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  *
  31  * lustre/osd-zfs/osd_object.c
  32  *
  33  * Author: Alex Zhuravlev <bzzz@whamcloud.com>
  34  * Author: Mike Pershin <tappro@whamcloud.com>
  35  * Author: Johann Lombardi <johann@whamcloud.com>
  36  */
  37
  38 #define DEBUG_SUBSYSTEM S_OSD
  39
  40 #include <libcfs/libcfs.h>
  41 #include <obd_support.h>
  42 #include <lustre_net.h>
  43 #include <obd.h>
  44 #include <obd_class.h>
  45 #include <lustre_disk.h>
  46 #include <lustre_fid.h>
  47
  48 #include "osd_internal.h"
  49
  50 #include <sys/dnode.h>
  51 #include <sys/dbuf.h>
  52 #include <sys/spa.h>
  53 #include <sys/stat.h>
  54 #include <sys/zap.h>
  55 #include <sys/spa_impl.h>
  56 #include <sys/zfs_znode.h>
  57 #include <sys/dmu_tx.h>
  58 #include <sys/dmu_objset.h>
  59 #include <sys/dsl_prop.h>
  60 #include <sys/sa_impl.h>
  61 #include <sys/txg.h>
  62
  63 char *osd_obj_tag = "osd_object";
  64 static int osd_object_sync_delay_us = -1;
  65
  66 static const struct dt_object_operations osd_obj_ops;
  67 static const struct lu_object_operations osd_lu_obj_ops;
  68 static const struct dt_object_operations osd_obj_otable_it_ops;
  69
  70 static void
  71 osd_object_sa_fini(struct osd_object *obj)
  72 {
  73         if (obj->oo_sa_hdl) {
  74                 sa_handle_destroy(obj->oo_sa_hdl);
  75                 obj->oo_sa_hdl = NULL;
  76         }
  77 }
  78
  79 static int
  80 osd_object_sa_init(struct osd_object *obj, struct osd_device *o)
  81 {
  82         int rc;
  83
  84         LASSERT(obj->oo_sa_hdl == NULL);
  85         LASSERT(obj->oo_dn != NULL);
  86
  87         rc = osd_sa_handle_get(obj);
  88         if (rc)
  89                 return rc;
  90
  91         /* Cache the xattr object id, valid for the life of the object */
  92         rc = -sa_lookup(obj->oo_sa_hdl, SA_ZPL_XATTR(o), &obj->oo_xattr, 8);
  93         if (rc == -ENOENT) {
  94                 obj->oo_xattr = ZFS_NO_OBJECT;
  95                 rc = 0;
  96         } else if (rc) {
  97                 osd_object_sa_fini(obj);
  98         }
  99
 100         return rc;
 101 }
 102
 103 /*
 104  * Add object to list of dirty objects in tx handle.
 105  */
 106 void osd_object_sa_dirty_add(struct osd_object *obj, struct osd_thandle *oh)
 107 {
 108         if (!list_empty(&obj->oo_sa_linkage))
 109                 return;
 110
 111         write_lock(&obj->oo_attr_lock);
 112         if (likely(list_empty(&obj->oo_sa_linkage)))
 113                 list_add(&obj->oo_sa_linkage, &oh->ot_sa_list);
 114         write_unlock(&obj->oo_attr_lock);
 115 }
 116
 117 /*
 118  * Release spill block dbuf hold for all dirty SAs.
 119  */
 120 void osd_object_sa_dirty_rele(const struct lu_env *env, struct osd_thandle *oh)
 121 {
 122         struct osd_object *obj;
 123
 124         while (!list_empty(&oh->ot_sa_list)) {
 125                 obj = list_entry(oh->ot_sa_list.next,
 126                                  struct osd_object, oo_sa_linkage);
 127                 write_lock(&obj->oo_attr_lock);
 128                 list_del_init(&obj->oo_sa_linkage);
 129                 write_unlock(&obj->oo_attr_lock);
 130                 if (obj->oo_late_xattr && obj->oo_destroyed == 0) {
 131                         /*
 132                          * take oo_guard to protect oo_sa_xattr buffer
 133                          * from concurrent update by osd_xattr_set()
 134                          */
 135                         LASSERT(oh->ot_assigned != 0);
 136                         down_write(&obj->oo_guard);
 137                         if (obj->oo_destroyed == 0) {
 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                         }
 143                         up_write(&obj->oo_guard);
 144                 }
 145                 sa_spill_rele(obj->oo_sa_hdl);
 146         }
 147 }
 148
 149 /*
 150  * Update the SA and add the object to the dirty list.
 151  */
 152 int osd_object_sa_update(struct osd_object *obj, sa_attr_type_t type,
 153                          void *buf, uint32_t buflen, struct osd_thandle *oh)
 154 {
 155         int rc;
 156
 157         LASSERT(obj->oo_sa_hdl != NULL);
 158         LASSERT(oh->ot_tx != NULL);
 159
 160         rc = -sa_update(obj->oo_sa_hdl, type, buf, buflen, oh->ot_tx);
 161         osd_object_sa_dirty_add(obj, oh);
 162
 163         return rc;
 164 }
 165
 166 /*
 167  * Bulk update the SA and add the object to the dirty list.
 168  */
 169 static int
 170 osd_object_sa_bulk_update(struct osd_object *obj, sa_bulk_attr_t *attrs,
 171                           int count, struct osd_thandle *oh)
 172 {
 173         int rc;
 174
 175         LASSERT(obj->oo_sa_hdl != NULL);
 176         LASSERT(oh->ot_tx != NULL);
 177
 178         rc = -sa_bulk_update(obj->oo_sa_hdl, attrs, count, oh->ot_tx);
 179         osd_object_sa_dirty_add(obj, oh);
 180
 181         return rc;
 182 }
 183
 184 /*
 185  * Retrieve the attributes of a DMU object
 186  */
 187 static int __osd_object_attr_get(const struct lu_env *env, struct osd_device *o,
 188                                  struct osd_object *obj, struct lu_attr *la)
 189 {
 190         struct osa_attr *osa = &osd_oti_get(env)->oti_osa;
 191         sa_bulk_attr_t *bulk = osd_oti_get(env)->oti_attr_bulk;
 192         struct lustre_mdt_attrs *lma;
 193         struct lu_buf buf;
 194         int cnt = 0;
 195         int              rc;
 196         ENTRY;
 197
 198         LASSERT(obj->oo_dn != NULL);
 199
 200         la->la_valid |= LA_ATIME | LA_MTIME | LA_CTIME | LA_BTIME | LA_MODE |
 201                         LA_TYPE | LA_SIZE | LA_UID | LA_GID | LA_FLAGS |
 202                         LA_NLINK;
 203
 204         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(o), NULL, osa->atime, 16);
 205         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(o), NULL, osa->mtime, 16);
 206         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(o), NULL, osa->ctime, 16);
 207         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CRTIME(o), NULL, osa->btime, 16);
 208         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(o), NULL, &osa->mode, 8);
 209         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(o), NULL, &osa->size, 8);
 210         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(o), NULL, &osa->nlink, 8);
 211         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(o), NULL, &osa->uid, 8);
 212         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(o), NULL, &osa->gid, 8);
 213         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(o), NULL, &osa->flags, 8);
 214         LASSERT(cnt <= ARRAY_SIZE(osd_oti_get(env)->oti_attr_bulk));
 215
 216         rc = -sa_bulk_lookup(obj->oo_sa_hdl, bulk, cnt);
 217         if (rc)
 218                 GOTO(out_sa, rc);
 219
 220 #ifdef ZFS_PROJINHERIT
 221         if (o->od_projectused_dn && osa->flags & ZFS_PROJID) {
 222                 rc = -sa_lookup(obj->oo_sa_hdl, SA_ZPL_PROJID(o),
 223                                 &osa->projid, 8);
 224                 if (rc)
 225                         GOTO(out_sa, rc);
 226
 227                 la->la_projid = osa->projid;
 228                 la->la_valid |= LA_PROJID;
 229                 obj->oo_with_projid = 1;
 230         } else {
 231                 la->la_projid = ZFS_DEFAULT_PROJID;
 232                 la->la_valid &= ~LA_PROJID;
 233         }
 234 #else
 235         la->la_projid = 0;
 236         la->la_valid &= ~LA_PROJID;
 237 #endif
 238
 239         la->la_atime = osa->atime[0];
 240         la->la_mtime = osa->mtime[0];
 241         la->la_ctime = osa->ctime[0];
 242         la->la_btime = osa->btime[0];
 243         la->la_mode = osa->mode;
 244         la->la_uid = osa->uid;
 245         la->la_gid = osa->gid;
 246         la->la_nlink = osa->nlink;
 247         la->la_flags = attrs_zfs2fs(osa->flags);
 248         la->la_size = osa->size;
 249
 250         /* Try to get extra flags from LMA */
 251         lma = (struct lustre_mdt_attrs *)osd_oti_get(env)->oti_buf;
 252         buf.lb_buf = lma;
 253         buf.lb_len = sizeof(osd_oti_get(env)->oti_buf);
 254         down_read(&obj->oo_guard);
 255         rc = osd_xattr_get_lma(env, obj, &buf);
 256         if (!rc) {
 257                 lma->lma_incompat = le32_to_cpu(lma->lma_incompat);
 258                 obj->oo_lma_flags =
 259                         lma_to_lustre_flags(lma->lma_incompat);
 260         } else if (rc == -ENODATA ||
 261                    !(S_ISDIR(la->la_mode) &&
 262                      dt_object_exists(&obj->oo_dt))) {
 263                 rc = 0;
 264         }
 265         up_read(&obj->oo_guard);
 266
 267         if (S_ISCHR(la->la_mode) || S_ISBLK(la->la_mode)) {
 268                 rc = -sa_lookup(obj->oo_sa_hdl, SA_ZPL_RDEV(o), &osa->rdev, 8);
 269                 if (rc)
 270                         GOTO(out_sa, rc);
 271                 la->la_rdev = osa->rdev;
 272                 la->la_valid |= LA_RDEV;
 273         }
 274 out_sa:
 275
 276         RETURN(rc);
 277 }
 278
 279 int __osd_obj2dnode(objset_t *os, uint64_t oid, dnode_t **dnp)
 280 {
 281         dmu_buf_t *db;
 282         dmu_buf_impl_t *dbi;
 283         int rc;
 284
 285         rc = -dmu_bonus_hold(os, oid, osd_obj_tag, &db);
 286         if (rc)
 287                 return rc;
 288
 289         dbi = (dmu_buf_impl_t *)db;
 290         DB_DNODE_ENTER(dbi);
 291         *dnp = DB_DNODE(dbi);
 292         DB_DNODE_EXIT(dbi);
 293         LASSERT(*dnp != NULL);
 294
 295         return 0;
 296 }
 297
 298 /*
 299  * Concurrency: no concurrent access is possible that early in object
 300  * life-cycle.
