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.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2009, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2012, 2014, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/osd-zfs/osd_object.c
  37  *
  38  * Author: Alex Zhuravlev <bzzz@whamcloud.com>
  39  * Author: Mike Pershin <tappro@whamcloud.com>
  40  * Author: Johann Lombardi <johann@whamcloud.com>
  41  */
  42
  43 #define DEBUG_SUBSYSTEM S_OSD
  44
  45 #include <lustre_ver.h>
  46 #include <libcfs/libcfs.h>
  47 #include <obd_support.h>
  48 #include <lustre_net.h>
  49 #include <obd.h>
  50 #include <obd_class.h>
  51 #include <lustre_disk.h>
  52 #include <lustre_fid.h>
  53
  54 #include "osd_internal.h"
  55
  56 #include <sys/dnode.h>
  57 #include <sys/dbuf.h>
  58 #include <sys/spa.h>
  59 #include <sys/stat.h>
  60 #include <sys/zap.h>
  61 #include <sys/spa_impl.h>
  62 #include <sys/zfs_znode.h>
  63 #include <sys/dmu_tx.h>
  64 #include <sys/dmu_objset.h>
  65 #include <sys/dsl_prop.h>
  66 #include <sys/sa_impl.h>
  67 #include <sys/txg.h>
  68
  69 char *osd_obj_tag = "osd_object";
  70
  71 static struct dt_object_operations osd_obj_ops;
  72 static struct lu_object_operations osd_lu_obj_ops;
  73 extern struct dt_body_operations osd_body_ops;
  74 static struct dt_object_operations osd_obj_otable_it_ops;
  75
  76 extern struct kmem_cache *osd_object_kmem;
  77
  78 static void
  79 osd_object_sa_fini(struct osd_object *obj)
  80 {
  81         if (obj->oo_sa_hdl) {
  82                 sa_handle_destroy(obj->oo_sa_hdl);
  83                 obj->oo_sa_hdl = NULL;
  84         }
  85 }
  86
  87 static int
  88 osd_object_sa_init(struct osd_object *obj, struct osd_device *o)
  89 {
  90         int rc;
  91
  92         LASSERT(obj->oo_sa_hdl == NULL);
  93         LASSERT(obj->oo_db != NULL);
  94
  95         rc = -sa_handle_get(o->od_os, obj->oo_db->db_object, obj,
  96                             SA_HDL_PRIVATE, &obj->oo_sa_hdl);
  97         if (rc)
  98                 return rc;
  99
 100         /* Cache the xattr object id, valid for the life of the object */
 101         rc = -sa_lookup(obj->oo_sa_hdl, SA_ZPL_XATTR(o), &obj->oo_xattr, 8);
 102         if (rc == -ENOENT) {
 103                 obj->oo_xattr = ZFS_NO_OBJECT;
 104                 rc = 0;
 105         } else if (rc) {
 106                 osd_object_sa_fini(obj);
 107         }
 108
 109         return rc;
 110 }
 111
 112 /*
 113  * Add object to list of dirty objects in tx handle.
 114  */
 115 static void
 116 osd_object_sa_dirty_add(struct osd_object *obj, struct osd_thandle *oh)
 117 {
 118         if (!list_empty(&obj->oo_sa_linkage))
 119                 return;
 120
 121         down(&oh->ot_sa_lock);
 122         write_lock(&obj->oo_attr_lock);
 123         if (likely(list_empty(&obj->oo_sa_linkage)))
 124                 list_add(&obj->oo_sa_linkage, &oh->ot_sa_list);
 125         write_unlock(&obj->oo_attr_lock);
 126         up(&oh->ot_sa_lock);
 127 }
 128
 129 /*
 130  * Release spill block dbuf hold for all dirty SAs.
 131  */
 132 void osd_object_sa_dirty_rele(struct osd_thandle *oh)
 133 {
 134         struct osd_object *obj;
 135
 136         down(&oh->ot_sa_lock);
 137         while (!list_empty(&oh->ot_sa_list)) {
 138                 obj = list_entry(oh->ot_sa_list.next,
 139                                  struct osd_object, oo_sa_linkage);
 140                 sa_spill_rele(obj->oo_sa_hdl);
 141                 write_lock(&obj->oo_attr_lock);
 142                 list_del_init(&obj->oo_sa_linkage);
 143                 write_unlock(&obj->oo_attr_lock);
 144         }
 145         up(&oh->ot_sa_lock);
 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 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_handle_t     *sa_hdl;
 191         sa_bulk_attr_t  *bulk;
 192         int              cnt = 0;
 193         int              rc;
 194         ENTRY;
 195
 196         LASSERT(obj->oo_db != NULL);
 197
 198         rc = -sa_handle_get(o->od_os, obj->oo_db->db_object, NULL,
 199                             SA_HDL_PRIVATE, &sa_hdl);
 200         if (rc)
 201                 RETURN(rc);
 202
 203         OBD_ALLOC(bulk, sizeof(sa_bulk_attr_t) * 9);
 204         if (bulk == NULL)
 205                 GOTO(out_sa, rc = -ENOMEM);
 206
 207         la->la_valid |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE | LA_TYPE |
 208                         LA_SIZE | LA_UID | LA_GID | LA_FLAGS | LA_NLINK;
 209
 210         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(o), NULL, osa->atime, 16);
 211         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(o), NULL, osa->mtime, 16);
 212         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(o), NULL, osa->ctime, 16);
 213         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(o), NULL, &osa->mode, 8);
 214         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(o), NULL, &osa->size, 8);
 215         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(o), NULL, &osa->nlink, 8);
 216         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(o), NULL, &osa->uid, 8);
 217         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(o), NULL, &osa->gid, 8);
 218         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(o), NULL, &osa->flags, 8);
 219
 220         rc = -sa_bulk_lookup(sa_hdl, bulk, cnt);
 221         if (rc)
 222                 GOTO(out_bulk, rc);
 223
 224         la->la_atime = osa->atime[0];
 225         la->la_mtime = osa->mtime[0];
 226         la->la_ctime = osa->ctime[0];
 227         la->la_mode = osa->mode;
 228         la->la_uid = osa->uid;
 229         la->la_gid = osa->gid;
 230         la->la_nlink = osa->nlink;
 231         la->la_flags = attrs_zfs2fs(osa->flags);
 232         la->la_size = osa->size;
 233
 234         if (S_ISCHR(la->la_mode) || S_ISBLK(la->la_mode)) {
 235                 rc = -sa_lookup(sa_hdl, SA_ZPL_RDEV(o), &osa->rdev, 8);
 236                 if (rc)
 237                         GOTO(out_bulk, rc);
 238                 la->la_rdev = osa->rdev;
 239                 la->la_valid |= LA_RDEV;
 240         }
 241 out_bulk:
 242         OBD_FREE(bulk, sizeof(sa_bulk_attr_t) * 9);
 243 out_sa:
 244         sa_handle_destroy(sa_hdl);
 245
 246         RETURN(rc);
 247 }
 248
 249 int __osd_obj2dbuf(const struct lu_env *env, objset_t *os,
 250                    uint64_t oid, dmu_buf_t **dbp)
 251 {
 252         dmu_object_info_t *doi = &osd_oti_get(env)->oti_doi;
 253         int rc;
 254
 255         rc = -sa_buf_hold(os, oid, osd_obj_tag, dbp);
 256         if (rc)
 257                 return rc;
 258
 259         dmu_object_info_from_db(*dbp, doi);
 260         if (unlikely (oid != DMU_USERUSED_OBJECT &&
 261             oid != DMU_GROUPUSED_OBJECT && doi->doi_bonus_type != DMU_OT_SA)) {
 262                 sa_buf_rele(*dbp, osd_obj_tag);
 263                 *dbp = NULL;
 264                 return -EINVAL;
 265         }
 266
 267         LASSERT(*dbp);
 268         LASSERT((*dbp)->db_object == oid);
 269         LASSERT((*dbp)->db_offset == -1);
 270         LASSERT((*dbp)->db_data != NULL);
 271
 272         return 0;
 273 }
 274
 275 /*
 276  * Concurrency: no concurrent access is possible that early in object
 277  * life-cycle.
