4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
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.
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).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2009, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2012, 2015, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/osd-zfs/osd_object.c
38 * Author: Alex Zhuravlev <bzzz@whamcloud.com>
39 * Author: Mike Pershin <tappro@whamcloud.com>
40 * Author: Johann Lombardi <johann@whamcloud.com>
43 #define DEBUG_SUBSYSTEM S_OSD
45 #include <lustre_ver.h>
46 #include <libcfs/libcfs.h>
47 #include <obd_support.h>
48 #include <lustre_net.h>
50 #include <obd_class.h>
51 #include <lustre_disk.h>
52 #include <lustre_fid.h>
54 #include "osd_internal.h"
56 #include <sys/dnode.h>
61 #include <sys/spa_impl.h>
62 #include <sys/zfs_znode.h>
63 #include <sys/dmu_tx.h>
64 #include <sys/dmu_objset.h>
65 #include <sys/dsl_prop.h>
66 #include <sys/sa_impl.h>
69 char *osd_obj_tag = "osd_object";
71 static struct dt_object_operations osd_obj_ops;
72 static struct lu_object_operations osd_lu_obj_ops;
73 extern struct dt_body_operations osd_body_ops;
74 static struct dt_object_operations osd_obj_otable_it_ops;
76 extern struct kmem_cache *osd_object_kmem;
79 osd_object_sa_fini(struct osd_object *obj)
82 sa_handle_destroy(obj->oo_sa_hdl);
83 obj->oo_sa_hdl = NULL;
88 osd_object_sa_init(struct osd_object *obj, struct osd_device *o)
92 LASSERT(obj->oo_sa_hdl == NULL);
93 LASSERT(obj->oo_db != NULL);
95 rc = -sa_handle_get(o->od_os, obj->oo_db->db_object, obj,
96 SA_HDL_PRIVATE, &obj->oo_sa_hdl);
100 /* Cache the xattr object id, valid for the life of the object */
101 rc = -sa_lookup(obj->oo_sa_hdl, SA_ZPL_XATTR(o), &obj->oo_xattr, 8);
103 obj->oo_xattr = ZFS_NO_OBJECT;
106 osd_object_sa_fini(obj);
113 * Add object to list of dirty objects in tx handle.
116 osd_object_sa_dirty_add(struct osd_object *obj, struct osd_thandle *oh)
118 if (!list_empty(&obj->oo_sa_linkage))
121 down(&oh->ot_sa_lock);
122 write_lock(&obj->oo_attr_lock);
123 if (likely(list_empty(&obj->oo_sa_linkage)))
124 list_add(&obj->oo_sa_linkage, &oh->ot_sa_list);
125 write_unlock(&obj->oo_attr_lock);
130 * Release spill block dbuf hold for all dirty SAs.
132 void osd_object_sa_dirty_rele(struct osd_thandle *oh)
134 struct osd_object *obj;
136 down(&oh->ot_sa_lock);
137 while (!list_empty(&oh->ot_sa_list)) {
138 obj = list_entry(oh->ot_sa_list.next,
139 struct osd_object, oo_sa_linkage);
140 sa_spill_rele(obj->oo_sa_hdl);
141 write_lock(&obj->oo_attr_lock);
142 list_del_init(&obj->oo_sa_linkage);
143 write_unlock(&obj->oo_attr_lock);
149 * Update the SA and add the object to the dirty list.
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)
156 LASSERT(obj->oo_sa_hdl != NULL);
157 LASSERT(oh->ot_tx != NULL);
159 rc = -sa_update(obj->oo_sa_hdl, type, buf, buflen, oh->ot_tx);
160 osd_object_sa_dirty_add(obj, oh);
166 * Bulk update the SA and add the object to the dirty list.
169 osd_object_sa_bulk_update(struct osd_object *obj, sa_bulk_attr_t *attrs,
170 int count, struct osd_thandle *oh)
174 LASSERT(obj->oo_sa_hdl != NULL);
175 LASSERT(oh->ot_tx != NULL);
177 rc = -sa_bulk_update(obj->oo_sa_hdl, attrs, count, oh->ot_tx);
178 osd_object_sa_dirty_add(obj, oh);
184 * Retrieve the attributes of a DMU object
186 int __osd_object_attr_get(const struct lu_env *env, struct osd_device *o,
187 struct osd_object *obj, struct lu_attr *la)
189 struct osa_attr *osa = &osd_oti_get(env)->oti_osa;
191 sa_bulk_attr_t *bulk;
196 LASSERT(obj->oo_db != NULL);
198 rc = -sa_handle_get(o->od_os, obj->oo_db->db_object, NULL,
199 SA_HDL_PRIVATE, &sa_hdl);
203 OBD_ALLOC(bulk, sizeof(sa_bulk_attr_t) * 9);
205 GOTO(out_sa, rc = -ENOMEM);
207 la->la_valid |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE | LA_TYPE |
208 LA_SIZE | LA_UID | LA_GID | LA_FLAGS | LA_NLINK;
210 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(o), NULL, osa->atime, 16);
211 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(o), NULL, osa->mtime, 16);
212 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(o), NULL, osa->ctime, 16);
213 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(o), NULL, &osa->mode, 8);
214 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(o), NULL, &osa->size, 8);
215 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(o), NULL, &osa->nlink, 8);
216 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(o), NULL, &osa->uid, 8);
217 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(o), NULL, &osa->gid, 8);
218 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(o), NULL, &osa->flags, 8);
220 rc = -sa_bulk_lookup(sa_hdl, bulk, cnt);
224 la->la_atime = osa->atime[0];
225 la->la_mtime = osa->mtime[0];
226 la->la_ctime = osa->ctime[0];
227 la->la_mode = osa->mode;
228 la->la_uid = osa->uid;
229 la->la_gid = osa->gid;
230 la->la_nlink = osa->nlink;
231 la->la_flags = attrs_zfs2fs(osa->flags);
232 la->la_size = osa->size;
234 /* Try to get extra flag from LMA. Right now, only LMAI_ORPHAN
235 * flags is stored in LMA, and it is only for orphan directory */
236 if (S_ISDIR(la->la_mode) && dt_object_exists(&obj->oo_dt)) {
237 struct osd_thread_info *info = osd_oti_get(env);
