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.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2009, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2012, 2016, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
32 * lustre/osd-zfs/osd_object.c
34 * Author: Alex Zhuravlev <bzzz@whamcloud.com>
35 * Author: Mike Pershin <tappro@whamcloud.com>
36 * Author: Johann Lombardi <johann@whamcloud.com>
39 #define DEBUG_SUBSYSTEM S_OSD
41 #include <lustre_ver.h>
42 #include <libcfs/libcfs.h>
43 #include <obd_support.h>
44 #include <lustre_net.h>
46 #include <obd_class.h>
47 #include <lustre_disk.h>
48 #include <lustre_fid.h>
50 #include "osd_internal.h"
52 #include <sys/dnode.h>
57 #include <sys/spa_impl.h>
58 #include <sys/zfs_znode.h>
59 #include <sys/dmu_tx.h>
60 #include <sys/dmu_objset.h>
61 #include <sys/dsl_prop.h>
62 #include <sys/sa_impl.h>
65 char *osd_obj_tag = "osd_object";
67 static struct dt_object_operations osd_obj_ops;
68 static struct lu_object_operations osd_lu_obj_ops;
69 extern struct dt_body_operations osd_body_ops;
70 static struct dt_object_operations osd_obj_otable_it_ops;
72 extern struct kmem_cache *osd_object_kmem;
75 osd_object_sa_fini(struct osd_object *obj)
78 sa_handle_destroy(obj->oo_sa_hdl);
79 obj->oo_sa_hdl = NULL;
84 osd_object_sa_init(struct osd_object *obj, struct osd_device *o)
88 LASSERT(obj->oo_sa_hdl == NULL);
89 LASSERT(obj->oo_dn != NULL);
91 rc = osd_sa_handle_get(obj);
95 /* Cache the xattr object id, valid for the life of the object */
96 rc = -sa_lookup(obj->oo_sa_hdl, SA_ZPL_XATTR(o), &obj->oo_xattr, 8);
98 obj->oo_xattr = ZFS_NO_OBJECT;
101 osd_object_sa_fini(obj);
108 * Add object to list of dirty objects in tx handle.
111 osd_object_sa_dirty_add(struct osd_object *obj, struct osd_thandle *oh)
113 if (!list_empty(&obj->oo_sa_linkage))
116 down(&oh->ot_sa_lock);
117 write_lock(&obj->oo_attr_lock);
118 if (likely(list_empty(&obj->oo_sa_linkage)))
119 list_add(&obj->oo_sa_linkage, &oh->ot_sa_list);
120 write_unlock(&obj->oo_attr_lock);
125 * Release spill block dbuf hold for all dirty SAs.
127 void osd_object_sa_dirty_rele(struct osd_thandle *oh)
129 struct osd_object *obj;
131 down(&oh->ot_sa_lock);
132 while (!list_empty(&oh->ot_sa_list)) {
133 obj = list_entry(oh->ot_sa_list.next,
134 struct osd_object, oo_sa_linkage);
135 sa_spill_rele(obj->oo_sa_hdl);
136 write_lock(&obj->oo_attr_lock);
137 list_del_init(&obj->oo_sa_linkage);
138 write_unlock(&obj->oo_attr_lock);
144 * Update the SA and add the object to the dirty list.
146 int osd_object_sa_update(struct osd_object *obj, sa_attr_type_t type,
147 void *buf, uint32_t buflen, struct osd_thandle *oh)
151 LASSERT(obj->oo_sa_hdl != NULL);
152 LASSERT(oh->ot_tx != NULL);
154 rc = -sa_update(obj->oo_sa_hdl, type, buf, buflen, oh->ot_tx);
155 osd_object_sa_dirty_add(obj, oh);
161 * Bulk update the SA and add the object to the dirty list.
