                  * the array size is PTLRPC_MAX_BRW_PAGES. It will also affect
1812                  * RDMA due to subpage transfer size */
1813                 rc = -dmu_object_set_blocksize(osd->od_os, dn->dn_object,
1814                                                PAGE_SIZE, 0, oh->ot_tx);
1815                 if (unlikely(rc)) {
1816                         CERROR("%s: can't change blocksize: %d\n",
1817                                osd->od_svname, rc);
1818                         return ERR_PTR(rc);
1819                 }
1820         } else if ((fid_is_llog(fid))) {
1821                 rc = -dmu_object_set_blocksize(osd->od_os, dn->dn_object,
1822                                                LLOG_MIN_CHUNK_SIZE, 0, oh->ot_tx);
1823                 if (unlikely(rc)) {
1824                         CERROR("%s: can't change blocksize: %d\n",
1825                                osd->od_svname, rc);
1826                         return ERR_PTR(rc);
1827                 }
1828         }
1829
1830         return dn;
1831 }
1832
1833 static dnode_t *osd_mksym(const struct lu_env *env, struct osd_object *obj,
1834                           struct lu_attr *la, struct osd_thandle *oh)
1835 {
1836         dnode_t *dn;
1837         int rc;
1838
1839         LASSERT(S_ISLNK(la->la_mode));
1840         rc = __osd_object_create(env, osd_obj2dev(obj), obj,
1841                                  lu_object_fid(&obj->oo_dt.do_lu),
1842                                  &dn, oh->ot_tx, la);
1843         if (rc)
1844                 return ERR_PTR(rc);
1845         return dn;
1846 }
1847
1848 static dnode_t *osd_mknod(const struct lu_env *env, struct osd_object *obj,
1849                           struct lu_attr *la, struct osd_thandle *oh)
1850 {
1851         dnode_t *dn;
1852         int rc;
1853
1854         if (S_ISCHR(la->la_mode) || S_ISBLK(la->la_mode))
1855                 la->la_valid |= LA_RDEV;
1856
1857         rc = __osd_object_create(env, osd_obj2dev(obj), obj,
1858                                  lu_object_fid(&obj->oo_dt.do_lu),
1859                                  &dn, oh->ot_tx, la);
1860         if (rc)
1861                 return ERR_PTR(rc);
1862         return dn;
1863 }
1864
1865 typedef dnode_t *(*osd_obj_type_f)(const struct lu_env *env,
1866                                    struct osd_object *obj,
1867                                    struct lu_attr *la,
1868                                    struct osd_thandle *oh);
1869
1870 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
1871 {
1872         osd_obj_type_f result;
1873
1874         switch (type) {
1875         case DFT_DIR:
1876                 result = osd_mkdir;
1877                 break;
1878         case DFT_INDEX:
1879                 result = osd_mkidx;
1880                 break;
1881         case DFT_REGULAR:
1882                 result = osd_mkreg;
1883                 break;
1884         case DFT_SYM:
1885                 result = osd_mksym;
1886                 break;
1887         case DFT_NODE:
1888                 result = osd_mknod;
1889                 break;
1890         default:
1891                 LBUG();
1892                 break;
1893         }
1894         return result;
1895 }
1896
1897 /*
1898  * Concurrency: @dt is write locked.
1899  */
1900 static int osd_create(const struct lu_env *env, struct dt_object *dt,
1901                       struct lu_attr *attr, struct dt_allocation_hint *hint,
1902                       struct dt_object_format *dof, struct thandle *th)
1903 {
1904         struct osd_thread_info  *info = osd_oti_get(env);
1905         struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
1906         struct zpl_direntry     *zde = &info->oti_zde.lzd_reg;
1907         const struct lu_fid     *fid = lu_object_fid(&dt->do_lu);
1908         struct osd_object       *obj = osd_dt_obj(dt);
1909         struct osd_device       *osd = osd_obj2dev(obj);
1910         char                    *buf = info->oti_str;
1911         struct osd_thandle      *oh;
1912         dnode_t *dn = NULL, *zdn = NULL;
1913         uint64_t                 zapid, parent = 0;
1914         int                      rc;
1915         __u32 compat = 0;
1916
1917         ENTRY;
1918
1919         LASSERT(!fid_is_acct(fid));
1920
1921         /* concurrent create declarations should not see
1922          * the object inconsistent (db, attr, etc).
