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_rwsem(&mo->oo_sem);
 295                 sema_init(&mo->oo_guard, 1);
 296                 rwlock_init(&mo->oo_attr_lock);
 297                 return l;
 298         } else {
 299                 return NULL;
 300         }
 301 }
 302
 303 /*
 304  * Concurrency: shouldn't matter.
 305  */
 306 int osd_object_init0(const struct lu_env *env, struct osd_object *obj)
 307 {
 308         struct osd_device       *osd = osd_obj2dev(obj);
 309         const struct lu_fid     *fid  = lu_object_fid(&obj->oo_dt.do_lu);
 310         int                      rc = 0;
 311         ENTRY;
 312
 313         if (obj->oo_db == NULL)
 314                 RETURN(0);
 315
 316         /* object exist */
 317
 318         rc = osd_object_sa_init(obj, osd);
 319         if (rc)
 320                 RETURN(rc);
 321
 322         /* cache attrs in object */
 323         rc = __osd_object_attr_get(env, osd, obj, &obj->oo_attr);
 324         if (rc)
 325                 RETURN(rc);
 326
 327         if (likely(!fid_is_acct(fid)))
 328                 /* no body operations for accounting objects */
 329                 obj->oo_dt.do_body_ops = &osd_body_ops;
 330
 331         /*
 332          * initialize object before marking it existing
 333          */
 334         obj->oo_dt.do_lu.lo_header->loh_attr |= obj->oo_attr.la_mode & S_IFMT;
 335
 336         smp_mb();
 337         obj->oo_dt.do_lu.lo_header->loh_attr |= LOHA_EXISTS;
 338
 339         RETURN(0);
 340 }
 341
 342 static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
 343 {
 344         struct osd_thread_info  *info = osd_oti_get(env);
 345         struct lu_buf           buf;
 346         int                     rc;
 347         struct lustre_mdt_attrs *lma;
 348         ENTRY;
 349
 350         CLASSERT(sizeof(info->oti_buf) >= sizeof(*lma));
 351         lma = (struct lustre_mdt_attrs *)info->oti_buf;
 352         buf.lb_buf = lma;
 353         buf.lb_len = sizeof(info->oti_buf);
 354
 355         rc = osd_xattr_get(env, &obj->oo_dt, &buf, XATTR_NAME_LMA);
 356         if (rc > 0) {
 357                 rc = 0;
 358                 lustre_lma_swab(lma);
 359                 if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
 360                              CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT))) {
 361                         CWARN("%s: unsupported incompat LMA feature(s) %#x for "
 362                               "fid = "DFID"\n", osd_obj2dev(obj)->od_svname,
 363                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
 364                               PFID(lu_object_fid(&obj->oo_dt.do_lu)));
 365                         rc = -EOPNOTSUPP;
 366                 }
 367         } else if (rc == -ENODATA) {
 368                 /* haven't initialize LMA xattr */
 369                 rc = 0;
 370         }
 371
 372         RETURN(rc);
 373 }
 374
 375 /*
 376  * Concurrency: no concurrent access is possible that early in object
 377  * life-cycle.
 378  */
 379 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
 380                            const struct lu_object_conf *conf)
 381 {
 382         struct osd_object       *obj = osd_obj(l);
 383         struct osd_device       *osd = osd_obj2dev(obj);
 384         uint64_t                 oid;
 385         int                      rc;
 386         ENTRY;
 387
 388         LASSERT(osd_invariant(obj));
 389
 390         if (fid_is_otable_it(&l->lo_header->loh_fid)) {
 391                 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
 392                 l->lo_header->loh_attr |= LOHA_EXISTS;
 393                 RETURN(0);
 394         }
 395
 396         rc = osd_fid_lookup(env, osd, lu_object_fid(l), &oid);
 397         if (rc == 0) {
 398                 LASSERT(obj->oo_db == NULL);
 399                 rc = __osd_obj2dbuf(env, osd->od_os, oid, &obj->oo_db);
 400                 if (rc != 0) {
 401                         CERROR("%s: lookup "DFID"/"LPX64" failed: rc = %d\n",
 402                                osd->od_svname, PFID(lu_object_fid(l)), oid, rc);
 403                         GOTO(out, rc);
 404                 }
 405                 LASSERT(obj->oo_db);
 406                 rc = osd_object_init0(env, obj);
 407                 if (rc != 0)
 408                         GOTO(out, rc);
 409
 410                 rc = osd_check_lma(env, obj);
 411                 if (rc != 0)
 412                         GOTO(out, rc);
 413         } else if (rc == -ENOENT) {
 414                 rc = 0;
 415         }
 416         LASSERT(osd_invariant(obj));
 417 out:
 418         RETURN(rc);
 419 }
 420
 421 /*
 422  * Concurrency: no concurrent access is possible that late in object
 423  * life-cycle.
 424  */
 425 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
 426 {
 427         struct osd_object *obj = osd_obj(l);
 428
 429         LASSERT(osd_invariant(obj));
 430
 431         dt_object_fini(&obj->oo_dt);
 432         OBD_SLAB_FREE_PTR(obj, osd_object_kmem);
 433 }
 434
 435 static void __osd_declare_object_destroy(const struct lu_env *env,
 436                                          struct osd_object *obj,
 437                                          struct osd_thandle *oh)
 438 {
 439         struct osd_device       *osd = osd_obj2dev(obj);
 440         dmu_buf_t               *db = obj->oo_db;
 441         zap_attribute_t         *za = &osd_oti_get(env)->oti_za;
 442         uint64_t                 oid = db->db_object, xid;
 443         dmu_tx_t                *tx = oh->ot_tx;
 444         zap_cursor_t            *zc;
 445         int                      rc = 0;
 446
 447         dmu_tx_hold_free(tx, oid, 0, DMU_OBJECT_END);
 448
 449         /* zap holding xattrs */
 450         if (obj->oo_xattr != ZFS_NO_OBJECT) {
 451                 oid = obj->oo_xattr;
 452
 453                 dmu_tx_hold_free(tx, oid, 0, DMU_OBJECT_END);
 454
 455                 rc = osd_zap_cursor_init(&zc, osd->od_os, oid, 0);
 456                 if (rc)
 457                         goto out;
 458
 459                 while ((rc = -zap_cursor_retrieve(zc, za)) == 0) {
 460                         BUG_ON(za->za_integer_length != sizeof(uint64_t));
 461                         BUG_ON(za->za_num_integers != 1);
 462
 463                         rc = -zap_lookup(osd->od_os, obj->oo_xattr, za->za_name,
 464                                          sizeof(uint64_t), 1, &xid);
 465                         if (rc) {
 466                                 CERROR("%s: xattr lookup failed: rc = %d\n",
 467                                        osd->od_svname, rc);
 468                                 goto out_err;
 469                         }
 470                         dmu_tx_hold_free(tx, xid, 0, DMU_OBJECT_END);
 471
 472                         zap_cursor_advance(zc);
 473                 }
 474                 if (rc == -ENOENT)
 475                         rc = 0;
 476 out_err:
 477                 osd_zap_cursor_fini(zc);
 478         }
 479 out:
 480         if (rc && tx->tx_err == 0)
 481                 tx->tx_err = -rc;
 482 }
 483
 484 static int osd_declare_object_destroy(const struct lu_env *env,
 485                                       struct dt_object *dt,
 486                                       struct thandle *th)
 487 {
 488         char                    *buf = osd_oti_get(env)->oti_str;
 489         const struct lu_fid     *fid = lu_object_fid(&dt->do_lu);
 490         struct osd_object       *obj = osd_dt_obj(dt);