164 osd_object_sa_bulk_update(struct osd_object *obj, sa_bulk_attr_t *attrs,
165 int count, struct osd_thandle *oh)
169 LASSERT(obj->oo_sa_hdl != NULL);
170 LASSERT(oh->ot_tx != NULL);
172 rc = -sa_bulk_update(obj->oo_sa_hdl, attrs, count, oh->ot_tx);
173 osd_object_sa_dirty_add(obj, oh);
179 * Retrieve the attributes of a DMU object
181 int __osd_object_attr_get(const struct lu_env *env, struct osd_device *o,
182 struct osd_object *obj, struct lu_attr *la)
184 struct osa_attr *osa = &osd_oti_get(env)->oti_osa;
185 sa_bulk_attr_t *bulk = osd_oti_get(env)->oti_attr_bulk;
190 LASSERT(obj->oo_dn != NULL);
192 la->la_valid |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE | LA_TYPE |
193 LA_SIZE | LA_UID | LA_GID | LA_FLAGS | LA_NLINK;
195 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(o), NULL, osa->atime, 16);
196 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(o), NULL, osa->mtime, 16);
197 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(o), NULL, osa->ctime, 16);
198 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(o), NULL, &osa->mode, 8);
199 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(o), NULL, &osa->size, 8);
200 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(o), NULL, &osa->nlink, 8);
201 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(o), NULL, &osa->uid, 8);
202 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(o), NULL, &osa->gid, 8);
203 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(o), NULL, &osa->flags, 8);
204 LASSERT(cnt <= ARRAY_SIZE(osd_oti_get(env)->oti_attr_bulk));
206 rc = -sa_bulk_lookup(obj->oo_sa_hdl, bulk, cnt);
210 la->la_atime = osa->atime[0];
211 la->la_mtime = osa->mtime[0];
212 la->la_ctime = osa->ctime[0];
213 la->la_mode = osa->mode;
214 la->la_uid = osa->uid;
215 la->la_gid = osa->gid;
216 la->la_nlink = osa->nlink;
217 la->la_flags = attrs_zfs2fs(osa->flags);
218 la->la_size = osa->size;
220 /* Try to get extra flag from LMA. Right now, only LMAI_ORPHAN
221 * flags is stored in LMA, and it is only for orphan directory */
222 if (S_ISDIR(la->la_mode) && dt_object_exists(&obj->oo_dt)) {
223 struct osd_thread_info *info = osd_oti_get(env);
224 struct lustre_mdt_attrs *lma;
227 lma = (struct lustre_mdt_attrs *)info->oti_buf;
229 buf.lb_len = sizeof(info->oti_buf);
230 rc = osd_xattr_get(env, &obj->oo_dt, &buf, XATTR_NAME_LMA);
233 lma->lma_incompat = le32_to_cpu(lma->lma_incompat);
235 lma_to_lustre_flags(lma->lma_incompat);
237 } else if (rc == -ENODATA) {
242 if (S_ISCHR(la->la_mode) || S_ISBLK(la->la_mode)) {
243 rc = -sa_lookup(obj->oo_sa_hdl, SA_ZPL_RDEV(o), &osa->rdev, 8);
246 la->la_rdev = osa->rdev;
247 la->la_valid |= LA_RDEV;
254 int __osd_obj2dnode(objset_t *os, uint64_t oid, dnode_t **dnp)
260 rc = -dmu_bonus_hold(os, oid, osd_obj_tag, &db);
264 dbi = (dmu_buf_impl_t *)db;
266 *dnp = DB_DNODE(dbi);
267 LASSERT(*dnp != NULL);
273 * Concurrency: no concurrent access is possible that early in object
276 struct lu_object *osd_object_alloc(const struct lu_env *env,
277 const struct lu_object_header *hdr,
280 struct osd_object *mo;
282 OBD_SLAB_ALLOC_PTR_GFP(mo, osd_object_kmem, GFP_NOFS);
286 l = &mo->oo_dt.do_lu;
287 dt_object_init(&mo->oo_dt, NULL, d);
288 mo->oo_dt.do_ops = &osd_obj_ops;
289 l->lo_ops = &osd_lu_obj_ops;
290 INIT_LIST_HEAD(&mo->oo_sa_linkage);
291 INIT_LIST_HEAD(&mo->oo_unlinked_linkage);
292 init_rwsem(&mo->oo_sem);
293 init_rwsem(&mo->oo_guard);
294 rwlock_init(&mo->oo_attr_lock);
295 mo->oo_destroy = OSD_DESTROY_NONE;
303 * Concurrency: shouldn't matter.
