lustre/osd-zfs/osd_object.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2009, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2012, 2017, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  * Lustre is a trademark of Sun Microsystems, Inc.
  31  *
  32  * lustre/osd-zfs/osd_object.c
  33  *
  34  * Author: Alex Zhuravlev <bzzz@whamcloud.com>
  35  * Author: Mike Pershin <tappro@whamcloud.com>
  36  * Author: Johann Lombardi <johann@whamcloud.com>
  37  */
  38
  39 #define DEBUG_SUBSYSTEM S_OSD
  40
  41 #include <libcfs/libcfs.h>
  42 #include <obd_support.h>
  43 #include <lustre_net.h>
  44 #include <obd.h>
  45 #include <obd_class.h>
  46 #include <lustre_disk.h>
  47 #include <lustre_fid.h>
  48
  49 #include "osd_internal.h"
  50
  51 #include <sys/dnode.h>
  52 #include <sys/dbuf.h>
  53 #include <sys/spa.h>
  54 #include <sys/stat.h>
  55 #include <sys/zap.h>
  56 #include <sys/spa_impl.h>
  57 #include <sys/zfs_znode.h>
  58 #include <sys/dmu_tx.h>
  59 #include <sys/dmu_objset.h>
  60 #include <sys/dsl_prop.h>
  61 #include <sys/sa_impl.h>
  62 #include <sys/txg.h>
  63
  64 char *osd_obj_tag = "osd_object";
  65 static int osd_object_sync_delay_us = -1;
  66
  67 static const struct dt_object_operations osd_obj_ops;
  68 static const struct lu_object_operations osd_lu_obj_ops;
  69 static const struct dt_object_operations osd_obj_otable_it_ops;
  70
  71 static void
  72 osd_object_sa_fini(struct osd_object *obj)
  73 {
  74         if (obj->oo_sa_hdl) {
  75                 sa_handle_destroy(obj->oo_sa_hdl);
  76                 obj->oo_sa_hdl = NULL;
  77         }
  78 }
  79
  80 static int
  81 osd_object_sa_init(struct osd_object *obj, struct osd_device *o)
  82 {
  83         int rc;
  84
  85         LASSERT(obj->oo_sa_hdl == NULL);
  86         LASSERT(obj->oo_dn != NULL);
  87
  88         rc = osd_sa_handle_get(obj);
  89         if (rc)
  90                 return rc;
  91
  92         /* Cache the xattr object id, valid for the life of the object */
  93         rc = -sa_lookup(obj->oo_sa_hdl, SA_ZPL_XATTR(o), &obj->oo_xattr, 8);
  94         if (rc == -ENOENT) {
  95                 obj->oo_xattr = ZFS_NO_OBJECT;
  96                 rc = 0;
  97         } else if (rc) {
  98                 osd_object_sa_fini(obj);
  99         }
 100
 101         return rc;
 102 }
 103
 104 /*
 105  * Add object to list of dirty objects in tx handle.
 106  */
 107 void osd_object_sa_dirty_add(struct osd_object *obj, struct osd_thandle *oh)
 108 {
 109         if (!list_empty(&obj->oo_sa_linkage))
 110                 return;
 111
 112         write_lock(&obj->oo_attr_lock);
 113         if (likely(list_empty(&obj->oo_sa_linkage)))
 114                 list_add(&obj->oo_sa_linkage, &oh->ot_sa_list);
 115         write_unlock(&obj->oo_attr_lock);
 116 }
 117
 118 /*
 119  * Release spill block dbuf hold for all dirty SAs.
 120  */
 121 void osd_object_sa_dirty_rele(const struct lu_env *env, struct osd_thandle *oh)
 122 {
 123         struct osd_object *obj;
 124
 125         while (!list_empty(&oh->ot_sa_list)) {
 126                 obj = list_entry(oh->ot_sa_list.next,
 127                                  struct osd_object, oo_sa_linkage);
 128                 write_lock(&obj->oo_attr_lock);
 129                 list_del_init(&obj->oo_sa_linkage);
 130                 write_unlock(&obj->oo_attr_lock);
 131                 if (obj->oo_late_xattr) {
 132                         /*
 133                          * take oo_guard to protect oo_sa_xattr buffer
 134                          * from concurrent update by osd_xattr_set()
 135                          */
 136                         LASSERT(oh->ot_assigned != 0);
 137                         down_write(&obj->oo_guard);
 138                         if (obj->oo_late_attr_set)
 139                                 __osd_sa_attr_init(env, obj, oh);
 140                         else if (obj->oo_late_xattr)
 141                                 __osd_sa_xattr_update(env, obj, oh);
 142                         up_write(&obj->oo_guard);
 143                 }
 144                 sa_spill_rele(obj->oo_sa_hdl);
 145         }
 146 }
 147
 148 /*
 149  * Update the SA and add the object to the dirty list.
 150  */
 151 int osd_object_sa_update(struct osd_object *obj, sa_attr_type_t type,
 152                          void *buf, uint32_t buflen, struct osd_thandle *oh)
 153 {
 154         int rc;
 155
 156         LASSERT(obj->oo_sa_hdl != NULL);
 157         LASSERT(oh->ot_tx != NULL);
 158
 159         rc = -sa_update(obj->oo_sa_hdl, type, buf, buflen, oh->ot_tx);
 160         osd_object_sa_dirty_add(obj, oh);
 161
 162         return rc;
 163 }
 164
 165 /*
 166  * Bulk update the SA and add the object to the dirty list.
 167  */
 168 static int
 169 osd_object_sa_bulk_update(struct osd_object *obj, sa_bulk_attr_t *attrs,
 170                           int count, struct osd_thandle *oh)
 171 {
 172         int rc;
 173
 174         LASSERT(obj->oo_sa_hdl != NULL);
 175         LASSERT(oh->ot_tx != NULL);
 176
 177         rc = -sa_bulk_update(obj->oo_sa_hdl, attrs, count, oh->ot_tx);
 178         osd_object_sa_dirty_add(obj, oh);
 179
 180         return rc;
 181 }
 182
 183 /*
 184  * Retrieve the attributes of a DMU object
 185  */
 186 static int __osd_object_attr_get(const struct lu_env *env, struct osd_device *o,
 187                                  struct osd_object *obj, struct lu_attr *la)
 188 {
 189         struct osa_attr *osa = &osd_oti_get(env)->oti_osa;
 190         sa_bulk_attr_t *bulk = osd_oti_get(env)->oti_attr_bulk;
 191         struct lustre_mdt_attrs *lma;
 192         struct lu_buf buf;
 193         int cnt = 0;
 194         int              rc;
 195         ENTRY;
 196
 197         LASSERT(obj->oo_dn != NULL);
 198
 199         la->la_valid |= LA_ATIME | LA_MTIME | LA_CTIME | LA_BTIME | LA_MODE |
 200                         LA_TYPE | LA_SIZE | LA_UID | LA_GID | LA_FLAGS |
 201                         LA_NLINK;
 202
 203         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(o), NULL, osa->atime, 16);
 204         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(o), NULL, osa->mtime, 16);
 205         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(o), NULL, osa->ctime, 16);
 206         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CRTIME(o), NULL, osa->btime, 16);
 207         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(o), NULL, &osa->mode, 8);
 208         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(o), NULL, &osa->size, 8);
 209         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(o), NULL, &osa->nlink, 8);
 210         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(o), NULL, &osa->uid, 8);
 211         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(o), NULL, &osa->gid, 8);
 212         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(o), NULL, &osa->flags, 8);
 213         LASSERT(cnt <= ARRAY_SIZE(osd_oti_get(env)->oti_attr_bulk));
 214
 215         rc = -sa_bulk_lookup(obj->oo_sa_hdl, bulk, cnt);
 216         if (rc)
 217                 GOTO(out_sa, rc);
 218
 219 #ifdef ZFS_PROJINHERIT
 220         if (o->od_projectused_dn && osa->flags & ZFS_PROJID) {
 221                 rc = -sa_lookup(obj->oo_sa_hdl, SA_ZPL_PROJID(o),
 222                                 &osa->projid, 8);
 223                 if (rc)
 224                         GOTO(out_sa, rc);
 225
 226                 la->la_projid = osa->projid;
 227                 la->la_valid |= LA_PROJID;
 228                 obj->oo_with_projid = 1;
 229         } else {
 230                 la->la_projid = ZFS_DEFAULT_PROJID;
 231                 la->la_valid &= ~LA_PROJID;
 232         }
 233 #else
 234         la->la_projid = 0;
 235         la->la_valid &= ~LA_PROJID;
 236 #endif
 237
 238         la->la_atime = osa->atime[0];
 239         la->la_mtime = osa->mtime[0];
 240         la->la_ctime = osa->ctime[0];
 241         la->la_btime = osa->btime[0];
 242         la->la_mode = osa->mode;
 243         la->la_uid = osa->uid;
 244         la->la_gid = osa->gid;
 245         la->la_nlink = osa->nlink;
 246         la->la_flags = attrs_zfs2fs(osa->flags);
 247         la->la_size = osa->size;
 248
 249         /* Try to get extra flags from LMA */
 250         lma = (struct lustre_mdt_attrs *)osd_oti_get(env)->oti_buf;
 251         buf.lb_buf = lma;
 252         buf.lb_len = sizeof(osd_oti_get(env)->oti_buf);
 253         down_read(&obj->oo_guard);
 254         rc = osd_xattr_get_lma(env, obj, &buf);
 255         if (!rc) {
 256                 lma->lma_incompat = le32_to_cpu(lma->lma_incompat);
 257                 obj->oo_lma_flags =
 258                         lma_to_lustre_flags(lma->lma_incompat);
 259         } else if (rc == -ENODATA ||
 260                    !(S_ISDIR(la->la_mode) &&
 261                      dt_object_exists(&obj->oo_dt))) {
 262                 rc = 0;
 263         }
 264         up_read(&obj->oo_guard);
 265
 266         if (S_ISCHR(la->la_mode) || S_ISBLK(la->la_mode)) {
 267                 rc = -sa_lookup(obj->oo_sa_hdl, SA_ZPL_RDEV(o), &osa->rdev, 8);
 268                 if (rc)
 269                         GOTO(out_sa, rc);
 270                 la->la_rdev = osa->rdev;
 271                 la->la_valid |= LA_RDEV;
 272         }
 273 out_sa:
 274
 275         RETURN(rc);
 276 }
 277
 278 int __osd_obj2dnode(objset_t *os, uint64_t oid, dnode_t **dnp)
 279 {
 280         dmu_buf_t *db;
 281         dmu_buf_impl_t *dbi;
 282         int rc;
 283
 284         rc = -dmu_bonus_hold(os, oid, osd_obj_tag, &db);
 285         if (rc)
 286                 return rc;
 287
 288         dbi = (dmu_buf_impl_t *)db;
 289         DB_DNODE_ENTER(dbi);
 290         *dnp = DB_DNODE(dbi);
 291         DB_DNODE_EXIT(dbi);
 292         LASSERT(*dnp != NULL);
 293
 294         return 0;
 295 }
 296
 297 /*
 298  * Concurrency: no concurrent access is possible that early in object
 299  * life-cycle.