 301  */
 302 struct lu_object *osd_object_alloc(const struct lu_env *env,
 303                                    const struct lu_object_header *hdr,
 304                                    struct lu_device *d)
 305 {
 306         struct osd_object *mo;
 307
 308         OBD_SLAB_ALLOC_PTR_GFP(mo, osd_object_kmem, GFP_NOFS);
 309         if (mo != NULL) {
 310                 struct lu_object *l;
 311                 struct lu_object_header *h;
 312                 struct osd_device *o = osd_dev(d);
 313
 314                 l = &mo->oo_dt.do_lu;
 315                 if (unlikely(o->od_in_init)) {
 316                         OBD_ALLOC_PTR(h);
 317                         if (!h) {
 318                                 OBD_FREE_PTR(mo);
 319                                 return NULL;
 320                         }
 321
 322                         lu_object_header_init(h);
 323                         lu_object_init(l, h, d);
 324                         lu_object_add_top(h, l);
 325                         mo->oo_header = h;
 326                 } else {
 327                         dt_object_init(&mo->oo_dt, NULL, d);
 328                         mo->oo_header = NULL;
 329                 }
 330
 331                 mo->oo_dt.do_ops = &osd_obj_ops;
 332                 l->lo_ops = &osd_lu_obj_ops;
 333                 INIT_LIST_HEAD(&mo->oo_sa_linkage);
 334                 INIT_LIST_HEAD(&mo->oo_unlinked_linkage);
 335                 init_rwsem(&mo->oo_sem);
 336                 init_rwsem(&mo->oo_guard);
 337                 rwlock_init(&mo->oo_attr_lock);
 338                 mo->oo_destroy = OSD_DESTROY_NONE;
 339                 return l;
 340         } else {
 341                 return NULL;
 342         }
 343 }
 344
 345 static void osd_obj_set_blksize(const struct lu_env *env,
 346                                 struct osd_device *osd, struct osd_object *obj)
 347 {
 348         const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
 349         dmu_tx_t *tx;
 350         dnode_t *dn = obj->oo_dn;
 351         uint32_t blksz;
 352         int rc = 0;
 353         ENTRY;
 354
 355         LASSERT(!osd_oti_get(env)->oti_in_trans);
 356
 357         tx = dmu_tx_create(osd->od_os);
 358         if (!tx) {
 359                 CERROR("%s: fail to create tx to set blksize for "DFID"\n",
 360                        osd->od_svname, PFID(fid));
 361                 RETURN_EXIT;
 362         }
 363
 364         dmu_tx_hold_bonus(tx, dn->dn_object);
 365         rc = -dmu_tx_assign(tx, TXG_WAIT);
 366         if (rc) {
 367                 dmu_tx_abort(tx);
 368                 CERROR("%s: fail to assign tx to set blksize for "DFID
 369                        ": rc = %d\n", osd->od_svname, PFID(fid), rc);
 370                 RETURN_EXIT;
 371         }
 372
 373         down_write(&obj->oo_guard);
 374         if (unlikely((1 << dn->dn_datablkshift) >= PAGE_SIZE))
 375                 GOTO(out, rc = 1);
 376
 377         blksz = dn->dn_datablksz;
 378         if (!is_power_of_2(blksz))
 379                 blksz = size_roundup_power2(blksz);
 380
 381         if (blksz > osd->od_max_blksz)
 382                 blksz = osd->od_max_blksz;
 383         else if (blksz < PAGE_SIZE)
 384                 blksz = PAGE_SIZE;
 385         rc = -dmu_object_set_blocksize(osd->od_os, dn->dn_object, blksz, 0, tx);
 386
 387         GOTO(out, rc);
 388
 389 out:
 390         up_write(&obj->oo_guard);
 391         if (rc) {
 392                 dmu_tx_abort(tx);
 393                 if (unlikely(obj->oo_dn->dn_maxblkid > 0))
 394                         rc = 1;
 395                 if (rc < 0)
 396                         CERROR("%s: fail to set blksize for "DFID": rc = %d\n",
 397                                osd->od_svname, PFID(fid), rc);
 398         } else {
 399                 dmu_tx_commit(tx);
 400                 CDEBUG(D_INODE, "%s: set blksize as %u for "DFID"\n",
 401                        osd->od_svname, blksz, PFID(fid));
 402         }
 403 }
 404
 405 /*
 406  * Concurrency: shouldn't matter.
 407  */
 408 static int osd_object_init0(const struct lu_env *env, struct osd_object *obj)
 409 {
 410         struct osd_device       *osd = osd_obj2dev(obj);
 411         const struct lu_fid     *fid = lu_object_fid(&obj->oo_dt.do_lu);
 412         int                      rc = 0;
 413         ENTRY;
 414
 415         LASSERT(obj->oo_dn);
 416
 417         rc = osd_object_sa_init(obj, osd);
 418         if (rc)
 419                 RETURN(rc);
 420
 421         /* cache attrs in object */
 422         rc = __osd_object_attr_get(env, osd, obj, &obj->oo_attr);
 423         if (rc)
 424                 RETURN(rc);
 425
 426         if (likely(!fid_is_acct(fid))) {
 427                 /* no body operations for accounting objects */
 428                 obj->oo_dt.do_body_ops = &osd_body_ops;
 429
 430                 if (S_ISREG(obj->oo_attr.la_mode) &&
 431                     obj->oo_dn->dn_maxblkid == 0 &&
 432                     (1 << obj->oo_dn->dn_datablkshift) < PAGE_SIZE &&
 433                     (fid_is_idif(fid) || fid_is_norm(fid) ||
 434                      fid_is_echo(fid)) &&
 435                     osd->od_is_ost && !osd->od_dt_dev.dd_rdonly)
 436                         osd_obj_set_blksize(env, osd, obj);
 437         }
 438
 439         /*
 440          * initialize object before marking it existing
 441          */
 442         obj->oo_dt.do_lu.lo_header->loh_attr |= obj->oo_attr.la_mode & S_IFMT;
 443
 444         smp_mb();
 445         obj->oo_dt.do_lu.lo_header->loh_attr |= LOHA_EXISTS;
 446
 447         RETURN(0);
 448 }
 449
 450 static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
 451 {
 452         struct osd_thread_info  *info = osd_oti_get(env);
 453         struct lu_buf           buf;
 454         int                     rc;
 455         struct lustre_mdt_attrs *lma;
 456         const struct lu_fid *rfid = lu_object_fid(&obj->oo_dt.do_lu);
 457         ENTRY;
 458
 459         BUILD_BUG_ON(sizeof(info->oti_buf) < sizeof(*lma));
 460         lma = (struct lustre_mdt_attrs *)info->oti_buf;
 461         buf.lb_buf = lma;
 462         buf.lb_len = sizeof(info->oti_buf);
 463
 464         rc = osd_xattr_get(env, &obj->oo_dt, &buf, XATTR_NAME_LMA);
 465         if (rc > 0) {
 466                 rc = 0;
 467                 lustre_lma_swab(lma);
 468                 if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
 469                              CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT))) {
 470                         CWARN("%s: unsupported incompat LMA feature(s) %#x for "
 471                               "fid = "DFID"\n", osd_obj2dev(obj)->od_svname,
 472                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
 473                               PFID(rfid));
 474                         rc = -EOPNOTSUPP;
 475                 } else if (unlikely(!lu_fid_eq(rfid, &lma->lma_self_fid))) {
 476                         CERROR("%s: FID-in-LMA "DFID" does not match the "
 477                               "object self-fid "DFID"\n",
 478                               osd_obj2dev(obj)->od_svname,
 479                               PFID(&lma->lma_self_fid), PFID(rfid));
 480                         rc = -EREMCHG;
 481                 } else {
 482                         struct osd_device *osd = osd_obj2dev(obj);
 483
 484                         if (lma->lma_compat & LMAC_STRIPE_INFO &&
 485                             osd->od_is_ost)
 486                                 obj->oo_pfid_in_lma = 1;
 487                         if (unlikely(lma->lma_incompat & LMAI_REMOTE_PARENT) &&
 488                             osd->od_remote_parent_dir != ZFS_NO_OBJECT)
 489                                 lu_object_set_agent_entry(&obj->oo_dt.do_lu);
 490                 }
 491         } else if (rc == -ENODATA) {
 492                 /* haven't initialize LMA xattr */
 493                 rc = 0;
 494         }
 495
 496         RETURN(rc);
 497 }
 498
 499 /**
 500  * Helper function to retrieve DMU object id from fid for accounting object
 501  */
 502 static dnode_t *osd_quota_fid2dmu(const struct osd_device *osd,
 503                                   const struct lu_fid *fid)
 504 {
 505         dnode_t *dn = NULL;
 506
 507         LASSERT(fid_is_acct(fid));
 508
 509         switch (fid_oid(fid)) {
 510         case ACCT_USER_OID:
 511                 dn = osd->od_userused_dn;
 512                 break;
 513         case ACCT_GROUP_OID:
 514                 dn = osd->od_groupused_dn;
 515                 break;
 516 #ifdef ZFS_PROJINHERIT
 517         case ACCT_PROJECT_OID:
 518                 dn = osd->od_projectused_dn;
 519                 break;
 520 #endif
 521         default:
 522                 break;
 523         }
 524
 525         return dn;
 526 }
 527
 528 /*
 529  * Concurrency: no concurrent access is possible that early in object
 530  * life-cycle.
 531  */
 532 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
 533                            const struct lu_object_conf *conf)
 534 {
 535         struct osd_object *obj = osd_obj(l);
 536         struct osd_device *osd = osd_obj2dev(obj);
 537         const struct lu_fid *fid = lu_object_fid(l);
 538         struct lustre_scrub *scrub = &osd->od_scrub;
 539         struct osd_thread_info *info = osd_oti_get(env);
 540         struct luz_direntry *zde = &info->oti_zde;
 541         struct osd_idmap_cache *idc;
 542         char *name = info->oti_str;
 543         uint64_t oid;
 544         int rc = 0;
 545         int rc1;
 546         bool remote = false;
 547         ENTRY;
 548
 549         LASSERT(osd_invariant(obj));
 550
 551         if (fid_is_otable_it(&l->lo_header->loh_fid)) {
 552                 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
 553                 l->lo_header->loh_attr |= LOHA_EXISTS;
 554
 555                 GOTO(out, rc = 0);
 556         }
 557
 558         if (conf && conf->loc_flags & LOC_F_NEW)
 559                 GOTO(out, rc = 0);
 560
 561         if (unlikely(fid_is_acct(fid))) {
 562                 obj->oo_dn = osd_quota_fid2dmu(osd, fid);
 563                 if (obj->oo_dn) {
 564                         obj->oo_dt.do_index_ops = &osd_acct_index_ops;
 565                         l->lo_header->loh_attr |= LOHA_EXISTS;
 566                 }
 567
 568                 GOTO(out, rc = 0);
 569         }
 570
 571         idc = osd_idc_find(env, osd, fid);
 572         if (idc && !idc->oic_remote && idc->oic_dnode != ZFS_NO_OBJECT) {
 573                 oid = idc->oic_dnode;
 574                 goto zget;
 575         }
 576
 577         rc = -ENOENT;
 578         if (!list_empty(&osd->od_scrub.os_inconsistent_items))
 579                 rc = osd_oii_lookup(osd, fid, &oid);
 580
 581         if (rc)
 582                 rc = osd_fid_lookup(env, osd, fid, &oid);
 583
 584         if (rc == -ENOENT) {
 585                 if (likely(!(fid_is_norm(fid) || fid_is_igif(fid)) ||
 586                            fid_is_on_ost(env, osd, fid) ||
 587                            !zfs_test_bit(osd_oi_fid2idx(osd, fid),
 588                                          scrub->os_file.sf_oi_bitmap)))
 589                         GOTO(out, rc = 0);
 590
 591                 rc = -EREMCHG;
 592                 goto trigger;
 593         }
 594
 595         if (rc)
 596                 GOTO(out, rc);
 597
 598 zget:
 599         LASSERT(obj->oo_dn == NULL);
 600
 601         rc = __osd_obj2dnode(osd->od_os, oid, &obj->oo_dn);
 602         /* EEXIST will be returned if object is being deleted in ZFS */
 603         if (rc == -EEXIST)
 604                 GOTO(out, rc = 0);
 605
 606         if (rc) {
 607                 CERROR("%s: lookup "DFID"/%#llx failed: rc = %d\n",
 608                        osd->od_svname, PFID(lu_object_fid(l)), oid, rc);
 609                 GOTO(out, rc);
 610         }
 611
 612         rc = osd_object_init0(env, obj);
 613         if (rc)
 614                 GOTO(out, rc);
 615
 616         if (unlikely(obj->oo_header))
 617                 GOTO(out, rc = 0);
 618
 619         rc = osd_check_lma(env, obj);
 620         if (rc != -EREMCHG)
 621                 GOTO(out, rc);
 622
 623         osd_scrub_refresh_mapping(env, osd, fid, oid, DTO_INDEX_DELETE, true,
 624                                   NULL);
 625
 626 trigger:
 627         /* We still have chance to get the valid dnode: for the object that is
 628          * referenced by remote name entry, the object on the local MDT will be
 629          * linked under the dir /REMOTE_PARENT_DIR with its FID string as name.