305 int osd_object_init0(const struct lu_env *env, struct osd_object *obj)
307 struct osd_device *osd = osd_obj2dev(obj);
308 const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
312 if (obj->oo_dn == NULL)
317 rc = osd_object_sa_init(obj, osd);
321 /* cache attrs in object */
322 rc = __osd_object_attr_get(env, osd, obj, &obj->oo_attr);
326 if (likely(!fid_is_acct(fid)))
327 /* no body operations for accounting objects */
328 obj->oo_dt.do_body_ops = &osd_body_ops;
331 * initialize object before marking it existing
333 obj->oo_dt.do_lu.lo_header->loh_attr |= obj->oo_attr.la_mode & S_IFMT;
336 obj->oo_dt.do_lu.lo_header->loh_attr |= LOHA_EXISTS;
341 static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
343 struct osd_thread_info *info = osd_oti_get(env);
346 struct lustre_mdt_attrs *lma;
349 CLASSERT(sizeof(info->oti_buf) >= sizeof(*lma));
350 lma = (struct lustre_mdt_attrs *)info->oti_buf;
352 buf.lb_len = sizeof(info->oti_buf);
354 rc = osd_xattr_get(env, &obj->oo_dt, &buf, XATTR_NAME_LMA);
357 lustre_lma_swab(lma);
358 if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
359 CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT))) {
360 CWARN("%s: unsupported incompat LMA feature(s) %#x for "
361 "fid = "DFID"\n", osd_obj2dev(obj)->od_svname,
362 lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
363 PFID(lu_object_fid(&obj->oo_dt.do_lu)));
366 } else if (rc == -ENODATA) {
367 /* haven't initialize LMA xattr */
375 * Helper function to retrieve DMU object id from fid for accounting object
377 static dnode_t *osd_quota_fid2dmu(const struct osd_device *osd,
378 const struct lu_fid *fid)
382 LASSERT(fid_is_acct(fid));
384 switch (fid_oid(fid)) {
386 dn = osd->od_userused_dn;
389 dn = osd->od_groupused_dn;
399 * Concurrency: no concurrent access is possible that early in object
402 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
403 const struct lu_object_conf *conf)
405 struct osd_object *obj = osd_obj(l);
406 struct osd_device *osd = osd_obj2dev(obj);
407 const struct lu_fid *fid = lu_object_fid(l);
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 if (conf != NULL && conf->loc_flags & LOC_F_NEW)
423 if (unlikely(fid_is_acct(fid))) {
424 obj->oo_dn = osd_quota_fid2dmu(osd, fid);
426 obj->oo_dt.do_index_ops = &osd_acct_index_ops;
427 l->lo_header->loh_attr |= LOHA_EXISTS;
433 rc = osd_fid_lookup(env, osd, fid, &oid);
435 LASSERT(obj->oo_dn == NULL);
436 rc = __osd_obj2dnode(osd->od_os, oid, &obj->oo_dn);
437 /* EEXIST will be returned if object is being deleted in ZFS */
443 CERROR("%s: lookup "DFID"/%#llx failed: rc = %d\n",
444 osd->od_svname, PFID(lu_object_fid(l)), oid, rc);
448 rc = osd_object_init0(env, obj);
452 rc = osd_check_lma(env, obj);
455 } else if (rc == -ENOENT) {
458 LASSERT(osd_invariant(obj));
464 * Concurrency: no concurrent access is possible that late in object
467 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
469 struct osd_object *obj = osd_obj(l);
471 LASSERT(osd_invariant(obj));
473 dt_object_fini(&obj->oo_dt);
474 OBD_SLAB_FREE_PTR(obj, osd_object_kmem);
478 osd_object_unlinked_add(struct osd_object *obj, struct osd_thandle *oh)
482 LASSERT(obj->oo_destroy == OSD_DESTROY_ASYNC);
484 /* the object is supposed to be exclusively locked by
485 * the caller (osd_destroy()), while the transaction
486 * (oh) is per-thread and not shared */
487 if (likely(list_empty(&obj->oo_unlinked_linkage))) {
488 list_add(&obj->oo_unlinked_linkage, &oh->ot_unlinked_list);
495 /* Default to max data size covered by a level-1 indirect block */
496 static unsigned long osd_sync_destroy_max_size =
497 1UL << (DN_MAX_INDBLKSHIFT - SPA_BLKPTRSHIFT + SPA_MAXBLOCKSHIFT);
498 module_param(osd_sync_destroy_max_size, ulong, 0444);
499 MODULE_PARM_DESC(osd_sync_destroy_max_size, "Maximum object size to use synchronous destroy.");
502 osd_object_set_destroy_type(struct osd_object *obj)
505 * Lock-less OST_WRITE can race with OST_DESTROY, so set destroy type
506 * only once and use it consistently thereafter.