238 struct lustre_mdt_attrs *lma;
241 lma = (struct lustre_mdt_attrs *)info->oti_buf;
243 buf.lb_len = sizeof(info->oti_buf);
244 rc = osd_xattr_get(env, &obj->oo_dt, &buf, XATTR_NAME_LMA);
247 lma->lma_incompat = le32_to_cpu(lma->lma_incompat);
249 lma_to_lustre_flags(lma->lma_incompat);
251 } else if (rc == -ENODATA) {
256 if (S_ISCHR(la->la_mode) || S_ISBLK(la->la_mode)) {
257 rc = -sa_lookup(sa_hdl, SA_ZPL_RDEV(o), &osa->rdev, 8);
260 la->la_rdev = osa->rdev;
261 la->la_valid |= LA_RDEV;
264 OBD_FREE(bulk, sizeof(sa_bulk_attr_t) * 9);
266 sa_handle_destroy(sa_hdl);
271 int __osd_obj2dbuf(const struct lu_env *env, objset_t *os,
272 uint64_t oid, dmu_buf_t **dbp)
274 dmu_object_info_t *doi = &osd_oti_get(env)->oti_doi;
277 rc = -sa_buf_hold(os, oid, osd_obj_tag, dbp);
281 dmu_object_info_from_db(*dbp, doi);
282 if (unlikely (oid != DMU_USERUSED_OBJECT &&
283 oid != DMU_GROUPUSED_OBJECT && doi->doi_bonus_type != DMU_OT_SA)) {
284 sa_buf_rele(*dbp, osd_obj_tag);
290 LASSERT((*dbp)->db_object == oid);
291 LASSERT((*dbp)->db_offset == -1);
292 LASSERT((*dbp)->db_data != NULL);
298 * Concurrency: no concurrent access is possible that early in object
301 struct lu_object *osd_object_alloc(const struct lu_env *env,
302 const struct lu_object_header *hdr,
305 struct osd_object *mo;
307 OBD_SLAB_ALLOC_PTR_GFP(mo, osd_object_kmem, GFP_NOFS);
311 l = &mo->oo_dt.do_lu;
312 dt_object_init(&mo->oo_dt, NULL, d);
313 mo->oo_dt.do_ops = &osd_obj_ops;
314 l->lo_ops = &osd_lu_obj_ops;
315 INIT_LIST_HEAD(&mo->oo_sa_linkage);
316 INIT_LIST_HEAD(&mo->oo_unlinked_linkage);
317 init_rwsem(&mo->oo_sem);
318 sema_init(&mo->oo_guard, 1);
319 rwlock_init(&mo->oo_attr_lock);
320 mo->oo_destroy = OSD_DESTROY_NONE;
328 * Concurrency: shouldn't matter.
330 int osd_object_init0(const struct lu_env *env, struct osd_object *obj)
332 struct osd_device *osd = osd_obj2dev(obj);
333 const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
337 if (obj->oo_db == NULL)
342 rc = osd_object_sa_init(obj, osd);
346 /* cache attrs in object */
347 rc = __osd_object_attr_get(env, osd, obj, &obj->oo_attr);
351 if (likely(!fid_is_acct(fid)))
352 /* no body operations for accounting objects */
353 obj->oo_dt.do_body_ops = &osd_body_ops;
356 * initialize object before marking it existing
358 obj->oo_dt.do_lu.lo_header->loh_attr |= obj->oo_attr.la_mode & S_IFMT;
361 obj->oo_dt.do_lu.lo_header->loh_attr |= LOHA_EXISTS;
366 static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
368 struct osd_thread_info *info = osd_oti_get(env);
371 struct lustre_mdt_attrs *lma;
374 CLASSERT(sizeof(info->oti_buf) >= sizeof(*lma));
375 lma = (struct lustre_mdt_attrs *)info->oti_buf;
377 buf.lb_len = sizeof(info->oti_buf);
379 rc = osd_xattr_get(env, &obj->oo_dt, &buf, XATTR_NAME_LMA);
382 lustre_lma_swab(lma);
383 if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
384 CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT))) {
385 CWARN("%s: unsupported incompat LMA feature(s) %#x for "
386 "fid = "DFID"\n", osd_obj2dev(obj)->od_svname,
387 lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
388 PFID(lu_object_fid(&obj->oo_dt.do_lu)));
391 } else if (rc == -ENODATA) {
392 /* haven't initialize LMA xattr */
400 * Concurrency: no concurrent access is possible that early in object
403 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
404 const struct lu_object_conf *conf)
406 struct osd_object *obj = osd_obj(l);
407 struct osd_device *osd = osd_obj2dev(obj);
412 LASSERT(osd_invariant(obj));
414 if (fid_is_otable_it(&l->lo_header->loh_fid)) {
415 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
416 l->lo_header->loh_attr |= LOHA_EXISTS;
420 rc = osd_fid_lookup(env, osd, lu_object_fid(l), &oid);
422 LASSERT(obj->oo_db == NULL);
423 rc = __osd_obj2dbuf(env, osd->od_os, oid, &obj->oo_db);
424 /* EEXIST will be returned if object is being deleted in ZFS */
430 CERROR("%s: lookup "DFID"/"LPX64" failed: rc = %d\n",
431 osd->od_svname, PFID(lu_object_fid(l)), oid, rc);
435 rc = osd_object_init0(env, obj);
439 rc = osd_check_lma(env, obj);
442 } else if (rc == -ENOENT) {
445 LASSERT(osd_invariant(obj));
451 * Concurrency: no concurrent access is possible that late in object
454 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
456 struct osd_object *obj = osd_obj(l);
458 LASSERT(osd_invariant(obj));
460 dt_object_fini(&obj->oo_dt);
461 OBD_SLAB_FREE_PTR(obj, osd_object_kmem);
465 osd_object_unlinked_add(struct osd_object *obj, struct osd_thandle *oh)
469 down(&obj->oo_guard);
471 LASSERT(obj->oo_destroy == OSD_DESTROY_ASYNC);
473 if (likely(list_empty(&obj->oo_unlinked_linkage))) {
474 list_add(&obj->oo_unlinked_linkage, &oh->ot_unlinked_list);
483 /* Default to max data size covered by a level-1 indirect block */
484 static unsigned long osd_sync_destroy_max_size =
485 1UL << (DN_MAX_INDBLKSHIFT - SPA_BLKPTRSHIFT + SPA_MAXBLOCKSHIFT);
486 CFS_MODULE_PARM(osd_sync_destroy_max_size, "ul", ulong, 0444,
487 "Maximum object size to use synchronous destroy.");
490 osd_object_set_destroy_type(struct osd_object *obj)
493 * Lock-less OST_WRITE can race with OST_DESTROY, so set destroy type
494 * only once and use it consistently thereafter.