 278  */
 279 struct lu_object *osd_object_alloc(const struct lu_env *env,
 280                                    const struct lu_object_header *hdr,
 281                                    struct lu_device *d)
 282 {
 283         struct osd_object *mo;
 284
 285         OBD_SLAB_ALLOC_PTR_GFP(mo, osd_object_kmem, GFP_NOFS);
 286         if (mo != NULL) {
 287                 struct lu_object *l;
 288
 289                 l = &mo->oo_dt.do_lu;
 290                 dt_object_init(&mo->oo_dt, NULL, d);
 291                 mo->oo_dt.do_ops = &osd_obj_ops;
 292                 l->lo_ops = &osd_lu_obj_ops;
 293                 INIT_LIST_HEAD(&mo->oo_sa_linkage);
 294                 INIT_LIST_HEAD(&mo->oo_unlinked_linkage);
 295                 init_rwsem(&mo->oo_sem);
 296                 sema_init(&mo->oo_guard, 1);
 297                 rwlock_init(&mo->oo_attr_lock);
 298                 mo->oo_destroy = OSD_DESTROY_NONE;
 299                 return l;
 300         } else {
 301                 return NULL;
 302         }
 303 }
 304
 305 /*
 306  * Concurrency: shouldn't matter.
 307  */
 308 int osd_object_init0(const struct lu_env *env, struct osd_object *obj)
 309 {
 310         struct osd_device       *osd = osd_obj2dev(obj);
 311         const struct lu_fid     *fid = lu_object_fid(&obj->oo_dt.do_lu);
 312         int                      rc = 0;
 313         ENTRY;
 314
 315         if (obj->oo_db == NULL)
 316                 RETURN(0);
 317
 318         /* object exist */
 319
 320         rc = osd_object_sa_init(obj, osd);
 321         if (rc)
 322                 RETURN(rc);
 323
 324         /* cache attrs in object */
 325         rc = __osd_object_attr_get(env, osd, obj, &obj->oo_attr);
 326         if (rc)
 327                 RETURN(rc);
 328
 329         if (likely(!fid_is_acct(fid)))
 330                 /* no body operations for accounting objects */
 331                 obj->oo_dt.do_body_ops = &osd_body_ops;
 332
 333         /*
 334          * initialize object before marking it existing
 335          */
 336         obj->oo_dt.do_lu.lo_header->loh_attr |= obj->oo_attr.la_mode & S_IFMT;
 337
 338         smp_mb();
 339         obj->oo_dt.do_lu.lo_header->loh_attr |= LOHA_EXISTS;
 340
 341         RETURN(0);
 342 }
 343
 344 static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
 345 {
 346         struct osd_thread_info  *info = osd_oti_get(env);
 347         struct lu_buf           buf;
 348         int                     rc;
 349         struct lustre_mdt_attrs *lma;
 350         ENTRY;
 351
 352         CLASSERT(sizeof(info->oti_buf) >= sizeof(*lma));
 353         lma = (struct lustre_mdt_attrs *)info->oti_buf;
 354         buf.lb_buf = lma;
 355         buf.lb_len = sizeof(info->oti_buf);
 356
 357         rc = osd_xattr_get(env, &obj->oo_dt, &buf, XATTR_NAME_LMA);
 358         if (rc > 0) {
 359                 rc = 0;
 360                 lustre_lma_swab(lma);
 361                 if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
 362                              CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT))) {
 363                         CWARN("%s: unsupported incompat LMA feature(s) %#x for "
 364                               "fid = "DFID"\n", osd_obj2dev(obj)->od_svname,
 365                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
 366                               PFID(lu_object_fid(&obj->oo_dt.do_lu)));
 367                         rc = -EOPNOTSUPP;
 368                 }
 369         } else if (rc == -ENODATA) {
 370                 /* haven't initialize LMA xattr */
 371                 rc = 0;
 372         }
 373
 374         RETURN(rc);
 375 }
 376
 377 /*
 378  * Concurrency: no concurrent access is possible that early in object
 379  * life-cycle.
 380  */
 381 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
 382                            const struct lu_object_conf *conf)
 383 {
 384         struct osd_object       *obj = osd_obj(l);
 385         struct osd_device       *osd = osd_obj2dev(obj);
 386         uint64_t                 oid;
 387         int                      rc;
 388         ENTRY;
 389
 390         LASSERT(osd_invariant(obj));
 391
 392         if (fid_is_otable_it(&l->lo_header->loh_fid)) {
 393                 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
 394                 l->lo_header->loh_attr |= LOHA_EXISTS;
 395                 RETURN(0);
 396         }
 397
 398         rc = osd_fid_lookup(env, osd, lu_object_fid(l), &oid);
 399         if (rc == 0) {
 400                 LASSERT(obj->oo_db == NULL);
 401                 rc = __osd_obj2dbuf(env, osd->od_os, oid, &obj->oo_db);
 402                 if (rc != 0) {
 403                         CERROR("%s: lookup "DFID"/"LPX64" failed: rc = %d\n",
 404                                osd->od_svname, PFID(lu_object_fid(l)), oid, rc);
 405                         GOTO(out, rc);
 406                 }
 407                 LASSERT(obj->oo_db);
 408                 rc = osd_object_init0(env, obj);
 409                 if (rc != 0)
 410                         GOTO(out, rc);
 411
 412                 rc = osd_check_lma(env, obj);
 413                 if (rc != 0)
 414                         GOTO(out, rc);
 415         } else if (rc == -ENOENT) {
 416                 rc = 0;
 417         }
 418         LASSERT(osd_invariant(obj));
 419 out:
 420         RETURN(rc);
 421 }
 422
 423 /*
 424  * Concurrency: no concurrent access is possible that late in object
 425  * life-cycle.
 426  */
 427 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
 428 {
 429         struct osd_object *obj = osd_obj(l);
 430
 431         LASSERT(osd_invariant(obj));
 432
 433         dt_object_fini(&obj->oo_dt);
 434         OBD_SLAB_FREE_PTR(obj, osd_object_kmem);
 435 }
 436
 437 static int
 438 osd_object_unlinked_add(struct osd_object *obj, struct osd_thandle *oh)
 439 {
 440         int rc = -EBUSY;
 441
 442         down(&obj->oo_guard);
 443
 444         LASSERT(obj->oo_destroy == OSD_DESTROY_ASYNC);
 445
 446         if (likely(list_empty(&obj->oo_unlinked_linkage))) {
 447                 list_add(&obj->oo_unlinked_linkage, &oh->ot_unlinked_list);
 448                 rc = 0;
 449         }
 450
 451         up(&obj->oo_guard);
 452
 453         return rc;
 454 }
 455
 456 /* Default to max data size covered by a level-1 indirect block */
 457 static unsigned long osd_sync_destroy_max_size =
 458         1UL << (DN_MAX_INDBLKSHIFT - SPA_BLKPTRSHIFT + SPA_MAXBLOCKSHIFT);
 459 CFS_MODULE_PARM(osd_sync_destroy_max_size, "ul", ulong, 0444,
 460                 "Maximum object size to use synchronous destroy.");
 461
 462 static inline void
 463 osd_object_set_destroy_type(struct osd_object *obj)
 464 {
 465         /*
 466          * Lock-less OST_WRITE can race with OST_DESTROY, so set destroy type
 467          * only once and use it consistently thereafter.