508 down_write(&obj->oo_guard);
509 if (obj->oo_destroy == OSD_DESTROY_NONE) {
510 if (obj->oo_attr.la_size <= osd_sync_destroy_max_size)
511 obj->oo_destroy = OSD_DESTROY_SYNC;
512 else /* Larger objects are destroyed asynchronously */
513 obj->oo_destroy = OSD_DESTROY_ASYNC;
515 up_write(&obj->oo_guard);
518 static int osd_declare_destroy(const struct lu_env *env, struct dt_object *dt,
521 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
522 struct osd_object *obj = osd_dt_obj(dt);
523 struct osd_device *osd = osd_obj2dev(obj);
524 struct osd_thandle *oh;
531 LASSERT(dt_object_exists(dt));
533 oh = container_of0(th, struct osd_thandle, ot_super);
534 LASSERT(oh->ot_tx != NULL);
536 /* declare that we'll remove object from fid-dnode mapping */
537 zapid = osd_get_name_n_idx(env, osd, fid, NULL, 0, &dn);
538 osd_tx_hold_zap(oh->ot_tx, zapid, dn, FALSE, NULL);
540 osd_declare_xattrs_destroy(env, obj, oh);
543 rc = osd_declare_quota(env, osd, obj->oo_attr.la_uid,
544 obj->oo_attr.la_gid, -1, oh, false, NULL, false);
548 /* data to be truncated */
549 rc = osd_declare_quota(env, osd, obj->oo_attr.la_uid,
550 obj->oo_attr.la_gid, 0, oh, true, NULL, false);
554 osd_object_set_destroy_type(obj);
555 if (obj->oo_destroy == OSD_DESTROY_SYNC)
556 dmu_tx_hold_free(oh->ot_tx, obj->oo_dn->dn_object,
559 osd_tx_hold_zap(oh->ot_tx, osd->od_unlinked->dn_object,
560 osd->od_unlinked, TRUE, NULL);
562 /* will help to find FID->ino when this object is being
563 * added to PENDING/ */
564 osd_idc_find_and_init(env, osd, obj);
569 static int osd_destroy(const struct lu_env *env, struct dt_object *dt,
572 struct osd_thread_info *info = osd_oti_get(env);
573 char *buf = info->oti_str;
574 struct osd_object *obj = osd_dt_obj(dt);
575 struct osd_device *osd = osd_obj2dev(obj);
576 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
577 struct osd_thandle *oh;
583 down_write(&obj->oo_guard);
585 if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
586 GOTO(out, rc = -ENOENT);
588 LASSERT(obj->oo_dn != NULL);
590 oh = container_of0(th, struct osd_thandle, ot_super);
592 LASSERT(oh->ot_tx != NULL);
594 /* remove obj ref from index dir (it depends) */
595 zapid = osd_get_name_n_idx(env, osd, fid, buf,
596 sizeof(info->oti_str), &zdn);
597 rc = osd_zap_remove(osd, zapid, zdn, buf, oh->ot_tx);
599 CERROR("%s: zap_remove(%s) failed: rc = %d\n",
600 osd->od_svname, buf, rc);
604 rc = osd_xattrs_destroy(env, obj, oh);
606 CERROR("%s: cannot destroy xattrs for %s: rc = %d\n",
607 osd->od_svname, buf, rc);
611 oid = obj->oo_dn->dn_object;
612 if (unlikely(obj->oo_destroy == OSD_DESTROY_NONE)) {
613 /* this may happen if the destroy wasn't declared
614 * e.g. when the object is created and then destroyed
615 * in the same transaction - we don't need additional
616 * space for destroy specifically */
617 LASSERT(obj->oo_attr.la_size <= osd_sync_destroy_max_size);
618 rc = -dmu_object_free(osd->od_os, oid, oh->ot_tx);
620 CERROR("%s: failed to free %s %llu: rc = %d\n",
621 osd->od_svname, buf, oid, rc);
622 } else if (obj->oo_destroy == OSD_DESTROY_SYNC) {
623 rc = -dmu_object_free(osd->od_os, oid, oh->ot_tx);
625 CERROR("%s: failed to free %s %llu: rc = %d\n",
626 osd->od_svname, buf, oid, rc);
627 } else { /* asynchronous destroy */
628 char *key = info->oti_key;
630 rc = osd_object_unlinked_add(obj, oh);
634 snprintf(key, sizeof(info->oti_key), "%llx", oid);
635 rc = osd_zap_add(osd, osd->od_unlinked->dn_object,
636 osd->od_unlinked, key, 8, 1, &oid, oh->ot_tx);
638 CERROR("%s: zap_add_int() failed %s %llu: rc = %d\n",
639 osd->od_svname, buf, oid, rc);
643 /* not needed in the cache anymore */
644 set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
646 obj->oo_destroyed = 1;
647 up_write(&obj->oo_guard);
651 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
653 struct osd_object *obj = osd_obj(l);
654 const struct lu_fid *fid = lu_object_fid(l);
657 if (likely(!fid_is_acct(fid))) {
658 osd_object_sa_fini(obj);
659 if (obj->oo_sa_xattr) {
660 nvlist_free(obj->oo_sa_xattr);
661 obj->oo_sa_xattr = NULL;
663 osd_dnode_rele(obj->oo_dn);
664 list_del(&obj->oo_sa_linkage);
671 * Concurrency: ->loo_object_release() is called under site spin-lock.