 491         struct osd_device       *osd = osd_obj2dev(obj);
 492         struct osd_thandle      *oh;
 493         uint64_t                 zapid;
 494         int                      rc;
 495         ENTRY;
 496
 497         LASSERT(th != NULL);
 498         LASSERT(dt_object_exists(dt));
 499
 500         oh = container_of0(th, struct osd_thandle, ot_super);
 501         LASSERT(oh->ot_tx != NULL);
 502
 503         /* declare that we'll destroy the object */
 504         __osd_declare_object_destroy(env, obj, oh);
 505
 506         /* declare that we'll remove object from fid-dnode mapping */
 507         zapid = osd_get_name_n_idx(env, osd, fid, buf);
 508         dmu_tx_hold_bonus(oh->ot_tx, zapid);
 509         dmu_tx_hold_zap(oh->ot_tx, zapid, 0, buf);
 510
 511         /* declare that we'll remove object from inode accounting ZAPs */
 512         dmu_tx_hold_bonus(oh->ot_tx, osd->od_iusr_oid);
 513         dmu_tx_hold_zap(oh->ot_tx, osd->od_iusr_oid, 0, buf);
 514         dmu_tx_hold_bonus(oh->ot_tx, osd->od_igrp_oid);
 515         dmu_tx_hold_zap(oh->ot_tx, osd->od_igrp_oid, 0, buf);
 516
 517         /* one less inode */
 518         rc = osd_declare_quota(env, osd, obj->oo_attr.la_uid,
 519                                obj->oo_attr.la_gid, -1, oh, false, NULL, false);
 520         if (rc)
 521                 RETURN(rc);
 522
 523         /* data to be truncated */
 524         rc = osd_declare_quota(env, osd, obj->oo_attr.la_uid,
 525                                obj->oo_attr.la_gid, 0, oh, true, NULL, false);
 526         RETURN(rc);
 527 }
 528
 529 /*
 530  * Delete a DMU object
 531  *
 532  * The transaction passed to this routine must have
 533  * dmu_tx_hold_free(tx, oid, 0, DMU_OBJECT_END) called
 534  * and then assigned to a transaction group.
 535  *
 536  * This will release db and set it to NULL to prevent further dbuf releases.
 537  */
 538 static int __osd_object_destroy(const struct lu_env *env,
 539                                 struct osd_object *obj,
 540                                 dmu_tx_t *tx, void *tag)
 541 {
 542         struct osd_device       *osd = osd_obj2dev(obj);
 543         uint64_t                 xid;
 544         zap_attribute_t         *za = &osd_oti_get(env)->oti_za;
 545         zap_cursor_t            *zc;
 546         int                      rc;
 547
 548         /* Assert that the transaction has been assigned to a
 549            transaction group. */
 550         LASSERT(tx->tx_txg != 0);
 551
 552         /* zap holding xattrs */
 553         if (obj->oo_xattr != ZFS_NO_OBJECT) {
 554                 rc = osd_zap_cursor_init(&zc, osd->od_os, obj->oo_xattr, 0);
 555                 if (rc)
 556                         return rc;
 557                 while ((rc = -zap_cursor_retrieve(zc, za)) == 0) {
 558                         BUG_ON(za->za_integer_length != sizeof(uint64_t));
 559                         BUG_ON(za->za_num_integers != 1);
 560
 561                         rc = -zap_lookup(osd->od_os, obj->oo_xattr, za->za_name,
 562                                          sizeof(uint64_t), 1, &xid);
 563                         if (rc) {
 564                                 CERROR("%s: lookup xattr %s failed: rc = %d\n",
 565                                        osd->od_svname, za->za_name, rc);
 566                                 continue;
 567                         }
 568                         rc = -dmu_object_free(osd->od_os, xid, tx);
 569                         if (rc)
 570                                 CERROR("%s: fetch xattr %s failed: rc = %d\n",
 571                                        osd->od_svname, za->za_name, rc);
 572                         zap_cursor_advance(zc);
 573                 }
 574                 osd_zap_cursor_fini(zc);
 575
 576                 rc = -dmu_object_free(osd->od_os, obj->oo_xattr, tx);
 577                 if (rc)
 578                         CERROR("%s: freeing xattr failed: rc = %d\n",
 579                                osd->od_svname, rc);
 580         }
 581
 582         return -dmu_object_free(osd->od_os, obj->oo_db->db_object, tx);
 583 }
 584
 585 static int osd_object_destroy(const struct lu_env *env,
 586                               struct dt_object *dt,
 587                               struct thandle *th)
 588 {
 589         char                    *buf = osd_oti_get(env)->oti_str;
 590         struct osd_object       *obj = osd_dt_obj(dt);
 591         struct osd_device       *osd = osd_obj2dev(obj);
 592         const struct lu_fid     *fid = lu_object_fid(&dt->do_lu);
 593         struct osd_thandle      *oh;
 594         int                      rc;
 595         uint64_t                 zapid;
 596         ENTRY;
 597
 598         LASSERT(obj->oo_db != NULL);
 599         LASSERT(dt_object_exists(dt));
 600         LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
 601
 602         oh = container_of0(th, struct osd_thandle, ot_super);
 603         LASSERT(oh != NULL);
 604         LASSERT(oh->ot_tx != NULL);
 605
 606         zapid = osd_get_name_n_idx(env, osd, fid, buf);
 607
 608         /* remove obj ref from index dir (it depends) */
 609         rc = -zap_remove(osd->od_os, zapid, buf, oh->ot_tx);
 610         if (rc) {
 611                 CERROR("%s: zap_remove() failed: rc = %d\n",
 612                        osd->od_svname, rc);
 613                 GOTO(out, rc);
 614         }
 615
 616         /* Remove object from inode accounting. It is not fatal for the destroy
 617          * operation if something goes wrong while updating accounting, but we
 618          * still log an error message to notify the administrator */
 619         rc = -zap_increment_int(osd->od_os, osd->od_iusr_oid,
 620                         obj->oo_attr.la_uid, -1, oh->ot_tx);
 621         if (rc)
 622                 CERROR("%s: failed to remove "DFID" from accounting ZAP for usr"
 623                         " %d: rc = %d\n", osd->od_svname, PFID(fid),
 624                         obj->oo_attr.la_uid, rc);
 625         rc = -zap_increment_int(osd->od_os, osd->od_igrp_oid,
 626                                 obj->oo_attr.la_gid, -1, oh->ot_tx);
 627         if (rc)
 628                 CERROR("%s: failed to remove "DFID" from accounting ZAP for grp"
 629                         " %d: rc = %d\n", osd->od_svname, PFID(fid),
 630                         obj->oo_attr.la_gid, rc);
 631
 632         /* kill object */
 633         rc = __osd_object_destroy(env, obj, oh->ot_tx, osd_obj_tag);
 634         if (rc) {
 635                 CERROR("%s: __osd_object_destroy() failed: rc = %d\n",
 636                        osd->od_svname, rc);
 637                 GOTO(out, rc);
 638         }
 639
 640 out:
 641         /* not needed in the cache anymore */
 642         set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
 643
 644         RETURN (0);
 645 }
 646
 647 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
 648 {
 649         struct osd_object *obj = osd_obj(l);
 650
 651         if (obj->oo_db != NULL) {
 652                 osd_object_sa_fini(obj);
 653                 if (obj->oo_sa_xattr) {
 654                         nvlist_free(obj->oo_sa_xattr);
 655                         obj->oo_sa_xattr = NULL;
 656                 }
 657                 sa_buf_rele(obj->oo_db, osd_obj_tag);
 658                 list_del(&obj->oo_sa_linkage);
 659                 obj->oo_db = NULL;
 660         }
 661 }
 662
 663 /*
 664  * Concurrency: ->loo_object_release() is called under site spin-lock.