 468          */
 469         down(&obj->oo_guard);
 470         if (obj->oo_destroy == OSD_DESTROY_NONE) {
 471                 if (obj->oo_attr.la_size <= osd_sync_destroy_max_size)
 472                         obj->oo_destroy = OSD_DESTROY_SYNC;
 473                 else /* Larger objects are destroyed asynchronously */
 474                         obj->oo_destroy = OSD_DESTROY_ASYNC;
 475         }
 476         up(&obj->oo_guard);
 477 }
 478
 479 static int osd_declare_object_destroy(const struct lu_env *env,
 480                                       struct dt_object *dt,
 481                                       struct thandle *th)
 482 {
 483         char                    *buf = osd_oti_get(env)->oti_str;
 484         const struct lu_fid     *fid = lu_object_fid(&dt->do_lu);
 485         struct osd_object       *obj = osd_dt_obj(dt);
 486         struct osd_device       *osd = osd_obj2dev(obj);
 487         struct osd_thandle      *oh;
 488         int                      rc;
 489         uint64_t                 zapid;
 490         ENTRY;
 491
 492         LASSERT(th != NULL);
 493         LASSERT(dt_object_exists(dt));
 494
 495         oh = container_of0(th, struct osd_thandle, ot_super);
 496         LASSERT(oh->ot_tx != NULL);
 497
 498         /* declare that we'll remove object from fid-dnode mapping */
 499         zapid = osd_get_name_n_idx(env, osd, fid, buf);
 500         dmu_tx_hold_bonus(oh->ot_tx, zapid);
 501         dmu_tx_hold_zap(oh->ot_tx, zapid, FALSE, buf);
 502
 503         osd_declare_xattrs_destroy(env, obj, oh);
 504
 505         /* declare that we'll remove object from inode accounting ZAPs */
 506         dmu_tx_hold_bonus(oh->ot_tx, osd->od_iusr_oid);
 507         dmu_tx_hold_zap(oh->ot_tx, osd->od_iusr_oid, FALSE, buf);
 508         dmu_tx_hold_bonus(oh->ot_tx, osd->od_igrp_oid);
 509         dmu_tx_hold_zap(oh->ot_tx, osd->od_igrp_oid, FALSE, buf);
 510
 511         /* one less inode */
 512         rc = osd_declare_quota(env, osd, obj->oo_attr.la_uid,
 513                                obj->oo_attr.la_gid, -1, oh, false, NULL, false);
 514         if (rc)
 515                 RETURN(rc);
 516
 517         /* data to be truncated */
 518         rc = osd_declare_quota(env, osd, obj->oo_attr.la_uid,
 519                                obj->oo_attr.la_gid, 0, oh, true, NULL, false);
 520         if (rc)
 521                 RETURN(rc);
 522
 523         osd_object_set_destroy_type(obj);
 524         if (obj->oo_destroy == OSD_DESTROY_SYNC)
 525                 dmu_tx_hold_free(oh->ot_tx, obj->oo_db->db_object,
 526                                  0, DMU_OBJECT_END);
 527         else
 528                 dmu_tx_hold_zap(oh->ot_tx, osd->od_unlinkedid, TRUE, NULL);
 529
 530         RETURN(0);
 531 }
 532
 533 static int osd_object_destroy(const struct lu_env *env,
 534                               struct dt_object *dt, struct thandle *th)
 535 {
 536         char                    *buf = osd_oti_get(env)->oti_str;
 537         struct osd_object       *obj = osd_dt_obj(dt);
 538         struct osd_device       *osd = osd_obj2dev(obj);
 539         const struct lu_fid     *fid = lu_object_fid(&dt->do_lu);
 540         struct osd_thandle      *oh;
 541         int                      rc;
 542         uint64_t                 oid, zapid;
 543         ENTRY;
 544
 545         LASSERT(obj->oo_db != NULL);
 546         LASSERT(dt_object_exists(dt));
 547         LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
 548
 549         oh = container_of0(th, struct osd_thandle, ot_super);
 550         LASSERT(oh != NULL);
 551         LASSERT(oh->ot_tx != NULL);
 552
 553         /* remove obj ref from index dir (it depends) */
 554         zapid = osd_get_name_n_idx(env, osd, fid, buf);
 555         rc = -zap_remove(osd->od_os, zapid, buf, oh->ot_tx);
 556         if (rc) {
 557                 CERROR("%s: zap_remove(%s) failed: rc = %d\n",
 558                        osd->od_svname, buf, rc);
 559                 GOTO(out, rc);
 560         }
 561
 562         rc = osd_xattrs_destroy(env, obj, oh);
 563         if (rc) {
 564                 CERROR("%s: cannot destroy xattrs for %s: rc = %d\n",
 565                        osd->od_svname, buf, rc);
 566                 GOTO(out, rc);
 567         }
 568
 569         /* Remove object from inode accounting. It is not fatal for the destroy
 570          * operation if something goes wrong while updating accounting, but we
 571          * still log an error message to notify the administrator */
 572         rc = -zap_increment_int(osd->od_os, osd->od_iusr_oid,
 573                                 obj->oo_attr.la_uid, -1, oh->ot_tx);
 574         if (rc)
 575                 CERROR("%s: failed to remove "DFID" from accounting ZAP for usr"
 576                        " %d: rc = %d\n", osd->od_svname, PFID(fid),
 577                        obj->oo_attr.la_uid, rc);
 578         rc = -zap_increment_int(osd->od_os, osd->od_igrp_oid,
 579                                 obj->oo_attr.la_gid, -1, oh->ot_tx);
 580         if (rc)
 581                 CERROR("%s: failed to remove "DFID" from accounting ZAP for grp"
 582                        " %d: rc = %d\n", osd->od_svname, PFID(fid),
 583                        obj->oo_attr.la_gid, rc);
 584
 585         oid = obj->oo_db->db_object;
 586         if (obj->oo_destroy == OSD_DESTROY_SYNC) {
 587                 rc = -dmu_object_free(osd->od_os, oid, oh->ot_tx);
 588                 if (rc)
 589                         CERROR("%s: failed to free %s "LPU64": rc = %d\n",
 590                                osd->od_svname, buf, oid, rc);
 591         } else { /* asynchronous destroy */
 592                 rc = osd_object_unlinked_add(obj, oh);
 593                 if (rc)
 594                         GOTO(out, rc);
 595
 596                 rc = -zap_add_int(osd->od_os, osd->od_unlinkedid,
 597                                   oid, oh->ot_tx);
 598                 if (rc)
 599                         CERROR("%s: zap_add_int() failed %s "LPU64": rc = %d\n",
 600                                osd->od_svname, buf, oid, rc);
 601         }
 602
 603 out:
 604         /* not needed in the cache anymore */
 605         set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
 606
 607         RETURN (0);
 608 }
 609
 610 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
 611 {
 612         struct osd_object *obj = osd_obj(l);
 613
 614         if (obj->oo_db != NULL) {
 615                 osd_object_sa_fini(obj);
 616                 if (obj->oo_sa_xattr) {
 617                         nvlist_free(obj->oo_sa_xattr);
 618                         obj->oo_sa_xattr = NULL;
 619                 }
 620                 sa_buf_rele(obj->oo_db, osd_obj_tag);
 621                 list_del(&obj->oo_sa_linkage);
 622                 obj->oo_db = NULL;
 623         }
 624 }
 625
 626 /*
 627  * Concurrency: ->loo_object_release() is called under site spin-lock.
 628  */
 629 static void osd_object_release(const struct lu_env *env,
 630                                struct lu_object *l)
 631 {
 632 }
 633
 634 /*
 635  * Concurrency: shouldn't matter.