673 static void osd_object_release(const struct lu_env *env,
679 * Concurrency: shouldn't matter.
681 static int osd_object_print(const struct lu_env *env, void *cookie,
682 lu_printer_t p, const struct lu_object *l)
684 struct osd_object *o = osd_obj(l);
686 return (*p)(env, cookie, LUSTRE_OSD_ZFS_NAME"-object@%p", o);
689 static void osd_read_lock(const struct lu_env *env, struct dt_object *dt,
692 struct osd_object *obj = osd_dt_obj(dt);
694 LASSERT(osd_invariant(obj));
696 down_read_nested(&obj->oo_sem, role);
699 static void osd_write_lock(const struct lu_env *env, struct dt_object *dt,
702 struct osd_object *obj = osd_dt_obj(dt);
704 LASSERT(osd_invariant(obj));
706 down_write_nested(&obj->oo_sem, role);
709 static void osd_read_unlock(const struct lu_env *env, struct dt_object *dt)
711 struct osd_object *obj = osd_dt_obj(dt);
713 LASSERT(osd_invariant(obj));
714 up_read(&obj->oo_sem);
717 static void osd_write_unlock(const struct lu_env *env, struct dt_object *dt)
719 struct osd_object *obj = osd_dt_obj(dt);
721 LASSERT(osd_invariant(obj));
722 up_write(&obj->oo_sem);
725 static int osd_write_locked(const struct lu_env *env, struct dt_object *dt)
727 struct osd_object *obj = osd_dt_obj(dt);
730 LASSERT(osd_invariant(obj));
732 if (down_write_trylock(&obj->oo_sem)) {
734 up_write(&obj->oo_sem);
739 static int osd_attr_get(const struct lu_env *env,
740 struct dt_object *dt,
741 struct lu_attr *attr)
743 struct osd_object *obj = osd_dt_obj(dt);
748 down_read(&obj->oo_guard);
750 if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
751 GOTO(out, rc = -ENOENT);
753 if (unlikely(fid_is_acct(lu_object_fid(&dt->do_lu))))
756 LASSERT(osd_invariant(obj));
759 read_lock(&obj->oo_attr_lock);
760 *attr = obj->oo_attr;
761 if (obj->oo_lma_flags & LUSTRE_ORPHAN_FL)
762 attr->la_flags |= LUSTRE_ORPHAN_FL;
763 read_unlock(&obj->oo_attr_lock);
765 /* with ZFS_DEBUG zrl_add_debug() called by DB_DNODE_ENTER()
766 * from within sa_object_size() can block on a mutex, so
767 * we can't call sa_object_size() holding rwlock */
768 sa_object_size(obj->oo_sa_hdl, &blksize, &blocks);
769 /* we do not control size of indices, so always calculate
770 * it from number of blocks reported by DMU */
771 if (S_ISDIR(attr->la_mode))
772 attr->la_size = 512 * blocks;
773 /* Block size may be not set; suggest maximal I/O transfers. */
775 blksize = osd_spa_maxblocksize(
776 dmu_objset_spa(osd_obj2dev(obj)->od_os));
778 attr->la_blksize = blksize;
779 attr->la_blocks = blocks;
780 attr->la_valid |= LA_BLOCKS | LA_BLKSIZE;
783 up_read(&obj->oo_guard);
787 /* Simple wrapper on top of qsd API which implement quota transfer for osd
788 * setattr needs. As a reminder, only the root user can change ownership of
789 * a file, that's why EDQUOT & EINPROGRESS errors are discarded */
790 static inline int qsd_transfer(const struct lu_env *env,
791 struct qsd_instance *qsd,
792 struct lquota_trans *trans, int qtype,
793 __u64 orig_id, __u64 new_id, __u64 bspace,
794 struct lquota_id_info *qi)
798 if (unlikely(qsd == NULL))
801 LASSERT(qtype >= 0 && qtype < LL_MAXQUOTAS);
802 qi->lqi_type = qtype;