 665  */
 666 static void osd_object_release(const struct lu_env *env,
 667                                struct lu_object *l)
 668 {
 669 }
 670
 671 /*
 672  * Concurrency: shouldn't matter.
 673  */
 674 static int osd_object_print(const struct lu_env *env, void *cookie,
 675                             lu_printer_t p, const struct lu_object *l)
 676 {
 677         struct osd_object *o = osd_obj(l);
 678
 679         return (*p)(env, cookie, LUSTRE_OSD_ZFS_NAME"-object@%p", o);
 680 }
 681
 682 static void osd_object_read_lock(const struct lu_env *env,
 683                                  struct dt_object *dt, unsigned role)
 684 {
 685         struct osd_object *obj = osd_dt_obj(dt);
 686
 687         LASSERT(osd_invariant(obj));
 688
 689         down_read(&obj->oo_sem);
 690 }
 691
 692 static void osd_object_write_lock(const struct lu_env *env,
 693                                   struct dt_object *dt, unsigned role)
 694 {
 695         struct osd_object *obj = osd_dt_obj(dt);
 696
 697         LASSERT(osd_invariant(obj));
 698
 699         down_write(&obj->oo_sem);
 700 }
 701
 702 static void osd_object_read_unlock(const struct lu_env *env,
 703                                    struct dt_object *dt)
 704 {
 705         struct osd_object *obj = osd_dt_obj(dt);
 706
 707         LASSERT(osd_invariant(obj));
 708         up_read(&obj->oo_sem);
 709 }
 710
 711 static void osd_object_write_unlock(const struct lu_env *env,
 712                                     struct dt_object *dt)
 713 {
 714         struct osd_object *obj = osd_dt_obj(dt);
 715
 716         LASSERT(osd_invariant(obj));
 717         up_write(&obj->oo_sem);
 718 }
 719
 720 static int osd_object_write_locked(const struct lu_env *env,
 721                                    struct dt_object *dt)
 722 {
 723         struct osd_object *obj = osd_dt_obj(dt);
 724         int rc = 1;
 725
 726         LASSERT(osd_invariant(obj));
 727
 728         if (down_write_trylock(&obj->oo_sem)) {
 729                 rc = 0;
 730                 up_write(&obj->oo_sem);
 731         }
 732         return rc;
 733 }
 734
 735 static int osd_attr_get(const struct lu_env *env,
 736                         struct dt_object *dt,
 737                         struct lu_attr *attr)
 738 {
 739         struct osd_object       *obj = osd_dt_obj(dt);
 740         uint64_t                 blocks;
 741         uint32_t                 blksize;
 742
 743         LASSERT(dt_object_exists(dt));
 744         LASSERT(osd_invariant(obj));
 745         LASSERT(obj->oo_db);
 746
 747         read_lock(&obj->oo_attr_lock);
 748         *attr = obj->oo_attr;
 749         read_unlock(&obj->oo_attr_lock);
 750
 751         /* with ZFS_DEBUG zrl_add_debug() called by DB_DNODE_ENTER()
 752          * from within sa_object_size() can block on a mutex, so
 753          * we can't call sa_object_size() holding rwlock */
 754         sa_object_size(obj->oo_sa_hdl, &blksize, &blocks);
 755         /* we do not control size of indices, so always calculate
 756          * it from number of blocks reported by DMU */
 757         if (S_ISDIR(attr->la_mode))
 758                 attr->la_size = 512 * blocks;
 759         /* Block size may be not set; suggest maximal I/O transfers. */
 760         if (blksize == 0)
 761                 blksize = osd_spa_maxblocksize(
 762                         dmu_objset_spa(osd_obj2dev(obj)->od_os));
 763
 764         attr->la_blksize = blksize;
 765         attr->la_blocks = blocks;
 766         attr->la_valid |= LA_BLOCKS | LA_BLKSIZE;
 767
 768         return 0;
 769 }
 770
 771 /* Simple wrapper on top of qsd API which implement quota transfer for osd
 772  * setattr needs. As a reminder, only the root user can change ownership of
 773  * a file, that's why EDQUOT & EINPROGRESS errors are discarded */
 774 static inline int qsd_transfer(const struct lu_env *env,
 775                                struct qsd_instance *qsd,
 776                                struct lquota_trans *trans, int qtype,
 777                                __u64 orig_id, __u64 new_id, __u64 bspace,
 778                                struct lquota_id_info *qi)
 779 {
 780         int     rc;
 781
 782         if (unlikely(qsd == NULL))
 783                 return 0;
 784
 785         LASSERT(qtype >= 0 && qtype < MAXQUOTAS);
 786         qi->lqi_type = qtype;
 787
 788         /* inode accounting */
 789         qi->lqi_is_blk = false;
 790
 791         /* one more inode for the new owner ... */
 792         qi->lqi_id.qid_uid = new_id;
 793         qi->lqi_space      = 1;
 794         rc = qsd_op_begin(env, qsd, trans, qi, NULL);
 795         if (rc == -EDQUOT || rc == -EINPROGRESS)
 796                 rc = 0;
 797         if (rc)
 798                 return rc;
 799
 800         /* and one less inode for the current id */
 801         qi->lqi_id.qid_uid = orig_id;;
 802         qi->lqi_space      = -1;
 803         /* can't get EDQUOT when reducing usage */
 804         rc = qsd_op_begin(env, qsd, trans, qi, NULL);
 805         if (rc == -EINPROGRESS)
 806                 rc = 0;
 807         if (rc)
 808                 return rc;
 809
 810         /* block accounting */
 811         qi->lqi_is_blk = true;
 812
 813         /* more blocks for the new owner ... */
 814         qi->lqi_id.qid_uid = new_id;
 815         qi->lqi_space      = bspace;
 816         rc = qsd_op_begin(env, qsd, trans, qi, NULL);
 817         if (rc == -EDQUOT || rc == -EINPROGRESS)
 818                 rc = 0;
 819         if (rc)
 820                 return rc;
 821
 822         /* and finally less blocks for the current owner */
 823         qi->lqi_id.qid_uid = orig_id;
 824         qi->lqi_space      = -bspace;
 825         rc = qsd_op_begin(env, qsd, trans, qi, NULL);
 826         /* can't get EDQUOT when reducing usage */
 827         if (rc == -EINPROGRESS)
 828                 rc = 0;
 829         return rc;
 830 }
 831
 832 static int osd_declare_attr_set(const struct lu_env *env,
 833                                 struct dt_object *dt,
 834                                 const struct lu_attr *attr,
 835                                 struct thandle *handle)
 836 {
 837         struct osd_thread_info  *info = osd_oti_get(env);
 838         char                    *buf = osd_oti_get(env)->oti_str;
 839         struct osd_object       *obj = osd_dt_obj(dt);
 840         struct osd_device       *osd = osd_obj2dev(obj);
 841         struct osd_thandle      *oh;
 842         uint64_t                 bspace;
 843         uint32_t                 blksize;
 844         int                      rc;
 845         ENTRY;
 846
 847         if (!dt_object_exists(dt)) {
 848                 /* XXX: sanity check that object creation is declared */
 849                 RETURN(0);
 850         }
 851
 852         LASSERT(handle != NULL);
 853         LASSERT(osd_invariant(obj));
 854
 855         oh = container_of0(handle, struct osd_thandle, ot_super);
 856