 300  */
 301 struct lu_object *osd_object_alloc(const struct lu_env *env,
 302                                    const struct lu_object_header *hdr,
 303                                    struct lu_device *d)
 304 {
 305         struct osd_object *mo;
 306
 307         OBD_SLAB_ALLOC_PTR_GFP(mo, osd_object_kmem, GFP_NOFS);
 308         if (mo != NULL) {
 309                 struct lu_object *l;
 310                 struct lu_object_header *h;
 311                 struct osd_device *o = osd_dev(d);
 312
 313                 l = &mo->oo_dt.do_lu;
 314                 if (unlikely(o->od_in_init)) {
 315                         OBD_ALLOC_PTR(h);
 316                         if (!h) {
 317                                 OBD_FREE_PTR(mo);
 318                                 return NULL;
 319                         }
 320
 321                         lu_object_header_init(h);
 322                         lu_object_init(l, h, d);
 323                         lu_object_add_top(h, l);
 324                         mo->oo_header = h;
 325                 } else {
 326                         dt_object_init(&mo->oo_dt, NULL, d);
 327                         mo->oo_header = NULL;
 328                 }
 329
 330                 mo->oo_dt.do_ops = &osd_obj_ops;
 331                 l->lo_ops = &osd_lu_obj_ops;
 332                 INIT_LIST_HEAD(&mo->oo_sa_linkage);
 333                 INIT_LIST_HEAD(&mo->oo_unlinked_linkage);
 334                 init_rwsem(&mo->oo_sem);
 335                 init_rwsem(&mo->oo_guard);
 336                 rwlock_init(&mo->oo_attr_lock);
 337                 mo->oo_destroy = OSD_DESTROY_NONE;
 338                 return l;
 339         } else {
 340                 return NULL;
 341         }
 342 }
 343
 344 static void osd_obj_set_blksize(const struct lu_env *env,
 345                                 struct osd_device *osd, struct osd_object *obj)
 346 {
 347         const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
 348         dmu_tx_t *tx;
 349         dnode_t *dn = obj->oo_dn;
 350         uint32_t blksz;
 351         int rc = 0;
 352         ENTRY;
 353
 354         LASSERT(!osd_oti_get(env)->oti_in_trans);
 355
 356         tx = dmu_tx_create(osd->od_os);
 357         if (!tx) {
 358                 CERROR("%s: fail to create tx to set blksize for "DFID"\n",
 359                        osd->od_svname, PFID(fid));
 360                 RETURN_EXIT;
 361         }
 362
 363         dmu_tx_hold_bonus(tx, dn->dn_object);
 364         rc = -dmu_tx_assign(tx, TXG_WAIT);
 365         if (rc) {
 366                 dmu_tx_abort(tx);
 367                 CERROR("%s: fail to assign tx to set blksize for "DFID
 368                        ": rc = %d\n", osd->od_svname, PFID(fid), rc);
 369                 RETURN_EXIT;
 370         }
 371
 372         down_write(&obj->oo_guard);
 373         if (unlikely((1 << dn->dn_datablkshift) >= PAGE_SIZE))
 374                 GOTO(out, rc = 1);
 375
 376         blksz = dn->dn_datablksz;
 377         if (!is_power_of_2(blksz))
 378                 blksz = size_roundup_power2(blksz);
 379
 380         if (blksz > osd->od_max_blksz)
 381                 blksz = osd->od_max_blksz;
 382         else if (blksz < PAGE_SIZE)
 383                 blksz = PAGE_SIZE;
 384         rc = -dmu_object_set_blocksize(osd->od_os, dn->dn_object, blksz, 0, tx);
 385
 386         GOTO(out, rc);
 387
 388 out:
 389         up_write(&obj->oo_guard);
 390         if (rc) {
 391                 dmu_tx_abort(tx);
 392                 if (unlikely(obj->oo_dn->dn_maxblkid > 0))
 393                         rc = 1;
 394                 if (rc < 0)
 395                         CERROR("%s: fail to set blksize for "DFID": rc = %d\n",
 396                                osd->od_svname, PFID(fid), rc);
 397         } else {
 398                 dmu_tx_commit(tx);
 399                 CDEBUG(D_INODE, "%s: set blksize as %u for "DFID"\n",
 400                        osd->od_svname, blksz, PFID(fid));
 401         }
 402 }
 403
 404 /*
 405  * Concurrency: shouldn't matter.
 406  */
 407 static int osd_object_init0(const struct lu_env *env, struct osd_object *obj)
 408 {
 409         struct osd_device       *osd = osd_obj2dev(obj);
 410         const struct lu_fid     *fid = lu_object_fid(&obj->oo_dt.do_lu);
 411         int                      rc = 0;
 412         ENTRY;
 413
 414         LASSERT(obj->oo_dn);
 415
 416         rc = osd_object_sa_init(obj, osd);
 417         if (rc)
 418                 RETURN(rc);
 419
 420         /* cache attrs in object */
 421         rc = __osd_object_attr_get(env, osd, obj, &obj->oo_attr);
 422         if (rc)
 423                 RETURN(rc);
 424
 425         if (likely(!fid_is_acct(fid))) {
 426                 /* no body operations for accounting objects */
 427                 obj->oo_dt.do_body_ops = &osd_body_ops;
 428
 429                 if (S_ISREG(obj->oo_attr.la_mode) &&
 430                     obj->oo_dn->dn_maxblkid == 0 &&
 431                     (1 << obj->oo_dn->dn_datablkshift) < PAGE_SIZE &&
 432                     (fid_is_idif(fid) || fid_is_norm(fid) ||
 433                      fid_is_echo(fid)) &&
 434                     osd->od_is_ost && !osd->od_dt_dev.dd_rdonly)
 435                         osd_obj_set_blksize(env, osd, obj);
 436         }
 437
 438         /*
 439          * initialize object before marking it existing
 440          */
 441         obj->oo_dt.do_lu.lo_header->loh_attr |= obj->oo_attr.la_mode & S_IFMT;
 442
 443         smp_mb();
 444         obj->oo_dt.do_lu.lo_header->loh_attr |= LOHA_EXISTS;
 445
 446         RETURN(0);
 447 }
 448
 449 static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
 450 {
 451         struct osd_thread_info  *info = osd_oti_get(env);
 452         struct lu_buf           buf;
 453         int                     rc;
 454         struct lustre_mdt_attrs *lma;
 455         const struct lu_fid *rfid = lu_object_fid(&obj->oo_dt.do_lu);
 456         ENTRY;
 457
 458         BUILD_BUG_ON(sizeof(info->oti_buf) < sizeof(*lma));
 459         lma = (struct lustre_mdt_attrs *)info->oti_buf;
 460         buf.lb_buf = lma;
 461         buf.lb_len = sizeof(info->oti_buf);
 462
 463         rc = osd_xattr_get(env, &obj->oo_dt, &buf, XATTR_NAME_LMA);
 464         if (rc > 0) {
 465                 rc = 0;
 466                 lustre_lma_swab(lma);
 467                 if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
 468                              CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT))) {
 469                         CWARN("%s: unsupported incompat LMA feature(s) %#x for "
 470                               "fid = "DFID"\n", osd_obj2dev(obj)->od_svname,
 471                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
 472                               PFID(rfid));
 473                         rc = -EOPNOTSUPP;
 474                 } else if (unlikely(!lu_fid_eq(rfid, &lma->lma_self_fid))) {
 475                         CERROR("%s: FID-in-LMA "DFID" does not match the "
 476                               "object self-fid "DFID"\n",
 477                               osd_obj2dev(obj)->od_svname,
 478                               PFID(&lma->lma_self_fid), PFID(rfid));
 479                         rc = -EREMCHG;
 480                 } else {
 481                         struct osd_device *osd = osd_obj2dev(obj);
 482
 483                         if (lma->lma_compat & LMAC_STRIPE_INFO &&
 484                             osd->od_is_ost)
 485                                 obj->oo_pfid_in_lma = 1;
 486                         if (unlikely(lma->lma_incompat & LMAI_REMOTE_PARENT) &&
 487                             osd->od_remote_parent_dir != ZFS_NO_OBJECT)
 488                                 lu_object_set_agent_entry(&obj->oo_dt.do_lu);
 489                 }
 490         } else if (rc == -ENODATA) {
 491                 /* haven't initialize LMA xattr */
 492                 rc = 0;
 493         }
 494
 495         RETURN(rc);
 496 }
 497
 498 /**
 499  * Helper function to retrieve DMU object id from fid for accounting object
 500  */
 501 static dnode_t *osd_quota_fid2dmu(const struct osd_device *osd,
 502                                   const struct lu_fid *fid)
 503 {
 504         dnode_t *dn = NULL;
 505
 506         LASSERT(fid_is_acct(fid));
 507
 508         switch (fid_oid(fid)) {
 509         case ACCT_USER_OID:
 510                 dn = osd->od_userused_dn;
 511                 break;
 512         case ACCT_GROUP_OID:
 513                 dn = osd->od_groupused_dn;
 514                 break;
 515 #ifdef ZFS_PROJINHERIT
 516         case ACCT_PROJECT_OID:
 517                 dn = osd->od_projectused_dn;
 518                 break;
 519 #endif
 520         default:
 521                 break;
 522         }
 523
 524         return dn;
 525 }
 526
 527 /*
 528  * Concurrency: no concurrent access is possible that early in object
 529  * life-cycle.
 530  */
 531 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
 532                            const struct lu_object_conf *conf)
 533 {
 534         struct osd_object *obj = osd_obj(l);
 535         struct osd_device *osd = osd_obj2dev(obj);
 536         const struct lu_fid *fid = lu_object_fid(l);
 537         struct lustre_scrub *scrub = &osd->od_scrub;
 538         struct osd_thread_info *info = osd_oti_get(env);
 539         struct luz_direntry *zde = &info->oti_zde;
 540         struct osd_idmap_cache *idc;
 541         char *name = info->oti_str;
 542         uint64_t oid;
 543         int rc = 0;
 544         int rc1;
 545         bool remote = false;
 546         ENTRY;
 547
 548         LASSERT(osd_invariant(obj));
 549
 550         if (fid_is_otable_it(&l->lo_header->loh_fid)) {
 551                 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
 552                 l->lo_header->loh_attr |= LOHA_EXISTS;
 553
 554                 GOTO(out, rc = 0);
 555         }
 556
 557         if (conf && conf->loc_flags & LOC_F_NEW)
 558                 GOTO(out, rc = 0);
 559
 560         if (unlikely(fid_is_acct(fid))) {
 561                 obj->oo_dn = osd_quota_fid2dmu(osd, fid);
 562                 if (obj->oo_dn) {
 563                         obj->oo_dt.do_index_ops = &osd_acct_index_ops;
 564                         l->lo_header->loh_attr |= LOHA_EXISTS;
 565                 }
 566
 567                 GOTO(out, rc = 0);
 568         }
 569
 570         idc = osd_idc_find(env, osd, fid);
 571         if (idc && !idc->oic_remote && idc->oic_dnode != ZFS_NO_OBJECT) {
 572                 oid = idc->oic_dnode;
 573                 goto zget;
 574         }
 575
 576         rc = -ENOENT;
 577         if (!list_empty(&osd->od_scrub.os_inconsistent_items))
 578                 rc = osd_oii_lookup(osd, fid, &oid);
 579
 580         if (rc)
 581                 rc = osd_fid_lookup(env, osd, fid, &oid);
 582
 583         if (rc == -ENOENT) {
 584                 if (likely(!(fid_is_norm(fid) || fid_is_igif(fid)) ||
 585                            fid_is_on_ost(env, osd, fid) ||
 586                            !zfs_test_bit(osd_oi_fid2idx(osd, fid),
 587                                          scrub->os_file.sf_oi_bitmap)))
 588                         GOTO(out, rc = 0);
 589
 590                 rc = -EREMCHG;
 591                 goto trigger;
 592         }
 593
 594         if (rc)
 595                 GOTO(out, rc);
 596
 597 zget:
 598         LASSERT(obj->oo_dn == NULL);
 599
 600         rc = __osd_obj2dnode(osd->od_os, oid, &obj->oo_dn);
 601         /* EEXIST will be returned if object is being deleted in ZFS */
 602         if (rc == -EEXIST)
 603                 GOTO(out, rc = 0);
 604
 605         if (rc) {
 606                 CERROR("%s: lookup "DFID"/%#llx failed: rc = %d\n",
 607                        osd->od_svname, PFID(lu_object_fid(l)), oid, rc);
 608                 GOTO(out, rc);
 609         }
 610
 611         rc = osd_object_init0(env, obj);
 612         if (rc)
 613                 GOTO(out, rc);
 614
 615         if (unlikely(obj->oo_header))
 616                 GOTO(out, rc = 0);
 617
 618         rc = osd_check_lma(env, obj);
 619         if (rc != -EREMCHG)
 620                 GOTO(out, rc);
 621
 622         osd_scrub_refresh_mapping(env, osd, fid, oid, DTO_INDEX_DELETE, true,
 623                                   NULL);
 624
 625 trigger:
 626         /* We still have chance to get the valid dnode: for the object that is
 627          * referenced by remote name entry, the object on the local MDT will be
 628          * linked under the dir /REMOTE_PARENT_DIR with its FID string as name.