 630          *
 631          * During the OI scrub, if we cannot find the OI mapping, we may still
 632          * have change to map the FID to local OID via lookup the dir
 633          * /REMOTE_PARENT_DIR. */
 634         if (!remote && !fid_is_on_ost(env, osd, fid)) {
 635                 osd_fid2str(name, fid, sizeof(info->oti_str));
 636                 rc = osd_zap_lookup(osd, osd->od_remote_parent_dir,
 637                                     NULL, name, 8, 3, (void *)zde);
 638                 if (!rc) {
 639                         oid = zde->lzd_reg.zde_dnode;
 640                         osd_dnode_rele(obj->oo_dn);
 641                         obj->oo_dn = NULL;
 642                         remote = true;
 643                         goto zget;
 644                 }
 645         }
 646
 647         /* The case someone triggered the OI scrub already. */
 648         if (scrub->os_running) {
 649                 if (!rc) {
 650                         LASSERT(remote);
 651
 652                         lu_object_set_agent_entry(l);
 653                         osd_oii_insert(env, osd, fid, oid, false);
 654                 } else {
 655                         rc = -EINPROGRESS;
 656                 }
 657
 658                 GOTO(out, rc);
 659         }
 660
 661         /* The case NOT allow to trigger OI scrub automatically. */
 662         if (osd->od_auto_scrub_interval == AS_NEVER)
 663                 GOTO(out, rc);
 664
 665         /* It is me to trigger the OI scrub. */
 666         rc1 = osd_scrub_start(env, osd, SS_CLEAR_DRYRUN |
 667                               SS_CLEAR_FAILOUT | SS_AUTO_FULL);
 668         CDEBUG_LIMIT(D_LFSCK | D_CONSOLE | D_WARNING,
 669                      "%s: trigger OI scrub by RPC for "DFID"/%#llx: rc = %d\n",
 670                      osd_name(osd), PFID(fid), oid, rc1);
 671         if (!rc) {
 672                 LASSERT(remote);
 673
 674                 lu_object_set_agent_entry(l);
 675                 if (!rc1)
 676                         osd_oii_insert(env, osd, fid, oid, false);
 677         } else {
 678                 if (!rc1)
 679                         rc = -EINPROGRESS;
 680                 else
 681                         rc = -EREMCHG;
 682         }
 683
 684         GOTO(out, rc);
 685
 686 out:
 687         RETURN(rc);
 688 }
 689
 690 /*
 691  * Concurrency: no concurrent access is possible that late in object
 692  * life-cycle.
 693  */
 694 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
 695 {
 696         struct osd_object *obj = osd_obj(l);
 697         struct lu_object_header *h = obj->oo_header;
 698
 699         LASSERT(osd_invariant(obj));
 700
 701         dt_object_fini(&obj->oo_dt);
 702         /* obj doesn't contain an lu_object_header, so we don't need call_rcu */
 703         OBD_SLAB_FREE_PTR(obj, osd_object_kmem);
 704         if (unlikely(h))
 705                 lu_object_header_free(h);
 706 }
 707
 708 static int
 709 osd_object_unlinked_add(struct osd_object *obj, struct osd_thandle *oh)
 710 {
 711         int rc = -EBUSY;
 712
 713         LASSERT(obj->oo_destroy == OSD_DESTROY_ASYNC);
 714
 715         /* the object is supposed to be exclusively locked by
 716          * the caller (osd_destroy()), while the transaction
 717          * (oh) is per-thread and not shared */
 718         if (likely(list_empty(&obj->oo_unlinked_linkage))) {
 719                 list_add(&obj->oo_unlinked_linkage, &oh->ot_unlinked_list);
 720                 rc = 0;
 721         }
 722
 723         return rc;
 724 }
 725
 726 /* Default to max data size covered by a level-1 indirect block */
 727 static unsigned long osd_sync_destroy_max_size =
 728         1UL << (DN_MAX_INDBLKSHIFT - SPA_BLKPTRSHIFT + SPA_MAXBLOCKSHIFT);
 729 module_param(osd_sync_destroy_max_size, ulong, 0444);
 730 MODULE_PARM_DESC(osd_sync_destroy_max_size, "Maximum object size to use synchronous destroy.");
 731
 732 static inline void
 733 osd_object_set_destroy_type(struct osd_object *obj)
 734 {
 735         /*
 736          * Lock-less OST_WRITE can race with OST_DESTROY, so set destroy type
 737          * only once and use it consistently thereafter.
 738          */
 739         down_write(&obj->oo_guard);
 740         if (obj->oo_destroy == OSD_DESTROY_NONE) {
 741                 if (obj->oo_attr.la_size <= osd_sync_destroy_max_size)
 742                         obj->oo_destroy = OSD_DESTROY_SYNC;
 743                 else /* Larger objects are destroyed asynchronously */
 744                         obj->oo_destroy = OSD_DESTROY_ASYNC;
 745         }
 746         up_write(&obj->oo_guard);
 747 }
 748
 749 static int osd_declare_destroy(const struct lu_env *env, struct dt_object *dt,
 750                                struct thandle *th)
 751 {
 752         const struct lu_fid     *fid = lu_object_fid(&dt->do_lu);
 753         struct osd_object       *obj = osd_dt_obj(dt);
 754         struct osd_device       *osd = osd_obj2dev(obj);
 755         struct osd_thandle      *oh;
 756         dnode_t *dn;
 757         int                      rc;
 758         uint64_t                 zapid;
 759         ENTRY;
 760
 761         LASSERT(th != NULL);
 762         LASSERT(dt_object_exists(dt));
 763
 764         oh = container_of(th, struct osd_thandle, ot_super);
 765         LASSERT(oh->ot_tx != NULL);
 766
 767         dmu_tx_mark_netfree(oh->ot_tx);
 768
 769         /* declare that we'll remove object from fid-dnode mapping */
 770         zapid = osd_get_name_n_idx(env, osd, fid, NULL, 0, &dn);
 771         osd_tx_hold_zap(oh->ot_tx, zapid, dn, FALSE, NULL);
 772
 773         osd_declare_xattrs_destroy(env, obj, oh);
 774
 775         /* one less inode */
 776         rc = osd_declare_quota(env, osd, obj->oo_attr.la_uid,
 777                                obj->oo_attr.la_gid, obj->oo_attr.la_projid,
 778                                -1, oh, NULL, OSD_QID_INODE);
 779         if (rc)
 780                 RETURN(rc);
 781
 782         /* data to be truncated */
 783         rc = osd_declare_quota(env, osd, obj->oo_attr.la_uid,
 784                                obj->oo_attr.la_gid, obj->oo_attr.la_projid,
 785                                0, oh, NULL, OSD_QID_BLK);
 786         if (rc)
 787                 RETURN(rc);
 788
 789         osd_object_set_destroy_type(obj);
 790         if (obj->oo_destroy == OSD_DESTROY_SYNC)
 791                 dmu_tx_hold_free(oh->ot_tx, obj->oo_dn->dn_object,
 792                                  0, DMU_OBJECT_END);
 793         else
 794                 osd_tx_hold_zap(oh->ot_tx, osd->od_unlinked->dn_object,
 795                                 osd->od_unlinked, TRUE, NULL);
 796
 797         /* remove agent entry (if have) from remote parent */
 798         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu))
 799                 osd_tx_hold_zap(oh->ot_tx, osd->od_remote_parent_dir,
 800                                 NULL, FALSE, NULL);
 801
 802         /* will help to find FID->ino when this object is being
 803          * added to PENDING/ */
 804         osd_idc_find_and_init(env, osd, obj);
 805
 806         RETURN(0);
 807 }
 808
 809 static int osd_destroy(const struct lu_env *env, struct dt_object *dt,
 810                        struct thandle *th)
 811 {
 812         struct osd_thread_info  *info = osd_oti_get(env);
 813         char                    *buf = info->oti_str;
 814         struct osd_object       *obj = osd_dt_obj(dt);
 815         struct osd_device       *osd = osd_obj2dev(obj);
 816         const struct lu_fid     *fid = lu_object_fid(&dt->do_lu);
 817         struct osd_thandle      *oh;
 818         int                      rc;
 819         uint64_t                 oid, zapid;
 820         dnode_t *zdn;
 821         ENTRY;
 822
 823         down_write(&obj->oo_guard);
 824
 825         if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
 826                 GOTO(out, rc = -ENOENT);
 827
 828         LASSERT(obj->oo_dn != NULL);
 829
 830         oh = container_of(th, struct osd_thandle, ot_super);
 831         LASSERT(oh != NULL);
 832         LASSERT(oh->ot_tx != NULL);
 833
 834         /* remove obj ref from index dir (it depends) */
 835         zapid = osd_get_name_n_idx(env, osd, fid, buf,
 836                                    sizeof(info->oti_str), &zdn);
 837         rc = osd_xattrs_destroy(env, obj, oh);
 838         if (rc) {
 839                 CERROR("%s: cannot destroy xattrs for %s: rc = %d\n",
 840                        osd->od_svname, buf, rc);
 841                 GOTO(out, rc);
 842         }
 843
 844         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu)) {
 845                 rc = osd_delete_from_remote_parent(env, osd, obj, oh, true);
 846                 if (rc)
 847                         GOTO(out, rc);
 848         }
 849
 850         oid = obj->oo_dn->dn_object;
 851         if (unlikely(obj->oo_destroy == OSD_DESTROY_NONE)) {
 852                 /* this may happen if the destroy wasn't declared
 853                  * e.g. when the object is created and then destroyed
 854                  * in the same transaction - we don't need additional
 855                  * space for destroy specifically */
 856                 LASSERT(obj->oo_attr.la_size <= osd_sync_destroy_max_size);
 857                 rc = -dmu_object_free(osd->od_os, oid, oh->ot_tx);
 858                 if (rc)
 859                         CERROR("%s: failed to free %s/%#llx: rc = %d\n",
 860                                osd->od_svname, buf, oid, rc);
 861         } else if (obj->oo_destroy == OSD_DESTROY_SYNC) {
 862                 rc = -dmu_object_free(osd->od_os, oid, oh->ot_tx);
 863                 if (rc)
 864                         CERROR("%s: failed to free %s/%#llx: rc = %d\n",
 865                                osd->od_svname, buf, oid, rc);
 866         } else { /* asynchronous destroy */
 867                 char *key = info->oti_key;
 868
 869                 rc = osd_object_unlinked_add(obj, oh);
 870                 if (rc)
 871                         GOTO(out, rc);
 872
 873                 snprintf(key, sizeof(info->oti_key), "%llx", oid);
 874                 rc = osd_zap_add(osd, osd->od_unlinked->dn_object,
 875                                  osd->od_unlinked, key, 8, 1, &oid, oh->ot_tx);
 876                 if (rc)
 877                         CERROR("%s: zap_add_int() failed %s/%#llx: rc = %d\n",
 878                                osd->od_svname, buf, oid, rc);
 879         }
 880
 881         /* Remove the OI mapping after the destroy to handle the race with
 882          * OI scrub that may insert missed OI mapping during the interval. */
 883         rc = osd_zap_remove(osd, zapid, zdn, buf, oh->ot_tx);
 884         if (unlikely(rc == -ENOENT))
 885                 rc = 0;
 886         if (rc)
 887                 CERROR("%s: zap_remove(%s) failed: rc = %d\n",
 888                        osd->od_svname, buf, rc);
 889
 890         GOTO(out, rc);
 891
 892 out:
 893         /* not needed in the cache anymore */
 894         set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
 895         if (rc == 0)
 896                 obj->oo_destroyed = 1;
 897         up_write(&obj->oo_guard);
 898         RETURN (0);
 899 }
 900
 901 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
 902 {
 903         struct osd_object *obj = osd_obj(l);
 904         const struct lu_fid *fid = lu_object_fid(l);
 905
 906         if (obj->oo_dn) {
 907                 if (likely(!fid_is_acct(fid))) {
 908                         osd_object_sa_fini(obj);
 909                         if (obj->oo_sa_xattr) {
 910                                 nvlist_free(obj->oo_sa_xattr);
 911                                 obj->oo_sa_xattr = NULL;
 912                         }
 913                         osd_dnode_rele(obj->oo_dn);
 914                         list_del(&obj->oo_sa_linkage);
 915                 }
 916                 obj->oo_dn = NULL;
 917         }
 918 }
 919
 920 /*
 921  * Concurrency: ->loo_object_release() is called under site spin-lock.