 857         LASSERT(obj->oo_sa_hdl != NULL);
 858         LASSERT(oh->ot_tx != NULL);
 859         dmu_tx_hold_sa(oh->ot_tx, obj->oo_sa_hdl, 0);
 860         if (oh->ot_tx->tx_err != 0)
 861                 RETURN(-oh->ot_tx->tx_err);
 862
 863         sa_object_size(obj->oo_sa_hdl, &blksize, &bspace);
 864         bspace = toqb(bspace * blksize);
 865
 866         if (attr && attr->la_valid & LA_UID) {
 867                 /* account for user inode tracking ZAP update */
 868                 dmu_tx_hold_bonus(oh->ot_tx, osd->od_iusr_oid);
 869                 dmu_tx_hold_zap(oh->ot_tx, osd->od_iusr_oid, TRUE, buf);
 870
 871                 /* quota enforcement for user */
 872                 if (attr->la_uid != obj->oo_attr.la_uid) {
 873                         rc = qsd_transfer(env, osd->od_quota_slave,
 874                                           &oh->ot_quota_trans, USRQUOTA,
 875                                           obj->oo_attr.la_uid, attr->la_uid,
 876                                           bspace, &info->oti_qi);
 877                         if (rc)
 878                                 RETURN(rc);
 879                 }
 880         }
 881         if (attr && attr->la_valid & LA_GID) {
 882                 /* account for user inode tracking ZAP update */
 883                 dmu_tx_hold_bonus(oh->ot_tx, osd->od_igrp_oid);
 884                 dmu_tx_hold_zap(oh->ot_tx, osd->od_igrp_oid, TRUE, buf);
 885
 886                 /* quota enforcement for group */
 887                 if (attr->la_gid != obj->oo_attr.la_gid) {
 888                         rc = qsd_transfer(env, osd->od_quota_slave,
 889                                           &oh->ot_quota_trans, GRPQUOTA,
 890                                           obj->oo_attr.la_gid, attr->la_gid,
 891                                           bspace, &info->oti_qi);
 892                         if (rc)
 893                                 RETURN(rc);
 894                 }
 895         }
 896
 897         RETURN(0);
 898 }
 899
 900 /*
 901  * Set the attributes of an object
 902  *
 903  * The transaction passed to this routine must have
 904  * dmu_tx_hold_bonus(tx, oid) called and then assigned
 905  * to a transaction group.
 906  */
 907 static int osd_attr_set(const struct lu_env *env, struct dt_object *dt,
 908                         const struct lu_attr *la, struct thandle *handle)
 909 {
 910         struct osd_object       *obj = osd_dt_obj(dt);
 911         struct osd_device       *osd = osd_obj2dev(obj);
 912         struct osd_thandle      *oh;
 913         struct osa_attr         *osa = &osd_oti_get(env)->oti_osa;
 914         sa_bulk_attr_t          *bulk;
 915         __u64                    valid = la->la_valid;
 916         int                      cnt;
 917         int                      rc = 0;
 918
 919         ENTRY;
 920         LASSERT(handle != NULL);
 921         LASSERT(dt_object_exists(dt));
 922         LASSERT(osd_invariant(obj));
 923         LASSERT(obj->oo_sa_hdl);
 924
 925         oh = container_of0(handle, struct osd_thandle, ot_super);
 926         /* Assert that the transaction has been assigned to a
 927            transaction group. */
 928         LASSERT(oh->ot_tx->tx_txg != 0);
 929
 930         /* Only allow set size for regular file */
 931         if (!S_ISREG(dt->do_lu.lo_header->loh_attr))
 932                 valid &= ~(LA_SIZE | LA_BLOCKS);
 933
 934         if (valid == 0)
 935                 RETURN(0);
 936
 937         OBD_ALLOC(bulk, sizeof(sa_bulk_attr_t) * 10);
 938         if (bulk == NULL)
 939                 RETURN(-ENOMEM);
 940
 941         /* do both accounting updates outside oo_attr_lock below */
 942         if ((valid & LA_UID) && (la->la_uid != obj->oo_attr.la_uid)) {
 943                 /* Update user accounting. Failure isn't fatal, but we still
 944                  * log an error message */
 945                 rc = -zap_increment_int(osd->od_os, osd->od_iusr_oid,
 946                                         la->la_uid, 1, oh->ot_tx);
 947                 if (rc)
 948                         CERROR("%s: failed to update accounting ZAP for user "
 949                                 "%d (%d)\n", osd->od_svname, la->la_uid, rc);
 950                 rc = -zap_increment_int(osd->od_os, osd->od_iusr_oid,
 951                                         obj->oo_attr.la_uid, -1, oh->ot_tx);
 952                 if (rc)
 953                         CERROR("%s: failed to update accounting ZAP for user "
 954                                 "%d (%d)\n", osd->od_svname,
 955                                 obj->oo_attr.la_uid, rc);
 956         }
 957         if ((valid & LA_GID) && (la->la_gid != obj->oo_attr.la_gid)) {
 958                 /* Update group accounting. Failure isn't fatal, but we still
 959                  * log an error message */
 960                 rc = -zap_increment_int(osd->od_os, osd->od_igrp_oid,
 961                                         la->la_gid, 1, oh->ot_tx);
 962                 if (rc)
 963                         CERROR("%s: failed to update accounting ZAP for user "
 964                                 "%d (%d)\n", osd->od_svname, la->la_gid, rc);
 965                 rc = -zap_increment_int(osd->od_os, osd->od_igrp_oid,
 966                                         obj->oo_attr.la_gid, -1, oh->ot_tx);
 967                 if (rc)
 968                         CERROR("%s: failed to update accounting ZAP for user "
 969                                 "%d (%d)\n", osd->od_svname,
 970                                 obj->oo_attr.la_gid, rc);
 971         }
 972
 973         write_lock(&obj->oo_attr_lock);
 974         cnt = 0;
 975         if (valid & LA_ATIME) {
 976                 osa->atime[0] = obj->oo_attr.la_atime = la->la_atime;
 977                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(osd), NULL,
 978                                  osa->atime, 16);
 979         }
 980         if (valid & LA_MTIME) {
 981                 osa->mtime[0] = obj->oo_attr.la_mtime = la->la_mtime;
 982                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(osd), NULL,
 983                                  osa->mtime, 16);
 984         }
 985         if (valid & LA_CTIME) {
 986                 osa->ctime[0] = obj->oo_attr.la_ctime = la->la_ctime;
 987                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(osd), NULL,
 988                                  osa->ctime, 16);
 989         }
 990         if (valid & LA_MODE) {
 991                 /* mode is stored along with type, so read it first */
 992                 obj->oo_attr.la_mode = (obj->oo_attr.la_mode & S_IFMT) |
 993                         (la->la_mode & ~S_IFMT);
 994                 osa->mode = obj->oo_attr.la_mode;
 995                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(osd), NULL,
 996                                  &osa->mode, 8);
 997         }
 998         if (valid & LA_SIZE) {