 629          *
 630          * During the OI scrub, if we cannot find the OI mapping, we may still
 631          * have change to map the FID to local OID via lookup the dir
 632          * /REMOTE_PARENT_DIR. */
 633         if (!remote && !fid_is_on_ost(env, osd, fid)) {
 634                 osd_fid2str(name, fid, sizeof(info->oti_str));
 635                 rc = osd_zap_lookup(osd, osd->od_remote_parent_dir,
 636                                     NULL, name, 8, 3, (void *)zde);
 637                 if (!rc) {
 638                         oid = zde->lzd_reg.zde_dnode;
 639                         osd_dnode_rele(obj->oo_dn);
 640                         obj->oo_dn = NULL;
 641                         remote = true;
 642                         goto zget;
 643                 }
 644         }
 645
 646         /* The case someone triggered the OI scrub already. */
 647         if (scrub->os_running) {
 648                 if (!rc) {
 649                         LASSERT(remote);
 650
 651                         lu_object_set_agent_entry(l);
 652                         osd_oii_insert(env, osd, fid, oid, false);
 653                 } else {
 654                         rc = -EINPROGRESS;
 655                 }
 656
 657                 GOTO(out, rc);
 658         }
 659
 660         /* The case NOT allow to trigger OI scrub automatically. */
 661         if (osd->od_auto_scrub_interval == AS_NEVER)
 662                 GOTO(out, rc);
 663
 664         /* It is me to trigger the OI scrub. */
 665         rc1 = osd_scrub_start(env, osd, SS_CLEAR_DRYRUN |
 666                               SS_CLEAR_FAILOUT | SS_AUTO_FULL);
 667         CDEBUG_LIMIT(D_LFSCK | D_CONSOLE | D_WARNING,
 668                      "%s: trigger OI scrub by RPC for "DFID"/%#llx: rc = %d\n",
 669                      osd_name(osd), PFID(fid), oid, rc1);
 670         if (!rc) {
 671                 LASSERT(remote);
 672
 673                 lu_object_set_agent_entry(l);
 674                 if (!rc1)
 675                         osd_oii_insert(env, osd, fid, oid, false);
 676         } else {
 677                 if (!rc1)
 678                         rc = -EINPROGRESS;
 679                 else
 680                         rc = -EREMCHG;
 681         }
 682
 683         GOTO(out, rc);
 684
 685 out:
 686         RETURN(rc);
 687 }
 688
 689 /*
 690  * Concurrency: no concurrent access is possible that late in object
 691  * life-cycle.
 692  */
 693 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
 694 {
 695         struct osd_object *obj = osd_obj(l);
 696         struct lu_object_header *h = obj->oo_header;
 697
 698         LASSERT(osd_invariant(obj));
 699
 700         dt_object_fini(&obj->oo_dt);
 701         /* obj doesn't contain an lu_object_header, so we don't need call_rcu */
 702         OBD_SLAB_FREE_PTR(obj, osd_object_kmem);
 703         if (unlikely(h))
 704                 lu_object_header_free(h);
 705 }
 706
 707 static int
 708 osd_object_unlinked_add(struct osd_object *obj, struct osd_thandle *oh)
 709 {
 710         int rc = -EBUSY;
 711
 712         LASSERT(obj->oo_destroy == OSD_DESTROY_ASYNC);
 713
 714         /* the object is supposed to be exclusively locked by
 715          * the caller (osd_destroy()), while the transaction
 716          * (oh) is per-thread and not shared */
 717         if (likely(list_empty(&obj->oo_unlinked_linkage))) {
 718                 list_add(&obj->oo_unlinked_linkage, &oh->ot_unlinked_list);
 719                 rc = 0;
 720         }
 721
 722         return rc;
 723 }
 724
 725 /* Default to max data size covered by a level-1 indirect block */
 726 static unsigned long osd_sync_destroy_max_size =
 727         1UL << (DN_MAX_INDBLKSHIFT - SPA_BLKPTRSHIFT + SPA_MAXBLOCKSHIFT);
 728 module_param(osd_sync_destroy_max_size, ulong, 0444);
 729 MODULE_PARM_DESC(osd_sync_destroy_max_size, "Maximum object size to use synchronous destroy.");
 730
 731 static inline void
 732 osd_object_set_destroy_type(struct osd_object *obj)
 733 {
 734         /*
 735          * Lock-less OST_WRITE can race with OST_DESTROY, so set destroy type
 736          * only once and use it consistently thereafter.
 737          */
 738         down_write(&obj->oo_guard);
 739         if (obj->oo_destroy == OSD_DESTROY_NONE) {
 740                 if (obj->oo_attr.la_size <= osd_sync_destroy_max_size)
 741                         obj->oo_destroy = OSD_DESTROY_SYNC;
 742                 else /* Larger objects are destroyed asynchronously */
 743                         obj->oo_destroy = OSD_DESTROY_ASYNC;
 744         }
 745         up_write(&obj->oo_guard);
 746 }
 747
 748 static int osd_declare_destroy(const struct lu_env *env, struct dt_object *dt,
 749                                struct thandle *th)
 750 {
 751         const struct lu_fid     *fid = lu_object_fid(&dt->do_lu);
 752         struct osd_object       *obj = osd_dt_obj(dt);
 753         struct osd_device       *osd = osd_obj2dev(obj);
 754         struct osd_thandle      *oh;
 755         dnode_t *dn;
 756         int                      rc;
 757         uint64_t                 zapid;
 758         ENTRY;
 759
 760         LASSERT(th != NULL);
 761         LASSERT(dt_object_exists(dt));
 762
 763         oh = container_of(th, struct osd_thandle, ot_super);
 764         LASSERT(oh->ot_tx != NULL);
 765
 766         dmu_tx_mark_netfree(oh->ot_tx);
 767
 768         /* declare that we'll remove object from fid-dnode mapping */
 769         zapid = osd_get_name_n_idx(env, osd, fid, NULL, 0, &dn);
 770         osd_tx_hold_zap(oh->ot_tx, zapid, dn, FALSE, NULL);
 771
 772         osd_declare_xattrs_destroy(env, obj, oh);
 773
 774         /* one less inode */
 775         rc = osd_declare_quota(env, osd, obj->oo_attr.la_uid,
 776                                obj->oo_attr.la_gid, obj->oo_attr.la_projid,
 777                                -1, oh, NULL, OSD_QID_INODE);
 778         if (rc)
 779                 RETURN(rc);
 780
 781         /* data to be truncated */
 782         rc = osd_declare_quota(env, osd, obj->oo_attr.la_uid,
 783                                obj->oo_attr.la_gid, obj->oo_attr.la_projid,
 784                                0, oh, NULL, OSD_QID_BLK);
 785         if (rc)
 786                 RETURN(rc);
 787
 788         osd_object_set_destroy_type(obj);
 789         if (obj->oo_destroy == OSD_DESTROY_SYNC)
 790                 dmu_tx_hold_free(oh->ot_tx, obj->oo_dn->dn_object,
 791                                  0, DMU_OBJECT_END);
 792         else
 793                 osd_tx_hold_zap(oh->ot_tx, osd->od_unlinked->dn_object,
 794                                 osd->od_unlinked, TRUE, NULL);
 795
 796         /* remove agent entry (if have) from remote parent */
 797         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu))
 798                 osd_tx_hold_zap(oh->ot_tx, osd->od_remote_parent_dir,
 799                                 NULL, FALSE, NULL);
 800
 801         /* will help to find FID->ino when this object is being
 802          * added to PENDING/ */
 803         osd_idc_find_and_init(env, osd, obj);
 804
 805         RETURN(0);
 806 }
 807
 808 static int osd_destroy(const struct lu_env *env, struct dt_object *dt,
 809                        struct thandle *th)
 810 {
 811         struct osd_thread_info  *info = osd_oti_get(env);
 812         char                    *buf = info->oti_str;
 813         struct osd_object       *obj = osd_dt_obj(dt);
 814         struct osd_device       *osd = osd_obj2dev(obj);
 815         const struct lu_fid     *fid = lu_object_fid(&dt->do_lu);
 816         struct osd_thandle      *oh;
 817         int                      rc;
 818         uint64_t                 oid, zapid;
 819         dnode_t *zdn;
 820         ENTRY;
 821
 822         down_write(&obj->oo_guard);
 823
 824         if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
 825                 GOTO(out, rc = -ENOENT);
 826
 827         LASSERT(obj->oo_dn != NULL);
 828
 829         oh = container_of(th, struct osd_thandle, ot_super);
 830         LASSERT(oh != NULL);
 831         LASSERT(oh->ot_tx != NULL);
 832
 833         /* remove obj ref from index dir (it depends) */
 834         zapid = osd_get_name_n_idx(env, osd, fid, buf,
 835                                    sizeof(info->oti_str), &zdn);
 836         rc = osd_xattrs_destroy(env, obj, oh);
 837         if (rc) {
 838                 CERROR("%s: cannot destroy xattrs for %s: rc = %d\n",
 839                        osd->od_svname, buf, rc);
 840                 GOTO(out, rc);
 841         }
 842
 843         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu)) {
 844                 rc = osd_delete_from_remote_parent(env, osd, obj, oh, true);
 845                 if (rc)
 846                         GOTO(out, rc);
 847         }
 848
 849         oid = obj->oo_dn->dn_object;
 850         if (unlikely(obj->oo_destroy == OSD_DESTROY_NONE)) {
 851                 /* this may happen if the destroy wasn't declared
 852                  * e.g. when the object is created and then destroyed
 853                  * in the same transaction - we don't need additional
 854                  * space for destroy specifically */
 855                 LASSERT(obj->oo_attr.la_size <= osd_sync_destroy_max_size);
 856                 rc = -dmu_object_free(osd->od_os, oid, oh->ot_tx);
 857                 if (rc)
 858                         CERROR("%s: failed to free %s/%#llx: rc = %d\n",
 859                                osd->od_svname, buf, oid, rc);
 860         } else if (obj->oo_destroy == OSD_DESTROY_SYNC) {
 861                 rc = -dmu_object_free(osd->od_os, oid, oh->ot_tx);
 862                 if (rc)
 863                         CERROR("%s: failed to free %s/%#llx: rc = %d\n",
 864                                osd->od_svname, buf, oid, rc);
 865         } else { /* asynchronous destroy */
 866                 char *key = info->oti_key;
 867
 868                 rc = osd_object_unlinked_add(obj, oh);
 869                 if (rc)
 870                         GOTO(out, rc);
 871
 872                 snprintf(key, sizeof(info->oti_key), "%llx", oid);
 873                 rc = osd_zap_add(osd, osd->od_unlinked->dn_object,
 874                                  osd->od_unlinked, key, 8, 1, &oid, oh->ot_tx);
 875                 if (rc)
 876                         CERROR("%s: zap_add_int() failed %s/%#llx: rc = %d\n",
 877                                osd->od_svname, buf, oid, rc);
 878         }
 879
 880         /* Remove the OI mapping after the destroy to handle the race with
 881          * OI scrub that may insert missed OI mapping during the interval. */
 882         rc = osd_zap_remove(osd, zapid, zdn, buf, oh->ot_tx);
 883         if (unlikely(rc == -ENOENT))
 884                 rc = 0;
 885         if (rc)
 886                 CERROR("%s: zap_remove(%s) failed: rc = %d\n",
 887                        osd->od_svname, buf, rc);
 888
 889         GOTO(out, rc);
 890
 891 out:
 892         /* not needed in the cache anymore */
 893         set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
 894         if (rc == 0)
 895                 obj->oo_destroyed = 1;
 896         up_write(&obj->oo_guard);
 897         RETURN (0);
 898 }
 899
 900 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
 901 {
 902         struct osd_object *obj = osd_obj(l);
 903         const struct lu_fid *fid = lu_object_fid(l);
 904
 905         if (obj->oo_dn) {
 906                 if (likely(!fid_is_acct(fid))) {
 907                         osd_object_sa_fini(obj);
 908                         if (obj->oo_sa_xattr) {
 909                                 nvlist_free(obj->oo_sa_xattr);
 910                                 obj->oo_sa_xattr = NULL;
 911                         }
 912                         osd_dnode_rele(obj->oo_dn);
 913                         list_del(&obj->oo_sa_linkage);
 914                 }
 915                 obj->oo_dn = NULL;
 916         }
 917 }
 918
 919 /*
 920  * Concurrency: ->loo_object_release() is called under site spin-lock.