 922  */
 923 static void osd_object_release(const struct lu_env *env,
 924                                struct lu_object *l)
 925 {
 926 }
 927
 928 /*
 929  * Concurrency: shouldn't matter.
 930  */
 931 static int osd_object_print(const struct lu_env *env, void *cookie,
 932                             lu_printer_t p, const struct lu_object *l)
 933 {
 934         struct osd_object *o = osd_obj(l);
 935
 936         return (*p)(env, cookie, LUSTRE_OSD_ZFS_NAME"-object@%p", o);
 937 }
 938
 939 static void osd_read_lock(const struct lu_env *env, struct dt_object *dt,
 940                           unsigned role)
 941 {
 942         struct osd_object *obj = osd_dt_obj(dt);
 943
 944         LASSERT(osd_invariant(obj));
 945
 946         down_read_nested(&obj->oo_sem, role);
 947 }
 948
 949 static void osd_write_lock(const struct lu_env *env, struct dt_object *dt,
 950                            unsigned role)
 951 {
 952         struct osd_object *obj = osd_dt_obj(dt);
 953
 954         LASSERT(osd_invariant(obj));
 955
 956         down_write_nested(&obj->oo_sem, role);
 957 }
 958
 959 static void osd_read_unlock(const struct lu_env *env, struct dt_object *dt)
 960 {
 961         struct osd_object *obj = osd_dt_obj(dt);
 962
 963         LASSERT(osd_invariant(obj));
 964         up_read(&obj->oo_sem);
 965 }
 966
 967 static void osd_write_unlock(const struct lu_env *env, struct dt_object *dt)
 968 {
 969         struct osd_object *obj = osd_dt_obj(dt);
 970
 971         LASSERT(osd_invariant(obj));
 972         up_write(&obj->oo_sem);
 973 }
 974
 975 static int osd_write_locked(const struct lu_env *env, struct dt_object *dt)
 976 {
 977         struct osd_object *obj = osd_dt_obj(dt);
 978         int rc = 1;
 979
 980         LASSERT(osd_invariant(obj));
 981
 982         if (down_write_trylock(&obj->oo_sem)) {
 983                 rc = 0;
 984                 up_write(&obj->oo_sem);
 985         }
 986         return rc;
 987 }
 988
 989 static int osd_attr_get(const struct lu_env *env, struct dt_object *dt,
 990                         struct lu_attr *attr)
 991 {
 992         struct osd_object *obj = osd_dt_obj(dt);
 993         struct osd_device *osd = osd_obj2dev(obj);
 994         uint64_t blocks;
 995         uint32_t blksize;
 996         int rc = 0;
 997
 998         down_read(&obj->oo_guard);
 999
1000         if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
1001                 GOTO(out, rc = -ENOENT);
1002
1003         if (unlikely(fid_is_acct(lu_object_fid(&dt->do_lu))))
1004                 GOTO(out, rc = 0);
1005
1006         LASSERT(osd_invariant(obj));
1007         LASSERT(obj->oo_dn);
1008
1009         read_lock(&obj->oo_attr_lock);
1010         *attr = obj->oo_attr;
1011         if (obj->oo_lma_flags & LUSTRE_ORPHAN_FL) {
1012                 attr->la_valid |= LA_FLAGS;
1013                 attr->la_flags |= LUSTRE_ORPHAN_FL;
1014         }
1015         if (obj->oo_lma_flags & LUSTRE_ENCRYPT_FL) {
1016                 attr->la_valid |= LA_FLAGS;
1017                 attr->la_flags |= LUSTRE_ENCRYPT_FL;
1018         }
1019         read_unlock(&obj->oo_attr_lock);
1020         if (attr->la_valid & LA_FLAGS && attr->la_flags & LUSTRE_ORPHAN_FL)
1021                 CDEBUG(D_INFO, "%s: set orphan flag on "DFID" (%#llx/%#x)\n",
1022                        osd_obj2dev(obj)->od_svname,
1023                        PFID(lu_object_fid(&dt->do_lu)),
1024                        attr->la_valid, obj->oo_lma_flags);
1025
1026         /* with ZFS_DEBUG zrl_add_debug() called by DB_DNODE_ENTER()
1027          * from within sa_object_size() can block on a mutex, so
1028          * we can't call sa_object_size() holding rwlock */
1029         sa_object_size(obj->oo_sa_hdl, &blksize, &blocks);
1030         /* we do not control size of indices, so always calculate
1031          * it from number of blocks reported by DMU */
1032         if (S_ISDIR(attr->la_mode)) {
1033                 attr->la_size = 512 * blocks;
1034                 rc = -zap_count(osd->od_os, obj->oo_dn->dn_object,
1035                                 &attr->la_dirent_count);
1036         }
1037         /* Block size may be not set; suggest maximal I/O transfers. */
1038         if (blksize == 0)
1039                 blksize = osd_spa_maxblocksize(
1040                         dmu_objset_spa(osd_obj2dev(obj)->od_os));
1041
1042         attr->la_blksize = blksize;
1043         attr->la_blocks = blocks;
1044         attr->la_valid |= LA_BLOCKS | LA_BLKSIZE;
1045
1046 out:
1047         up_read(&obj->oo_guard);
1048         return rc;
1049 }
1050
1051 /* Simple wrapper on top of qsd API which implement quota transfer for osd
1052  * setattr needs. As a reminder, only the root user can change ownership of
1053  * a file, that's why EDQUOT & EINPROGRESS errors are discarded */
1054 static inline int qsd_transfer(const struct lu_env *env,
1055                                struct qsd_instance *qsd,
1056                                struct lquota_trans *trans, int qtype,
1057                                __u64 orig_id, __u64 new_id, __u64 bspace,
1058                                struct lquota_id_info *qi, bool ignore_edquot)
1059 {
1060         int     rc;
1061
1062         if (unlikely(qsd == NULL))
1063                 return 0;
1064
1065         LASSERT(qtype >= 0 && qtype < LL_MAXQUOTAS);
1066         qi->lqi_type = qtype;
1067
1068         /* inode accounting */
1069         qi->lqi_is_blk = false;
1070
1071         /* one more inode for the new owner ... */
1072         qi->lqi_id.qid_uid = new_id;
1073         qi->lqi_space      = 1;
1074         rc = qsd_op_begin(env, qsd, trans, qi, NULL);
1075         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
1076                 rc = 0;
1077         if (rc)
1078                 return rc;
1079
1080         /* and one less inode for the current id */
1081         qi->lqi_id.qid_uid = orig_id;;
1082         qi->lqi_space      = -1;
1083         /* can't get EDQUOT when reducing usage */
1084         rc = qsd_op_begin(env, qsd, trans, qi, NULL);
1085         if (rc == -EINPROGRESS)
1086                 rc = 0;
1087         if (rc)
1088                 return rc;
1089
1090         /* block accounting */
1091         qi->lqi_is_blk = true;
1092
1093         /* more blocks for the new owner ... */
1094         qi->lqi_id.qid_uid = new_id;
1095         qi->lqi_space      = bspace;
1096         rc = qsd_op_begin(env, qsd, trans, qi, NULL);
1097         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
1098                 rc = 0;
1099         if (rc)
1100                 return rc;
1101
1102         /* and finally less blocks for the current owner */
1103         qi->lqi_id.qid_uid = orig_id;
1104         qi->lqi_space      = -bspace;
1105         rc = qsd_op_begin(env, qsd, trans, qi, NULL);
1106         /* can't get EDQUOT when reducing usage */
1107         if (rc == -EINPROGRESS)
1108                 rc = 0;
1109         return rc;
1110 }
1111
1112 static int osd_declare_attr_set(const struct lu_env *env,
1113                                 struct dt_object *dt,
1114                                 const struct lu_attr *attr,
1115                                 struct thandle *handle)
1116 {
1117         struct osd_thread_info  *info = osd_oti_get(env);
1118         struct osd_object       *obj = osd_dt_obj(dt);
1119         struct osd_device       *osd = osd_obj2dev(obj);
1120         dmu_tx_hold_t           *txh;
1121         struct osd_thandle      *oh;
1122         uint64_t                 bspace;
1123         uint32_t                 blksize;
1124         int                      rc = 0;
1125         bool                     found;
1126         ENTRY;
1127
1128
1129         LASSERT(handle != NULL);
1130         LASSERT(osd_invariant(obj));
1131
1132         oh = container_of(handle, struct osd_thandle, ot_super);
1133
1134         down_read(&obj->oo_guard);
1135         if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
1136                 GOTO(out_sem, rc = 0);
1137
1138         LASSERT(obj->oo_sa_hdl != NULL);
1139         LASSERT(oh->ot_tx != NULL);
1140         /* regular attributes are part of the bonus buffer */
1141         /* let's check whether this object is already part of
1142          * transaction.. */
1143         found = false;
1144         for (txh = list_head(&oh->ot_tx->tx_holds); txh;
1145              txh = list_next(&oh->ot_tx->tx_holds, txh)) {
1146                 if (txh->txh_dnode == NULL)
1147                         continue;
1148                 if (txh->txh_dnode->dn_object != obj->oo_dn->dn_object)
1149                         continue;
1150                 /* this object is part of the transaction already
1151                  * we don't need to declare bonus again */
1152                 found = true;
1153                 break;
1154         }
1155         if (!found)
1156                 dmu_tx_hold_bonus(oh->ot_tx, obj->oo_dn->dn_object);
1157         if (oh->ot_tx->tx_err != 0)
1158                 GOTO(out_sem, rc = -oh->ot_tx->tx_err);
1159
1160         if (attr && attr->la_valid & LA_FLAGS) {
1161                 /* LMA is usually a part of bonus, no need to declare
1162                  * anything else */
1163         }
1164
1165         if (attr && (attr->la_valid & (LA_UID | LA_GID | LA_PROJID))) {
1166                 sa_object_size(obj->oo_sa_hdl, &blksize, &bspace);
1167                 bspace = toqb(bspace * 512);
1168
1169                 CDEBUG(D_QUOTA,
1170                        "%s: enforce quota on UID %u, GID %u, the quota space is %lld (%u)\n",
1171                        osd->od_svname,
1172                        attr->la_uid, attr->la_gid, bspace, blksize);
1173         }
1174         /* to preserve locking order - qsd_transfer() may need to flush
1175          * currently running transaction when we're out of quota. */
1176         up_read(&obj->oo_guard);
1177
1178         if (attr && attr->la_valid & LA_UID) {
1179                 /* quota enforcement for user */
1180                 if (attr->la_uid != obj->oo_attr.la_uid) {
1181                         rc = qsd_transfer(env, osd_def_qsd(osd),
1182                                           &oh->ot_quota_trans, USRQUOTA,
1183                                           obj->oo_attr.la_uid, attr->la_uid,
1184                                           bspace, &info->oti_qi, true);
1185                         if (rc)
1186                                 GOTO(out, rc);
1187                 }
1188         }
1189         if (attr && attr->la_valid & LA_GID) {
1190                 /* quota enforcement for group */
1191                 if (attr->la_gid != obj->oo_attr.la_gid) {
1192                         rc = qsd_transfer(env, osd_def_qsd(osd),
1193                                           &oh->ot_quota_trans, GRPQUOTA,
1194                                           obj->oo_attr.la_gid, attr->la_gid,
1195                                           bspace, &info->oti_qi,
1196                                           !(attr->la_flags &
1197                                                         LUSTRE_SET_SYNC_FL));
1198                         if (rc)
1199                                 GOTO(out, rc);
1200                 }
1201         }
1202 #ifdef ZFS_PROJINHERIT
1203         if (attr && attr->la_valid & LA_PROJID) {
1204                 /* quota enforcement for project */
1205                 if (attr->la_projid != obj->oo_attr.la_projid) {
1206                         if (!osd->od_projectused_dn)
1207                                 GOTO(out, rc = -EOPNOTSUPP);
1208
1209                         if (!projid_valid(make_kprojid(&init_user_ns, attr->la_projid)))
1210                                 GOTO(out, rc = -EINVAL);
1211
1212                         /* Usually, if project quota is upgradable for the
1213                          * device, then the upgrade will be done before or when
1214                          * mount the device. So when we come here, this project
1215                          * should have project ID attribute already (that is
1216                          * zero by default).  Otherwise, there was something
1217                          * wrong during the former upgrade, let's return failure
1218                          * to report that.