 999                 osa->size = obj->oo_attr.la_size = la->la_size;
1000                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(osd), NULL,
1001                                  &osa->size, 8);
1002         }
1003         if (valid & LA_NLINK) {
1004                 osa->nlink = obj->oo_attr.la_nlink = la->la_nlink;
1005                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(osd), NULL,
1006                                  &osa->nlink, 8);
1007         }
1008         if (valid & LA_RDEV) {
1009                 osa->rdev = obj->oo_attr.la_rdev = la->la_rdev;
1010                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_RDEV(osd), NULL,
1011                                  &osa->rdev, 8);
1012         }
1013         if (valid & LA_FLAGS) {
1014                 osa->flags = attrs_fs2zfs(la->la_flags);
1015                 /* many flags are not supported by zfs, so ensure a good cached
1016                  * copy */
1017                 obj->oo_attr.la_flags = attrs_zfs2fs(osa->flags);
1018                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(osd), NULL,
1019                                  &osa->flags, 8);
1020         }
1021         if (valid & LA_UID) {
1022                 osa->uid = obj->oo_attr.la_uid = la->la_uid;
1023                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(osd), NULL,
1024                                  &osa->uid, 8);
1025         }
1026         if (valid & LA_GID) {
1027                 osa->gid = obj->oo_attr.la_gid = la->la_gid;
1028                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(osd), NULL,
1029                                  &osa->gid, 8);
1030         }
1031         obj->oo_attr.la_valid |= valid;
1032         write_unlock(&obj->oo_attr_lock);
1033
1034         rc = osd_object_sa_bulk_update(obj, bulk, cnt, oh);
1035
1036         OBD_FREE(bulk, sizeof(sa_bulk_attr_t) * 10);
1037         RETURN(rc);
1038 }
1039
1040 /*
1041  * Object creation.
1042  *
1043  * XXX temporary solution.
1044  */
1045
1046 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
1047                         struct dt_object *parent, struct dt_object *child,
1048                         umode_t child_mode)
1049 {
1050         LASSERT(ah);
1051
1052         ah->dah_parent = parent;
1053         ah->dah_mode = child_mode;
1054 }
1055
1056 static int osd_declare_object_create(const struct lu_env *env,
1057                                      struct dt_object *dt,
1058                                      struct lu_attr *attr,
1059                                      struct dt_allocation_hint *hint,
1060                                      struct dt_object_format *dof,
1061                                      struct thandle *handle)
1062 {
1063         char                    *buf = osd_oti_get(env)->oti_str;
1064         const struct lu_fid     *fid = lu_object_fid(&dt->do_lu);
1065         struct osd_object       *obj = osd_dt_obj(dt);
1066         struct osd_device       *osd = osd_obj2dev(obj);
1067         struct osd_thandle      *oh;
1068         uint64_t                 zapid;
1069         int                      rc;
1070         ENTRY;
1071
1072         LASSERT(dof);
1073
1074         switch (dof->dof_type) {
1075                 case DFT_REGULAR:
1076                 case DFT_SYM:
1077                 case DFT_NODE:
1078                         if (obj->oo_dt.do_body_ops == NULL)
1079                                 obj->oo_dt.do_body_ops = &osd_body_ops;
1080                         break;
1081                 default:
1082                         break;
1083         }
1084
1085         LASSERT(handle != NULL);
1086         oh = container_of0(handle, struct osd_thandle, ot_super);
1087         LASSERT(oh->ot_tx != NULL);
1088
1089         switch (dof->dof_type) {
1090                 case DFT_DIR:
1091                         dt->do_index_ops = &osd_dir_ops;
1092                 case DFT_INDEX:
1093                         /* for zap create */
1094                         dmu_tx_hold_zap(oh->ot_tx, DMU_NEW_OBJECT, 1, NULL);
1095                         break;
1096                 case DFT_REGULAR:
1097                 case DFT_SYM:
1098                 case DFT_NODE:
1099                         /* first, we'll create new object */
1100                         dmu_tx_hold_bonus(oh->ot_tx, DMU_NEW_OBJECT);
1101                         break;
1102
1103                 default:
1104                         LBUG();
1105                         break;
1106         }
1107
1108         /* and we'll add it to some mapping */
1109         zapid = osd_get_name_n_idx(env, osd, fid, buf);
1110         dmu_tx_hold_bonus(oh->ot_tx, zapid);
1111         dmu_tx_hold_zap(oh->ot_tx, zapid, TRUE, buf);
1112
1113         /* we will also update inode accounting ZAPs */
1114         dmu_tx_hold_bonus(oh->ot_tx, osd->od_iusr_oid);
1115         dmu_tx_hold_zap(oh->ot_tx, osd->od_iusr_oid, TRUE, buf);
1116         dmu_tx_hold_bonus(oh->ot_tx, osd->od_igrp_oid);
1117         dmu_tx_hold_zap(oh->ot_tx, osd->od_igrp_oid, TRUE, buf);
1118
1119         dmu_tx_hold_sa_create(oh->ot_tx, ZFS_SA_BASE_ATTR_SIZE);
1120
1121         __osd_xattr_declare_set(env, obj, sizeof(struct lustre_mdt_attrs),
1122                                 XATTR_NAME_LMA, oh);
1123
1124         rc = osd_declare_quota(env, osd, attr->la_uid, attr->la_gid, 1, oh,
1125                                false, NULL, false);
1126         RETURN(rc);
1127 }
1128
1129 int __osd_attr_init(const struct lu_env *env, struct osd_device *osd,
1130                     uint64_t oid, dmu_tx_t *tx, struct lu_attr *la,
1131                     uint64_t parent)
1132 {
1133         sa_bulk_attr_t  *bulk;
1134         sa_handle_t     *sa_hdl;
1135         struct osa_attr *osa = &osd_oti_get(env)->oti_osa;
1136         uint64_t         gen;
1137         uint64_t         crtime[2];
1138         timestruc_t      now;
1139         int              cnt;
1140         int              rc;
1141
1142         gethrestime(&now);
1143         gen = dmu_tx_get_txg(tx);
1144
1145         ZFS_TIME_ENCODE(&now, crtime);
1146
1147         osa->atime[0] = la->la_atime;
1148         osa->ctime[0] = la->la_ctime;
1149         osa->mtime[0] = la->la_mtime;
1150         osa->mode = la->la_mode;
1151         osa->uid = la->la_uid;
1152         osa->gid = la->la_gid;
1153         osa->rdev = la->la_rdev;
1154         osa->nlink = la->la_nlink;
1155         osa->flags = attrs_fs2zfs(la->la_flags);
1156         osa->size  = la->la_size;
1157
1158         /* Now add in all of the "SA" attributes */
1159         rc = -sa_handle_get(osd->od_os, oid, NULL, SA_HDL_PRIVATE, &sa_hdl);
1160         if (rc)
1161                 return rc;
1162
1163         OBD_ALLOC(bulk, sizeof(sa_bulk_attr_t) * 13);
1164         if (bulk == NULL) {
1165                 rc = -ENOMEM;
1166                 goto out;
1167         }
1168         /*
1169          * we need to create all SA below upon object create.