804 /* inode accounting */
805 qi->lqi_is_blk = false;
807 /* one more inode for the new owner ... */
808 qi->lqi_id.qid_uid = new_id;
810 rc = qsd_op_begin(env, qsd, trans, qi, NULL);
811 if (rc == -EDQUOT || rc == -EINPROGRESS)
816 /* and one less inode for the current id */
817 qi->lqi_id.qid_uid = orig_id;;
819 /* can't get EDQUOT when reducing usage */
820 rc = qsd_op_begin(env, qsd, trans, qi, NULL);
821 if (rc == -EINPROGRESS)
826 /* block accounting */
827 qi->lqi_is_blk = true;
829 /* more blocks for the new owner ... */
830 qi->lqi_id.qid_uid = new_id;
831 qi->lqi_space = bspace;
832 rc = qsd_op_begin(env, qsd, trans, qi, NULL);
833 if (rc == -EDQUOT || rc == -EINPROGRESS)
838 /* and finally less blocks for the current owner */
839 qi->lqi_id.qid_uid = orig_id;
840 qi->lqi_space = -bspace;
841 rc = qsd_op_begin(env, qsd, trans, qi, NULL);
842 /* can't get EDQUOT when reducing usage */
843 if (rc == -EINPROGRESS)
848 static int osd_declare_attr_set(const struct lu_env *env,
849 struct dt_object *dt,
850 const struct lu_attr *attr,
851 struct thandle *handle)
853 struct osd_thread_info *info = osd_oti_get(env);
854 struct osd_object *obj = osd_dt_obj(dt);
855 struct osd_device *osd = osd_obj2dev(obj);
857 struct osd_thandle *oh;
865 LASSERT(handle != NULL);
866 LASSERT(osd_invariant(obj));
868 oh = container_of0(handle, struct osd_thandle, ot_super);
870 down_read(&obj->oo_guard);
871 if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
874 LASSERT(obj->oo_sa_hdl != NULL);
875 LASSERT(oh->ot_tx != NULL);
876 /* regular attributes are part of the bonus buffer */
877 /* let's check whether this object is already part of
880 for (txh = list_head(&oh->ot_tx->tx_holds); txh;
881 txh = list_next(&oh->ot_tx->tx_holds, txh)) {
882 if (txh->txh_dnode == NULL)
884 if (txh->txh_dnode->dn_object != obj->oo_dn->dn_object)
886 /* this object is part of the transaction already
887 * we don't need to declare bonus again */
892 dmu_tx_hold_bonus(oh->ot_tx, obj->oo_dn->dn_object);
893 if (oh->ot_tx->tx_err != 0)
894 GOTO(out, rc = -oh->ot_tx->tx_err);
896 if (attr && attr->la_valid & LA_FLAGS) {
897 /* LMA is usually a part of bonus, no need to declare
901 if (attr && (attr->la_valid & (LA_UID | LA_GID))) {
902 sa_object_size(obj->oo_sa_hdl, &blksize, &bspace);
903 bspace = toqb(bspace * blksize);
906 if (attr && attr->la_valid & LA_UID) {
907 /* quota enforcement for user */
908 if (attr->la_uid != obj->oo_attr.la_uid) {
909 rc = qsd_transfer(env, osd->od_quota_slave,
910 &oh->ot_quota_trans, USRQUOTA,
911 obj->oo_attr.la_uid, attr->la_uid,
912 bspace, &info->oti_qi);
917 if (attr && attr->la_valid & LA_GID) {
918 /* quota enforcement for group */
919 if (attr->la_gid != obj->oo_attr.la_gid) {
920 rc = qsd_transfer(env, osd->od_quota_slave,
921 &oh->ot_quota_trans, GRPQUOTA,
922 obj->oo_attr.la_gid, attr->la_gid,
923 bspace, &info->oti_qi);
930 up_read(&obj->oo_guard);
935 * Set the attributes of an object
937 * The transaction passed to this routine must have
938 * dmu_tx_hold_bonus(tx, oid) called and then assigned
939 * to a transaction group.