 921  */
 922 static void osd_object_release(const struct lu_env *env,
 923                                struct lu_object *l)
 924 {
 925 }
 926
 927 /*
 928  * Concurrency: shouldn't matter.
 929  */
 930 static int osd_object_print(const struct lu_env *env, void *cookie,
 931                             lu_printer_t p, const struct lu_object *l)
 932 {
 933         struct osd_object *o = osd_obj(l);
 934
 935         return (*p)(env, cookie, LUSTRE_OSD_ZFS_NAME"-object@%p", o);
 936 }
 937
 938 static void osd_read_lock(const struct lu_env *env, struct dt_object *dt,
 939                           unsigned role)
 940 {
 941         struct osd_object *obj = osd_dt_obj(dt);
 942
 943         LASSERT(osd_invariant(obj));
 944
 945         down_read_nested(&obj->oo_sem, role);
 946 }
 947
 948 static void osd_write_lock(const struct lu_env *env, struct dt_object *dt,
 949                            unsigned role)
 950 {
 951         struct osd_object *obj = osd_dt_obj(dt);
 952
 953         LASSERT(osd_invariant(obj));
 954
 955         down_write_nested(&obj->oo_sem, role);
 956 }
 957
 958 static void osd_read_unlock(const struct lu_env *env, struct dt_object *dt)
 959 {
 960         struct osd_object *obj = osd_dt_obj(dt);
 961
 962         LASSERT(osd_invariant(obj));
 963         up_read(&obj->oo_sem);
 964 }
 965
 966 static void osd_write_unlock(const struct lu_env *env, struct dt_object *dt)
 967 {
 968         struct osd_object *obj = osd_dt_obj(dt);
 969
 970         LASSERT(osd_invariant(obj));
 971         up_write(&obj->oo_sem);
 972 }
 973
 974 static int osd_write_locked(const struct lu_env *env, struct dt_object *dt)
 975 {
 976         struct osd_object *obj = osd_dt_obj(dt);
 977         int rc = 1;
 978
 979         LASSERT(osd_invariant(obj));
 980
 981         if (down_write_trylock(&obj->oo_sem)) {
 982                 rc = 0;
 983                 up_write(&obj->oo_sem);
 984         }
 985         return rc;
 986 }
 987
 988 static int osd_attr_get(const struct lu_env *env, struct dt_object *dt,
 989                         struct lu_attr *attr)
 990 {
 991         struct osd_object *obj = osd_dt_obj(dt);
 992         struct osd_device *osd = osd_obj2dev(obj);
 993         uint64_t blocks;
 994         uint32_t blksize;
 995         int rc = 0;
 996
 997         down_read(&obj->oo_guard);
 998
 999         if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
1000                 GOTO(out, rc = -ENOENT);
1001
1002         if (unlikely(fid_is_acct(lu_object_fid(&dt->do_lu))))
1003                 GOTO(out, rc = 0);
1004
1005         LASSERT(osd_invariant(obj));
1006         LASSERT(obj->oo_dn);
1007
1008         read_lock(&obj->oo_attr_lock);
1009         *attr = obj->oo_attr;
1010         if (obj->oo_lma_flags & LUSTRE_ORPHAN_FL) {
1011                 attr->la_valid |= LA_FLAGS;
1012                 attr->la_flags |= LUSTRE_ORPHAN_FL;
1013         }
1014         if (obj->oo_lma_flags & LUSTRE_ENCRYPT_FL) {
1015                 attr->la_valid |= LA_FLAGS;
1016                 attr->la_flags |= LUSTRE_ENCRYPT_FL;
1017         }
1018         read_unlock(&obj->oo_attr_lock);
1019         if (attr->la_valid & LA_FLAGS && attr->la_flags & LUSTRE_ORPHAN_FL)
1020                 CDEBUG(D_INFO, "%s: set orphan flag on "DFID" (%#llx/%#x)\n",
1021                        osd_obj2dev(obj)->od_svname,
1022                        PFID(lu_object_fid(&dt->do_lu)),
1023                        attr->la_valid, obj->oo_lma_flags);
1024
1025         /* with ZFS_DEBUG zrl_add_debug() called by DB_DNODE_ENTER()
1026          * from within sa_object_size() can block on a mutex, so
1027          * we can't call sa_object_size() holding rwlock */
1028         sa_object_size(obj->oo_sa_hdl, &blksize, &blocks);
1029         /* we do not control size of indices, so always calculate
1030          * it from number of blocks reported by DMU */
1031         if (S_ISDIR(attr->la_mode)) {
1032                 attr->la_size = 512 * blocks;
1033                 rc = -zap_count(osd->od_os, obj->oo_dn->dn_object,
1034                                 &attr->la_dirent_count);
1035         }
1036         /* Block size may be not set; suggest maximal I/O transfers. */
1037         if (blksize == 0)
1038                 blksize = osd_spa_maxblocksize(
1039                         dmu_objset_spa(osd_obj2dev(obj)->od_os));
1040
1041         attr->la_blksize = blksize;
1042         attr->la_blocks = blocks;
1043         attr->la_valid |= LA_BLOCKS | LA_BLKSIZE;
1044
1045 out:
1046         up_read(&obj->oo_guard);
1047         return rc;
1048 }
1049
1050 /* Simple wrapper on top of qsd API which implement quota transfer for osd
1051  * setattr needs. As a reminder, only the root user can change ownership of
1052  * a file, that's why EDQUOT & EINPROGRESS errors are discarded */
1053 static inline int qsd_transfer(const struct lu_env *env,
1054                                struct qsd_instance *qsd,
1055                                struct lquota_trans *trans, int qtype,
1056                                __u64 orig_id, __u64 new_id, __u64 bspace,
1057                                struct lquota_id_info *qi, bool ignore_edquot)
1058 {
1059         int     rc;
1060
1061         if (unlikely(qsd == NULL))
1062                 return 0;
1063
1064         LASSERT(qtype >= 0 && qtype < LL_MAXQUOTAS);
1065         qi->lqi_type = qtype;
1066
1067         /* inode accounting */
1068         qi->lqi_is_blk = false;
1069
1070         /* one more inode for the new owner ... */
1071         qi->lqi_id.qid_uid = new_id;
1072         qi->lqi_space      = 1;
1073         rc = qsd_op_begin(env, qsd, trans, qi, NULL);
1074         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
1075                 rc = 0;
1076         if (rc)
1077                 return rc;
1078
1079         /* and one less inode for the current id */
1080         qi->lqi_id.qid_uid = orig_id;;
1081         qi->lqi_space      = -1;
1082         /* can't get EDQUOT when reducing usage */
1083         rc = qsd_op_begin(env, qsd, trans, qi, NULL);
1084         if (rc == -EINPROGRESS)
1085                 rc = 0;
1086         if (rc)
1087                 return rc;
1088
1089         /* block accounting */
1090         qi->lqi_is_blk = true;
1091
1092         /* more blocks for the new owner ... */
1093         qi->lqi_id.qid_uid = new_id;
1094         qi->lqi_space      = bspace;
1095         rc = qsd_op_begin(env, qsd, trans, qi, NULL);
1096         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
1097                 rc = 0;
1098         if (rc)
1099                 return rc;
1100
1101         /* and finally less blocks for the current owner */
1102         qi->lqi_id.qid_uid = orig_id;
1103         qi->lqi_space      = -bspace;
1104         rc = qsd_op_begin(env, qsd, trans, qi, NULL);
1105         /* can't get EDQUOT when reducing usage */
1106         if (rc == -EINPROGRESS)
1107                 rc = 0;
1108         return rc;
1109 }
1110
1111 static int osd_declare_attr_set(const struct lu_env *env,
1112                                 struct dt_object *dt,
1113                                 const struct lu_attr *attr,
1114                                 struct thandle *handle)
1115 {
1116         struct osd_thread_info  *info = osd_oti_get(env);
1117         struct osd_object       *obj = osd_dt_obj(dt);
1118         struct osd_device       *osd = osd_obj2dev(obj);
1119         dmu_tx_hold_t           *txh;
1120         struct osd_thandle      *oh;
1121         uint64_t                 bspace;
1122         uint32_t                 blksize;
1123         int                      rc = 0;
1124         bool                     found;
1125         ENTRY;
1126
1127
1128         LASSERT(handle != NULL);
1129         LASSERT(osd_invariant(obj));
1130
1131         oh = container_of(handle, struct osd_thandle, ot_super);
1132
1133         down_read(&obj->oo_guard);
1134         if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
1135                 GOTO(out, rc = 0);
1136
1137         LASSERT(obj->oo_sa_hdl != NULL);
1138         LASSERT(oh->ot_tx != NULL);
1139         /* regular attributes are part of the bonus buffer */
1140         /* let's check whether this object is already part of
1141          * transaction.. */
1142         found = false;
1143         for (txh = list_head(&oh->ot_tx->tx_holds); txh;
1144              txh = list_next(&oh->ot_tx->tx_holds, txh)) {
1145                 if (txh->txh_dnode == NULL)
1146                         continue;
1147                 if (txh->txh_dnode->dn_object != obj->oo_dn->dn_object)
1148                         continue;
1149                 /* this object is part of the transaction already
1150                  * we don't need to declare bonus again */
1151                 found = true;
1152                 break;
1153         }
1154         if (!found)
1155                 dmu_tx_hold_bonus(oh->ot_tx, obj->oo_dn->dn_object);
1156         if (oh->ot_tx->tx_err != 0)
1157                 GOTO(out, rc = -oh->ot_tx->tx_err);
1158
1159         if (attr && attr->la_valid & LA_FLAGS) {
1160                 /* LMA is usually a part of bonus, no need to declare
1161                  * anything else */
1162         }
1163
1164         if (attr && (attr->la_valid & (LA_UID | LA_GID | LA_PROJID))) {
1165                 sa_object_size(obj->oo_sa_hdl, &blksize, &bspace);
1166                 bspace = toqb(bspace * 512);
1167
1168                 CDEBUG(D_QUOTA,
1169                        "%s: enforce quota on UID %u, GID %u, the quota space is %lld (%u)\n",
1170                        osd->od_svname,
1171                        attr->la_uid, attr->la_gid, bspace, blksize);
1172         }
1173
1174         if (attr && attr->la_valid & LA_UID) {
1175                 /* quota enforcement for user */
1176                 if (attr->la_uid != obj->oo_attr.la_uid) {
1177                         rc = qsd_transfer(env, osd_def_qsd(osd),
1178                                           &oh->ot_quota_trans, USRQUOTA,
1179                                           obj->oo_attr.la_uid, attr->la_uid,
1180                                           bspace, &info->oti_qi, true);
1181                         if (rc)
1182                                 GOTO(out, rc);
1183                 }
1184         }
1185         if (attr && attr->la_valid & LA_GID) {
1186                 /* quota enforcement for group */
1187                 if (attr->la_gid != obj->oo_attr.la_gid) {
1188                         rc = qsd_transfer(env, osd_def_qsd(osd),
1189                                           &oh->ot_quota_trans, GRPQUOTA,
1190                                           obj->oo_attr.la_gid, attr->la_gid,
1191                                           bspace, &info->oti_qi,
1192                                           !(attr->la_flags &
1193                                                         LUSTRE_SET_SYNC_FL));
1194                         if (rc)
1195                                 GOTO(out, rc);
1196                 }
1197         }
1198 #ifdef ZFS_PROJINHERIT
1199         if (attr && attr->la_valid & LA_PROJID) {
1200                 /* quota enforcement for project */
1201                 if (attr->la_projid != obj->oo_attr.la_projid) {
1202                         if (!osd->od_projectused_dn)
1203                                 GOTO(out, rc = -EOPNOTSUPP);
1204
1205                         /* Usually, if project quota is upgradable for the
1206                          * device, then the upgrade will be done before or when
1207                          * mount the device. So when we come here, this project
1208                          * should have project ID attribute already (that is
1209                          * zero by default).  Otherwise, there was something
1210                          * wrong during the former upgrade, let's return failure
1211                          * to report that.