1170          *
1171          * XXX The attribute order matters since the accounting callback relies
1172          * on static offsets (i.e. SA_*_OFFSET, see zfs_space_delta_cb()) to
1173          * look up the UID/GID attributes. Moreover, the callback does not seem
1174          * to support the spill block.
1175          * We define attributes in the same order as SA_*_OFFSET in order to
1176          * work around the problem. See ORI-610.
1177          */
1178         cnt = 0;
1179         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(osd), NULL, &osa->mode, 8);
1180         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(osd), NULL, &osa->size, 8);
1181         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GEN(osd), NULL, &gen, 8);
1182         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(osd), NULL, &osa->uid, 8);
1183         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(osd), NULL, &osa->gid, 8);
1184         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_PARENT(osd), NULL, &parent, 8);
1185         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(osd), NULL, &osa->flags, 8);
1186         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(osd), NULL, osa->atime, 16);
1187         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(osd), NULL, osa->mtime, 16);
1188         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(osd), NULL, osa->ctime, 16);
1189         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CRTIME(osd), NULL, crtime, 16);
1190         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(osd), NULL, &osa->nlink, 8);
1191         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_RDEV(osd), NULL, &osa->rdev, 8);
1192
1193         rc = -sa_replace_all_by_template(sa_hdl, bulk, cnt, tx);
1194
1195         OBD_FREE(bulk, sizeof(sa_bulk_attr_t) * 13);
1196 out:
1197         sa_handle_destroy(sa_hdl);
1198         return rc;
1199 }
1200
1201 /*
1202  * The transaction passed to this routine must have
1203  * dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT) called and then assigned
1204  * to a transaction group.
1205  */
1206 int __osd_object_create(const struct lu_env *env, struct osd_object *obj,
1207                         dmu_buf_t **dbp, dmu_tx_t *tx, struct lu_attr *la,
1208                         uint64_t parent)
1209 {
1210         uint64_t             oid;
1211         int                  rc;
1212         struct osd_device   *osd = osd_obj2dev(obj);
1213         const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
1214         dmu_object_type_t    type = DMU_OT_PLAIN_FILE_CONTENTS;
1215
1216         /* Assert that the transaction has been assigned to a
1217            transaction group. */
1218         LASSERT(tx->tx_txg != 0);
1219
1220         /* Use DMU_OTN_UINT8_METADATA for local objects so their data blocks
1221          * would get an additional ditto copy */
1222         if (unlikely(S_ISREG(la->la_mode) &&
1223                      fid_seq_is_local_file(fid_seq(fid))))
1224                 type = DMU_OTN_UINT8_METADATA;
1225
1226         /* Create a new DMU object. */
1227         oid = dmu_object_alloc(osd->od_os, type, 0,
1228                                DMU_OT_SA, DN_MAX_BONUSLEN, tx);
1229         rc = -sa_buf_hold(osd->od_os, oid, osd_obj_tag, dbp);
1230         LASSERTF(rc == 0, "sa_buf_hold "LPU64" failed: %d\n", oid, rc);
1231
1232         LASSERT(la->la_valid & LA_MODE);
1233         la->la_size = 0;
1234         la->la_nlink = 1;
1235
1236         rc = __osd_attr_init(env, osd, oid, tx, la, parent);
1237         if (rc != 0) {
1238                 sa_buf_rele(*dbp, osd_obj_tag);
1239                 *dbp = NULL;
1240                 dmu_object_free(osd->od_os, oid, tx);
1241                 return rc;
1242         }
1243
1244         return 0;
1245 }
1246
1247 /*
1248  * The transaction passed to this routine must have
1249  * dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, ...) called and then assigned
1250  * to a transaction group.
1251  *
1252  * Using ZAP_FLAG_HASH64 will force the ZAP to always be a FAT ZAP.
1253  * This is fine for directories today, because storing the FID in the dirent
1254  * will also require a FAT ZAP.  If there is a new type of micro ZAP created
1255  * then we might need to re-evaluate the use of this flag and instead do
1256  * a conversion from the different internal ZAP hash formats being used. */
1257 int __osd_zap_create(const struct lu_env *env, struct osd_device *osd,
1258                      dmu_buf_t **zap_dbp, dmu_tx_t *tx,
1259                      struct lu_attr *la, uint64_t parent, zap_flags_t flags)
1260 {
1261         uint64_t oid;
1262         int      rc;
1263
1264         /* Assert that the transaction has been assigned to a
1265            transaction group. */
1266         LASSERT(tx->tx_txg != 0);
1267
1268         oid = zap_create_flags(osd->od_os, 0, flags | ZAP_FLAG_HASH64,
1269                                DMU_OT_DIRECTORY_CONTENTS,
1270                                14, /* == ZFS fzap_default_block_shift */
1271                                DN_MAX_INDBLKSHIFT, /* indirect block shift */
1272                                DMU_OT_SA, DN_MAX_BONUSLEN, tx);
1273
1274         rc = -sa_buf_hold(osd->od_os, oid, osd_obj_tag, zap_dbp);
1275         if (rc)
1276                 return rc;
1277
1278         LASSERT(la->la_valid & LA_MODE);
1279         la->la_size = 2;
1280         la->la_nlink = 1;
1281
1282         return __osd_attr_init(env, osd, oid, tx, la, parent);
1283 }
1284
1285 static dmu_buf_t *osd_mkidx(const struct lu_env *env, struct osd_object *obj,
1286                             struct lu_attr *la, uint64_t parent,
1287                             struct osd_thandle *oh)
1288 {
1289         dmu_buf_t *db;
1290         int        rc;
1291
1292         /* Index file should be created as regular file in order not to confuse
1293          * ZPL which could interpret them as directory.