941 static int osd_attr_set(const struct lu_env *env, struct dt_object *dt,
942 const struct lu_attr *la, struct thandle *handle)
944 struct osd_thread_info *info = osd_oti_get(env);
945 sa_bulk_attr_t *bulk = osd_oti_get(env)->oti_attr_bulk;
946 struct osd_object *obj = osd_dt_obj(dt);
947 struct osd_device *osd = osd_obj2dev(obj);
948 struct osd_thandle *oh;
949 struct osa_attr *osa = &info->oti_osa;
950 __u64 valid = la->la_valid;
956 down_read(&obj->oo_guard);
957 if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
958 GOTO(out, rc = -ENOENT);
960 LASSERT(handle != NULL);
961 LASSERT(osd_invariant(obj));
962 LASSERT(obj->oo_sa_hdl);
964 oh = container_of0(handle, struct osd_thandle, ot_super);
965 /* Assert that the transaction has been assigned to a
966 transaction group. */
967 LASSERT(oh->ot_tx->tx_txg != 0);
969 /* Only allow set size for regular file */
970 if (!S_ISREG(dt->do_lu.lo_header->loh_attr))
971 valid &= ~(LA_SIZE | LA_BLOCKS);
973 if (valid & LA_CTIME && la->la_ctime == obj->oo_attr.la_ctime)
976 if (valid & LA_MTIME && la->la_mtime == obj->oo_attr.la_mtime)
979 if (valid & LA_ATIME && la->la_atime == obj->oo_attr.la_atime)
985 if (valid & LA_FLAGS) {
986 struct lustre_mdt_attrs *lma;
989 if (la->la_flags & LUSTRE_LMA_FL_MASKS) {
990 CLASSERT(sizeof(info->oti_buf) >= sizeof(*lma));
991 lma = (struct lustre_mdt_attrs *)&info->oti_buf;
993 buf.lb_len = sizeof(info->oti_buf);
994 rc = osd_xattr_get(env, &obj->oo_dt, &buf,
998 le32_to_cpu(lma->lma_incompat);
1000 lustre_to_lma_flags(la->la_flags);
1002 cpu_to_le32(lma->lma_incompat);
1004 buf.lb_len = sizeof(*lma);
1005 rc = osd_xattr_set_internal(env, obj, &buf,
1011 CWARN("%s: failed to set LMA flags: rc = %d\n",
1012 osd->od_svname, rc);
1018 write_lock(&obj->oo_attr_lock);
1020 if (valid & LA_ATIME) {
1021 osa->atime[0] = obj->oo_attr.la_atime = la->la_atime;
1022 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(osd), NULL,
1025 if (valid & LA_MTIME) {
1026 osa->mtime[0] = obj->oo_attr.la_mtime = la->la_mtime;
1027 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(osd), NULL,
1030 if (valid & LA_CTIME) {
1031 osa->ctime[0] = obj->oo_attr.la_ctime = la->la_ctime;
1032 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(osd), NULL,
1035 if (valid & LA_MODE) {
1036 /* mode is stored along with type, so read it first */
1037 obj->oo_attr.la_mode = (obj->oo_attr.la_mode & S_IFMT) |
1038 (la->la_mode & ~S_IFMT);
1039 osa->mode = obj->oo_attr.la_mode;
1040 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(osd), NULL,
1043 if (valid & LA_SIZE) {
1044 osa->size = obj->oo_attr.la_size = la->la_size;
1045 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(osd), NULL,
1048 if (valid & LA_NLINK) {
1049 osa->nlink = obj->oo_attr.la_nlink = la->la_nlink;
1050 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(osd), NULL,
1053 if (valid & LA_RDEV) {
1054 osa->rdev = obj->oo_attr.la_rdev = la->la_rdev;
1055 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_RDEV(osd), NULL,
1058 if (valid & LA_FLAGS) {
1059 osa->flags = attrs_fs2zfs(la->la_flags);
1060 /* many flags are not supported by zfs, so ensure a good cached
1062 obj->oo_attr.la_flags = attrs_zfs2fs(osa->flags);
1063 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(osd), NULL,
1066 if (valid & LA_UID) {
1067 osa->uid = obj->oo_attr.la_uid = la->la_uid;
1068 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(osd), NULL,
1071 if (valid & LA_GID) {
1072 osa->gid = obj->oo_attr.la_gid = la->la_gid;
1073 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(osd), NULL,
1076 obj->oo_attr.la_valid |= valid;
1077 write_unlock(&obj->oo_attr_lock);
1079 LASSERT(cnt <= ARRAY_SIZE(osd_oti_get(env)->oti_attr_bulk));
1080 rc = osd_object_sa_bulk_update(obj, bulk, cnt, oh);
1083 up_read(&obj->oo_guard);
1090 * XXX temporary solution.