 636  */
 637 static int osd_object_print(const struct lu_env *env, void *cookie,
 638                             lu_printer_t p, const struct lu_object *l)
 639 {
 640         struct osd_object *o = osd_obj(l);
 641
 642         return (*p)(env, cookie, LUSTRE_OSD_ZFS_NAME"-object@%p", o);
 643 }
 644
 645 static void osd_object_read_lock(const struct lu_env *env,
 646                                  struct dt_object *dt, unsigned role)
 647 {
 648         struct osd_object *obj = osd_dt_obj(dt);
 649
 650         LASSERT(osd_invariant(obj));
 651
 652         down_read(&obj->oo_sem);
 653 }
 654
 655 static void osd_object_write_lock(const struct lu_env *env,
 656                                   struct dt_object *dt, unsigned role)
 657 {
 658         struct osd_object *obj = osd_dt_obj(dt);
 659
 660         LASSERT(osd_invariant(obj));
 661
 662         down_write(&obj->oo_sem);
 663 }
 664
 665 static void osd_object_read_unlock(const struct lu_env *env,
 666                                    struct dt_object *dt)
 667 {
 668         struct osd_object *obj = osd_dt_obj(dt);
 669
 670         LASSERT(osd_invariant(obj));
 671         up_read(&obj->oo_sem);
 672 }
 673
 674 static void osd_object_write_unlock(const struct lu_env *env,
 675                                     struct dt_object *dt)
 676 {
 677         struct osd_object *obj = osd_dt_obj(dt);
 678
 679         LASSERT(osd_invariant(obj));
 680         up_write(&obj->oo_sem);
 681 }
 682
 683 static int osd_object_write_locked(const struct lu_env *env,
 684                                    struct dt_object *dt)
 685 {
 686         struct osd_object *obj = osd_dt_obj(dt);
 687         int rc = 1;
 688
 689         LASSERT(osd_invariant(obj));
 690
 691         if (down_write_trylock(&obj->oo_sem)) {
 692                 rc = 0;
 693                 up_write(&obj->oo_sem);
 694         }
 695         return rc;
 696 }
 697
 698 static int osd_attr_get(const struct lu_env *env,
 699                         struct dt_object *dt,
 700                         struct lu_attr *attr)
 701 {
 702         struct osd_object       *obj = osd_dt_obj(dt);
 703         uint64_t                 blocks;
 704         uint32_t                 blksize;
 705
 706         LASSERT(dt_object_exists(dt));
 707         LASSERT(osd_invariant(obj));
 708         LASSERT(obj->oo_db);
 709
 710         read_lock(&obj->oo_attr_lock);
 711         *attr = obj->oo_attr;
 712         read_unlock(&obj->oo_attr_lock);
 713
 714         /* with ZFS_DEBUG zrl_add_debug() called by DB_DNODE_ENTER()
 715          * from within sa_object_size() can block on a mutex, so
 716          * we can't call sa_object_size() holding rwlock */
 717         sa_object_size(obj->oo_sa_hdl, &blksize, &blocks);
 718         /* we do not control size of indices, so always calculate
 719          * it from number of blocks reported by DMU */
 720         if (S_ISDIR(attr->la_mode))
 721                 attr->la_size = 512 * blocks;
 722         /* Block size may be not set; suggest maximal I/O transfers. */
 723         if (blksize == 0)
 724                 blksize = osd_spa_maxblocksize(
 725                         dmu_objset_spa(osd_obj2dev(obj)->od_os));
 726
 727         attr->la_blksize = blksize;
 728         attr->la_blocks = blocks;
 729         attr->la_valid |= LA_BLOCKS | LA_BLKSIZE;
 730
 731         return 0;
 732 }
 733
 734 /* Simple wrapper on top of qsd API which implement quota transfer for osd
 735  * setattr needs. As a reminder, only the root user can change ownership of
 736  * a file, that's why EDQUOT & EINPROGRESS errors are discarded */
 737 static inline int qsd_transfer(const struct lu_env *env,
 738                                struct qsd_instance *qsd,
 739                                struct lquota_trans *trans, int qtype,
 740                                __u64 orig_id, __u64 new_id, __u64 bspace,
 741                                struct lquota_id_info *qi)
 742 {
 743         int     rc;
 744
 745         if (unlikely(qsd == NULL))
 746                 return 0;
 747
 748         LASSERT(qtype >= 0 && qtype < MAXQUOTAS);
 749         qi->lqi_type = qtype;
 750
 751         /* inode accounting */
 752         qi->lqi_is_blk = false;
 753
 754         /* one more inode for the new owner ... */
 755         qi->lqi_id.qid_uid = new_id;
 756         qi->lqi_space      = 1;
 757         rc = qsd_op_begin(env, qsd, trans, qi, NULL);
 758         if (rc == -EDQUOT || rc == -EINPROGRESS)
 759                 rc = 0;
 760         if (rc)
 761                 return rc;
 762
 763         /* and one less inode for the current id */
 764         qi->lqi_id.qid_uid = orig_id;;
 765         qi->lqi_space      = -1;
 766         /* can't get EDQUOT when reducing usage */
 767         rc = qsd_op_begin(env, qsd, trans, qi, NULL);
 768         if (rc == -EINPROGRESS)
 769                 rc = 0;
 770         if (rc)
 771                 return rc;
 772
 773         /* block accounting */
 774         qi->lqi_is_blk = true;
 775
 776         /* more blocks for the new owner ... */
 777         qi->lqi_id.qid_uid = new_id;
 778         qi->lqi_space      = bspace;
 779         rc = qsd_op_begin(env, qsd, trans, qi, NULL);
 780         if (rc == -EDQUOT || rc == -EINPROGRESS)
 781                 rc = 0;
 782         if (rc)
 783                 return rc;
 784
 785         /* and finally less blocks for the current owner */
 786         qi->lqi_id.qid_uid = orig_id;
 787         qi->lqi_space      = -bspace;
 788         rc = qsd_op_begin(env, qsd, trans, qi, NULL);
 789         /* can't get EDQUOT when reducing usage */
 790         if (rc == -EINPROGRESS)
 791                 rc = 0;
 792         return rc;
 793 }
 794
 795 static int osd_declare_attr_set(const struct lu_env *env,
 796                                 struct dt_object *dt,
 797                                 const struct lu_attr *attr,
 798                                 struct thandle *handle)
 799 {
 800         struct osd_thread_info  *info = osd_oti_get(env);
 801         char                    *buf = osd_oti_get(env)->oti_str;
 802         struct osd_object       *obj = osd_dt_obj(dt);
 803         struct osd_device       *osd = osd_obj2dev(obj);
 804         struct osd_thandle      *oh;
 805         uint64_t                 bspace;
 806         uint32_t                 blksize;
 807         int                      rc;
 808         ENTRY;
 809
 810         if (!dt_object_exists(dt)) {
 811                 /* XXX: sanity check that object creation is declared */
 812                 RETURN(0);
 813         }
 814
 815         LASSERT(handle != NULL);
 816         LASSERT(osd_invariant(obj));
 817
 818         oh = container_of0(handle, struct osd_thandle, ot_super);
 819
 820         LASSERT(obj->oo_sa_hdl != NULL);
 821         LASSERT(oh->ot_tx != NULL);
 822         dmu_tx_hold_sa(oh->ot_tx, obj->oo_sa_hdl, 0);
 823         if (oh->ot_tx->tx_err != 0)
 824                 RETURN(-oh->ot_tx->tx_err);
 825
 826         sa_object_size(obj->oo_sa_hdl, &blksize, &bspace);
 827         bspace = toqb(bspace * blksize);
 828
 829         if (attr && attr->la_valid & LA_UID) {
 830                 /* account for user inode tracking ZAP update */
 831                 dmu_tx_hold_bonus(oh->ot_tx, osd->od_iusr_oid);
 832                 dmu_tx_hold_zap(oh->ot_tx, osd->od_iusr_oid, TRUE, buf);
 833
 834                 /* quota enforcement for user */
 835                 if (attr->la_uid != obj->oo_attr.la_uid) {
 836                         rc = qsd_transfer(env, osd->od_quota_slave,
 837                                           &oh->ot_quota_trans, USRQUOTA,
 838                                           obj->oo_attr.la_uid, attr->la_uid,
 839                                           bspace, &info->oti_qi);
 840                         if (rc)
 841                                 RETURN(rc);
 842                 }
 843         }
 844         if (attr && attr->la_valid & LA_GID) {
 845                 /* account for user inode tracking ZAP update */
 846                 dmu_tx_hold_bonus(oh->ot_tx, osd->od_igrp_oid);
 847                 dmu_tx_hold_zap(oh->ot_tx, osd->od_igrp_oid, TRUE, buf);
 848
 849                 /* quota enforcement for group */
 850                 if (attr->la_gid != obj->oo_attr.la_gid) {
 851                         rc = qsd_transfer(env, osd->od_quota_slave,
 852                                           &oh->ot_quota_trans, GRPQUOTA,
 853                                           obj->oo_attr.la_gid, attr->la_gid,
 854                                           bspace, &info->oti_qi);
 855                         if (rc)
 856                                 RETURN(rc);
 857                 }
 858         }
 859
 860         RETURN(0);
 861 }
 862
 863 /*
 864  * Set the attributes of an object
 865  *
 866  * The transaction passed to this routine must have
 867  * dmu_tx_hold_bonus(tx, oid) called and then assigned
 868  * to a transaction group.