496 down(&obj->oo_guard);
497 if (obj->oo_destroy == OSD_DESTROY_NONE) {
498 if (obj->oo_attr.la_size <= osd_sync_destroy_max_size)
499 obj->oo_destroy = OSD_DESTROY_SYNC;
500 else /* Larger objects are destroyed asynchronously */
501 obj->oo_destroy = OSD_DESTROY_ASYNC;
506 static int osd_declare_object_destroy(const struct lu_env *env,
507 struct dt_object *dt,
510 char *buf = osd_oti_get(env)->oti_str;
511 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
512 struct osd_object *obj = osd_dt_obj(dt);
513 struct osd_device *osd = osd_obj2dev(obj);
514 struct osd_thandle *oh;
520 LASSERT(dt_object_exists(dt));
522 oh = container_of0(th, struct osd_thandle, ot_super);
523 LASSERT(oh->ot_tx != NULL);
525 /* declare that we'll remove object from fid-dnode mapping */
526 zapid = osd_get_name_n_idx(env, osd, fid, buf);
527 dmu_tx_hold_bonus(oh->ot_tx, zapid);
528 dmu_tx_hold_zap(oh->ot_tx, zapid, FALSE, buf);
530 osd_declare_xattrs_destroy(env, obj, oh);
532 /* declare that we'll remove object from inode accounting ZAPs */
533 dmu_tx_hold_bonus(oh->ot_tx, osd->od_iusr_oid);
534 dmu_tx_hold_zap(oh->ot_tx, osd->od_iusr_oid, FALSE, buf);
535 dmu_tx_hold_bonus(oh->ot_tx, osd->od_igrp_oid);
536 dmu_tx_hold_zap(oh->ot_tx, osd->od_igrp_oid, FALSE, buf);
539 rc = osd_declare_quota(env, osd, obj->oo_attr.la_uid,
540 obj->oo_attr.la_gid, -1, oh, false, NULL, false);
544 /* data to be truncated */
545 rc = osd_declare_quota(env, osd, obj->oo_attr.la_uid,
546 obj->oo_attr.la_gid, 0, oh, true, NULL, false);
550 osd_object_set_destroy_type(obj);
551 if (obj->oo_destroy == OSD_DESTROY_SYNC)
552 dmu_tx_hold_free(oh->ot_tx, obj->oo_db->db_object,
555 dmu_tx_hold_zap(oh->ot_tx, osd->od_unlinkedid, TRUE, NULL);
560 static int osd_object_destroy(const struct lu_env *env,
561 struct dt_object *dt, struct thandle *th)
563 char *buf = osd_oti_get(env)->oti_str;
564 struct osd_object *obj = osd_dt_obj(dt);
565 struct osd_device *osd = osd_obj2dev(obj);
566 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
567 struct osd_thandle *oh;
572 LASSERT(obj->oo_db != NULL);
573 LASSERT(dt_object_exists(dt));
574 LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
576 oh = container_of0(th, struct osd_thandle, ot_super);
578 LASSERT(oh->ot_tx != NULL);
580 /* remove obj ref from index dir (it depends) */
581 zapid = osd_get_name_n_idx(env, osd, fid, buf);
582 rc = -zap_remove(osd->od_os, zapid, buf, oh->ot_tx);
584 CERROR("%s: zap_remove(%s) failed: rc = %d\n",
585 osd->od_svname, buf, rc);
589 rc = osd_xattrs_destroy(env, obj, oh);
591 CERROR("%s: cannot destroy xattrs for %s: rc = %d\n",
592 osd->od_svname, buf, rc);
596 /* Remove object from inode accounting. It is not fatal for the destroy
597 * operation if something goes wrong while updating accounting, but we
598 * still log an error message to notify the administrator */
599 rc = -zap_increment_int(osd->od_os, osd->od_iusr_oid,
600 obj->oo_attr.la_uid, -1, oh->ot_tx);
602 CERROR("%s: failed to remove "DFID" from accounting ZAP for usr"
603 " %d: rc = %d\n", osd->od_svname, PFID(fid),
604 obj->oo_attr.la_uid, rc);
605 rc = -zap_increment_int(osd->od_os, osd->od_igrp_oid,
606 obj->oo_attr.la_gid, -1, oh->ot_tx);
608 CERROR("%s: failed to remove "DFID" from accounting ZAP for grp"
609 " %d: rc = %d\n", osd->od_svname, PFID(fid),
610 obj->oo_attr.la_gid, rc);
612 oid = obj->oo_db->db_object;
613 if (unlikely(obj->oo_destroy == OSD_DESTROY_NONE)) {
614 /* this may happen if the destroy wasn't declared
615 * e.g. when the object is created and then destroyed
616 * in the same transaction - we don't need additional
617 * space for destroy specifically */
618 LASSERT(obj->oo_attr.la_size <= osd_sync_destroy_max_size);
619 rc = -dmu_object_free(osd->od_os, oid, oh->ot_tx);
621 CERROR("%s: failed to free %s "LPU64": rc = %d\n",
622 osd->od_svname, buf, oid, rc);
623 } else if (obj->oo_destroy == OSD_DESTROY_SYNC) {
624 rc = -dmu_object_free(osd->od_os, oid, oh->ot_tx);
626 CERROR("%s: failed to free %s "LPU64": rc = %d\n",
627 osd->od_svname, buf, oid, rc);
628 } else { /* asynchronous destroy */
629 rc = osd_object_unlinked_add(obj, oh);
633 rc = -zap_add_int(osd->od_os, osd->od_unlinkedid,
636 CERROR("%s: zap_add_int() failed %s "LPU64": rc = %d\n",
637 osd->od_svname, buf, oid, rc);
641 /* not needed in the cache anymore */
642 set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
644 obj->oo_destroyed = 1;
648 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
650 struct osd_object *obj = osd_obj(l);
652 if (obj->oo_db != NULL) {
653 osd_object_sa_fini(obj);
654 if (obj->oo_sa_xattr) {
655 nvlist_free(obj->oo_sa_xattr);
656 obj->oo_sa_xattr = NULL;
658 sa_buf_rele(obj->oo_db, osd_obj_tag);
659 list_del(&obj->oo_sa_linkage);
665 * Concurrency: ->loo_object_release() is called under site spin-lock.
667 static void osd_object_release(const struct lu_env *env,
673 * Concurrency: shouldn't matter.