1219                          *
1220                          * Please note that, different from other attributes,
1221                          * you can NOT simply set the project ID attribute under
1222                          * such case, because adding (NOT change) project ID
1223                          * attribute needs to change the object's attribute
1224                          * layout to match zfs backend quota accounting
1225                          * requirement. */
1226                         if (unlikely(!obj->oo_with_projid))
1227                                 GOTO(out, rc = -ENXIO);
1228
1229                         rc = qsd_transfer(env, osd_def_qsd(osd),
1230                                           &oh->ot_quota_trans, PRJQUOTA,
1231                                           obj->oo_attr.la_projid,
1232                                           attr->la_projid, bspace,
1233                                           &info->oti_qi, true);
1234                         if (rc)
1235                                 GOTO(out, rc);
1236                 }
1237         }
1238 #endif
1239 out:
1240         RETURN(rc);
1241 out_sem:
1242         up_read(&obj->oo_guard);
1243         RETURN(rc);
1244 }
1245
1246 /*
1247  * Set the attributes of an object
1248  *
1249  * The transaction passed to this routine must have
1250  * dmu_tx_hold_bonus(tx, oid) called and then assigned
1251  * to a transaction group.
1252  */
1253 static int osd_attr_set(const struct lu_env *env, struct dt_object *dt,
1254                         const struct lu_attr *la, struct thandle *handle)
1255 {
1256         struct osd_thread_info  *info = osd_oti_get(env);
1257         sa_bulk_attr_t          *bulk = osd_oti_get(env)->oti_attr_bulk;
1258         struct osd_object       *obj = osd_dt_obj(dt);
1259         struct osd_device       *osd = osd_obj2dev(obj);
1260         struct osd_thandle      *oh;
1261         struct osa_attr         *osa = &info->oti_osa;
1262         __u64                    valid = la->la_valid;
1263         int                      cnt;
1264         int                      rc = 0;
1265
1266         ENTRY;
1267
1268         down_read(&obj->oo_guard);
1269         if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
1270                 GOTO(out, rc = -ENOENT);
1271
1272         LASSERT(handle != NULL);
1273         LASSERT(osd_invariant(obj));
1274         LASSERT(obj->oo_sa_hdl);
1275
1276         oh = container_of(handle, struct osd_thandle, ot_super);
1277         /* Assert that the transaction has been assigned to a
1278            transaction group. */
1279         LASSERT(oh->ot_tx->tx_txg != 0);
1280
1281         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING) && !osd->od_is_ost) {
1282                 struct zpl_direntry *zde = &info->oti_zde.lzd_reg;
1283                 char *buf = info->oti_str;
1284                 dnode_t *zdn = NULL;
1285                 uint64_t zapid;
1286
1287                 zapid = osd_get_name_n_idx(env, osd, lu_object_fid(&dt->do_lu),
1288                                            buf, sizeof(info->oti_str), &zdn);
1289                 rc = osd_zap_lookup(osd, zapid, zdn, buf, 8,
1290                                     sizeof(*zde) / 8, zde);
1291                 if (!rc) {
1292                         zde->zde_dnode -= 1;
1293                         rc = -zap_update(osd->od_os, zapid, buf, 8,
1294                                          sizeof(*zde) / 8, zde, oh->ot_tx);
1295                 }
1296                 if (rc > 0)
1297                         rc = 0;
1298                 GOTO(out, rc);
1299         }
1300
1301         /* Only allow set size for regular file */
1302         if (!S_ISREG(dt->do_lu.lo_header->loh_attr))
1303                 valid &= ~(LA_SIZE | LA_BLOCKS);
1304
1305         if (valid & LA_CTIME && la->la_ctime == obj->oo_attr.la_ctime)
1306                 valid &= ~LA_CTIME;
1307
1308         if (valid & LA_MTIME && la->la_mtime == obj->oo_attr.la_mtime)
1309                 valid &= ~LA_MTIME;
1310
1311         if (valid & LA_ATIME && la->la_atime == obj->oo_attr.la_atime)
1312                 valid &= ~LA_ATIME;
1313
1314         if (valid == 0)
1315                 GOTO(out, rc = 0);
1316
1317         if (valid & LA_FLAGS) {
1318                 struct lustre_mdt_attrs *lma;
1319                 struct lu_buf buf;
1320                 int size = 0;
1321
1322                 if (la->la_flags & LUSTRE_LMA_FL_MASKS) {
1323                         LASSERT(!obj->oo_pfid_in_lma);
1324                         BUILD_BUG_ON(sizeof(info->oti_buf) < sizeof(*lma));
1325                         lma = (struct lustre_mdt_attrs *)&info->oti_buf;
1326                         buf.lb_buf = lma;
1327                         buf.lb_len = sizeof(info->oti_buf);
1328
1329                         /* Please do NOT call osd_xattr_get() directly, that
1330                          * will cause recursive down_read() on oo_guard. */
1331                         rc = osd_xattr_get_internal(env, obj, &buf,
1332                                                     XATTR_NAME_LMA, &size);
1333                         if (!rc && unlikely(size < sizeof(*lma))) {
1334                                 rc = -EINVAL;
1335                         } else if (!rc) {
1336                                 lma->lma_incompat =
1337                                         le32_to_cpu(lma->lma_incompat);
1338                                 lma->lma_incompat |=
1339                                         lustre_to_lma_flags(la->la_flags);
1340                                 lma->lma_incompat =
1341                                         cpu_to_le32(lma->lma_incompat);
1342                                 buf.lb_buf = lma;
1343                                 buf.lb_len = sizeof(*lma);
1344                                 rc = osd_xattr_set_internal(env, obj, &buf,
1345                                                             XATTR_NAME_LMA,
1346                                                             LU_XATTR_REPLACE,
1347                                                             oh);
1348                         }
1349                         if (rc < 0) {
1350                                 CWARN("%s: failed to set LMA flags: rc = %d\n",
1351                                        osd->od_svname, rc);
1352                                 GOTO(out, rc);
1353                         } else {
1354                                 obj->oo_lma_flags =
1355                                         la->la_flags & LUSTRE_LMA_FL_MASKS;
1356                         }
1357                 }
1358         }
1359
1360         write_lock(&obj->oo_attr_lock);
1361         cnt = 0;
1362
1363         if (valid & LA_PROJID) {
1364 #ifdef ZFS_PROJINHERIT
1365                 if (osd->od_projectused_dn) {
1366                         LASSERT(obj->oo_with_projid);
1367
1368                         osa->projid = obj->oo_attr.la_projid = la->la_projid;
1369                         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_PROJID(osd), NULL,
1370                                          &osa->projid, 8);
1371                 } else
1372 #endif
1373                         valid &= ~LA_PROJID;
1374         }
1375
1376         if (valid & LA_ATIME) {
1377                 osa->atime[0] = obj->oo_attr.la_atime = la->la_atime;
1378                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(osd), NULL,
1379                                  osa->atime, 16);
1380         }
1381         if (valid & LA_MTIME) {
1382                 osa->mtime[0] = obj->oo_attr.la_mtime = la->la_mtime;
1383                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(osd), NULL,
1384                                  osa->mtime, 16);
1385         }
1386         if (valid & LA_CTIME) {
1387                 osa->ctime[0] = obj->oo_attr.la_ctime = la->la_ctime;
1388                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(osd), NULL,
1389                                  osa->ctime, 16);
1390         }
1391         if (valid & LA_MODE) {
1392                 /* mode is stored along with type, so read it first */
1393                 obj->oo_attr.la_mode = (obj->oo_attr.la_mode & S_IFMT) |
1394                         (la->la_mode & ~S_IFMT);
1395                 osa->mode = obj->oo_attr.la_mode;
1396                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(osd), NULL,
1397                                  &osa->mode, 8);
1398         }
1399         if (valid & LA_SIZE) {
1400                 osa->size = obj->oo_attr.la_size = la->la_size;
1401                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(osd), NULL,
1402                                  &osa->size, 8);
1403         }
1404         if (valid & LA_NLINK) {
1405                 osa->nlink = obj->oo_attr.la_nlink = la->la_nlink;
1406                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(osd), NULL,
1407                                  &osa->nlink, 8);
1408         }
1409         if (valid & LA_RDEV) {
1410                 osa->rdev = obj->oo_attr.la_rdev = la->la_rdev;
1411                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_RDEV(osd), NULL,
1412                                  &osa->rdev, 8);
1413         }
1414         if (valid & LA_FLAGS) {
1415                 osa->flags = attrs_fs2zfs(la->la_flags);
1416                 /* many flags are not supported by zfs, so ensure a good cached
1417                  * copy */
1418                 obj->oo_attr.la_flags = attrs_zfs2fs(osa->flags);
1419 #ifdef ZFS_PROJINHERIT
1420                 if (obj->oo_with_projid && osd->od_projectused_dn)
1421                         osa->flags |= ZFS_PROJID;
1422 #endif
1423                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(osd), NULL,
1424                                  &osa->flags, 8);
1425         }
1426         if (valid & LA_UID) {
1427                 osa->uid = obj->oo_attr.la_uid = la->la_uid;
1428                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(osd), NULL,
1429                                  &osa->uid, 8);
1430         }
1431         if (valid & LA_GID) {
1432                 osa->gid = obj->oo_attr.la_gid = la->la_gid;
1433                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(osd), NULL,
1434                                  &osa->gid, 8);
1435         }
1436         obj->oo_attr.la_valid |= valid;
1437         write_unlock(&obj->oo_attr_lock);
1438
1439         LASSERT(cnt <= ARRAY_SIZE(osd_oti_get(env)->oti_attr_bulk));
1440         rc = osd_object_sa_bulk_update(obj, bulk, cnt, oh);
1441
1442 out:
1443         up_read(&obj->oo_guard);
1444         RETURN(rc);
1445 }
1446
1447 /*
1448  * Object creation.