 869  */
 870 static int osd_attr_set(const struct lu_env *env, struct dt_object *dt,
 871                         const struct lu_attr *la, struct thandle *handle)
 872 {
 873         struct osd_object       *obj = osd_dt_obj(dt);
 874         struct osd_device       *osd = osd_obj2dev(obj);
 875         struct osd_thandle      *oh;
 876         struct osa_attr         *osa = &osd_oti_get(env)->oti_osa;
 877         sa_bulk_attr_t          *bulk;
 878         __u64                    valid = la->la_valid;
 879         int                      cnt;
 880         int                      rc = 0;
 881
 882         ENTRY;
 883         LASSERT(handle != NULL);
 884         LASSERT(dt_object_exists(dt));
 885         LASSERT(osd_invariant(obj));
 886         LASSERT(obj->oo_sa_hdl);
 887
 888         oh = container_of0(handle, struct osd_thandle, ot_super);
 889         /* Assert that the transaction has been assigned to a
 890            transaction group. */
 891         LASSERT(oh->ot_tx->tx_txg != 0);
 892
 893         /* Only allow set size for regular file */
 894         if (!S_ISREG(dt->do_lu.lo_header->loh_attr))
 895                 valid &= ~(LA_SIZE | LA_BLOCKS);
 896
 897         if (valid == 0)
 898                 RETURN(0);
 899
 900         OBD_ALLOC(bulk, sizeof(sa_bulk_attr_t) * 10);
 901         if (bulk == NULL)
 902                 RETURN(-ENOMEM);
 903
 904         /* do both accounting updates outside oo_attr_lock below */
 905         if ((valid & LA_UID) && (la->la_uid != obj->oo_attr.la_uid)) {
 906                 /* Update user accounting. Failure isn't fatal, but we still
 907                  * log an error message */
 908                 rc = -zap_increment_int(osd->od_os, osd->od_iusr_oid,
 909                                         la->la_uid, 1, oh->ot_tx);
 910                 if (rc)
 911                         CERROR("%s: failed to update accounting ZAP for user "
 912                                 "%d (%d)\n", osd->od_svname, la->la_uid, rc);
 913                 rc = -zap_increment_int(osd->od_os, osd->od_iusr_oid,
 914                                         obj->oo_attr.la_uid, -1, oh->ot_tx);
 915                 if (rc)
 916                         CERROR("%s: failed to update accounting ZAP for user "
 917                                 "%d (%d)\n", osd->od_svname,
 918                                 obj->oo_attr.la_uid, rc);
 919         }
 920         if ((valid & LA_GID) && (la->la_gid != obj->oo_attr.la_gid)) {
 921                 /* Update group accounting. Failure isn't fatal, but we still
 922                  * log an error message */
 923                 rc = -zap_increment_int(osd->od_os, osd->od_igrp_oid,
 924                                         la->la_gid, 1, oh->ot_tx);
 925                 if (rc)
 926                         CERROR("%s: failed to update accounting ZAP for user "
 927                                 "%d (%d)\n", osd->od_svname, la->la_gid, rc);
 928                 rc = -zap_increment_int(osd->od_os, osd->od_igrp_oid,
 929                                         obj->oo_attr.la_gid, -1, oh->ot_tx);
 930                 if (rc)
 931                         CERROR("%s: failed to update accounting ZAP for user "
 932                                 "%d (%d)\n", osd->od_svname,
 933                                 obj->oo_attr.la_gid, rc);
 934         }
 935
 936         write_lock(&obj->oo_attr_lock);
 937         cnt = 0;
 938         if (valid & LA_ATIME) {
 939                 osa->atime[0] = obj->oo_attr.la_atime = la->la_atime;
 940                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(osd), NULL,
 941                                  osa->atime, 16);
 942         }
 943         if (valid & LA_MTIME) {
 944                 osa->mtime[0] = obj->oo_attr.la_mtime = la->la_mtime;
 945                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(osd), NULL,
 946                                  osa->mtime, 16);
 947         }
 948         if (valid & LA_CTIME) {
 949                 osa->ctime[0] = obj->oo_attr.la_ctime = la->la_ctime;
 950                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(osd), NULL,
 951                                  osa->ctime, 16);
 952         }
 953         if (valid & LA_MODE) {
 954                 /* mode is stored along with type, so read it first */
 955                 obj->oo_attr.la_mode = (obj->oo_attr.la_mode & S_IFMT) |
 956                         (la->la_mode & ~S_IFMT);
 957                 osa->mode = obj->oo_attr.la_mode;
 958                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(osd), NULL,
 959                                  &osa->mode, 8);
 960         }
 961         if (valid & LA_SIZE) {
 962                 osa->size = obj->oo_attr.la_size = la->la_size;
 963                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(osd), NULL,
 964                                  &osa->size, 8);
 965         }
 966         if (valid & LA_NLINK) {
 967                 osa->nlink = obj->oo_attr.la_nlink = la->la_nlink;
 968                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(osd), NULL,
 969                                  &osa->nlink, 8);
 970         }
 971         if (valid & LA_RDEV) {
 972                 osa->rdev = obj->oo_attr.la_rdev = la->la_rdev;
 973                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_RDEV(osd), NULL,
 974                                  &osa->rdev, 8);
 975         }
 976         if (valid & LA_FLAGS) {
 977                 osa->flags = attrs_fs2zfs(la->la_flags);
 978                 /* many flags are not supported by zfs, so ensure a good cached
 979                  * copy */
 980                 obj->oo_attr.la_flags = attrs_zfs2fs(osa->flags);
 981                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(osd), NULL,
 982                                  &osa->flags, 8);
 983         }
 984         if (valid & LA_UID) {
 985                 osa->uid = obj->oo_attr.la_uid = la->la_uid;
 986                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(osd), NULL,
 987                                  &osa->uid, 8);
 988         }
 989         if (valid & LA_GID) {
 990                 osa->gid = obj->oo_attr.la_gid = la->la_gid;
 991                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(osd), NULL,
 992                                  &osa->gid, 8);
 993         }
 994         obj->oo_attr.la_valid |= valid;
 995         write_unlock(&obj->oo_attr_lock);
 996
 997         rc = osd_object_sa_bulk_update(obj, bulk, cnt, oh);
 998
 999         OBD_FREE(bulk, sizeof(sa_bulk_attr_t) * 10);
1000         RETURN(rc);
1001 }
1002
1003 /*
1004  * Object creation.
1005  *
1006  * XXX temporary solution.