675 static int osd_object_print(const struct lu_env *env, void *cookie,
676 lu_printer_t p, const struct lu_object *l)
678 struct osd_object *o = osd_obj(l);
680 return (*p)(env, cookie, LUSTRE_OSD_ZFS_NAME"-object@%p", o);
683 static void osd_object_read_lock(const struct lu_env *env,
684 struct dt_object *dt, unsigned role)
686 struct osd_object *obj = osd_dt_obj(dt);
688 LASSERT(osd_invariant(obj));
690 down_read_nested(&obj->oo_sem, role);
693 static void osd_object_write_lock(const struct lu_env *env,
694 struct dt_object *dt, unsigned role)
696 struct osd_object *obj = osd_dt_obj(dt);
698 LASSERT(osd_invariant(obj));
700 down_write_nested(&obj->oo_sem, role);
703 static void osd_object_read_unlock(const struct lu_env *env,
704 struct dt_object *dt)
706 struct osd_object *obj = osd_dt_obj(dt);
708 LASSERT(osd_invariant(obj));
709 up_read(&obj->oo_sem);
712 static void osd_object_write_unlock(const struct lu_env *env,
713 struct dt_object *dt)
715 struct osd_object *obj = osd_dt_obj(dt);
717 LASSERT(osd_invariant(obj));
718 up_write(&obj->oo_sem);
721 static int osd_object_write_locked(const struct lu_env *env,
722 struct dt_object *dt)
724 struct osd_object *obj = osd_dt_obj(dt);
727 LASSERT(osd_invariant(obj));
729 if (down_write_trylock(&obj->oo_sem)) {
731 up_write(&obj->oo_sem);
736 static int osd_attr_get(const struct lu_env *env,
737 struct dt_object *dt,
738 struct lu_attr *attr)
740 struct osd_object *obj = osd_dt_obj(dt);
744 if (unlikely(!dt_object_exists(dt)))
746 if (unlikely(obj->oo_destroyed))
749 LASSERT(osd_invariant(obj));
752 read_lock(&obj->oo_attr_lock);
753 *attr = obj->oo_attr;
754 if (obj->oo_lma_flags & LUSTRE_ORPHAN_FL)
755 attr->la_flags |= LUSTRE_ORPHAN_FL;
756 read_unlock(&obj->oo_attr_lock);
758 /* with ZFS_DEBUG zrl_add_debug() called by DB_DNODE_ENTER()
759 * from within sa_object_size() can block on a mutex, so
760 * we can't call sa_object_size() holding rwlock */
761 sa_object_size(obj->oo_sa_hdl, &blksize, &blocks);
762 /* we do not control size of indices, so always calculate
763 * it from number of blocks reported by DMU */
764 if (S_ISDIR(attr->la_mode))
765 attr->la_size = 512 * blocks;
766 /* Block size may be not set; suggest maximal I/O transfers. */
768 blksize = osd_spa_maxblocksize(
769 dmu_objset_spa(osd_obj2dev(obj)->od_os));
771 attr->la_blksize = blksize;
772 attr->la_blocks = blocks;
773 attr->la_valid |= LA_BLOCKS | LA_BLKSIZE;
778 /* Simple wrapper on top of qsd API which implement quota transfer for osd
779 * setattr needs. As a reminder, only the root user can change ownership of
780 * a file, that's why EDQUOT & EINPROGRESS errors are discarded */
781 static inline int qsd_transfer(const struct lu_env *env,
782 struct qsd_instance *qsd,
783 struct lquota_trans *trans, int qtype,
784 __u64 orig_id, __u64 new_id, __u64 bspace,
785 struct lquota_id_info *qi)
789 if (unlikely(qsd == NULL))
792 LASSERT(qtype >= 0 && qtype < MAXQUOTAS);
793 qi->lqi_type = qtype;
795 /* inode accounting */
796 qi->lqi_is_blk = false;
798 /* one more inode for the new owner ... */
799 qi->lqi_id.qid_uid = new_id;
801 rc = qsd_op_begin(env, qsd, trans, qi, NULL);
802 if (rc == -EDQUOT || rc == -EINPROGRESS)
807 /* and one less inode for the current id */
808 qi->lqi_id.qid_uid = orig_id;;
810 /* can't get EDQUOT when reducing usage */
811 rc = qsd_op_begin(env, qsd, trans, qi, NULL);
812 if (rc == -EINPROGRESS)
817 /* block accounting */
818 qi->lqi_is_blk = true;
820 /* more blocks for the new owner ... */
821 qi->lqi_id.qid_uid = new_id;
822 qi->lqi_space = bspace;
823 rc = qsd_op_begin(env, qsd, trans, qi, NULL);
824 if (rc == -EDQUOT || rc == -EINPROGRESS)
829 /* and finally less blocks for the current owner */
830 qi->lqi_id.qid_uid = orig_id;
831 qi->lqi_space = -bspace;
832 rc = qsd_op_begin(env, qsd, trans, qi, NULL);
833 /* can't get EDQUOT when reducing usage */
834 if (rc == -EINPROGRESS)
839 static int osd_declare_attr_set(const struct lu_env *env,
840 struct dt_object *dt,
841 const struct lu_attr *attr,
842 struct thandle *handle)
844 struct osd_thread_info *info = osd_oti_get(env);
845 char *buf = osd_oti_get(env)->oti_str;
846 struct osd_object *obj = osd_dt_obj(dt);
847 struct osd_device *osd = osd_obj2dev(obj);
848 struct osd_thandle *oh;
854 if (!dt_object_exists(dt)) {
855 /* XXX: sanity check that object creation is declared */
859 LASSERT(handle != NULL);
860 LASSERT(osd_invariant(obj));
862 oh = container_of0(handle, struct osd_thandle, ot_super);
864 LASSERT(obj->oo_sa_hdl != NULL);
865 LASSERT(oh->ot_tx != NULL);
866 dmu_tx_hold_sa(oh->ot_tx, obj->oo_sa_hdl, 0);
867 if (oh->ot_tx->tx_err != 0)
868 RETURN(-oh->ot_tx->tx_err);
870 sa_object_size(obj->oo_sa_hdl, &blksize, &bspace);
871 bspace = toqb(bspace * blksize);
873 __osd_xattr_declare_set(env, obj, sizeof(struct lustre_mdt_attrs),
876 if (attr && attr->la_valid & LA_UID) {
877 /* account for user inode tracking ZAP update */
878 dmu_tx_hold_bonus(oh->ot_tx, osd->od_iusr_oid);
879 dmu_tx_hold_zap(oh->ot_tx, osd->od_iusr_oid, TRUE, buf);
881 /* quota enforcement for user */
882 if (attr->la_uid != obj->oo_attr.la_uid) {
883 rc = qsd_transfer(env, osd->od_quota_slave,
884 &oh->ot_quota_trans, USRQUOTA,
885 obj->oo_attr.la_uid, attr->la_uid,
886 bspace, &info->oti_qi);
891 if (attr && attr->la_valid & LA_GID) {
892 /* account for user inode tracking ZAP update */
893 dmu_tx_hold_bonus(oh->ot_tx, osd->od_igrp_oid);
894 dmu_tx_hold_zap(oh->ot_tx, osd->od_igrp_oid, TRUE, buf);
896 /* quota enforcement for group */
897 if (attr->la_gid != obj->oo_attr.la_gid) {
898 rc = qsd_transfer(env, osd->od_quota_slave,
899 &oh->ot_quota_trans, GRPQUOTA,
900 obj->oo_attr.la_gid, attr->la_gid,
901 bspace, &info->oti_qi);
911 * Set the attributes of an object
913 * The transaction passed to this routine must have
914 * dmu_tx_hold_bonus(tx, oid) called and then assigned
915 * to a transaction group.