1449  *
1450  * XXX temporary solution.
1451  */
1452
1453 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
1454                         struct dt_object *parent, struct dt_object *child,
1455                         umode_t child_mode)
1456 {
1457         LASSERT(ah);
1458
1459         ah->dah_parent = parent;
1460         ah->dah_mode = child_mode;
1461
1462         if (parent != NULL && !dt_object_remote(parent)) {
1463                 /* will help to find FID->ino at dt_insert("..") */
1464                 struct osd_object *pobj = osd_dt_obj(parent);
1465
1466                 osd_idc_find_and_init(env, osd_obj2dev(pobj), pobj);
1467         }
1468 }
1469
1470 static int osd_declare_create(const struct lu_env *env, struct dt_object *dt,
1471                               struct lu_attr *attr,
1472                               struct dt_allocation_hint *hint,
1473                               struct dt_object_format *dof,
1474                               struct thandle *handle)
1475 {
1476         const struct lu_fid     *fid = lu_object_fid(&dt->do_lu);
1477         struct osd_object       *obj = osd_dt_obj(dt);
1478         struct osd_device       *osd = osd_obj2dev(obj);
1479         struct osd_thandle      *oh;
1480         uint64_t                 zapid;
1481         dnode_t                 *dn;
1482         int                      rc, dnode_size;
1483         ENTRY;
1484
1485         LASSERT(dof);
1486
1487         switch (dof->dof_type) {
1488                 case DFT_REGULAR:
1489                 case DFT_SYM:
1490                 case DFT_NODE:
1491                         if (obj->oo_dt.do_body_ops == NULL)
1492                                 obj->oo_dt.do_body_ops = &osd_body_ops;
1493                         break;
1494                 default:
1495                         break;
1496         }
1497
1498         LASSERT(handle != NULL);
1499         oh = container_of(handle, struct osd_thandle, ot_super);
1500         LASSERT(oh->ot_tx != NULL);
1501
1502         /* this is the minimum set of EAs on every Lustre object */
1503         obj->oo_ea_in_bonus = OSD_BASE_EA_IN_BONUS;
1504         /* reserve 32 bytes for extra stuff like ACLs */
1505         dnode_size = size_roundup_power2(obj->oo_ea_in_bonus + 32);
1506
1507         switch (dof->dof_type) {
1508                 case DFT_DIR:
1509                         dt->do_index_ops = &osd_dir_ops;
1510                         /* fallthrough */
1511                 case DFT_INDEX:
1512                         /* for zap create */
1513                         dmu_tx_hold_zap(oh->ot_tx, DMU_NEW_OBJECT, FALSE, NULL);
1514                         dmu_tx_hold_sa_create(oh->ot_tx, dnode_size);
1515                         break;
1516                 case DFT_REGULAR:
1517                 case DFT_SYM:
1518                 case DFT_NODE:
1519                         /* first, we'll create new object */
1520                         dmu_tx_hold_sa_create(oh->ot_tx, dnode_size);
1521                         break;
1522
1523                 default:
1524                         LBUG();
1525                         break;
1526         }
1527
1528         /* and we'll add it to some mapping */
1529         zapid = osd_get_name_n_idx(env, osd, fid, NULL, 0, &dn);
1530         osd_tx_hold_zap(oh->ot_tx, zapid, dn, TRUE, NULL);
1531
1532         /* will help to find FID->ino mapping at dt_insert() */
1533         osd_idc_find_and_init(env, osd, obj);
1534
1535         rc = osd_declare_quota(env, osd, attr->la_uid, attr->la_gid,
1536                                attr->la_projid, 1, oh, NULL, OSD_QID_INODE);
1537
1538         RETURN(rc);
1539 }
1540
1541 int __osd_attr_init(const struct lu_env *env, struct osd_device *osd,
1542                     struct osd_object *obj, sa_handle_t *sa_hdl, dmu_tx_t *tx,
1543                     struct lu_attr *la, uint64_t parent,
1544                     nvlist_t *xattr)
1545 {
1546         sa_bulk_attr_t *bulk = osd_oti_get(env)->oti_attr_bulk;
1547         struct osa_attr *osa = &osd_oti_get(env)->oti_osa;
1548         uint64_t gen;
1549         inode_timespec_t now;
1550         int cnt;
1551         int rc;
1552         char *dxattr = NULL;
1553         size_t sa_size;
1554
1555
1556         LASSERT(sa_hdl);
1557
1558         gen = dmu_tx_get_txg(tx);
1559         gethrestime(&now);
1560         ZFS_TIME_ENCODE(&now, osa->btime);
1561
1562         osa->atime[0] = la->la_atime;
1563         osa->ctime[0] = la->la_ctime;
1564         osa->mtime[0] = la->la_mtime;
1565         osa->mode = la->la_mode;
1566         osa->uid = la->la_uid;
1567         osa->gid = la->la_gid;
1568         osa->rdev = la->la_rdev;
1569         osa->nlink = la->la_nlink;
1570         if (la->la_valid & LA_FLAGS)
1571                 osa->flags = attrs_fs2zfs(la->la_flags);
1572         else
1573                 osa->flags = 0;
1574         osa->size  = la->la_size;
1575 #ifdef ZFS_PROJINHERIT
1576         if (osd->od_projectused_dn) {
1577                 if (la->la_valid & LA_PROJID)
1578                         osa->projid = la->la_projid;
1579                 else
1580                         osa->projid = ZFS_DEFAULT_PROJID;
1581                 osa->flags |= ZFS_PROJID;
1582                 if (obj)
1583                         obj->oo_with_projid = 1;
1584         } else {
1585                 osa->flags &= ~ZFS_PROJID;
1586         }
1587 #endif
1588
1589         /*
1590          * we need to create all SA below upon object create.
1591          *
1592          * XXX The attribute order matters since the accounting callback relies
1593          * on static offsets (i.e. SA_*_OFFSET, see zfs_space_delta_cb()) to
1594          * look up the UID/GID/PROJID attributes. Moreover, the callback does
1595          * not seem to support the spill block.
1596          * We define attributes in the same order as SA_*_OFFSET in order to
1597          * work around the problem. See ORI-610.
1598          */
1599         cnt = 0;
1600         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(osd), NULL, &osa->mode, 8);
1601         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(osd), NULL, &osa->size, 8);
1602         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GEN(osd), NULL, &gen, 8);
1603         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(osd), NULL, &osa->uid, 8);
1604         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(osd), NULL, &osa->gid, 8);
1605         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_PARENT(osd), NULL, &parent, 8);
1606         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(osd), NULL, &osa->flags, 8);
1607         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(osd), NULL, osa->atime, 16);
1608         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(osd), NULL, osa->mtime, 16);
1609         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(osd), NULL, osa->ctime, 16);
1610         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CRTIME(osd), NULL, osa->btime, 16);
1611         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(osd), NULL, &osa->nlink, 8);
1612 #ifdef ZFS_PROJINHERIT
1613         if (osd->od_projectused_dn)
1614                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_PROJID(osd), NULL,
1615                                  &osa->projid, 8);
1616 #endif
1617         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_RDEV(osd), NULL, &osa->rdev, 8);
1618         LASSERT(cnt <= ARRAY_SIZE(osd_oti_get(env)->oti_attr_bulk));
1619
1620         if (xattr) {
1621                 rc = -nvlist_size(xattr, &sa_size, NV_ENCODE_XDR);
1622                 LASSERT(rc == 0);
1623
1624                 dxattr = osd_zio_buf_alloc(sa_size);
1625                 LASSERT(dxattr);
1626
1627                 rc = -nvlist_pack(xattr, &dxattr, &sa_size,
1628                                 NV_ENCODE_XDR, KM_SLEEP);
1629                 LASSERT(rc == 0);
1630
1631                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_DXATTR(osd),
1632                                 NULL, dxattr, sa_size);
1633         }
1634
1635         rc = -sa_replace_all_by_template(sa_hdl, bulk, cnt, tx);
1636         if (dxattr)
1637                 osd_zio_buf_free(dxattr, sa_size);
1638
1639         return rc;
1640 }
1641
1642 int osd_find_new_dnode(const struct lu_env *env, dmu_tx_t *tx,
1643                        uint64_t oid, dnode_t **dnp)
1644 {
1645         dmu_tx_hold_t *txh;
1646         int rc = 0;
1647
1648         /* take dnode_t from tx to save on dnode#->dnode_t lookup */
1649         for (txh = list_tail(&tx->tx_holds); txh;
1650              txh = list_prev(&tx->tx_holds, txh)) {
1651                 dnode_t *dn = txh->txh_dnode;
1652                 dmu_buf_impl_t *db;
1653
1654                 if (dn == NULL)
1655                         continue;
1656                 if (dn->dn_object != oid)
1657                         continue;
1658                 db = dn->dn_bonus;
1659                 if (db == NULL) {
1660                         rw_enter(&dn->dn_struct_rwlock, RW_WRITER);
1661                         if (dn->dn_bonus == NULL)
1662                                 dbuf_create_bonus(dn);
1663                         rw_exit(&dn->dn_struct_rwlock);
1664                 }
1665                 db = dn->dn_bonus;
1666                 LASSERT(db);
1667                 LASSERT(dn->dn_handle);
1668                 DB_DNODE_ENTER(db);
1669                 if (zfs_refcount_add(&db->db_holds, osd_obj_tag) == 1) {
1670                         zfs_refcount_add(&dn->dn_holds, osd_obj_tag);
1671                         atomic_inc_32(&dn->dn_dbufs_count);
1672                 }
1673                 *dnp = dn;
1674                 DB_DNODE_EXIT(db);
1675                 dbuf_read(db, NULL, DB_RF_MUST_SUCCEED | DB_RF_NOPREFETCH);
1676                 break;
1677         }
1678
1679         if (unlikely(*dnp == NULL))
1680                 rc = __osd_obj2dnode(tx->tx_objset, oid, dnp);
1681
1682         return rc;
1683 }
1684
1685 #ifdef HAVE_DMU_OBJECT_ALLOC_DNSIZE
1686 int osd_find_dnsize(struct osd_device *osd, int ea_in_bonus)
1687 {
1688         int dnsize;
1689
1690         if (osd->od_dnsize == ZFS_DNSIZE_AUTO) {
1691                 dnsize = DNODE_MIN_SIZE;
1692                 do {
1693                         if (DN_BONUS_SIZE(dnsize) >= ea_in_bonus + 32)
1694                                 break;
1695                         dnsize <<= 1;
1696                 } while (dnsize < DNODE_MAX_SIZE);
1697                 if (dnsize > DNODE_MAX_SIZE)
1698                         dnsize = DNODE_MAX_SIZE;
1699         } else if (osd->od_dnsize == ZFS_DNSIZE_1K) {
1700                 dnsize = 1024;
1701         } else if (osd->od_dnsize == ZFS_DNSIZE_2K) {
1702                 dnsize = 2048;
1703         } else if (osd->od_dnsize == ZFS_DNSIZE_4K) {
1704                 dnsize = 4096;
1705         } else if (osd->od_dnsize == ZFS_DNSIZE_8K) {
1706                 dnsize = 8192;
1707         } else if (osd->od_dnsize == ZFS_DNSIZE_16K) {
1708                 dnsize = 16384;
1709         } else {
1710                 dnsize = DNODE_MIN_SIZE;
1711         }
1712         return dnsize;
1713 }
1714 #endif
1715
1716 /*
1717  * The transaction passed to this routine must have
1718  * dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT) called and then assigned
1719  * to a transaction group.