1007  */
1008
1009 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
1010                         struct dt_object *parent, struct dt_object *child,
1011                         umode_t child_mode)
1012 {
1013         LASSERT(ah);
1014
1015         ah->dah_parent = parent;
1016         ah->dah_mode = child_mode;
1017 }
1018
1019 static int osd_declare_object_create(const struct lu_env *env,
1020                                      struct dt_object *dt,
1021                                      struct lu_attr *attr,
1022                                      struct dt_allocation_hint *hint,
1023                                      struct dt_object_format *dof,
1024                                      struct thandle *handle)
1025 {
1026         char                    *buf = osd_oti_get(env)->oti_str;
1027         const struct lu_fid     *fid = lu_object_fid(&dt->do_lu);
1028         struct osd_object       *obj = osd_dt_obj(dt);
1029         struct osd_device       *osd = osd_obj2dev(obj);
1030         struct osd_thandle      *oh;
1031         uint64_t                 zapid;
1032         int                      rc;
1033         ENTRY;
1034
1035         LASSERT(dof);
1036
1037         switch (dof->dof_type) {
1038                 case DFT_REGULAR:
1039                 case DFT_SYM:
1040                 case DFT_NODE:
1041                         if (obj->oo_dt.do_body_ops == NULL)
1042                                 obj->oo_dt.do_body_ops = &osd_body_ops;
1043                         break;
1044                 default:
1045                         break;
1046         }
1047
1048         LASSERT(handle != NULL);
1049         oh = container_of0(handle, struct osd_thandle, ot_super);
1050         LASSERT(oh->ot_tx != NULL);
1051
1052         switch (dof->dof_type) {
1053                 case DFT_DIR:
1054                         dt->do_index_ops = &osd_dir_ops;
1055                 case DFT_INDEX:
1056                         /* for zap create */
1057                         dmu_tx_hold_zap(oh->ot_tx, DMU_NEW_OBJECT, 1, NULL);
1058                         break;
1059                 case DFT_REGULAR:
1060                 case DFT_SYM:
1061                 case DFT_NODE:
1062                         /* first, we'll create new object */
1063                         dmu_tx_hold_bonus(oh->ot_tx, DMU_NEW_OBJECT);
1064                         break;
1065
1066                 default:
1067                         LBUG();
1068                         break;
1069         }
1070
1071         /* and we'll add it to some mapping */
1072         zapid = osd_get_name_n_idx(env, osd, fid, buf);
1073         dmu_tx_hold_bonus(oh->ot_tx, zapid);
1074         dmu_tx_hold_zap(oh->ot_tx, zapid, TRUE, buf);
1075
1076         /* we will also update inode accounting ZAPs */
1077         dmu_tx_hold_bonus(oh->ot_tx, osd->od_iusr_oid);
1078         dmu_tx_hold_zap(oh->ot_tx, osd->od_iusr_oid, TRUE, buf);
1079         dmu_tx_hold_bonus(oh->ot_tx, osd->od_igrp_oid);
1080         dmu_tx_hold_zap(oh->ot_tx, osd->od_igrp_oid, TRUE, buf);
1081
1082         dmu_tx_hold_sa_create(oh->ot_tx, ZFS_SA_BASE_ATTR_SIZE);
1083
1084         __osd_xattr_declare_set(env, obj, sizeof(struct lustre_mdt_attrs),
1085                                 XATTR_NAME_LMA, oh);
1086
1087         rc = osd_declare_quota(env, osd, attr->la_uid, attr->la_gid, 1, oh,
1088                                false, NULL, false);
1089         RETURN(rc);
1090 }
1091
1092 int __osd_attr_init(const struct lu_env *env, struct osd_device *osd,
1093                     uint64_t oid, dmu_tx_t *tx, struct lu_attr *la,
1094                     uint64_t parent)
1095 {
1096         sa_bulk_attr_t  *bulk;
1097         sa_handle_t     *sa_hdl;
1098         struct osa_attr *osa = &osd_oti_get(env)->oti_osa;
1099         uint64_t         gen;
1100         uint64_t         crtime[2];
1101         timestruc_t      now;
1102         int              cnt;
1103         int              rc;
1104
1105         gethrestime(&now);
1106         gen = dmu_tx_get_txg(tx);
1107
1108         ZFS_TIME_ENCODE(&now, crtime);
1109
1110         osa->atime[0] = la->la_atime;
1111         osa->ctime[0] = la->la_ctime;
1112         osa->mtime[0] = la->la_mtime;
1113         osa->mode = la->la_mode;
1114         osa->uid = la->la_uid;
1115         osa->gid = la->la_gid;
1116         osa->rdev = la->la_rdev;
1117         osa->nlink = la->la_nlink;
1118         osa->flags = attrs_fs2zfs(la->la_flags);
1119         osa->size  = la->la_size;
1120
1121         /* Now add in all of the "SA" attributes */
1122         rc = -sa_handle_get(osd->od_os, oid, NULL, SA_HDL_PRIVATE, &sa_hdl);
1123         if (rc)
1124                 return rc;
1125
1126         OBD_ALLOC(bulk, sizeof(sa_bulk_attr_t) * 13);
1127         if (bulk == NULL) {
1128                 rc = -ENOMEM;
1129                 goto out;
1130         }
1131         /*
1132          * we need to create all SA below upon object create.
1133          *
1134          * XXX The attribute order matters since the accounting callback relies
1135          * on static offsets (i.e. SA_*_OFFSET, see zfs_space_delta_cb()) to
1136          * look up the UID/GID attributes. Moreover, the callback does not seem
1137          * to support the spill block.
1138          * We define attributes in the same order as SA_*_OFFSET in order to
1139          * work around the problem. See ORI-610.
1140          */
1141         cnt = 0;
1142         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(osd), NULL, &osa->mode, 8);
1143         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(osd), NULL, &osa->size, 8);
1144         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GEN(osd), NULL, &gen, 8);
1145         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(osd), NULL, &osa->uid, 8);
1146         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(osd), NULL, &osa->gid, 8);
1147         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_PARENT(osd), NULL, &parent, 8);
1148         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(osd), NULL, &osa->flags, 8);
1149         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(osd), NULL, osa->atime, 16);
1150         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(osd), NULL, osa->mtime, 16);
1151         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(osd), NULL, osa->ctime, 16);
1152         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CRTIME(osd), NULL, crtime, 16);
1153         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(osd), NULL, &osa->nlink, 8);
1154         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_RDEV(osd), NULL, &osa->rdev, 8);
1155
1156         rc = -sa_replace_all_by_template(sa_hdl, bulk, cnt, tx);
1157
1158         OBD_FREE(bulk, sizeof(sa_bulk_attr_t) * 13);
1159 out:
1160         sa_handle_destroy(sa_hdl);
1161         return rc;
1162 }
1163
1164 /*
1165  * The transaction passed to this routine must have
1166  * dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT) called and then assigned
1167  * to a transaction group.
1168  */
1169 int __osd_object_create(const struct lu_env *env, struct osd_object *obj,
1170                         dmu_buf_t **dbp, dmu_tx_t *tx, struct lu_attr *la,
1171                         uint64_t parent)
1172 {
1173         uint64_t             oid;
1174         int                  rc;
1175         struct osd_device   *osd = osd_obj2dev(obj);
1176         const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
1177         dmu_object_type_t    type = DMU_OT_PLAIN_FILE_CONTENTS;
1178
1179         /* Assert that the transaction has been assigned to a
1180            transaction group. */
1181         LASSERT(tx->tx_txg != 0);
1182
1183         /* Use DMU_OTN_UINT8_METADATA for local objects so their data blocks
1184          * would get an additional ditto copy */
1185         if (unlikely(S_ISREG(la->la_mode) &&
1186                      fid_seq_is_local_file(fid_seq(fid))))
1187                 type = DMU_OTN_UINT8_METADATA;
1188
1189         /* Create a new DMU object. */
1190         oid = dmu_object_alloc(osd->od_os, type, 0,
1191                                DMU_OT_SA, DN_MAX_BONUSLEN, tx);
1192         rc = -sa_buf_hold(osd->od_os, oid, osd_obj_tag, dbp);
1193         LASSERTF(rc == 0, "sa_buf_hold "LPU64" failed: %d\n", oid, rc);
1194
1195         LASSERT(la->la_valid & LA_MODE);
1196         la->la_size = 0;
1197         la->la_nlink = 1;
1198
1199         rc = __osd_attr_init(env, osd, oid, tx, la, parent);
1200         if (rc != 0) {
1201                 sa_buf_rele(*dbp, osd_obj_tag);
1202                 *dbp = NULL;
1203                 dmu_object_free(osd->od_os, oid, tx);
1204                 return rc;
1205         }
1206
1207         return 0;
1208 }
1209
1210 /*
1211  * The transaction passed to this routine must have
1212  * dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, ...) called and then assigned
1213  * to a transaction group.
1214  *
1215  * Using ZAP_FLAG_HASH64 will force the ZAP to always be a FAT ZAP.