1923          * in regular cases acquisition should be cheap */
1924         down_write(&obj->oo_guard);
1925
1926         if (unlikely(dt_object_exists(dt)))
1927                 GOTO(out, rc = -EEXIST);
1928
1929         LASSERT(osd_invariant(obj));
1930         LASSERT(dof != NULL);
1931
1932         LASSERT(th != NULL);
1933         oh = container_of0(th, struct osd_thandle, ot_super);
1934
1935         LASSERT(obj->oo_dn == NULL);
1936
1937         /* to follow ZFS on-disk format we need
1938          * to initialize parent dnode properly */
1939         if (hint != NULL && hint->dah_parent != NULL &&
1940             !dt_object_remote(hint->dah_parent))
1941                 parent = osd_dt_obj(hint->dah_parent)->oo_dn->dn_object;
1942
1943         /* we may fix some attributes, better do not change the source */
1944         obj->oo_attr = *attr;
1945         obj->oo_attr.la_valid |= LA_SIZE | LA_NLINK | LA_TYPE;
1946
1947 #ifdef ZFS_PROJINHERIT
1948         if (osd->od_projectused_dn) {
1949                 if (!(obj->oo_attr.la_valid & LA_PROJID))
1950                         obj->oo_attr.la_projid = ZFS_DEFAULT_PROJID;
1951                 obj->oo_with_projid = 1;
1952         }
1953 #endif
1954
1955         dn = osd_create_type_f(dof->dof_type)(env, obj, &obj->oo_attr, oh);
1956         if (IS_ERR(dn)) {
1957                 rc = PTR_ERR(dn);
1958                 dn = NULL;
1959                 GOTO(out, rc);
1960         }
1961
1962         zde->zde_pad = 0;
1963         zde->zde_dnode = dn->dn_object;
1964         zde->zde_type = IFTODT(attr->la_mode & S_IFMT);
1965
1966         zapid = osd_get_name_n_idx(env, osd, fid, buf,
1967                                    sizeof(info->oti_str), &zdn);
1968         if (CFS_FAIL_CHECK(OBD_FAIL_OSD_NO_OI_ENTRY) ||
1969             (osd->od_is_ost && OBD_FAIL_CHECK(OBD_FAIL_OSD_COMPAT_NO_ENTRY)))
1970                 goto skip_add;
1971
1972         if (osd->od_is_ost && OBD_FAIL_CHECK(OBD_FAIL_OSD_COMPAT_INVALID_ENTRY))
1973                 zde->zde_dnode++;
1974
1975         rc = osd_zap_add(osd, zapid, zdn, buf, 8, 1, zde, oh->ot_tx);
1976         if (rc)
1977                 GOTO(out, rc);
1978
1979 skip_add:
1980         obj->oo_dn = dn;
1981         /* Now add in all of the "SA" attributes */
1982         rc = osd_sa_handle_get(obj);
1983         if (rc)
1984                 GOTO(out, rc);
1985
1986         rc = -nvlist_alloc(&obj->oo_sa_xattr, NV_UNIQUE_NAME, KM_SLEEP);
1987         if (rc)
1988                 GOTO(out, rc);
1989
1990         /* initialize LMA */
1991         if (fid_is_idif(fid) || (fid_is_norm(fid) && osd->od_is_ost))
1992                 compat |= LMAC_FID_ON_OST;
1993         lustre_lma_init(lma, fid, compat, 0);
1994         lustre_lma_swab(lma);
1995         rc = -nvlist_add_byte_array(obj->oo_sa_xattr, XATTR_NAME_LMA,
1996                                     (uchar_t *)lma, sizeof(*lma));