1294          * We set ZAP_FLAG_UINT64_KEY to let ZFS know than we are going to use
1295          * binary keys */
1296         LASSERT(S_ISREG(la->la_mode));
1297         rc = __osd_zap_create(env, osd_obj2dev(obj), &db, oh->ot_tx, la, parent,
1298                               ZAP_FLAG_UINT64_KEY);
1299         if (rc)
1300                 return ERR_PTR(rc);
1301         return db;
1302 }
1303
1304 static dmu_buf_t *osd_mkdir(const struct lu_env *env, struct osd_object *obj,
1305                             struct lu_attr *la, uint64_t parent,
1306                             struct osd_thandle *oh)
1307 {
1308         dmu_buf_t *db;
1309         int        rc;
1310
1311         LASSERT(S_ISDIR(la->la_mode));
1312         rc = __osd_zap_create(env, osd_obj2dev(obj), &db,
1313                               oh->ot_tx, la, parent, 0);
1314         if (rc)
1315                 return ERR_PTR(rc);
1316         return db;
1317 }
1318
1319 static dmu_buf_t *osd_mkreg(const struct lu_env *env, struct osd_object *obj,
1320                             struct lu_attr *la, uint64_t parent,
1321                             struct osd_thandle *oh)
1322 {
1323         dmu_buf_t         *db;
1324         int                rc;
1325         struct osd_device *osd = osd_obj2dev(obj);
1326
1327         LASSERT(S_ISREG(la->la_mode));
1328         rc = __osd_object_create(env, obj, &db, oh->ot_tx, la, parent);
1329         if (rc)
1330                 return ERR_PTR(rc);
1331
1332         /*
1333          * XXX: This heuristic is non-optimal.  It would be better to
1334          * increase the blocksize up to osd->od_max_blksz during the write.
1335          * This is exactly how the ZPL behaves and it ensures that the right
1336          * blocksize is selected based on the file size rather than the
1337          * making broad assumptions based on the osd type.
1338          */
1339         if (!lu_device_is_md(osd2lu_dev(osd))) {
1340                 rc = -dmu_object_set_blocksize(osd->od_os, db->db_object,
1341                                                osd->od_max_blksz, 0, oh->ot_tx);
1342                 if (unlikely(rc)) {
1343                         CERROR("%s: can't change blocksize: %d\n",
1344                                osd->od_svname, rc);
1345                         return ERR_PTR(rc);
1346                 }
1347         }
1348
1349         return db;
1350 }
1351
1352 static dmu_buf_t *osd_mksym(const struct lu_env *env, struct osd_object *obj,
1353                             struct lu_attr *la, uint64_t parent,
1354                             struct osd_thandle *oh)
1355 {
1356         dmu_buf_t *db;
1357         int        rc;
1358
1359         LASSERT(S_ISLNK(la->la_mode));
1360         rc = __osd_object_create(env, obj, &db, oh->ot_tx, la, parent);
1361         if (rc)
1362                 return ERR_PTR(rc);
1363         return db;
1364 }
1365
1366 static dmu_buf_t *osd_mknod(const struct lu_env *env, struct osd_object *obj,
1367                             struct lu_attr *la, uint64_t parent,
1368                             struct osd_thandle *oh)
1369 {
1370         dmu_buf_t *db;
1371         int        rc;
1372
1373         la->la_valid = LA_MODE;
1374         if (S_ISCHR(la->la_mode) || S_ISBLK(la->la_mode))
1375                 la->la_valid |= LA_RDEV;
1376
1377         rc = __osd_object_create(env, obj, &db, oh->ot_tx, la, parent);
1378         if (rc)
1379                 return ERR_PTR(rc);
1380         return db;
1381 }
1382
1383 typedef dmu_buf_t *(*osd_obj_type_f)(const struct lu_env *env,
1384                                      struct osd_object *obj,
1385                                      struct lu_attr *la,
1386                                      uint64_t parent,
1387                                      struct osd_thandle *oh);
1388
1389 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
1390 {
1391         osd_obj_type_f result;
1392
1393         switch (type) {
1394         case DFT_DIR:
1395                 result = osd_mkdir;
1396                 break;
1397         case DFT_INDEX:
1398                 result = osd_mkidx;
1399                 break;
1400         case DFT_REGULAR:
1401                 result = osd_mkreg;
1402                 break;
1403         case DFT_SYM:
1404                 result = osd_mksym;
1405                 break;
1406         case DFT_NODE:
1407                 result = osd_mknod;
1408                 break;
1409         default:
1410                 LBUG();
1411                 break;
1412         }
1413         return result;
1414 }
1415
1416 /*
1417  * Primitives for directory (i.e. ZAP) handling
1418  */
1419 static inline int osd_init_lma(const struct lu_env *env, struct osd_object *obj,
1420                                const struct lu_fid *fid, struct osd_thandle *oh)
1421 {
1422         struct osd_thread_info  *info = osd_oti_get(env);
1423         struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
1424         struct lu_buf            buf;
1425         int rc;
1426
1427         lustre_lma_init(lma, fid, 0, 0);
1428         lustre_lma_swab(lma);
1429         buf.lb_buf = lma;
1430         buf.lb_len = sizeof(*lma);
1431
1432         rc = osd_xattr_set_internal(env, obj, &buf, XATTR_NAME_LMA,
1433                                     LU_XATTR_CREATE, oh);
1434
1435         return rc;
1436 }
1437
1438 /*
1439  * Concurrency: @dt is write locked.
1440  */
1441 static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
1442                              struct lu_attr *attr,
1443                              struct dt_allocation_hint *hint,
1444                              struct dt_object_format *dof,
1445                              struct thandle *th)
1446 {
1447         struct zpl_direntry     *zde = &osd_oti_get(env)->oti_zde.lzd_reg;
1448         const struct lu_fid     *fid = lu_object_fid(&dt->do_lu);
1449         struct osd_object       *obj = osd_dt_obj(dt);
1450         struct osd_device       *osd = osd_obj2dev(obj);
1451         char                    *buf = osd_oti_get(env)->oti_str;
1452         struct osd_thandle      *oh;
1453         dmu_buf_t               *db;
1454         uint64_t                 zapid;
1455         int                      rc;
1456
1457         ENTRY;
1458
1459         /* concurrent create declarations should not see
1460          * the object inconsistent (db, attr, etc).
1461          * in regular cases acquisition should be cheap */
1462         down(&obj->oo_guard);
1463
1464         LASSERT(osd_invariant(obj));
1465         LASSERT(!dt_object_exists(dt));
1466         LASSERT(dof != NULL);
1467
1468         LASSERT(th != NULL);
1469         oh = container_of0(th, struct osd_thandle, ot_super);
1470
1471         /*
1472          * XXX missing: Quote handling.
1473          */
1474
1475         LASSERT(obj->oo_db == NULL);
1476
1477         /* to follow ZFS on-disk format we need
1478          * to initialize parent dnode properly */
1479         zapid = 0;
1480         if (hint && hint->dah_parent)
1481                 zapid = osd_dt_obj(hint->dah_parent)->oo_db->db_object;
1482
1483         db = osd_create_type_f(dof->dof_type)(env, obj, attr, zapid, oh);
1484         if (IS_ERR(db))
1485                 GOTO(out, rc = PTR_ERR(db));
1486
1487         zde->zde_pad = 0;
1488         zde->zde_dnode = db->db_object;
1489         zde->zde_type = IFTODT(attr->la_mode & S_IFMT);
1490
1491         zapid = osd_get_name_n_idx(env, osd, fid, buf);
1492
1493         rc = -zap_add(osd->od_os, zapid, buf, 8, 1, zde, oh->ot_tx);
1494         if (rc)
1495                 GOTO(out, rc);
1496
1497         /* Add new object to inode accounting.