1212                          *
1213                          * Please note that, different from other attributes,
1214                          * you can NOT simply set the project ID attribute under
1215                          * such case, because adding (NOT change) project ID
1216                          * attribute needs to change the object's attribute
1217                          * layout to match zfs backend quota accounting
1218                          * requirement. */
1219                         if (unlikely(!obj->oo_with_projid))
1220                                 GOTO(out, rc = -ENXIO);
1221
1222                         rc = qsd_transfer(env, osd_def_qsd(osd),
1223                                           &oh->ot_quota_trans, PRJQUOTA,
1224                                           obj->oo_attr.la_projid,
1225                                           attr->la_projid, bspace,
1226                                           &info->oti_qi, true);
1227                         if (rc)
1228                                 GOTO(out, rc);
1229                 }
1230         }
1231 #endif
1232 out:
1233         up_read(&obj->oo_guard);
1234         RETURN(rc);
1235 }
1236
1237 /*
1238  * Set the attributes of an object
1239  *
1240  * The transaction passed to this routine must have
1241  * dmu_tx_hold_bonus(tx, oid) called and then assigned
1242  * to a transaction group.
1243  */
1244 static int osd_attr_set(const struct lu_env *env, struct dt_object *dt,
1245                         const struct lu_attr *la, struct thandle *handle)
1246 {
1247         struct osd_thread_info  *info = osd_oti_get(env);
1248         sa_bulk_attr_t          *bulk = osd_oti_get(env)->oti_attr_bulk;
1249         struct osd_object       *obj = osd_dt_obj(dt);
1250         struct osd_device       *osd = osd_obj2dev(obj);
1251         struct osd_thandle      *oh;
1252         struct osa_attr         *osa = &info->oti_osa;
1253         __u64                    valid = la->la_valid;
1254         int                      cnt;
1255         int                      rc = 0;
1256
1257         ENTRY;
1258
1259         down_read(&obj->oo_guard);
1260         if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
1261                 GOTO(out, rc = -ENOENT);
1262
1263         LASSERT(handle != NULL);
1264         LASSERT(osd_invariant(obj));
1265         LASSERT(obj->oo_sa_hdl);
1266
1267         oh = container_of(handle, struct osd_thandle, ot_super);
1268         /* Assert that the transaction has been assigned to a
1269            transaction group. */
1270         LASSERT(oh->ot_tx->tx_txg != 0);
1271
1272         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING) && !osd->od_is_ost) {
1273                 struct zpl_direntry *zde = &info->oti_zde.lzd_reg;
1274                 char *buf = info->oti_str;
1275                 dnode_t *zdn = NULL;
1276                 uint64_t zapid;
1277
1278                 zapid = osd_get_name_n_idx(env, osd, lu_object_fid(&dt->do_lu),
1279                                            buf, sizeof(info->oti_str), &zdn);
1280                 rc = osd_zap_lookup(osd, zapid, zdn, buf, 8,
1281                                     sizeof(*zde) / 8, zde);
1282                 if (!rc) {
1283                         zde->zde_dnode -= 1;
1284                         rc = -zap_update(osd->od_os, zapid, buf, 8,
1285                                          sizeof(*zde) / 8, zde, oh->ot_tx);
1286                 }
1287                 if (rc > 0)
1288                         rc = 0;
1289                 GOTO(out, rc);
1290         }
1291
1292         /* Only allow set size for regular file */
1293         if (!S_ISREG(dt->do_lu.lo_header->loh_attr))
1294                 valid &= ~(LA_SIZE | LA_BLOCKS);
1295
1296         if (valid & LA_CTIME && la->la_ctime == obj->oo_attr.la_ctime)
1297                 valid &= ~LA_CTIME;
1298
1299         if (valid & LA_MTIME && la->la_mtime == obj->oo_attr.la_mtime)
1300                 valid &= ~LA_MTIME;
1301
1302         if (valid & LA_ATIME && la->la_atime == obj->oo_attr.la_atime)
1303                 valid &= ~LA_ATIME;
1304
1305         if (valid == 0)
1306                 GOTO(out, rc = 0);
1307
1308         if (valid & LA_FLAGS) {
1309                 struct lustre_mdt_attrs *lma;
1310                 struct lu_buf buf;
1311                 int size = 0;
1312
1313                 if (la->la_flags & LUSTRE_LMA_FL_MASKS) {
1314                         LASSERT(!obj->oo_pfid_in_lma);
1315                         BUILD_BUG_ON(sizeof(info->oti_buf) < sizeof(*lma));
1316                         lma = (struct lustre_mdt_attrs *)&info->oti_buf;
1317                         buf.lb_buf = lma;
1318                         buf.lb_len = sizeof(info->oti_buf);
1319
1320                         /* Please do NOT call osd_xattr_get() directly, that
1321                          * will cause recursive down_read() on oo_guard. */
1322                         rc = osd_xattr_get_internal(env, obj, &buf,
1323                                                     XATTR_NAME_LMA, &size);
1324                         if (!rc && unlikely(size < sizeof(*lma))) {
1325                                 rc = -EINVAL;
1326                         } else if (!rc) {
1327                                 lma->lma_incompat =
1328                                         le32_to_cpu(lma->lma_incompat);
1329                                 lma->lma_incompat |=
1330                                         lustre_to_lma_flags(la->la_flags);
1331                                 lma->lma_incompat =
1332                                         cpu_to_le32(lma->lma_incompat);
1333                                 buf.lb_buf = lma;
1334                                 buf.lb_len = sizeof(*lma);
1335                                 rc = osd_xattr_set_internal(env, obj, &buf,
1336                                                             XATTR_NAME_LMA,
1337                                                             LU_XATTR_REPLACE,
1338                                                             oh);
1339                         }
1340                         if (rc < 0) {
1341                                 CWARN("%s: failed to set LMA flags: rc = %d\n",
1342                                        osd->od_svname, rc);
1343                                 GOTO(out, rc);
1344                         } else {
1345                                 obj->oo_lma_flags =
1346                                         la->la_flags & LUSTRE_LMA_FL_MASKS;
1347                         }
1348                 }
1349         }
1350
1351         write_lock(&obj->oo_attr_lock);
1352         cnt = 0;
1353
1354         if (valid & LA_PROJID) {
1355 #ifdef ZFS_PROJINHERIT
1356                 if (osd->od_projectused_dn) {
1357                         LASSERT(obj->oo_with_projid);
1358
1359                         osa->projid = obj->oo_attr.la_projid = la->la_projid;
1360                         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_PROJID(osd), NULL,
1361                                          &osa->projid, 8);
1362                 } else
1363 #endif
1364                         valid &= ~LA_PROJID;
1365         }
1366
1367         if (valid & LA_ATIME) {
1368                 osa->atime[0] = obj->oo_attr.la_atime = la->la_atime;
1369                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(osd), NULL,
1370                                  osa->atime, 16);
1371         }
1372         if (valid & LA_MTIME) {
1373                 osa->mtime[0] = obj->oo_attr.la_mtime = la->la_mtime;
1374                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(osd), NULL,
1375                                  osa->mtime, 16);
1376         }
1377         if (valid & LA_CTIME) {
1378                 osa->ctime[0] = obj->oo_attr.la_ctime = la->la_ctime;
1379                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(osd), NULL,
1380                                  osa->ctime, 16);
1381         }
1382         if (valid & LA_MODE) {
1383                 /* mode is stored along with type, so read it first */
1384                 obj->oo_attr.la_mode = (obj->oo_attr.la_mode & S_IFMT) |
1385                         (la->la_mode & ~S_IFMT);
1386                 osa->mode = obj->oo_attr.la_mode;
1387                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(osd), NULL,
1388                                  &osa->mode, 8);
1389         }
1390         if (valid & LA_SIZE) {
1391                 osa->size = obj->oo_attr.la_size = la->la_size;
1392                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(osd), NULL,
1393                                  &osa->size, 8);
1394         }
1395         if (valid & LA_NLINK) {
1396                 osa->nlink = obj->oo_attr.la_nlink = la->la_nlink;
1397                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(osd), NULL,
1398                                  &osa->nlink, 8);
1399         }
1400         if (valid & LA_RDEV) {
1401                 osa->rdev = obj->oo_attr.la_rdev = la->la_rdev;
1402                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_RDEV(osd), NULL,
1403                                  &osa->rdev, 8);
1404         }
1405         if (valid & LA_FLAGS) {
1406                 osa->flags = attrs_fs2zfs(la->la_flags);
1407                 /* many flags are not supported by zfs, so ensure a good cached
1408                  * copy */
1409                 obj->oo_attr.la_flags = attrs_zfs2fs(osa->flags);
1410 #ifdef ZFS_PROJINHERIT
1411                 if (obj->oo_with_projid && osd->od_projectused_dn)
1412                         osa->flags |= ZFS_PROJID;
1413 #endif
1414                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(osd), NULL,
1415                                  &osa->flags, 8);
1416         }
1417         if (valid & LA_UID) {
1418                 osa->uid = obj->oo_attr.la_uid = la->la_uid;
1419                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(osd), NULL,
1420                                  &osa->uid, 8);
1421         }
1422         if (valid & LA_GID) {
1423                 osa->gid = obj->oo_attr.la_gid = la->la_gid;
1424                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(osd), NULL,
1425                                  &osa->gid, 8);
1426         }
1427         obj->oo_attr.la_valid |= valid;
1428         write_unlock(&obj->oo_attr_lock);
1429
1430         LASSERT(cnt <= ARRAY_SIZE(osd_oti_get(env)->oti_attr_bulk));
1431         rc = osd_object_sa_bulk_update(obj, bulk, cnt, oh);
1432
1433 out:
1434         up_read(&obj->oo_guard);
1435         RETURN(rc);
1436 }
1437
1438 /*
1439  * Object creation.