1997         if (rc)
1998                 GOTO(out, rc);
1999
2000         /* configure new osd object */
2001         obj->oo_parent = parent != 0 ? parent : zapid;
2002         obj->oo_late_attr_set = 1;
2003         rc = __osd_sa_xattr_schedule_update(env, obj, oh);
2004         if (rc)
2005                 GOTO(out, rc);
2006
2007         /* XXX: oo_lma_flags */
2008         obj->oo_dt.do_lu.lo_header->loh_attr |= obj->oo_attr.la_mode & S_IFMT;
2009         if (likely(!fid_is_acct(lu_object_fid(&obj->oo_dt.do_lu))))
2010                 /* no body operations for accounting objects */
2011                 obj->oo_dt.do_body_ops = &osd_body_ops;
2012
2013         osd_idc_find_and_init(env, osd, obj);
2014
2015 out:
2016         if (unlikely(rc && dn)) {
2017                 dmu_object_free(osd->od_os, dn->dn_object, oh->ot_tx);
2018                 osd_dnode_rele(dn);
2019                 obj->oo_dn = NULL;
2020         } else if (!rc) {
2021                 obj->oo_dt.do_lu.lo_header->loh_attr |= LOHA_EXISTS;
2022         }
2023         up_write(&obj->oo_guard);
2024         RETURN(rc);
2025 }
2026
2027 static int osd_declare_ref_add(const struct lu_env *env, struct dt_object *dt,
2028                                struct thandle *th)
2029 {
2030         osd_idc_find_and_init(env, osd_dev(dt->do_lu.lo_dev), osd_dt_obj(dt));
2031         return osd_declare_attr_set(env, dt, NULL, th);
2032 }
2033
2034 /*
2035  * Concurrency: @dt is write locked.
2036  */
2037 static int osd_ref_add(const struct lu_env *env, struct dt_object *dt,
2038                        struct thandle *handle)
2039 {
2040         struct osd_object       *obj = osd_dt_obj(dt);
2041         struct osd_thandle      *oh;
2042         struct osd_device       *osd = osd_obj2dev(obj);
2043         uint64_t                 nlink;
2044         int rc;
2045
2046         ENTRY;
2047
2048         down_read(&obj->oo_guard);
2049         if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
2050                 GOTO(out, rc = -ENOENT);
2051
2052         LASSERT(osd_invariant(obj));
2053         LASSERT(obj->oo_sa_hdl != NULL);
2054
2055         oh = container_of0(handle, struct osd_thandle, ot_super);
2056
2057         write_lock(&obj->oo_attr_lock);
2058         nlink = ++obj->oo_attr.la_nlink;
2059         write_unlock(&obj->oo_attr_lock);
2060
2061         rc = osd_object_sa_update(obj, SA_ZPL_LINKS(osd), &nlink, 8, oh);
2062
2063 out:
2064         up_read(&obj->oo_guard);
2065         RETURN(rc);
2066 }
2067
2068 static int osd_declare_ref_del(const struct lu_env *env, struct dt_object *dt,
2069                                struct thandle *handle)
2070 {
2071         osd_idc_find_and_init(env, osd_dev(dt->do_lu.lo_dev), osd_dt_obj(dt));
2072         return osd_declare_attr_set(env, dt, NULL, handle);
2073 }
2074
2075 /*
2076  * Concurrency: @dt is write locked.