1720  */
1721 int __osd_object_create(const struct lu_env *env, struct osd_device *osd,
1722                         struct osd_object *obj, const struct lu_fid *fid,
1723                         dnode_t **dnp, dmu_tx_t *tx, struct lu_attr *la)
1724 {
1725         dmu_object_type_t type = DMU_OT_PLAIN_FILE_CONTENTS;
1726         uint64_t oid;
1727         int size;
1728
1729         /* Use DMU_OTN_UINT8_METADATA for local objects so their data blocks
1730          * would get an additional ditto copy */
1731         if (unlikely(S_ISREG(la->la_mode) &&
1732                      fid_seq_is_local_file(fid_seq(fid))))
1733                 type = DMU_OTN_UINT8_METADATA;
1734
1735         /* Create a new DMU object using the default dnode size. */
1736         if (obj)
1737                 size = obj->oo_ea_in_bonus;
1738         else
1739                 size = OSD_BASE_EA_IN_BONUS;
1740         oid = osd_dmu_object_alloc(osd->od_os, type, 0,
1741                                    osd_find_dnsize(osd, size), tx);
1742
1743         LASSERT(la->la_valid & LA_MODE);
1744         la->la_size = 0;
1745         la->la_nlink = 1;
1746
1747         return osd_find_new_dnode(env, tx, oid, dnp);
1748 }
1749
1750 /*
1751  * The transaction passed to this routine must have
1752  * dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, ...) called and then assigned
1753  * to a transaction group.
1754  *
1755  * Using ZAP_FLAG_HASH64 will force the ZAP to always be a FAT ZAP.
1756  * This is fine for directories today, because storing the FID in the dirent
1757  * will also require a FAT ZAP.  If there is a new type of micro ZAP created
1758  * then we might need to re-evaluate the use of this flag and instead do
1759  * a conversion from the different internal ZAP hash formats being used. */
1760 int __osd_zap_create(const struct lu_env *env, struct osd_device *osd,
1761                      dnode_t **dnp, dmu_tx_t *tx, struct lu_attr *la,
1762                      unsigned dnsize, zap_flags_t flags)
1763 {
1764         uint64_t oid;
1765
1766         /* Assert that the transaction has been assigned to a
1767            transaction group. */
1768         LASSERT(tx->tx_txg != 0);
1769         *dnp = NULL;
1770
1771         oid = osd_zap_create_flags(osd->od_os, 0, flags | ZAP_FLAG_HASH64,
1772                                    DMU_OT_DIRECTORY_CONTENTS,
1773                                    14, /* == ZFS fzap_default_blockshift */
1774                                    DN_MAX_INDBLKSHIFT, /* indirect blockshift */
1775                                    dnsize, tx);
1776
1777         la->la_size = 2;
1778         la->la_nlink = 1;
1779
1780         return osd_find_new_dnode(env, tx, oid, dnp);
1781 }
1782
1783 static dnode_t *osd_mkidx(const struct lu_env *env, struct osd_object *obj,
1784                           struct lu_attr *la, struct osd_thandle *oh)
1785 {
1786         struct osd_device *osd = osd_obj2dev(obj);
1787         dnode_t *dn;
1788         int rc;
1789
1790         /* Index file should be created as regular file in order not to confuse
1791          * ZPL which could interpret them as directory.
1792          * We set ZAP_FLAG_UINT64_KEY to let ZFS know than we are going to use
1793          * binary keys */
1794         LASSERT(S_ISREG(la->la_mode));
1795         rc = __osd_zap_create(env, osd, &dn, oh->ot_tx, la,
1796                 osd_find_dnsize(osd, obj->oo_ea_in_bonus), ZAP_FLAG_UINT64_KEY);
1797         if (rc)
1798                 return ERR_PTR(rc);
1799         return dn;
1800 }
1801
1802 static dnode_t *osd_mkdir(const struct lu_env *env, struct osd_object *obj,
1803                           struct lu_attr *la, struct osd_thandle *oh)
1804 {
1805         struct osd_device *osd = osd_obj2dev(obj);
1806         dnode_t *dn;
1807         int rc;
1808
1809         LASSERT(S_ISDIR(la->la_mode));
1810         rc = __osd_zap_create(env, osd, &dn, oh->ot_tx, la,
1811                               osd_find_dnsize(osd, obj->oo_ea_in_bonus), 0);
1812         if (rc)
1813                 return ERR_PTR(rc);
1814         return dn;
1815 }
1816
1817 static dnode_t *osd_mkreg(const struct lu_env *env, struct osd_object *obj,
1818                           struct lu_attr *la, struct osd_thandle *oh)
1819 {
1820         const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
1821         struct osd_device *osd = osd_obj2dev(obj);
1822         dnode_t *dn;
1823         int rc;
1824
1825         LASSERT(S_ISREG(la->la_mode));
1826         rc = __osd_object_create(env, osd, obj, fid, &dn, oh->ot_tx, la);
1827         if (rc)
1828                 return ERR_PTR(rc);
1829
1830         if ((fid_is_idif(fid) || fid_is_norm(fid) || fid_is_echo(fid))) {
1831                 /* The minimum block size must be at least page size otherwise
1832                  * it will break the assumption in tgt_thread_big_cache where
1833                  * the array size is PTLRPC_MAX_BRW_PAGES. It will also affect
1834                  * RDMA due to subpage transfer size */
1835                 rc = -dmu_object_set_blocksize(osd->od_os, dn->dn_object,
1836                                                PAGE_SIZE, 0, oh->ot_tx);
1837                 if (unlikely(rc)) {
1838                         CERROR("%s: can't change blocksize: %d\n",
1839                                osd->od_svname, rc);
1840                         return ERR_PTR(rc);
1841                 }
1842         } else if ((fid_is_llog(fid))) {
1843                 rc = -dmu_object_set_blocksize(osd->od_os, dn->dn_object,
1844                                                LLOG_MIN_CHUNK_SIZE, 0, oh->ot_tx);
1845                 if (unlikely(rc)) {
1846                         CERROR("%s: can't change blocksize: %d\n",
1847                                osd->od_svname, rc);
1848                         return ERR_PTR(rc);
1849                 }
1850         }
1851
1852         return dn;
1853 }
1854
1855 static dnode_t *osd_mksym(const struct lu_env *env, struct osd_object *obj,
1856                           struct lu_attr *la, struct osd_thandle *oh)
1857 {
1858         dnode_t *dn;
1859         int rc;
1860
1861         LASSERT(S_ISLNK(la->la_mode));
1862         rc = __osd_object_create(env, osd_obj2dev(obj), obj,
1863                                  lu_object_fid(&obj->oo_dt.do_lu),
1864                                  &dn, oh->ot_tx, la);
1865         if (rc)
1866                 return ERR_PTR(rc);
1867         return dn;
1868 }
1869
1870 static dnode_t *osd_mknod(const struct lu_env *env, struct osd_object *obj,
1871                           struct lu_attr *la, struct osd_thandle *oh)
1872 {
1873         dnode_t *dn;
1874         int rc;
1875
1876         if (S_ISCHR(la->la_mode) || S_ISBLK(la->la_mode))
1877                 la->la_valid |= LA_RDEV;
1878
1879         rc = __osd_object_create(env, osd_obj2dev(obj), obj,
1880                                  lu_object_fid(&obj->oo_dt.do_lu),
1881                                  &dn, oh->ot_tx, la);
1882         if (rc)
1883                 return ERR_PTR(rc);
1884         return dn;
1885 }
1886
1887 typedef dnode_t *(*osd_obj_type_f)(const struct lu_env *env,
1888                                    struct osd_object *obj,
1889                                    struct lu_attr *la,
1890                                    struct osd_thandle *oh);
1891
1892 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
1893 {
1894         osd_obj_type_f result;
1895
1896         switch (type) {
1897         case DFT_DIR:
1898                 result = osd_mkdir;
1899                 break;
1900         case DFT_INDEX:
1901                 result = osd_mkidx;
1902                 break;
1903         case DFT_REGULAR:
1904                 result = osd_mkreg;
1905                 break;
1906         case DFT_SYM:
1907                 result = osd_mksym;
1908                 break;
1909         case DFT_NODE:
1910                 result = osd_mknod;
1911                 break;
1912         default:
1913                 LBUG();
1914                 break;
1915         }
1916         return result;
1917 }
1918
1919 /*
1920  * Concurrency: @dt is write locked.
1921  */
1922 static int osd_create(const struct lu_env *env, struct dt_object *dt,
1923                       struct lu_attr *attr, struct dt_allocation_hint *hint,
1924                       struct dt_object_format *dof, struct thandle *th)
1925 {
1926         struct osd_thread_info  *info = osd_oti_get(env);
1927         struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
1928         struct zpl_direntry     *zde = &info->oti_zde.lzd_reg;
1929         const struct lu_fid     *fid = lu_object_fid(&dt->do_lu);
1930         struct osd_object       *obj = osd_dt_obj(dt);
1931         struct osd_device       *osd = osd_obj2dev(obj);
1932         char                    *buf = info->oti_str;
1933         struct osd_thandle      *oh;
1934         dnode_t *dn = NULL, *zdn = NULL;
1935         uint64_t                 zapid, parent = 0;
1936         int                      rc;
1937         __u32 compat = 0;
1938
1939         ENTRY;
1940
1941         LASSERT(!fid_is_acct(fid));
1942
1943         /* concurrent create declarations should not see
1944          * the object inconsistent (db, attr, etc).