1498          * Errors are not considered as fatal */
1499         rc = -zap_increment_int(osd->od_os, osd->od_iusr_oid,
1500                                 (attr->la_valid & LA_UID) ? attr->la_uid : 0, 1,
1501                                 oh->ot_tx);
1502         if (rc)
1503                 CERROR("%s: failed to add "DFID" to accounting ZAP for usr %d "
1504                         "(%d)\n", osd->od_svname, PFID(fid), attr->la_uid, rc);
1505         rc = -zap_increment_int(osd->od_os, osd->od_igrp_oid,
1506                                 (attr->la_valid & LA_GID) ? attr->la_gid : 0, 1,
1507                                 oh->ot_tx);
1508         if (rc)
1509                 CERROR("%s: failed to add "DFID" to accounting ZAP for grp %d "
1510                         "(%d)\n", osd->od_svname, PFID(fid), attr->la_gid, rc);
1511
1512         /* configure new osd object */
1513         obj->oo_db = db;
1514         rc = osd_object_init0(env, obj);
1515         LASSERT(ergo(rc == 0, dt_object_exists(dt)));
1516         LASSERT(osd_invariant(obj));
1517
1518         rc = osd_init_lma(env, obj, fid, oh);
1519         if (rc) {
1520                 CERROR("%s: can not set LMA on "DFID": rc = %d\n",
1521                        osd->od_svname, PFID(fid), rc);
1522                 /* ignore errors during LMA initialization */
1523                 rc = 0;
1524         }
1525
1526 out:
1527         up(&obj->oo_guard);
1528         RETURN(rc);
1529 }
1530
1531 static int osd_declare_object_ref_add(const struct lu_env *env,
1532                                       struct dt_object *dt,
1533                                       struct thandle *th)
1534 {
1535         return osd_declare_attr_set(env, dt, NULL, th);
1536 }
1537
1538 /*
1539  * Concurrency: @dt is write locked.
1540  */
1541 static int osd_object_ref_add(const struct lu_env *env,
1542                               struct dt_object *dt,
1543                               struct thandle *handle)
1544 {
1545         struct osd_object       *obj = osd_dt_obj(dt);
1546         struct osd_thandle      *oh;
1547         struct osd_device       *osd = osd_obj2dev(obj);
1548         uint64_t                 nlink;
1549         int rc;
1550
1551         ENTRY;
1552
1553         LASSERT(osd_invariant(obj));
1554         LASSERT(dt_object_exists(dt));
1555         LASSERT(obj->oo_sa_hdl != NULL);
1556
1557         oh = container_of0(handle, struct osd_thandle, ot_super);
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_declare_object_ref_del(const struct lu_env *env,
1568                                       struct dt_object *dt,
1569                                       struct thandle *handle)
1570 {
1571         return osd_declare_attr_set(env, dt, NULL, handle);
1572 }
1573
1574 /*
1575  * Concurrency: @dt is write locked.
1576  */
1577 static int osd_object_ref_del(const struct lu_env *env,
1578                               struct dt_object *dt,
1579                               struct thandle *handle)
1580 {
1581         struct osd_object       *obj = osd_dt_obj(dt);
1582         struct osd_thandle      *oh;
1583         struct osd_device       *osd = osd_obj2dev(obj);
1584         uint64_t                 nlink;
1585         int                      rc;
1586
1587         ENTRY;
1588
1589         LASSERT(osd_invariant(obj));
1590         LASSERT(dt_object_exists(dt));
1591         LASSERT(obj->oo_sa_hdl != NULL);
1592
1593         oh = container_of0(handle, struct osd_thandle, ot_super);
1594         LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
1595
1596         write_lock(&obj->oo_attr_lock);
1597         nlink = --obj->oo_attr.la_nlink;
1598         write_unlock(&obj->oo_attr_lock);
1599
1600         rc = osd_object_sa_update(obj, SA_ZPL_LINKS(osd), &nlink, 8, oh);
1601         return rc;
1602 }
1603
1604 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
1605                            __u64 start, __u64 end)
1606 {
1607         struct osd_device *osd = osd_obj2dev(osd_dt_obj(dt));
1608         ENTRY;
1609
1610         /* XXX: no other option than syncing the whole filesystem until we
1611          * support ZIL.  If the object tracked the txg that it was last
1612          * modified in, it could pass that txg here instead of "0".  Maybe
1613          * the changes are already committed, so no wait is needed at all? */
1614         txg_wait_synced(dmu_objset_pool(osd->od_os), 0ULL);
1615
1616         RETURN(0);
1617 }
1618
1619 static struct dt_object_operations osd_obj_ops = {
1620         .do_read_lock           = osd_object_read_lock,
1621         .do_write_lock          = osd_object_write_lock,
1622         .do_read_unlock         = osd_object_read_unlock,
1623         .do_write_unlock        = osd_object_write_unlock,
1624         .do_write_locked        = osd_object_write_locked,
1625         .do_attr_get            = osd_attr_get,
1626         .do_declare_attr_set    = osd_declare_attr_set,
1627         .do_attr_set            = osd_attr_set,
1628         .do_ah_init             = osd_ah_init,
1629         .do_declare_create      = osd_declare_object_create,
1630         .do_create              = osd_object_create,
1631         .do_declare_destroy     = osd_declare_object_destroy,
1632         .do_destroy             = osd_object_destroy,
1633         .do_index_try           = osd_index_try,
1634         .do_declare_ref_add     = osd_declare_object_ref_add,
1635         .do_ref_add             = osd_object_ref_add,
1636         .do_declare_ref_del     = osd_declare_object_ref_del,
1637         .do_ref_del             = osd_object_ref_del,
1638         .do_xattr_get           = osd_xattr_get,
1639         .do_declare_xattr_set   = osd_declare_xattr_set,
1640         .do_xattr_set           = osd_xattr_set,
1641         .do_declare_xattr_del   = osd_declare_xattr_del,
1642         .do_xattr_del           = osd_xattr_del,
1643         .do_xattr_list          = osd_xattr_list,
1644         .do_object_sync         = osd_object_sync,
1645 };
1646
1647 static struct lu_object_operations osd_lu_obj_ops = {
1648         .loo_object_init        = osd_object_init,
1649         .loo_object_delete      = osd_object_delete,
1650         .loo_object_release     = osd_object_release,
1651         .loo_object_free        = osd_object_free,
1652         .loo_object_print       = osd_object_print,
1653         .loo_object_invariant   = osd_object_invariant,
1654 };
1655
1656 static int osd_otable_it_attr_get(const struct lu_env *env,
1657                                 struct dt_object *dt,
1658                                 struct lu_attr *attr)
1659 {
1660         attr->la_valid = 0;
1661         return 0;
1662 }
1663
1664 static struct dt_object_operations osd_obj_otable_it_ops = {
1665         .do_attr_get    = osd_otable_it_attr_get,
1666         .do_index_try   = osd_index_try,
1667 };