2077  */
2078 static int osd_ref_del(const struct lu_env *env, struct dt_object *dt,
2079                        struct thandle *handle)
2080 {
2081         struct osd_object       *obj = osd_dt_obj(dt);
2082         struct osd_thandle      *oh;
2083         struct osd_device       *osd = osd_obj2dev(obj);
2084         uint64_t                 nlink;
2085         int                      rc;
2086
2087         ENTRY;
2088
2089         down_read(&obj->oo_guard);
2090
2091         if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
2092                 GOTO(out, rc = -ENOENT);
2093
2094         LASSERT(osd_invariant(obj));
2095         LASSERT(obj->oo_sa_hdl != NULL);
2096
2097         oh = container_of0(handle, struct osd_thandle, ot_super);
2098         LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
2099
2100         write_lock(&obj->oo_attr_lock);
2101         nlink = --obj->oo_attr.la_nlink;
2102         write_unlock(&obj->oo_attr_lock);
2103
2104         rc = osd_object_sa_update(obj, SA_ZPL_LINKS(osd), &nlink, 8, oh);
2105
2106 out:
2107         up_read(&obj->oo_guard);
2108         RETURN(rc);
2109 }
2110
2111 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
2112                            __u64 start, __u64 end)
2113 {
2114         struct osd_device *osd = osd_obj2dev(osd_dt_obj(dt));
2115         struct dmu_buf_impl *db = osd_dt_obj(dt)->oo_dn->dn_dbuf;
2116         uint64_t txg = 0;
2117         ENTRY;
2118
2119         if (osd->od_dt_dev.dd_rdonly)
2120                 RETURN(0);
2121
2122         mutex_enter(&db->db_mtx);
2123         if (db->db_last_dirty)
2124                 txg = db->db_last_dirty->dr_txg;
2125         mutex_exit(&db->db_mtx);
2126
2127         if (txg) {
2128                 /* the object is dirty or being synced */
2129                 if (osd_object_sync_delay_us < 0)
2130                         txg_wait_synced(dmu_objset_pool(osd->od_os), txg);
2131                 else
2132                         udelay(osd_object_sync_delay_us);
2133         }
2134
2135         RETURN(0);
2136 }
2137
2138 static int osd_invalidate(const struct lu_env *env, struct dt_object *dt)
2139 {
2140         return 0;
2141 }
2142
2143 static struct dt_object_operations osd_obj_ops = {
2144         .do_read_lock           = osd_read_lock,
2145         .do_write_lock          = osd_write_lock,
2146         .do_read_unlock         = osd_read_unlock,
2147         .do_write_unlock        = osd_write_unlock,
2148         .do_write_locked        = osd_write_locked,
2149         .do_attr_get            = osd_attr_get,
2150         .do_declare_attr_set    = osd_declare_attr_set,
2151         .do_attr_set            = osd_attr_set,
2152         .do_ah_init             = osd_ah_init,
2153         .do_declare_create      = osd_declare_create,
2154         .do_create              = osd_create,
2155         .do_declare_destroy     = osd_declare_destroy,
2156         .do_destroy             = osd_destroy,
2157         .do_index_try           = osd_index_try,
2158         .do_declare_ref_add     = osd_declare_ref_add,
2159         .do_ref_add             = osd_ref_add,
2160         .do_declare_ref_del     = osd_declare_ref_del,
2161         .do_ref_del             = osd_ref_del,
2162         .do_xattr_get           = osd_xattr_get,
2163         .do_declare_xattr_set   = osd_declare_xattr_set,
2164         .do_xattr_set           = osd_xattr_set,
2165         .do_declare_xattr_del   = osd_declare_xattr_del,
2166         .do_xattr_del           = osd_xattr_del,
2167         .do_xattr_list          = osd_xattr_list,
2168         .do_object_sync         = osd_object_sync,
2169         .do_invalidate          = osd_invalidate,
2170 };
2171
2172 static struct lu_object_operations osd_lu_obj_ops = {
2173         .loo_object_init        = osd_object_init,
2174         .loo_object_delete      = osd_object_delete,
2175         .loo_object_release     = osd_object_release,
2176         .loo_object_free        = osd_object_free,
2177         .loo_object_print       = osd_object_print,
2178         .loo_object_invariant   = osd_object_invariant,
2179 };
2180
2181 static int osd_otable_it_attr_get(const struct lu_env *env,
2182                                 struct dt_object *dt,
2183                                 struct lu_attr *attr)
2184 {
2185         attr->la_valid = 0;
2186         return 0;
2187 }
2188
2189 static struct dt_object_operations osd_obj_otable_it_ops = {
2190         .do_attr_get            = osd_otable_it_attr_get,
2191         .do_index_try           = osd_index_try,
2192 };
2193
2194 module_param(osd_object_sync_delay_us, int, 0644);
2195 MODULE_PARM_DESC(osd_object_sync_delay_us,
2196                  "If zero or larger delay N usec instead of doing object sync");