917 static int osd_attr_set(const struct lu_env *env, struct dt_object *dt,
918 const struct lu_attr *la, struct thandle *handle)
920 struct osd_thread_info *info = osd_oti_get(env);
921 struct osd_object *obj = osd_dt_obj(dt);
922 struct osd_device *osd = osd_obj2dev(obj);
923 struct osd_thandle *oh;
924 struct osa_attr *osa = &info->oti_osa;
925 sa_bulk_attr_t *bulk;
926 __u64 valid = la->la_valid;
931 LASSERT(handle != NULL);
932 LASSERT(dt_object_exists(dt));
933 LASSERT(osd_invariant(obj));
934 LASSERT(obj->oo_sa_hdl);
936 oh = container_of0(handle, struct osd_thandle, ot_super);
937 /* Assert that the transaction has been assigned to a
938 transaction group. */
939 LASSERT(oh->ot_tx->tx_txg != 0);
941 /* Only allow set size for regular file */
942 if (!S_ISREG(dt->do_lu.lo_header->loh_attr))
943 valid &= ~(LA_SIZE | LA_BLOCKS);
948 if (valid & LA_FLAGS) {
949 struct lustre_mdt_attrs *lma;
952 if (la->la_flags & LUSTRE_LMA_FL_MASKS) {
953 CLASSERT(sizeof(info->oti_buf) >= sizeof(*lma));
954 lma = (struct lustre_mdt_attrs *)&info->oti_buf;
956 buf.lb_len = sizeof(info->oti_buf);
957 rc = osd_xattr_get(env, &obj->oo_dt, &buf,
961 le32_to_cpu(lma->lma_incompat);
963 lustre_to_lma_flags(la->la_flags);
965 cpu_to_le32(lma->lma_incompat);
967 buf.lb_len = sizeof(*lma);
968 rc = osd_xattr_set_internal(env, obj, &buf,
974 CWARN("%s: failed to set LMA flags: rc = %d\n",
981 OBD_ALLOC(bulk, sizeof(sa_bulk_attr_t) * 10);
985 /* do both accounting updates outside oo_attr_lock below */
986 if ((valid & LA_UID) && (la->la_uid != obj->oo_attr.la_uid)) {
987 /* Update user accounting. Failure isn't fatal, but we still
988 * log an error message */
989 rc = -zap_increment_int(osd->od_os, osd->od_iusr_oid,
990 la->la_uid, 1, oh->ot_tx);
992 CERROR("%s: failed to update accounting ZAP for user "
993 "%d (%d)\n", osd->od_svname, la->la_uid, rc);
994 rc = -zap_increment_int(osd->od_os, osd->od_iusr_oid,
995 obj->oo_attr.la_uid, -1, oh->ot_tx);
997 CERROR("%s: failed to update accounting ZAP for user "
998 "%d (%d)\n", osd->od_svname,
999 obj->oo_attr.la_uid, rc);
1001 if ((valid & LA_GID) && (la->la_gid != obj->oo_attr.la_gid)) {
1002 /* Update group accounting. Failure isn't fatal, but we still
1003 * log an error message */
1004 rc = -zap_increment_int(osd->od_os, osd->od_igrp_oid,
1005 la->la_gid, 1, oh->ot_tx);
1007 CERROR("%s: failed to update accounting ZAP for user "
1008 "%d (%d)\n", osd->od_svname, la->la_gid, rc);
1009 rc = -zap_increment_int(osd->od_os, osd->od_igrp_oid,
1010 obj->oo_attr.la_gid, -1, oh->ot_tx);
1012 CERROR("%s: failed to update accounting ZAP for user "
1013 "%d (%d)\n", osd->od_svname,
1014 obj->oo_attr.la_gid, rc);
1017 write_lock(&obj->oo_attr_lock);
1019 if (valid & LA_ATIME) {
1020 osa->atime[0] = obj->oo_attr.la_atime = la->la_atime;
1021 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(osd), NULL,
1024 if (valid & LA_MTIME) {
1025 osa->mtime[0] = obj->oo_attr.la_mtime = la->la_mtime;
1026 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(osd), NULL,
1029 if (valid & LA_CTIME) {
1030 osa->ctime[0] = obj->oo_attr.la_ctime = la->la_ctime;
1031 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(osd), NULL,
1034 if (valid & LA_MODE) {
1035 /* mode is stored along with type, so read it first */
1036 obj->oo_attr.la_mode = (obj->oo_attr.la_mode & S_IFMT) |
1037 (la->la_mode & ~S_IFMT);
1038 osa->mode = obj->oo_attr.la_mode;
1039 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(osd), NULL,
1042 if (valid & LA_SIZE) {
1043 osa->size = obj->oo_attr.la_size = la->la_size;
1044 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(osd), NULL,
1047 if (valid & LA_NLINK) {
1048 osa->nlink = obj->oo_attr.la_nlink = la->la_nlink;
1049 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(osd), NULL,
1052 if (valid & LA_RDEV) {
1053 osa->rdev = obj->oo_attr.la_rdev = la->la_rdev;
1054 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_RDEV(osd), NULL,
1057 if (valid & LA_FLAGS) {
1058 osa->flags = attrs_fs2zfs(la->la_flags);
1059 /* many flags are not supported by zfs, so ensure a good cached
1061 obj->oo_attr.la_flags = attrs_zfs2fs(osa->flags);
1062 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(osd), NULL,
1065 if (valid & LA_UID) {
1066 osa->uid = obj->oo_attr.la_uid = la->la_uid;
1067 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(osd), NULL,
1070 if (valid & LA_GID) {
1071 osa->gid = obj->oo_attr.la_gid = la->la_gid;
1072 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(osd), NULL,
1075 obj->oo_attr.la_valid |= valid;
1076 write_unlock(&obj->oo_attr_lock);
1078 rc = osd_object_sa_bulk_update(obj, bulk, cnt, oh);
1080 OBD_FREE(bulk, sizeof(sa_bulk_attr_t) * 10);
1087 * XXX temporary solution.