1440  *
1441  * XXX temporary solution.
1442  */
1443
1444 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
1445                         struct dt_object *parent, struct dt_object *child,
1446                         umode_t child_mode)
1447 {
1448         LASSERT(ah);
1449
1450         ah->dah_parent = parent;
1451         ah->dah_mode = child_mode;
1452
1453         if (parent != NULL && !dt_object_remote(parent)) {
1454                 /* will help to find FID->ino at dt_insert("..") */
1455                 struct osd_object *pobj = osd_dt_obj(parent);
1456
1457                 osd_idc_find_and_init(env, osd_obj2dev(pobj), pobj);
1458         }
1459 }
1460
1461 static int osd_declare_create(const struct lu_env *env, struct dt_object *dt,
1462                               struct lu_attr *attr,
1463                               struct dt_allocation_hint *hint,
1464                               struct dt_object_format *dof,
1465                               struct thandle *handle)
1466 {
1467         const struct lu_fid     *fid = lu_object_fid(&dt->do_lu);
1468         struct osd_object       *obj = osd_dt_obj(dt);
1469         struct osd_device       *osd = osd_obj2dev(obj);
1470         struct osd_thandle      *oh;
1471         uint64_t                 zapid;
1472         dnode_t                 *dn;
1473         int                      rc, dnode_size;
1474         ENTRY;
1475
1476         LASSERT(dof);
1477
1478         switch (dof->dof_type) {
1479                 case DFT_REGULAR:
1480                 case DFT_SYM:
1481                 case DFT_NODE:
1482                         if (obj->oo_dt.do_body_ops == NULL)
1483                                 obj->oo_dt.do_body_ops = &osd_body_ops;
1484                         break;
1485                 default:
1486                         break;
1487         }
1488
1489         LASSERT(handle != NULL);
1490         oh = container_of(handle, struct osd_thandle, ot_super);
1491         LASSERT(oh->ot_tx != NULL);
1492
1493         /* this is the minimum set of EAs on every Lustre object */
1494         obj->oo_ea_in_bonus = OSD_BASE_EA_IN_BONUS;
1495         /* reserve 32 bytes for extra stuff like ACLs */
1496         dnode_size = size_roundup_power2(obj->oo_ea_in_bonus + 32);
1497
1498         switch (dof->dof_type) {
1499                 case DFT_DIR:
1500                         dt->do_index_ops = &osd_dir_ops;
1501                         /* fallthrough */
1502                 case DFT_INDEX:
1503                         /* for zap create */
1504                         dmu_tx_hold_zap(oh->ot_tx, DMU_NEW_OBJECT, FALSE, NULL);
1505                         dmu_tx_hold_sa_create(oh->ot_tx, dnode_size);
1506                         break;
1507                 case DFT_REGULAR:
1508                 case DFT_SYM:
1509                 case DFT_NODE:
1510                         /* first, we'll create new object */
1511                         dmu_tx_hold_sa_create(oh->ot_tx, dnode_size);
1512                         break;
1513
1514                 default:
1515                         LBUG();
1516                         break;
1517         }
1518
1519         /* and we'll add it to some mapping */
1520         zapid = osd_get_name_n_idx(env, osd, fid, NULL, 0, &dn);
1521         osd_tx_hold_zap(oh->ot_tx, zapid, dn, TRUE, NULL);
1522
1523         /* will help to find FID->ino mapping at dt_insert() */
1524         osd_idc_find_and_init(env, osd, obj);
1525
1526         rc = osd_declare_quota(env, osd, attr->la_uid, attr->la_gid,
1527                                attr->la_projid, 1, oh, NULL, OSD_QID_INODE);
1528
1529         RETURN(rc);
1530 }
1531
1532 int __osd_attr_init(const struct lu_env *env, struct osd_device *osd,
1533                     struct osd_object *obj, sa_handle_t *sa_hdl, dmu_tx_t *tx,
1534                     struct lu_attr *la, uint64_t parent,
1535                     nvlist_t *xattr)
1536 {
1537         sa_bulk_attr_t *bulk = osd_oti_get(env)->oti_attr_bulk;
1538         struct osa_attr *osa = &osd_oti_get(env)->oti_osa;
1539         uint64_t gen;
1540         inode_timespec_t now;
1541         int cnt;
1542         int rc;
1543         char *dxattr = NULL;
1544         size_t sa_size;
1545
1546
1547         LASSERT(sa_hdl);
1548
1549         gen = dmu_tx_get_txg(tx);
1550         gethrestime(&now);
1551         ZFS_TIME_ENCODE(&now, osa->btime);
1552
1553         osa->atime[0] = la->la_atime;
1554         osa->ctime[0] = la->la_ctime;
1555         osa->mtime[0] = la->la_mtime;
1556         osa->mode = la->la_mode;
1557         osa->uid = la->la_uid;
1558         osa->gid = la->la_gid;
1559         osa->rdev = la->la_rdev;
1560         osa->nlink = la->la_nlink;
1561         if (la->la_valid & LA_FLAGS)
1562                 osa->flags = attrs_fs2zfs(la->la_flags);
1563         else
1564                 osa->flags = 0;
1565         osa->size  = la->la_size;
1566 #ifdef ZFS_PROJINHERIT
1567         if (osd->od_projectused_dn) {
1568                 if (la->la_valid & LA_PROJID)
1569                         osa->projid = la->la_projid;
1570                 else
1571                         osa->projid = ZFS_DEFAULT_PROJID;
1572                 osa->flags |= ZFS_PROJID;
1573                 if (obj)
1574                         obj->oo_with_projid = 1;
1575         } else {
1576                 osa->flags &= ~ZFS_PROJID;
1577         }
1578 #endif
1579
1580         /*
1581          * we need to create all SA below upon object create.
1582          *
1583          * XXX The attribute order matters since the accounting callback relies
1584          * on static offsets (i.e. SA_*_OFFSET, see zfs_space_delta_cb()) to
1585          * look up the UID/GID/PROJID attributes. Moreover, the callback does
1586          * not seem to support the spill block.
1587          * We define attributes in the same order as SA_*_OFFSET in order to
1588          * work around the problem. See ORI-610.
1589          */
1590         cnt = 0;
1591         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(osd), NULL, &osa->mode, 8);
1592         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(osd), NULL, &osa->size, 8);
1593         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GEN(osd), NULL, &gen, 8);
1594         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(osd), NULL, &osa->uid, 8);
1595         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(osd), NULL, &osa->gid, 8);
1596         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_PARENT(osd), NULL, &parent, 8);
1597         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(osd), NULL, &osa->flags, 8);
1598         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(osd), NULL, osa->atime, 16);
1599         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(osd), NULL, osa->mtime, 16);
1600         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(osd), NULL, osa->ctime, 16);
1601         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CRTIME(osd), NULL, osa->btime, 16);
1602         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(osd), NULL, &osa->nlink, 8);
1603 #ifdef ZFS_PROJINHERIT
1604         if (osd->od_projectused_dn)
1605                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_PROJID(osd), NULL,
1606                                  &osa->projid, 8);
1607 #endif
1608         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_RDEV(osd), NULL, &osa->rdev, 8);
1609         LASSERT(cnt <= ARRAY_SIZE(osd_oti_get(env)->oti_attr_bulk));
1610
1611         if (xattr) {
1612                 rc = -nvlist_size(xattr, &sa_size, NV_ENCODE_XDR);
1613                 LASSERT(rc == 0);
1614
1615                 dxattr = osd_zio_buf_alloc(sa_size);
1616                 LASSERT(dxattr);
1617
1618                 rc = -nvlist_pack(xattr, &dxattr, &sa_size,
1619                                 NV_ENCODE_XDR, KM_SLEEP);
1620                 LASSERT(rc == 0);
1621
1622                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_DXATTR(osd),
1623                                 NULL, dxattr, sa_size);
1624         }
1625
1626         rc = -sa_replace_all_by_template(sa_hdl, bulk, cnt, tx);
1627         if (dxattr)
1628                 osd_zio_buf_free(dxattr, sa_size);
1629
1630         return rc;
1631 }
1632
1633 int osd_find_new_dnode(const struct lu_env *env, dmu_tx_t *tx,
1634                        uint64_t oid, dnode_t **dnp)
1635 {
1636         dmu_tx_hold_t *txh;
1637         int rc = 0;
1638
1639         /* take dnode_t from tx to save on dnode#->dnode_t lookup */
1640         for (txh = list_tail(&tx->tx_holds); txh;
1641              txh = list_prev(&tx->tx_holds, txh)) {
1642                 dnode_t *dn = txh->txh_dnode;
1643                 dmu_buf_impl_t *db;
1644
1645                 if (dn == NULL)
1646                         continue;
1647                 if (dn->dn_object != oid)
1648                         continue;
1649                 db = dn->dn_bonus;
1650                 if (db == NULL) {
1651                         rw_enter(&dn->dn_struct_rwlock, RW_WRITER);
1652                         if (dn->dn_bonus == NULL)
1653                                 dbuf_create_bonus(dn);
1654                         rw_exit(&dn->dn_struct_rwlock);
1655                 }
1656                 db = dn->dn_bonus;
1657                 LASSERT(db);
1658                 LASSERT(dn->dn_handle);
1659                 DB_DNODE_ENTER(db);
1660                 if (zfs_refcount_add(&db->db_holds, osd_obj_tag) == 1) {
1661                         zfs_refcount_add(&dn->dn_holds, osd_obj_tag);
1662                         atomic_inc_32(&dn->dn_dbufs_count);
1663                 }
1664                 *dnp = dn;
1665                 DB_DNODE_EXIT(db);
1666                 dbuf_read(db, NULL, DB_RF_MUST_SUCCEED | DB_RF_NOPREFETCH);
1667                 break;
1668         }
1669
1670         if (unlikely(*dnp == NULL))
1671                 rc = __osd_obj2dnode(tx->tx_objset, oid, dnp);
1672
1673         return rc;
1674 }
1675
1676 #ifdef HAVE_DMU_OBJECT_ALLOC_DNSIZE
1677 int osd_find_dnsize(struct osd_device *osd, int ea_in_bonus)
1678 {
1679         int dnsize;
1680
1681         if (osd->od_dnsize == ZFS_DNSIZE_AUTO) {
1682                 dnsize = DNODE_MIN_SIZE;
1683                 do {
1684                         if (DN_BONUS_SIZE(dnsize) >= ea_in_bonus + 32)
1685                                 break;
1686                         dnsize <<= 1;
1687                 } while (dnsize < DNODE_MAX_SIZE);
1688                 if (dnsize > DNODE_MAX_SIZE)
1689                         dnsize = DNODE_MAX_SIZE;
1690         } else if (osd->od_dnsize == ZFS_DNSIZE_1K) {
1691                 dnsize = 1024;
1692         } else if (osd->od_dnsize == ZFS_DNSIZE_2K) {
1693                 dnsize = 2048;
1694         } else if (osd->od_dnsize == ZFS_DNSIZE_4K) {
1695                 dnsize = 4096;
1696         } else if (osd->od_dnsize == ZFS_DNSIZE_8K) {
1697                 dnsize = 8192;
1698         } else if (osd->od_dnsize == ZFS_DNSIZE_16K) {
1699                 dnsize = 16384;
1700         } else {
1701                 dnsize = DNODE_MIN_SIZE;
1702         }
1703         return dnsize;
1704 }
1705 #endif
1706
1707 /*
1708  * The transaction passed to this routine must have
1709  * dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT) called and then assigned
1710  * to a transaction group.