1216  * This is fine for directories today, because storing the FID in the dirent
1217  * will also require a FAT ZAP.  If there is a new type of micro ZAP created
1218  * then we might need to re-evaluate the use of this flag and instead do
1219  * a conversion from the different internal ZAP hash formats being used. */
1220 int __osd_zap_create(const struct lu_env *env, struct osd_device *osd,
1221                      dmu_buf_t **zap_dbp, dmu_tx_t *tx,
1222                      struct lu_attr *la, uint64_t parent, zap_flags_t flags)
1223 {
1224         uint64_t oid;
1225         int      rc;
1226
1227         /* Assert that the transaction has been assigned to a
1228            transaction group. */
1229         LASSERT(tx->tx_txg != 0);
1230
1231         oid = zap_create_flags(osd->od_os, 0, flags | ZAP_FLAG_HASH64,
1232                                DMU_OT_DIRECTORY_CONTENTS,
1233                                14, /* == ZFS fzap_default_block_shift */
1234                                DN_MAX_INDBLKSHIFT, /* indirect block shift */
1235                                DMU_OT_SA, DN_MAX_BONUSLEN, tx);
1236
1237         rc = -sa_buf_hold(osd->od_os, oid, osd_obj_tag, zap_dbp);
1238         if (rc)
1239                 return rc;
1240
1241         LASSERT(la->la_valid & LA_MODE);
1242         la->la_size = 2;
1243         la->la_nlink = 1;
1244
1245         return __osd_attr_init(env, osd, oid, tx, la, parent);
1246 }
1247
1248 static dmu_buf_t *osd_mkidx(const struct lu_env *env, struct osd_object *obj,
1249                             struct lu_attr *la, uint64_t parent,
1250                             struct osd_thandle *oh)
1251 {
1252         dmu_buf_t *db;
1253         int        rc;
1254
1255         /* Index file should be created as regular file in order not to confuse
1256          * ZPL which could interpret them as directory.
1257          * We set ZAP_FLAG_UINT64_KEY to let ZFS know than we are going to use
1258          * binary keys */
1259         LASSERT(S_ISREG(la->la_mode));
1260         rc = __osd_zap_create(env, osd_obj2dev(obj), &db, oh->ot_tx, la, parent,
1261                               ZAP_FLAG_UINT64_KEY);
1262         if (rc)
1263                 return ERR_PTR(rc);
1264         return db;
1265 }
1266
1267 static dmu_buf_t *osd_mkdir(const struct lu_env *env, struct osd_object *obj,
1268                             struct lu_attr *la, uint64_t parent,
1269                             struct osd_thandle *oh)
1270 {
1271         dmu_buf_t *db;
1272         int        rc;
1273
1274         LASSERT(S_ISDIR(la->la_mode));
1275         rc = __osd_zap_create(env, osd_obj2dev(obj), &db,
1276                               oh->ot_tx, la, parent, 0);
1277         if (rc)
1278                 return ERR_PTR(rc);
1279         return db;
1280 }
1281
1282 static dmu_buf_t *osd_mkreg(const struct lu_env *env, struct osd_object *obj,
1283                             struct lu_attr *la, uint64_t parent,
1284                             struct osd_thandle *oh)
1285 {
1286         dmu_buf_t         *db;
1287         int                rc;
1288         struct osd_device *osd = osd_obj2dev(obj);
1289
1290         LASSERT(S_ISREG(la->la_mode));
1291         rc = __osd_object_create(env, obj, &db, oh->ot_tx, la, parent);
1292         if (rc)
1293                 return ERR_PTR(rc);
1294
1295         /*
1296          * XXX: This heuristic is non-optimal.  It would be better to
1297          * increase the blocksize up to osd->od_max_blksz during the write.
1298          * This is exactly how the ZPL behaves and it ensures that the right
1299          * blocksize is selected based on the file size rather than the
1300          * making broad assumptions based on the osd type.
1301          */
1302         if (!lu_device_is_md(osd2lu_dev(osd))) {
1303                 rc = -dmu_object_set_blocksize(osd->od_os, db->db_object,
1304                                                osd->od_max_blksz, 0, oh->ot_tx);
1305                 if (unlikely(rc)) {
1306                         CERROR("%s: can't change blocksize: %d\n",
1307                                osd->od_svname, rc);
1308                         return ERR_PTR(rc);
1309                 }
1310         }
1311
1312         return db;
1313 }
1314
1315 static dmu_buf_t *osd_mksym(const struct lu_env *env, struct osd_object *obj,
1316                             struct lu_attr *la, uint64_t parent,
1317                             struct osd_thandle *oh)
1318 {
1319         dmu_buf_t *db;
1320         int        rc;
1321
1322         LASSERT(S_ISLNK(la->la_mode));
1323         rc = __osd_object_create(env, obj, &db, oh->ot_tx, la, parent);
1324         if (rc)
1325                 return ERR_PTR(rc);
1326         return db;
1327 }
1328
1329 static dmu_buf_t *osd_mknod(const struct lu_env *env, struct osd_object *obj,
1330                             struct lu_attr *la, uint64_t parent,
1331                             struct osd_thandle *oh)
1332 {
1333         dmu_buf_t *db;
1334         int        rc;
1335
1336         la->la_valid = LA_MODE;
1337         if (S_ISCHR(la->la_mode) || S_ISBLK(la->la_mode))
1338                 la->la_valid |= LA_RDEV;
1339
1340         rc = __osd_object_create(env, obj, &db, oh->ot_tx, la, parent);
1341         if (rc)
1342                 return ERR_PTR(rc);
1343         return db;
1344 }
1345
1346 typedef dmu_buf_t *(*osd_obj_type_f)(const struct lu_env *env,
1347                                      struct osd_object *obj,
1348                                      struct lu_attr *la,
1349                                      uint64_t parent,
1350                                      struct osd_thandle *oh);
1351
1352 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
1353 {
1354         osd_obj_type_f result;
1355
1356         switch (type) {
1357         case DFT_DIR:
1358                 result = osd_mkdir;
1359                 break;
1360         case DFT_INDEX:
1361                 result = osd_mkidx;
1362                 break;
1363         case DFT_REGULAR:
1364                 result = osd_mkreg;
1365                 break;
1366         case DFT_SYM:
1367                 result = osd_mksym;
1368                 break;
1369         case DFT_NODE:
1370                 result = osd_mknod;
1371                 break;
1372         default:
1373                 LBUG();
1374                 break;
1375         }
1376         return result;
1377 }
1378
1379 /*
1380  * Primitives for directory (i.e. ZAP) handling
1381  */
1382 static inline int osd_init_lma(const struct lu_env *env, struct osd_object *obj,
1383                                const struct lu_fid *fid, struct osd_thandle *oh)
1384 {
1385         struct osd_thread_info  *info = osd_oti_get(env);
1386         struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
1387         struct lu_buf            buf;
1388         int rc;
1389
1390         lustre_lma_init(lma, fid, 0, 0);
1391         lustre_lma_swab(lma);
1392         buf.lb_buf = lma;
1393         buf.lb_len = sizeof(*lma);
1394
1395         rc = osd_xattr_set_internal(env, obj, &buf, XATTR_NAME_LMA,
1396                                     LU_XATTR_CREATE, oh);
1397
1398         return rc;
1399 }
1400
1401 /*
1402  * Concurrency: @dt is write locked.
1403  */
1404 static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
1405                              struct lu_attr *attr,
1406                              struct dt_allocation_hint *hint,
1407                              struct dt_object_format *dof,
1408                              struct thandle *th)
1409 {
1410         struct zpl_direntry     *zde = &osd_oti_get(env)->oti_zde.lzd_reg;
1411         const struct lu_fid     *fid = lu_object_fid(&dt->do_lu);
1412         struct osd_object       *obj = osd_dt_obj(dt);
1413         struct osd_device       *osd = osd_obj2dev(obj);
1414         char                    *buf = osd_oti_get(env)->oti_str;
1415         struct osd_thandle      *oh;
1416         dmu_buf_t               *db;
1417         uint64_t                 zapid;
1418         int                      rc;
1419
1420         ENTRY;
1421
1422         /* concurrent create declarations should not see
1423          * the object inconsistent (db, attr, etc).
1424          * in regular cases acquisition should be cheap */
1425         down(&obj->oo_guard);
1426
1427         LASSERT(osd_invariant(obj));
1428         LASSERT(!dt_object_exists(dt));
1429         LASSERT(dof != NULL);
1430
1431         LASSERT(th != NULL);
1432         oh = container_of0(th, struct osd_thandle, ot_super);
1433
1434         /*
1435          * XXX missing: Quote handling.