1945          * in regular cases acquisition should be cheap */
1946         down_write(&obj->oo_guard);
1947
1948         if (unlikely(dt_object_exists(dt)))
1949                 GOTO(out, rc = -EEXIST);
1950
1951         LASSERT(osd_invariant(obj));
1952         LASSERT(dof != NULL);
1953
1954         LASSERT(th != NULL);
1955         oh = container_of(th, struct osd_thandle, ot_super);
1956
1957         LASSERT(obj->oo_dn == NULL);
1958
1959         /* to follow ZFS on-disk format we need
1960          * to initialize parent dnode properly */
1961         if (hint != NULL && hint->dah_parent != NULL &&
1962             !dt_object_remote(hint->dah_parent))
1963                 parent = osd_dt_obj(hint->dah_parent)->oo_dn->dn_object;
1964
1965         /* we may fix some attributes, better do not change the source */
1966         obj->oo_attr = *attr;
1967         obj->oo_attr.la_size = 0;
1968         obj->oo_attr.la_nlink = 0;
1969         obj->oo_attr.la_valid |= LA_SIZE | LA_NLINK | LA_TYPE;
1970
1971 #ifdef ZFS_PROJINHERIT
1972         if (osd->od_projectused_dn) {
1973                 if (!(obj->oo_attr.la_valid & LA_PROJID))
1974                         obj->oo_attr.la_projid = ZFS_DEFAULT_PROJID;
1975                 obj->oo_with_projid = 1;
1976         }
1977 #endif
1978
1979         dn = osd_create_type_f(dof->dof_type)(env, obj, &obj->oo_attr, oh);
1980         if (IS_ERR(dn)) {
1981                 rc = PTR_ERR(dn);
1982                 dn = NULL;
1983                 GOTO(out, rc);
1984         }
1985
1986         zde->zde_pad = 0;
1987         zde->zde_dnode = dn->dn_object;
1988         zde->zde_type = S_DT(attr->la_mode & S_IFMT);
1989
1990         zapid = osd_get_name_n_idx(env, osd, fid, buf,
1991                                    sizeof(info->oti_str), &zdn);
1992         if (CFS_FAIL_CHECK(OBD_FAIL_OSD_NO_OI_ENTRY) ||
1993             (osd->od_is_ost && OBD_FAIL_CHECK(OBD_FAIL_OSD_COMPAT_NO_ENTRY)))
1994                 goto skip_add;
1995
1996         if (osd->od_is_ost && OBD_FAIL_CHECK(OBD_FAIL_OSD_COMPAT_INVALID_ENTRY))
1997                 zde->zde_dnode++;
1998
1999         rc = osd_zap_add(osd, zapid, zdn, buf, 8, 1, zde, oh->ot_tx);
2000         if (rc)
2001                 GOTO(out, rc);
2002
2003 skip_add:
2004         obj->oo_dn = dn;
2005         /* Now add in all of the "SA" attributes */
2006         rc = osd_sa_handle_get(obj);
2007         if (rc)
2008                 GOTO(out, rc);
2009
2010         rc = -nvlist_alloc(&obj->oo_sa_xattr, NV_UNIQUE_NAME, KM_SLEEP);
2011         if (rc)
2012                 GOTO(out, rc);
2013
2014         /* initialize LMA */
2015         if (fid_is_idif(fid) || (fid_is_norm(fid) && osd->od_is_ost))
2016                 compat |= LMAC_FID_ON_OST;
2017         lustre_lma_init(lma, fid, compat, 0);
2018         lustre_lma_swab(lma);
2019         rc = -nvlist_add_byte_array(obj->oo_sa_xattr, XATTR_NAME_LMA,
2020                                     (uchar_t *)lma, sizeof(*lma));
2021         if (rc)
2022                 GOTO(out, rc);
2023
2024         /* configure new osd object */
2025         obj->oo_parent = parent != 0 ? parent : zapid;
2026         obj->oo_late_attr_set = 1;
2027         rc = __osd_sa_xattr_schedule_update(env, obj, oh);
2028         if (rc)
2029                 GOTO(out, rc);
2030
2031         /* XXX: oo_lma_flags */
2032         obj->oo_dt.do_lu.lo_header->loh_attr |= obj->oo_attr.la_mode & S_IFMT;
2033         if (likely(!fid_is_acct(lu_object_fid(&obj->oo_dt.do_lu))))
2034                 /* no body operations for accounting objects */
2035                 obj->oo_dt.do_body_ops = &osd_body_ops;
2036
2037         osd_idc_find_and_init(env, osd, obj);
2038
2039 out:
2040         if (unlikely(rc && dn)) {
2041                 dmu_object_free(osd->od_os, dn->dn_object, oh->ot_tx);
2042                 osd_dnode_rele(dn);
2043                 obj->oo_dn = NULL;
2044         } else if (!rc) {
2045                 obj->oo_dt.do_lu.lo_header->loh_attr |= LOHA_EXISTS;
2046         }
2047         up_write(&obj->oo_guard);
2048         RETURN(rc);
2049 }
2050
2051 static int osd_declare_ref_add(const struct lu_env *env, struct dt_object *dt,
2052                                struct thandle *th)
2053 {
2054         osd_idc_find_and_init(env, osd_dev(dt->do_lu.lo_dev), osd_dt_obj(dt));
2055         return osd_declare_attr_set(env, dt, NULL, th);
2056 }
2057
2058 /*
2059  * Concurrency: @dt is write locked.
2060  */
2061 static int osd_ref_add(const struct lu_env *env, struct dt_object *dt,
2062                        struct thandle *handle)
2063 {
2064         struct osd_object       *obj = osd_dt_obj(dt);
2065         struct osd_thandle      *oh;
2066         struct osd_device       *osd = osd_obj2dev(obj);
2067         uint64_t                 nlink;
2068         int rc;
2069
2070         ENTRY;
2071
2072         down_read(&obj->oo_guard);
2073         if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
2074                 GOTO(out, rc = -ENOENT);
2075
2076         LASSERT(osd_invariant(obj));
2077         LASSERT(obj->oo_sa_hdl != NULL);
2078
2079         oh = container_of(handle, struct osd_thandle, ot_super);
2080
2081         write_lock(&obj->oo_attr_lock);
2082         nlink = ++obj->oo_attr.la_nlink;
2083         write_unlock(&obj->oo_attr_lock);
2084
2085         rc = osd_object_sa_update(obj, SA_ZPL_LINKS(osd), &nlink, 8, oh);
2086
2087 out:
2088         up_read(&obj->oo_guard);
2089         RETURN(rc);
2090 }
2091
2092 static int osd_declare_ref_del(const struct lu_env *env, struct dt_object *dt,
2093                                struct thandle *handle)
2094 {
2095         osd_idc_find_and_init(env, osd_dev(dt->do_lu.lo_dev), osd_dt_obj(dt));
2096         return osd_declare_attr_set(env, dt, NULL, handle);
2097 }
2098
2099 /*
2100  * Concurrency: @dt is write locked.
2101  */
2102 static int osd_ref_del(const struct lu_env *env, struct dt_object *dt,
2103                        struct thandle *handle)
2104 {
2105         struct osd_object       *obj = osd_dt_obj(dt);
2106         struct osd_thandle      *oh;
2107         struct osd_device       *osd = osd_obj2dev(obj);
2108         uint64_t                 nlink;
2109         int                      rc;
2110
2111         ENTRY;
2112
2113         down_read(&obj->oo_guard);
2114
2115         if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
2116                 GOTO(out, rc = -ENOENT);
2117
2118         LASSERT(osd_invariant(obj));
2119         LASSERT(obj->oo_sa_hdl != NULL);
2120
2121         oh = container_of(handle, struct osd_thandle, ot_super);
2122         LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
2123
2124         write_lock(&obj->oo_attr_lock);
2125         nlink = --obj->oo_attr.la_nlink;
2126         write_unlock(&obj->oo_attr_lock);
2127
2128         rc = osd_object_sa_update(obj, SA_ZPL_LINKS(osd), &nlink, 8, oh);
2129
2130 out:
2131         up_read(&obj->oo_guard);
2132         RETURN(rc);
2133 }
2134
2135 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
2136                            __u64 start, __u64 end)
2137 {
2138         struct osd_device *osd = osd_obj2dev(osd_dt_obj(dt));
2139         uint64_t txg = 0;
2140         ENTRY;
2141
2142         if (osd->od_dt_dev.dd_rdonly)
2143                 RETURN(0);
2144
2145         txg = osd_db_dirty_txg(osd_dt_obj(dt)->oo_dn->dn_dbuf);
2146         if (txg) {
2147                 /* the object is dirty or being synced */
2148                 if (osd_object_sync_delay_us < 0)
2149                         txg_wait_synced(dmu_objset_pool(osd->od_os), txg);
2150                 else
2151                         udelay(osd_object_sync_delay_us);
2152         }
2153
2154         RETURN(0);
2155 }
2156
2157 static int osd_invalidate(const struct lu_env *env, struct dt_object *dt)
2158 {
2159         return 0;
2160 }
2161
2162 static bool osd_check_stale(struct dt_object *dt)
2163 {
2164         return false;
2165 }
2166
2167 static const struct dt_object_operations osd_obj_ops = {
2168         .do_read_lock           = osd_read_lock,
2169         .do_write_lock          = osd_write_lock,
2170         .do_read_unlock         = osd_read_unlock,
2171         .do_write_unlock        = osd_write_unlock,
2172         .do_write_locked        = osd_write_locked,
2173         .do_attr_get            = osd_attr_get,
2174         .do_declare_attr_set    = osd_declare_attr_set,
2175         .do_attr_set            = osd_attr_set,
2176         .do_ah_init             = osd_ah_init,
2177         .do_declare_create      = osd_declare_create,
2178         .do_create              = osd_create,
2179         .do_declare_destroy     = osd_declare_destroy,
2180         .do_destroy             = osd_destroy,
2181         .do_index_try           = osd_index_try,
2182         .do_declare_ref_add     = osd_declare_ref_add,
2183         .do_ref_add             = osd_ref_add,
2184         .do_declare_ref_del     = osd_declare_ref_del,
2185         .do_ref_del             = osd_ref_del,
2186         .do_xattr_get           = osd_xattr_get,
2187         .do_declare_xattr_set   = osd_declare_xattr_set,
2188         .do_xattr_set           = osd_xattr_set,
2189         .do_declare_xattr_del   = osd_declare_xattr_del,
2190         .do_xattr_del           = osd_xattr_del,
2191         .do_xattr_list          = osd_xattr_list,
2192         .do_object_sync         = osd_object_sync,
2193         .do_invalidate          = osd_invalidate,
2194         .do_check_stale         = osd_check_stale,
2195 };
2196
2197 static const struct lu_object_operations osd_lu_obj_ops = {
2198         .loo_object_init        = osd_object_init,
2199         .loo_object_delete      = osd_object_delete,
2200         .loo_object_release     = osd_object_release,
2201         .loo_object_free        = osd_object_free,
2202         .loo_object_print       = osd_object_print,
2203         .loo_object_invariant   = osd_object_invariant,
2204 };
2205
2206 static int osd_otable_it_attr_get(const struct lu_env *env,
2207                                 struct dt_object *dt,
2208                                 struct lu_attr *attr)
2209 {
2210         attr->la_valid = 0;
2211         return 0;
2212 }
2213
2214 static const struct dt_object_operations osd_obj_otable_it_ops = {
2215         .do_attr_get            = osd_otable_it_attr_get,
2216         .do_index_try           = osd_index_try,
2217 };
2218
2219 module_param(osd_object_sync_delay_us, int, 0644);
2220 MODULE_PARM_DESC(osd_object_sync_delay_us,
2221                  "If zero or larger delay N usec instead of doing object sync");