1711  */
1712 int __osd_object_create(const struct lu_env *env, struct osd_device *osd,
1713                         struct osd_object *obj, const struct lu_fid *fid,
1714                         dnode_t **dnp, dmu_tx_t *tx, struct lu_attr *la)
1715 {
1716         dmu_object_type_t type = DMU_OT_PLAIN_FILE_CONTENTS;
1717         uint64_t oid;
1718         int size;
1719
1720         /* Use DMU_OTN_UINT8_METADATA for local objects so their data blocks
1721          * would get an additional ditto copy */
1722         if (unlikely(S_ISREG(la->la_mode) &&
1723                      fid_seq_is_local_file(fid_seq(fid))))
1724                 type = DMU_OTN_UINT8_METADATA;
1725
1726         /* Create a new DMU object using the default dnode size. */
1727         if (obj)
1728                 size = obj->oo_ea_in_bonus;
1729         else
1730                 size = OSD_BASE_EA_IN_BONUS;
1731         oid = osd_dmu_object_alloc(osd->od_os, type, 0,
1732                                    osd_find_dnsize(osd, size), tx);
1733
1734         LASSERT(la->la_valid & LA_MODE);
1735         la->la_size = 0;
1736         la->la_nlink = 1;
1737
1738         return osd_find_new_dnode(env, tx, oid, dnp);
1739 }
1740
1741 /*
1742  * The transaction passed to this routine must have
1743  * dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, ...) called and then assigned
1744  * to a transaction group.
1745  *
1746  * Using ZAP_FLAG_HASH64 will force the ZAP to always be a FAT ZAP.
1747  * This is fine for directories today, because storing the FID in the dirent
1748  * will also require a FAT ZAP.  If there is a new type of micro ZAP created
1749  * then we might need to re-evaluate the use of this flag and instead do
1750  * a conversion from the different internal ZAP hash formats being used. */
1751 int __osd_zap_create(const struct lu_env *env, struct osd_device *osd,
1752                      dnode_t **dnp, dmu_tx_t *tx, struct lu_attr *la,
1753                      unsigned dnsize, zap_flags_t flags)
1754 {
1755         uint64_t oid;
1756
1757         /* Assert that the transaction has been assigned to a
1758            transaction group. */
1759         LASSERT(tx->tx_txg != 0);
1760         *dnp = NULL;
1761
1762         oid = osd_zap_create_flags(osd->od_os, 0, flags | ZAP_FLAG_HASH64,
1763                                    DMU_OT_DIRECTORY_CONTENTS,
1764                                    14, /* == ZFS fzap_default_blockshift */
1765                                    DN_MAX_INDBLKSHIFT, /* indirect blockshift */
1766                                    dnsize, tx);
1767
1768         la->la_size = 2;
1769         la->la_nlink = 1;
1770
1771         return osd_find_new_dnode(env, tx, oid, dnp);
1772 }
1773
1774 static dnode_t *osd_mkidx(const struct lu_env *env, struct osd_object *obj,
1775                           struct lu_attr *la, struct osd_thandle *oh)
1776 {
1777         struct osd_device *osd = osd_obj2dev(obj);
1778         dnode_t *dn;
1779         int rc;
1780
1781         /* Index file should be created as regular file in order not to confuse
1782          * ZPL which could interpret them as directory.
1783          * We set ZAP_FLAG_UINT64_KEY to let ZFS know than we are going to use
1784          * binary keys */
1785         LASSERT(S_ISREG(la->la_mode));
1786         rc = __osd_zap_create(env, osd, &dn, oh->ot_tx, la,
1787                 osd_find_dnsize(osd, obj->oo_ea_in_bonus), ZAP_FLAG_UINT64_KEY);
1788         if (rc)
1789                 return ERR_PTR(rc);
1790         return dn;
1791 }
1792
1793 static dnode_t *osd_mkdir(const struct lu_env *env, struct osd_object *obj,
1794                           struct lu_attr *la, struct osd_thandle *oh)
1795 {
1796         struct osd_device *osd = osd_obj2dev(obj);
1797         dnode_t *dn;
1798         int rc;
1799
1800         LASSERT(S_ISDIR(la->la_mode));
1801         rc = __osd_zap_create(env, osd, &dn, oh->ot_tx, la,
1802                               osd_find_dnsize(osd, obj->oo_ea_in_bonus), 0);
1803         if (rc)
1804                 return ERR_PTR(rc);
1805         return dn;
1806 }
1807
1808 static dnode_t *osd_mkreg(const struct lu_env *env, struct osd_object *obj,
1809                           struct lu_attr *la, struct osd_thandle *oh)
1810 {
1811         const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
1812         struct osd_device *osd = osd_obj2dev(obj);
1813         dnode_t *dn;
1814         int rc;
1815
1816         LASSERT(S_ISREG(la->la_mode));
1817         rc = __osd_object_create(env, osd, obj, fid, &dn, oh->ot_tx, la);
1818         if (rc)
1819                 return ERR_PTR(rc);
1820
1821         if ((fid_is_idif(fid) || fid_is_norm(fid) || fid_is_echo(fid))) {
1822                 /* The minimum block size must be at least page size otherwise
1823                  * it will break the assumption in tgt_thread_big_cache where
1824                  * the array size is PTLRPC_MAX_BRW_PAGES. It will also affect
1825                  * RDMA due to subpage transfer size */
1826                 rc = -dmu_object_set_blocksize(osd->od_os, dn->dn_object,
1827                                                PAGE_SIZE, 0, oh->ot_tx);
1828                 if (unlikely(rc)) {
1829                         CERROR("%s: can't change blocksize: %d\n",
1830                                osd->od_svname, rc);
1831                         return ERR_PTR(rc);
1832                 }
1833         } else if ((fid_is_llog(fid))) {
1834                 rc = -dmu_object_set_blocksize(osd->od_os, dn->dn_object,
1835                                                LLOG_MIN_CHUNK_SIZE, 0, oh->ot_tx);
1836                 if (unlikely(rc)) {
1837                         CERROR("%s: can't change blocksize: %d\n",
1838                                osd->od_svname, rc);
1839                         return ERR_PTR(rc);
1840                 }
1841         }
1842
1843         return dn;
1844 }
1845
1846 static dnode_t *osd_mksym(const struct lu_env *env, struct osd_object *obj,
1847                           struct lu_attr *la, struct osd_thandle *oh)
1848 {
1849         dnode_t *dn;
1850         int rc;
1851
1852         LASSERT(S_ISLNK(la->la_mode));
1853         rc = __osd_object_create(env, osd_obj2dev(obj), obj,
1854                                  lu_object_fid(&obj->oo_dt.do_lu),
1855                                  &dn, oh->ot_tx, la);
1856         if (rc)
1857                 return ERR_PTR(rc);
1858         return dn;
1859 }
1860
1861 static dnode_t *osd_mknod(const struct lu_env *env, struct osd_object *obj,
1862                           struct lu_attr *la, struct osd_thandle *oh)
1863 {
1864         dnode_t *dn;
1865         int rc;
1866
1867         if (S_ISCHR(la->la_mode) || S_ISBLK(la->la_mode))
1868                 la->la_valid |= LA_RDEV;
1869
1870         rc = __osd_object_create(env, osd_obj2dev(obj), obj,
1871                                  lu_object_fid(&obj->oo_dt.do_lu),
1872                                  &dn, oh->ot_tx, la);
1873         if (rc)
1874                 return ERR_PTR(rc);
1875         return dn;
1876 }
1877
1878 typedef dnode_t *(*osd_obj_type_f)(const struct lu_env *env,
1879                                    struct osd_object *obj,
1880                                    struct lu_attr *la,
1881                                    struct osd_thandle *oh);
1882
1883 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
1884 {
1885         osd_obj_type_f result;
1886
1887         switch (type) {
1888         case DFT_DIR:
1889                 result = osd_mkdir;
1890                 break;
1891         case DFT_INDEX:
1892                 result = osd_mkidx;
1893                 break;
1894         case DFT_REGULAR:
1895                 result = osd_mkreg;
1896                 break;
1897         case DFT_SYM:
1898                 result = osd_mksym;
1899                 break;
1900         case DFT_NODE:
1901                 result = osd_mknod;
1902                 break;
1903         default:
1904                 LBUG();
1905                 break;
1906         }
1907         return result;
1908 }
1909
1910 /*
1911  * Concurrency: @dt is write locked.
1912  */
1913 static int osd_create(const struct lu_env *env, struct dt_object *dt,
1914                       struct lu_attr *attr, struct dt_allocation_hint *hint,
1915                       struct dt_object_format *dof, struct thandle *th)
1916 {
1917         struct osd_thread_info  *info = osd_oti_get(env);
1918         struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
1919         struct zpl_direntry     *zde = &info->oti_zde.lzd_reg;
1920         const struct lu_fid     *fid = lu_object_fid(&dt->do_lu);
1921         struct osd_object       *obj = osd_dt_obj(dt);
1922         struct osd_device       *osd = osd_obj2dev(obj);
1923         char                    *buf = info->oti_str;
1924         struct osd_thandle      *oh;
1925         dnode_t *dn = NULL, *zdn = NULL;
1926         uint64_t                 zapid, parent = 0;
1927         int                      rc;
1928         __u32 compat = 0;
1929
1930         ENTRY;
1931
1932         LASSERT(!fid_is_acct(fid));
1933
1934         /* concurrent create declarations should not see
1935          * the object inconsistent (db, attr, etc).