1090 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
1091 struct dt_object *parent, struct dt_object *child,
1096 ah->dah_parent = parent;
1097 ah->dah_mode = child_mode;
1100 static int osd_declare_object_create(const struct lu_env *env,
1101 struct dt_object *dt,
1102 struct lu_attr *attr,
1103 struct dt_allocation_hint *hint,
1104 struct dt_object_format *dof,
1105 struct thandle *handle)
1107 char *buf = osd_oti_get(env)->oti_str;
1108 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1109 struct osd_object *obj = osd_dt_obj(dt);
1110 struct osd_device *osd = osd_obj2dev(obj);
1111 struct osd_thandle *oh;
1118 switch (dof->dof_type) {
1122 if (obj->oo_dt.do_body_ops == NULL)
1123 obj->oo_dt.do_body_ops = &osd_body_ops;
1129 LASSERT(handle != NULL);
1130 oh = container_of0(handle, struct osd_thandle, ot_super);
1131 LASSERT(oh->ot_tx != NULL);
1133 switch (dof->dof_type) {
1135 dt->do_index_ops = &osd_dir_ops;
1137 /* for zap create */
1138 dmu_tx_hold_zap(oh->ot_tx, DMU_NEW_OBJECT, 1, NULL);
1143 /* first, we'll create new object */
1144 dmu_tx_hold_bonus(oh->ot_tx, DMU_NEW_OBJECT);
1152 /* and we'll add it to some mapping */
1153 zapid = osd_get_name_n_idx(env, osd, fid, buf);
1154 dmu_tx_hold_bonus(oh->ot_tx, zapid);
1155 dmu_tx_hold_zap(oh->ot_tx, zapid, TRUE, buf);
1157 /* we will also update inode accounting ZAPs */
1158 dmu_tx_hold_bonus(oh->ot_tx, osd->od_iusr_oid);
1159 dmu_tx_hold_zap(oh->ot_tx, osd->od_iusr_oid, TRUE, buf);
1160 dmu_tx_hold_bonus(oh->ot_tx, osd->od_igrp_oid);
1161 dmu_tx_hold_zap(oh->ot_tx, osd->od_igrp_oid, TRUE, buf);
1163 dmu_tx_hold_sa_create(oh->ot_tx, ZFS_SA_BASE_ATTR_SIZE);
1165 __osd_xattr_declare_set(env, obj, sizeof(struct lustre_mdt_attrs),
1166 XATTR_NAME_LMA, oh);
1168 rc = osd_declare_quota(env, osd, attr->la_uid, attr->la_gid, 1, oh,
1169 false, NULL, false);
1173 int __osd_attr_init(const struct lu_env *env, struct osd_device *osd,
1174 uint64_t oid, dmu_tx_t *tx, struct lu_attr *la,
1177 sa_bulk_attr_t *bulk;
1178 sa_handle_t *sa_hdl;
1179 struct osa_attr *osa = &osd_oti_get(env)->oti_osa;
1187 gen = dmu_tx_get_txg(tx);
1189 ZFS_TIME_ENCODE(&now, crtime);
1191 osa->atime[0] = la->la_atime;
1192 osa->ctime[0] = la->la_ctime;
1193 osa->mtime[0] = la->la_mtime;
1194 osa->mode = la->la_mode;
1195 osa->uid = la->la_uid;
1196 osa->gid = la->la_gid;
1197 osa->rdev = la->la_rdev;
1198 osa->nlink = la->la_nlink;
1199 osa->flags = attrs_fs2zfs(la->la_flags);
1200 osa->size = la->la_size;
1202 /* Now add in all of the "SA" attributes */
1203 rc = -sa_handle_get(osd->od_os, oid, NULL, SA_HDL_PRIVATE, &sa_hdl);
1207 OBD_ALLOC(bulk, sizeof(sa_bulk_attr_t) * 13);
1213 * we need to create all SA below upon object create.
1215 * XXX The attribute order matters since the accounting callback relies
1216 * on static offsets (i.e. SA_*_OFFSET, see zfs_space_delta_cb()) to
1217 * look up the UID/GID attributes. Moreover, the callback does not seem
1218 * to support the spill block.
1219 * We define attributes in the same order as SA_*_OFFSET in order to
1220 * work around the problem. See ORI-610.
1223 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(osd), NULL, &osa->mode, 8);
1224 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(osd), NULL, &osa->size, 8);
1225 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GEN(osd), NULL, &gen, 8);
1226 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(osd), NULL, &osa->uid, 8);
1227 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(osd), NULL, &osa->gid, 8);
1228 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_PARENT(osd), NULL, &parent, 8);
1229 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(osd), NULL, &osa->flags, 8);
1230 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(osd), NULL, osa->atime, 16);
1231 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(osd), NULL, osa->mtime, 16);
1232 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(osd), NULL, osa->ctime, 16);
1233 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CRTIME(osd), NULL, crtime, 16);
1234 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(osd), NULL, &osa->nlink, 8);
1235 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_RDEV(osd), NULL, &osa->rdev, 8);
1237 rc = -sa_replace_all_by_template(sa_hdl, bulk, cnt, tx);
1239 OBD_FREE(bulk, sizeof(sa_bulk_attr_t) * 13);
1241 sa_handle_destroy(sa_hdl);
1246 * The transaction passed to this routine must have
1247 * dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT) called and then assigned
1248 * to a transaction group.
1250 int __osd_object_create(const struct lu_env *env, struct osd_object *obj,
1251 dmu_buf_t **dbp, dmu_tx_t *tx, struct lu_attr *la,
1256 struct osd_device *osd = osd_obj2dev(obj);
1257 const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
1258 dmu_object_type_t type = DMU_OT_PLAIN_FILE_CONTENTS;
1260 /* Assert that the transaction has been assigned to a
1261 transaction group. */
1262 LASSERT(tx->tx_txg != 0);
1264 /* Use DMU_OTN_UINT8_METADATA for local objects so their data blocks
1265 * would get an additional ditto copy */
1266 if (unlikely(S_ISREG(la->la_mode) &&
1267 fid_seq_is_local_file(fid_seq(fid))))
1268 type = DMU_OTN_UINT8_METADATA;
1270 /* Create a new DMU object. */
1271 oid = dmu_object_alloc(osd->od_os, type, 0,
1272 DMU_OT_SA, DN_MAX_BONUSLEN, tx);
1273 rc = -sa_buf_hold(osd->od_os, oid, osd_obj_tag, dbp);
1274 LASSERTF(rc == 0, "sa_buf_hold "LPU64" failed: %d\n", oid, rc);
1276 LASSERT(la->la_valid & LA_MODE);
1280 rc = __osd_attr_init(env, osd, oid, tx, la, parent);
1282 sa_buf_rele(*dbp, osd_obj_tag);
1284 dmu_object_free(osd->od_os, oid, tx);
1292 * The transaction passed to this routine must have
1293 * dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, ...) called and then assigned
1294 * to a transaction group.