1436          */
1437
1438         LASSERT(obj->oo_db == NULL);
1439
1440         /* to follow ZFS on-disk format we need
1441          * to initialize parent dnode properly */
1442         zapid = 0;
1443         if (hint && hint->dah_parent)
1444                 zapid = osd_dt_obj(hint->dah_parent)->oo_db->db_object;
1445
1446         db = osd_create_type_f(dof->dof_type)(env, obj, attr, zapid, oh);
1447         if (IS_ERR(db))
1448                 GOTO(out, rc = PTR_ERR(db));
1449
1450         zde->zde_pad = 0;
1451         zde->zde_dnode = db->db_object;
1452         zde->zde_type = IFTODT(attr->la_mode & S_IFMT);
1453
1454         zapid = osd_get_name_n_idx(env, osd, fid, buf);
1455
1456         rc = -zap_add(osd->od_os, zapid, buf, 8, 1, zde, oh->ot_tx);
1457         if (rc)
1458                 GOTO(out, rc);
1459
1460         /* Add new object to inode accounting.
1461          * Errors are not considered as fatal */
1462         rc = -zap_increment_int(osd->od_os, osd->od_iusr_oid,
1463                                 (attr->la_valid & LA_UID) ? attr->la_uid : 0, 1,
1464                                 oh->ot_tx);
1465         if (rc)
1466                 CERROR("%s: failed to add "DFID" to accounting ZAP for usr %d "
1467                         "(%d)\n", osd->od_svname, PFID(fid), attr->la_uid, rc);
1468         rc = -zap_increment_int(osd->od_os, osd->od_igrp_oid,
1469                                 (attr->la_valid & LA_GID) ? attr->la_gid : 0, 1,
1470                                 oh->ot_tx);
1471         if (rc)
1472                 CERROR("%s: failed to add "DFID" to accounting ZAP for grp %d "
1473                         "(%d)\n", osd->od_svname, PFID(fid), attr->la_gid, rc);
1474
1475         /* configure new osd object */
1476         obj->oo_db = db;
1477         rc = osd_object_init0(env, obj);
1478         LASSERT(ergo(rc == 0, dt_object_exists(dt)));
1479         LASSERT(osd_invariant(obj));
1480
1481         rc = osd_init_lma(env, obj, fid, oh);
1482         if (rc) {
1483                 CERROR("%s: can not set LMA on "DFID": rc = %d\n",
1484                        osd->od_svname, PFID(fid), rc);
1485                 /* ignore errors during LMA initialization */
1486                 rc = 0;
1487         }
1488
1489 out:
1490         up(&obj->oo_guard);
1491         RETURN(rc);
1492 }
1493
1494 static int osd_declare_object_ref_add(const struct lu_env *env,
1495                                       struct dt_object *dt,
1496                                       struct thandle *th)
1497 {
1498         return osd_declare_attr_set(env, dt, NULL, th);
1499 }
1500
1501 /*
1502  * Concurrency: @dt is write locked.
1503  */
1504 static int osd_object_ref_add(const struct lu_env *env,
1505                               struct dt_object *dt,
1506                               struct thandle *handle)
1507 {
1508         struct osd_object       *obj = osd_dt_obj(dt);
1509         struct osd_thandle      *oh;
1510         struct osd_device       *osd = osd_obj2dev(obj);
1511         uint64_t                 nlink;
1512         int rc;
1513
1514         ENTRY;
1515
1516         LASSERT(osd_invariant(obj));
1517         LASSERT(dt_object_exists(dt));
1518         LASSERT(obj->oo_sa_hdl != NULL);
1519
1520         oh = container_of0(handle, struct osd_thandle, ot_super);
1521
1522         write_lock(&obj->oo_attr_lock);
1523         nlink = ++obj->oo_attr.la_nlink;
1524         write_unlock(&obj->oo_attr_lock);
1525
1526         rc = osd_object_sa_update(obj, SA_ZPL_LINKS(osd), &nlink, 8, oh);
1527         return rc;
1528 }
1529
1530 static int osd_declare_object_ref_del(const struct lu_env *env,
1531                                       struct dt_object *dt,
1532                                       struct thandle *handle)
1533 {
1534         return osd_declare_attr_set(env, dt, NULL, handle);
1535 }
1536
1537 /*
1538  * Concurrency: @dt is write locked.
1539  */
1540 static int osd_object_ref_del(const struct lu_env *env,
1541                               struct dt_object *dt,
1542                               struct thandle *handle)
1543 {
1544         struct osd_object       *obj = osd_dt_obj(dt);
1545         struct osd_thandle      *oh;
1546         struct osd_device       *osd = osd_obj2dev(obj);
1547         uint64_t                 nlink;
1548         int                      rc;
1549
1550         ENTRY;
1551
1552         LASSERT(osd_invariant(obj));
1553         LASSERT(dt_object_exists(dt));
1554         LASSERT(obj->oo_sa_hdl != NULL);
1555
1556         oh = container_of0(handle, struct osd_thandle, ot_super);
1557         LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
1558
1559         write_lock(&obj->oo_attr_lock);
1560         nlink = --obj->oo_attr.la_nlink;
1561         write_unlock(&obj->oo_attr_lock);
1562
1563         rc = osd_object_sa_update(obj, SA_ZPL_LINKS(osd), &nlink, 8, oh);
1564         return rc;
1565 }
1566
1567 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
1568                            __u64 start, __u64 end)
1569 {
1570         struct osd_device *osd = osd_obj2dev(osd_dt_obj(dt));
1571         ENTRY;
1572
1573         /* XXX: no other option than syncing the whole filesystem until we
1574          * support ZIL.  If the object tracked the txg that it was last
1575          * modified in, it could pass that txg here instead of "0".  Maybe
1576          * the changes are already committed, so no wait is needed at all? */
1577         txg_wait_synced(dmu_objset_pool(osd->od_os), 0ULL);
1578
1579         RETURN(0);
1580 }
1581
1582 static struct dt_object_operations osd_obj_ops = {
1583         .do_read_lock           = osd_object_read_lock,
1584         .do_write_lock          = osd_object_write_lock,
1585         .do_read_unlock         = osd_object_read_unlock,
1586         .do_write_unlock        = osd_object_write_unlock,
1587         .do_write_locked        = osd_object_write_locked,
1588         .do_attr_get            = osd_attr_get,
1589         .do_declare_attr_set    = osd_declare_attr_set,
1590         .do_attr_set            = osd_attr_set,
1591         .do_ah_init             = osd_ah_init,
1592         .do_declare_create      = osd_declare_object_create,
1593         .do_create              = osd_object_create,
1594         .do_declare_destroy     = osd_declare_object_destroy,
1595         .do_destroy             = osd_object_destroy,
1596         .do_index_try           = osd_index_try,
1597         .do_declare_ref_add     = osd_declare_object_ref_add,
1598         .do_ref_add             = osd_object_ref_add,
1599         .do_declare_ref_del     = osd_declare_object_ref_del,
1600         .do_ref_del             = osd_object_ref_del,
1601         .do_xattr_get           = osd_xattr_get,
1602         .do_declare_xattr_set   = osd_declare_xattr_set,
1603         .do_xattr_set           = osd_xattr_set,
1604         .do_declare_xattr_del   = osd_declare_xattr_del,
1605         .do_xattr_del           = osd_xattr_del,
1606         .do_xattr_list          = osd_xattr_list,
1607         .do_object_sync         = osd_object_sync,
1608 };
1609
1610 static struct lu_object_operations osd_lu_obj_ops = {
1611         .loo_object_init        = osd_object_init,
1612         .loo_object_delete      = osd_object_delete,
1613         .loo_object_release     = osd_object_release,
1614         .loo_object_free        = osd_object_free,
1615         .loo_object_print       = osd_object_print,
1616         .loo_object_invariant   = osd_object_invariant,
1617 };
1618
1619 static int osd_otable_it_attr_get(const struct lu_env *env,
1620                                 struct dt_object *dt,
1621                                 struct lu_attr *attr)
1622 {
1623         attr->la_valid = 0;
1624         return 0;
1625 }
1626
1627 static struct dt_object_operations osd_obj_otable_it_ops = {
1628         .do_attr_get    = osd_otable_it_attr_get,
1629         .do_index_try   = osd_index_try,
1630 };