1936          * in regular cases acquisition should be cheap */
1937         down_write(&obj->oo_guard);
1938
1939         if (unlikely(dt_object_exists(dt)))
1940                 GOTO(out, rc = -EEXIST);
1941
1942         LASSERT(osd_invariant(obj));
1943         LASSERT(dof != NULL);
1944
1945         LASSERT(th != NULL);
1946         oh = container_of(th, struct osd_thandle, ot_super);
1947
1948         LASSERT(obj->oo_dn == NULL);
1949
1950         /* to follow ZFS on-disk format we need
1951          * to initialize parent dnode properly */
1952         if (hint != NULL && hint->dah_parent != NULL &&
1953             !dt_object_remote(hint->dah_parent))
1954                 parent = osd_dt_obj(hint->dah_parent)->oo_dn->dn_object;
1955
1956         /* we may fix some attributes, better do not change the source */
1957         obj->oo_attr = *attr;
1958         obj->oo_attr.la_size = 0;
1959         obj->oo_attr.la_nlink = 0;
1960         obj->oo_attr.la_valid |= LA_SIZE | LA_NLINK | LA_TYPE;
1961
1962 #ifdef ZFS_PROJINHERIT
1963         if (osd->od_projectused_dn) {
1964                 if (!(obj->oo_attr.la_valid & LA_PROJID))
1965                         obj->oo_attr.la_projid = ZFS_DEFAULT_PROJID;
1966                 obj->oo_with_projid = 1;
1967         }
1968 #endif
1969
1970         dn = osd_create_type_f(dof->dof_type)(env, obj, &obj->oo_attr, oh);
1971         if (IS_ERR(dn)) {
1972                 rc = PTR_ERR(dn);
1973                 dn = NULL;
1974                 GOTO(out, rc);
1975         }
1976
1977         zde->zde_pad = 0;
1978         zde->zde_dnode = dn->dn_object;
1979         zde->zde_type = S_DT(attr->la_mode & S_IFMT);
1980
1981         zapid = osd_get_name_n_idx(env, osd, fid, buf,
1982                                    sizeof(info->oti_str), &zdn);
1983         if (CFS_FAIL_CHECK(OBD_FAIL_OSD_NO_OI_ENTRY) ||
1984             (osd->od_is_ost && OBD_FAIL_CHECK(OBD_FAIL_OSD_COMPAT_NO_ENTRY)))
1985                 goto skip_add;
1986
1987         if (osd->od_is_ost && OBD_FAIL_CHECK(OBD_FAIL_OSD_COMPAT_INVALID_ENTRY))
1988                 zde->zde_dnode++;
1989
1990         rc = osd_zap_add(osd, zapid, zdn, buf, 8, 1, zde, oh->ot_tx);
1991         if (rc)
1992                 GOTO(out, rc);
1993
1994 skip_add:
1995         obj->oo_dn = dn;
1996         /* Now add in all of the "SA" attributes */
1997         rc = osd_sa_handle_get(obj);
1998         if (rc)
1999                 GOTO(out, rc);
2000
2001         rc = -nvlist_alloc(&obj->oo_sa_xattr, NV_UNIQUE_NAME, KM_SLEEP);
2002         if (rc)
2003                 GOTO(out, rc);
2004
2005         /* initialize LMA */
2006         if (fid_is_idif(fid) || (fid_is_norm(fid) && osd->od_is_ost))
2007                 compat |= LMAC_FID_ON_OST;
2008         lustre_lma_init(lma, fid, compat, 0);
2009         lustre_lma_swab(lma);
2010         rc = -nvlist_add_byte_array(obj->oo_sa_xattr, XATTR_NAME_LMA,
2011                                     (uchar_t *)lma, sizeof(*lma));
2012         if (rc)
2013                 GOTO(out, rc);
2014
2015         /* configure new osd object */
2016         obj->oo_parent = parent != 0 ? parent : zapid;
2017         obj->oo_late_attr_set = 1;
2018         rc = __osd_sa_xattr_schedule_update(env, obj, oh);
2019         if (rc)
2020                 GOTO(out, rc);
2021
2022         /* XXX: oo_lma_flags */
2023         obj->oo_dt.do_lu.lo_header->loh_attr |= obj->oo_attr.la_mode & S_IFMT;
2024         if (likely(!fid_is_acct(lu_object_fid(&obj->oo_dt.do_lu))))
2025                 /* no body operations for accounting objects */
2026                 obj->oo_dt.do_body_ops = &osd_body_ops;
2027
2028         osd_idc_find_and_init(env, osd, obj);
2029
2030 out:
2031         if (unlikely(rc && dn)) {
2032                 dmu_object_free(osd->od_os, dn->dn_object, oh->ot_tx);
2033                 osd_dnode_rele(dn);
2034                 obj->oo_dn = NULL;
2035         } else if (!rc) {
2036                 obj->oo_dt.do_lu.lo_header->loh_attr |= LOHA_EXISTS;
2037         }
2038         up_write(&obj->oo_guard);
2039         RETURN(rc);
2040 }
2041
2042 static int osd_declare_ref_add(const struct lu_env *env, struct dt_object *dt,
2043                                struct thandle *th)
2044 {
2045         osd_idc_find_and_init(env, osd_dev(dt->do_lu.lo_dev), osd_dt_obj(dt));
2046         return osd_declare_attr_set(env, dt, NULL, th);
2047 }
2048
2049 /*
2050  * Concurrency: @dt is write locked.
2051  */
2052 static int osd_ref_add(const struct lu_env *env, struct dt_object *dt,
2053                        struct thandle *handle)
2054 {
2055         struct osd_object       *obj = osd_dt_obj(dt);
2056         struct osd_thandle      *oh;
2057         struct osd_device       *osd = osd_obj2dev(obj);
2058         uint64_t                 nlink;
2059         int rc;
2060
2061         ENTRY;
2062
2063         down_read(&obj->oo_guard);
2064         if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
2065                 GOTO(out, rc = -ENOENT);
2066
2067         LASSERT(osd_invariant(obj));
2068         LASSERT(obj->oo_sa_hdl != NULL);
2069
2070         oh = container_of(handle, struct osd_thandle, ot_super);
2071
2072         write_lock(&obj->oo_attr_lock);
2073         nlink = ++obj->oo_attr.la_nlink;
2074         write_unlock(&obj->oo_attr_lock);
2075
2076         rc = osd_object_sa_update(obj, SA_ZPL_LINKS(osd), &nlink, 8, oh);
2077
2078 out:
2079         up_read(&obj->oo_guard);
2080         RETURN(rc);
2081 }
2082
2083 static int osd_declare_ref_del(const struct lu_env *env, struct dt_object *dt,
2084                                struct thandle *handle)
2085 {
2086         osd_idc_find_and_init(env, osd_dev(dt->do_lu.lo_dev), osd_dt_obj(dt));
2087         return osd_declare_attr_set(env, dt, NULL, handle);
2088 }
2089
2090 /*
2091  * Concurrency: @dt is write locked.
2092  */
2093 static int osd_ref_del(const struct lu_env *env, struct dt_object *dt,
2094                        struct thandle *handle)
2095 {
2096         struct osd_object       *obj = osd_dt_obj(dt);
2097         struct osd_thandle      *oh;
2098         struct osd_device       *osd = osd_obj2dev(obj);
2099         uint64_t                 nlink;
2100         int                      rc;
2101
2102         ENTRY;
2103
2104         down_read(&obj->oo_guard);
2105
2106         if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
2107                 GOTO(out, rc = -ENOENT);
2108
2109         LASSERT(osd_invariant(obj));
2110         LASSERT(obj->oo_sa_hdl != NULL);
2111
2112         oh = container_of(handle, struct osd_thandle, ot_super);
2113         LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
2114
2115         write_lock(&obj->oo_attr_lock);
2116         nlink = --obj->oo_attr.la_nlink;
2117         write_unlock(&obj->oo_attr_lock);
2118
2119         rc = osd_object_sa_update(obj, SA_ZPL_LINKS(osd), &nlink, 8, oh);
2120
2121 out:
2122         up_read(&obj->oo_guard);
2123         RETURN(rc);
2124 }
2125
2126 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
2127                            __u64 start, __u64 end)
2128 {
2129         struct osd_device *osd = osd_obj2dev(osd_dt_obj(dt));
2130         uint64_t txg = 0;
2131         ENTRY;
2132
2133         if (osd->od_dt_dev.dd_rdonly)
2134                 RETURN(0);
2135
2136         txg = osd_db_dirty_txg(osd_dt_obj(dt)->oo_dn->dn_dbuf);
2137         if (txg) {
2138                 /* the object is dirty or being synced */
2139                 if (osd_object_sync_delay_us < 0)
2140                         txg_wait_synced(dmu_objset_pool(osd->od_os), txg);
2141                 else
2142                         udelay(osd_object_sync_delay_us);
2143         }
2144
2145         RETURN(0);
2146 }
2147
2148 static int osd_invalidate(const struct lu_env *env, struct dt_object *dt)
2149 {
2150         return 0;
2151 }
2152
2153 static bool osd_check_stale(struct dt_object *dt)
2154 {
2155         return false;
2156 }
2157
2158 static const struct dt_object_operations osd_obj_ops = {
2159         .do_read_lock           = osd_read_lock,
2160         .do_write_lock          = osd_write_lock,
2161         .do_read_unlock         = osd_read_unlock,
2162         .do_write_unlock        = osd_write_unlock,
2163         .do_write_locked        = osd_write_locked,
2164         .do_attr_get            = osd_attr_get,
2165         .do_declare_attr_set    = osd_declare_attr_set,
2166         .do_attr_set            = osd_attr_set,
2167         .do_ah_init             = osd_ah_init,
2168         .do_declare_create      = osd_declare_create,
2169         .do_create              = osd_create,
2170         .do_declare_destroy     = osd_declare_destroy,
2171         .do_destroy             = osd_destroy,
2172         .do_index_try           = osd_index_try,
2173         .do_declare_ref_add     = osd_declare_ref_add,
2174         .do_ref_add             = osd_ref_add,
2175         .do_declare_ref_del     = osd_declare_ref_del,
2176         .do_ref_del             = osd_ref_del,
2177         .do_xattr_get           = osd_xattr_get,
2178         .do_declare_xattr_set   = osd_declare_xattr_set,
2179         .do_xattr_set           = osd_xattr_set,
2180         .do_declare_xattr_del   = osd_declare_xattr_del,
2181         .do_xattr_del           = osd_xattr_del,
2182         .do_xattr_list          = osd_xattr_list,
2183         .do_object_sync         = osd_object_sync,
2184         .do_invalidate          = osd_invalidate,
2185         .do_check_stale         = osd_check_stale,
2186 };
2187
2188 static const struct lu_object_operations osd_lu_obj_ops = {
2189         .loo_object_init        = osd_object_init,
2190         .loo_object_delete      = osd_object_delete,
2191         .loo_object_release     = osd_object_release,
2192         .loo_object_free        = osd_object_free,
2193         .loo_object_print       = osd_object_print,
2194         .loo_object_invariant   = osd_object_invariant,
2195 };
2196
2197 static int osd_otable_it_attr_get(const struct lu_env *env,
2198                                 struct dt_object *dt,
2199                                 struct lu_attr *attr)
2200 {
2201         attr->la_valid = 0;
2202         return 0;
2203 }
2204
2205 static const struct dt_object_operations osd_obj_otable_it_ops = {
2206         .do_attr_get            = osd_otable_it_attr_get,
2207         .do_index_try           = osd_index_try,
2208 };
2209
2210 module_param(osd_object_sync_delay_us, int, 0644);
2211 MODULE_PARM_DESC(osd_object_sync_delay_us,
2212                  "If zero or larger delay N usec instead of doing object sync");