1296 * Using ZAP_FLAG_HASH64 will force the ZAP to always be a FAT ZAP.
1297 * This is fine for directories today, because storing the FID in the dirent
1298 * will also require a FAT ZAP. If there is a new type of micro ZAP created
1299 * then we might need to re-evaluate the use of this flag and instead do
1300 * a conversion from the different internal ZAP hash formats being used. */
1301 int __osd_zap_create(const struct lu_env *env, struct osd_device *osd,
1302 dmu_buf_t **zap_dbp, dmu_tx_t *tx,
1303 struct lu_attr *la, uint64_t parent, zap_flags_t flags)
1308 /* Assert that the transaction has been assigned to a
1309 transaction group. */
1310 LASSERT(tx->tx_txg != 0);
1312 oid = zap_create_flags(osd->od_os, 0, flags | ZAP_FLAG_HASH64,
1313 DMU_OT_DIRECTORY_CONTENTS,
1314 14, /* == ZFS fzap_default_block_shift */
1315 DN_MAX_INDBLKSHIFT, /* indirect block shift */
1316 DMU_OT_SA, DN_MAX_BONUSLEN, tx);
1318 rc = -sa_buf_hold(osd->od_os, oid, osd_obj_tag, zap_dbp);
1322 LASSERT(la->la_valid & LA_MODE);
1326 return __osd_attr_init(env, osd, oid, tx, la, parent);
1329 static dmu_buf_t *osd_mkidx(const struct lu_env *env, struct osd_object *obj,
1330 struct lu_attr *la, uint64_t parent,
1331 struct osd_thandle *oh)
1336 /* Index file should be created as regular file in order not to confuse
1337 * ZPL which could interpret them as directory.
1338 * We set ZAP_FLAG_UINT64_KEY to let ZFS know than we are going to use
1340 LASSERT(S_ISREG(la->la_mode));
1341 rc = __osd_zap_create(env, osd_obj2dev(obj), &db, oh->ot_tx, la, parent,
1342 ZAP_FLAG_UINT64_KEY);
1348 static dmu_buf_t *osd_mkdir(const struct lu_env *env, struct osd_object *obj,
1349 struct lu_attr *la, uint64_t parent,
1350 struct osd_thandle *oh)
1355 LASSERT(S_ISDIR(la->la_mode));
1356 rc = __osd_zap_create(env, osd_obj2dev(obj), &db,
1357 oh->ot_tx, la, parent, 0);
1363 static dmu_buf_t *osd_mkreg(const struct lu_env *env, struct osd_object *obj,
1364 struct lu_attr *la, uint64_t parent,
1365 struct osd_thandle *oh)
1369 struct osd_device *osd = osd_obj2dev(obj);
1371 LASSERT(S_ISREG(la->la_mode));
1372 rc = __osd_object_create(env, obj, &db, oh->ot_tx, la, parent);
1377 * XXX: This heuristic is non-optimal. It would be better to
1378 * increase the blocksize up to osd->od_max_blksz during the write.
1379 * This is exactly how the ZPL behaves and it ensures that the right
1380 * blocksize is selected based on the file size rather than the
1381 * making broad assumptions based on the osd type.
1383 if (!lu_device_is_md(osd2lu_dev(osd))) {
1384 rc = -dmu_object_set_blocksize(osd->od_os, db->db_object,
1385 osd->od_max_blksz, 0, oh->ot_tx);
1387 CERROR("%s: can't change blocksize: %d\n",
1388 osd->od_svname, rc);
1396 static dmu_buf_t *osd_mksym(const struct lu_env *env, struct osd_object *obj,
1397 struct lu_attr *la, uint64_t parent,
1398 struct osd_thandle *oh)
1403 LASSERT(S_ISLNK(la->la_mode));
1404 rc = __osd_object_create(env, obj, &db, oh->ot_tx, la, parent);
1410 static dmu_buf_t *osd_mknod(const struct lu_env *env, struct osd_object *obj,
1411 struct lu_attr *la, uint64_t parent,
1412 struct osd_thandle *oh)
1417 la->la_valid = LA_MODE;
1418 if (S_ISCHR(la->la_mode) || S_ISBLK(la->la_mode))
1419 la->la_valid |= LA_RDEV;
1421 rc = __osd_object_create(env, obj, &db, oh->ot_tx, la, parent);
1427 typedef dmu_buf_t *(*osd_obj_type_f)(const struct lu_env *env,
1428 struct osd_object *obj,
1431 struct osd_thandle *oh);
1433 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
1435 osd_obj_type_f result;
1461 * Primitives for directory (i.e. ZAP) handling
1463 static inline int osd_init_lma(const struct lu_env *env, struct osd_object *obj,
1464 const struct lu_fid *fid, struct osd_thandle *oh)
1466 struct osd_thread_info *info = osd_oti_get(env);
1467 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
1471 lustre_lma_init(lma, fid, 0, 0);
1472 lustre_lma_swab(lma);
1474 buf.lb_len = sizeof(*lma);
1476 rc = osd_xattr_set_internal(env, obj, &buf, XATTR_NAME_LMA,
1477 LU_XATTR_CREATE, oh);
1483 * Concurrency: @dt is write locked.
1485 static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
1486 struct lu_attr *attr,
1487 struct dt_allocation_hint *hint,
1488 struct dt_object_format *dof,
1491 struct zpl_direntry *zde = &osd_oti_get(env)->oti_zde.lzd_reg;
1492 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1493 struct osd_object *obj = osd_dt_obj(dt);
1494 struct osd_device *osd = osd_obj2dev(obj);
1495 char *buf = osd_oti_get(env)->oti_str;
1496 struct osd_thandle *oh;
1503 /* concurrent create declarations should not see
1504 * the object inconsistent (db, attr, etc).
1505 * in regular cases acquisition should be cheap */
1506 down(&obj->oo_guard);
1508 LASSERT(osd_invariant(obj));
1509 LASSERT(!dt_object_exists(dt));
1510 LASSERT(dof != NULL);
1512 LASSERT(th != NULL);
1513 oh = container_of0(th, struct osd_thandle, ot_super);
1516 * XXX missing: Quote handling.
1519 LASSERT(obj->oo_db == NULL);
1521 /* to follow ZFS on-disk format we need
1522 * to initialize parent dnode properly */
1524 if (hint != NULL && hint->dah_parent != NULL &&
1525 !dt_object_remote(hint->dah_parent))
1526 zapid = osd_dt_obj(hint->dah_parent)->oo_db->db_object;
1528 db = osd_create_type_f(dof->dof_type)(env, obj, attr, zapid, oh);
1530 GOTO(out, rc = PTR_ERR(db));
1533 zde->zde_dnode = db->db_object;
1534 zde->zde_type = IFTODT(attr->la_mode & S_IFMT);
1536 zapid = osd_get_name_n_idx(env, osd, fid, buf);
1538 rc = -zap_add(osd->od_os, zapid, buf, 8, 1, zde, oh->ot_tx);
1542 /* Add new object to inode accounting.
1543 * Errors are not considered as fatal */
1544 rc = -zap_increment_int(osd->od_os, osd->od_iusr_oid,
1545 (attr->la_valid & LA_UID) ? attr->la_uid : 0, 1,
1548 CERROR("%s: failed to add "DFID" to accounting ZAP for usr %d "
1549 "(%d)\n", osd->od_svname, PFID(fid), attr->la_uid, rc);
1550 rc = -zap_increment_int(osd->od_os, osd->od_igrp_oid,
1551 (attr->la_valid & LA_GID) ? attr->la_gid : 0, 1,
1554 CERROR("%s: failed to add "DFID" to accounting ZAP for grp %d "
1555 "(%d)\n", osd->od_svname, PFID(fid), attr->la_gid, rc);
1557 /* configure new osd object */
1559 rc = osd_object_init0(env, obj);
1560 LASSERT(ergo(rc == 0, dt_object_exists(dt)));
1561 LASSERT(osd_invariant(obj));
1563 rc = osd_init_lma(env, obj, fid, oh);
1565 CERROR("%s: can not set LMA on "DFID": rc = %d\n",
1566 osd->od_svname, PFID(fid), rc);
1567 /* ignore errors during LMA initialization */
1576 static int osd_declare_object_ref_add(const struct lu_env *env,
1577 struct dt_object *dt,
1580 return osd_declare_attr_set(env, dt, NULL, th);
1584 * Concurrency: @dt is write locked.
1586 static int osd_object_ref_add(const struct lu_env *env,
1587 struct dt_object *dt,
1588 struct thandle *handle)
1590 struct osd_object *obj = osd_dt_obj(dt);
1591 struct osd_thandle *oh;
1592 struct osd_device *osd = osd_obj2dev(obj);
1598 if (!dt_object_exists(dt))
1601 LASSERT(osd_invariant(obj));
1602 LASSERT(obj->oo_sa_hdl != NULL);
1604 oh = container_of0(handle, struct osd_thandle, ot_super);
1606 write_lock(&obj->oo_attr_lock);
1607 nlink = ++obj->oo_attr.la_nlink;
1608 write_unlock(&obj->oo_attr_lock);
1610 rc = osd_object_sa_update(obj, SA_ZPL_LINKS(osd), &nlink, 8, oh);
1614 static int osd_declare_object_ref_del(const struct lu_env *env,
1615 struct dt_object *dt,
1616 struct thandle *handle)
1618 return osd_declare_attr_set(env, dt, NULL, handle);
1622 * Concurrency: @dt is write locked.
1624 static int osd_object_ref_del(const struct lu_env *env,
1625 struct dt_object *dt,
1626 struct thandle *handle)
1628 struct osd_object *obj = osd_dt_obj(dt);
1629 struct osd_thandle *oh;
1630 struct osd_device *osd = osd_obj2dev(obj);
1636 LASSERT(osd_invariant(obj));
1637 LASSERT(dt_object_exists(dt));
1638 LASSERT(obj->oo_sa_hdl != NULL);
1640 oh = container_of0(handle, struct osd_thandle, ot_super);
1641 LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
1643 write_lock(&obj->oo_attr_lock);
1644 nlink = --obj->oo_attr.la_nlink;
1645 write_unlock(&obj->oo_attr_lock);
1647 rc = osd_object_sa_update(obj, SA_ZPL_LINKS(osd), &nlink, 8, oh);
1651 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
1652 __u64 start, __u64 end)
1654 struct osd_device *osd = osd_obj2dev(osd_dt_obj(dt));
1657 /* XXX: no other option than syncing the whole filesystem until we
1658 * support ZIL. If the object tracked the txg that it was last
1659 * modified in, it could pass that txg here instead of "0". Maybe
1660 * the changes are already committed, so no wait is needed at all? */
1661 txg_wait_synced(dmu_objset_pool(osd->od_os), 0ULL);
1666 static struct dt_object_operations osd_obj_ops = {
1667 .do_read_lock = osd_object_read_lock,
1668 .do_write_lock = osd_object_write_lock,
1669 .do_read_unlock = osd_object_read_unlock,
1670 .do_write_unlock = osd_object_write_unlock,
1671 .do_write_locked = osd_object_write_locked,
1672 .do_attr_get = osd_attr_get,
1673 .do_declare_attr_set = osd_declare_attr_set,
1674 .do_attr_set = osd_attr_set,
1675 .do_ah_init = osd_ah_init,
1676 .do_declare_create = osd_declare_object_create,
1677 .do_create = osd_object_create,
1678 .do_declare_destroy = osd_declare_object_destroy,
1679 .do_destroy = osd_object_destroy,
1680 .do_index_try = osd_index_try,
1681 .do_declare_ref_add = osd_declare_object_ref_add,
1682 .do_ref_add = osd_object_ref_add,
1683 .do_declare_ref_del = osd_declare_object_ref_del,
1684 .do_ref_del = osd_object_ref_del,
1685 .do_xattr_get = osd_xattr_get,
1686 .do_declare_xattr_set = osd_declare_xattr_set,
1687 .do_xattr_set = osd_xattr_set,
1688 .do_declare_xattr_del = osd_declare_xattr_del,
1689 .do_xattr_del = osd_xattr_del,
1690 .do_xattr_list = osd_xattr_list,
1691 .do_object_sync = osd_object_sync,
1694 static struct lu_object_operations osd_lu_obj_ops = {
1695 .loo_object_init = osd_object_init,
1696 .loo_object_delete = osd_object_delete,
1697 .loo_object_release = osd_object_release,
1698 .loo_object_free = osd_object_free,
1699 .loo_object_print = osd_object_print,
1700 .loo_object_invariant = osd_object_invariant,
1703 static int osd_otable_it_attr_get(const struct lu_env *env,
1704 struct dt_object *dt,
1705 struct lu_attr *attr)
1711 static struct dt_object_operations osd_obj_otable_it_ops = {
1712 .do_attr_get = osd_otable_it_attr_get,
1713 .do_index_try = osd_index_try,