1093 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
1094 struct dt_object *parent, struct dt_object *child,
1099 ah->dah_parent = parent;
1100 ah->dah_mode = child_mode;
1102 if (parent != NULL && !dt_object_remote(parent)) {
1103 /* will help to find FID->ino at dt_insert("..") */
1104 struct osd_object *pobj = osd_dt_obj(parent);
1106 osd_idc_find_and_init(env, osd_obj2dev(pobj), pobj);
1110 static int osd_declare_create(const struct lu_env *env, struct dt_object *dt,
1111 struct lu_attr *attr,
1112 struct dt_allocation_hint *hint,
1113 struct dt_object_format *dof,
1114 struct thandle *handle)
1116 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1117 struct osd_object *obj = osd_dt_obj(dt);
1118 struct osd_device *osd = osd_obj2dev(obj);
1119 struct osd_thandle *oh;
1127 switch (dof->dof_type) {
1131 if (obj->oo_dt.do_body_ops == NULL)
1132 obj->oo_dt.do_body_ops = &osd_body_ops;
1138 LASSERT(handle != NULL);
1139 oh = container_of0(handle, struct osd_thandle, ot_super);
1140 LASSERT(oh->ot_tx != NULL);
1142 /* this is the minimum set of EAs on every Lustre object */
1143 obj->oo_ea_in_bonus = ZFS_SA_BASE_ATTR_SIZE +
1144 sizeof(__u64) + /* VBR VERSION */
1145 sizeof(struct lustre_mdt_attrs); /* LMA */
1146 /* reserve 32 bytes for extra stuff like ACLs */
1147 dnode_size = size_roundup_power2(obj->oo_ea_in_bonus + 32);
1149 switch (dof->dof_type) {
1151 dt->do_index_ops = &osd_dir_ops;
1153 /* for zap create */
1154 dmu_tx_hold_zap(oh->ot_tx, DMU_NEW_OBJECT, FALSE, NULL);
1155 dmu_tx_hold_sa_create(oh->ot_tx, dnode_size);
1160 /* first, we'll create new object */
1161 dmu_tx_hold_sa_create(oh->ot_tx, dnode_size);
1169 /* and we'll add it to some mapping */
1170 zapid = osd_get_name_n_idx(env, osd, fid, NULL, 0, &dn);
1171 osd_tx_hold_zap(oh->ot_tx, zapid, dn, TRUE, NULL);
1173 /* will help to find FID->ino mapping at dt_insert() */
1174 osd_idc_find_and_init(env, osd, obj);
1176 rc = osd_declare_quota(env, osd, attr->la_uid, attr->la_gid, 1, oh,
1177 false, NULL, false);
1182 int __osd_attr_init(const struct lu_env *env, struct osd_device *osd,
1183 sa_handle_t *sa_hdl, dmu_tx_t *tx,
1184 struct lu_attr *la, uint64_t parent)
1186 sa_bulk_attr_t *bulk = osd_oti_get(env)->oti_attr_bulk;
1187 struct osa_attr *osa = &osd_oti_get(env)->oti_osa;
1196 gen = dmu_tx_get_txg(tx);
1198 ZFS_TIME_ENCODE(&now, crtime);
1200 osa->atime[0] = la->la_atime;
1201 osa->ctime[0] = la->la_ctime;
1202 osa->mtime[0] = la->la_mtime;
1203 osa->mode = la->la_mode;
1204 osa->uid = la->la_uid;
1205 osa->gid = la->la_gid;
1206 osa->rdev = la->la_rdev;
1207 osa->nlink = la->la_nlink;
1208 osa->flags = attrs_fs2zfs(la->la_flags);
1209 osa->size = la->la_size;
1212 * we need to create all SA below upon object create.
1214 * XXX The attribute order matters since the accounting callback relies
1215 * on static offsets (i.e. SA_*_OFFSET, see zfs_space_delta_cb()) to
1216 * look up the UID/GID attributes. Moreover, the callback does not seem
1217 * to support the spill block.
1218 * We define attributes in the same order as SA_*_OFFSET in order to
1219 * work around the problem. See ORI-610.
1222 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(osd), NULL, &osa->mode, 8);
1223 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(osd), NULL, &osa->size, 8);
1224 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GEN(osd), NULL, &gen, 8);
1225 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(osd), NULL, &osa->uid, 8);
1226 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(osd), NULL, &osa->gid, 8);
1227 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_PARENT(osd), NULL, &parent, 8);
1228 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(osd), NULL, &osa->flags, 8);
1229 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(osd), NULL, osa->atime, 16);
1230 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(osd), NULL, osa->mtime, 16);
1231 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(osd), NULL, osa->ctime, 16);
1232 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CRTIME(osd), NULL, crtime, 16);
1233 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(osd), NULL, &osa->nlink, 8);
1234 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_RDEV(osd), NULL, &osa->rdev, 8);
1235 LASSERT(cnt <= ARRAY_SIZE(osd_oti_get(env)->oti_attr_bulk));
1237 rc = -sa_replace_all_by_template(sa_hdl, bulk, cnt, tx);
1242 static int osd_find_new_dnode(const struct lu_env *env, dmu_tx_t *tx,
1243 uint64_t oid, dnode_t **dnp)
1248 /* take dnode_t from tx to save on dnode#->dnode_t lookup */
1249 for (txh = list_tail(&tx->tx_holds); txh;
1250 txh = list_prev(&tx->tx_holds, txh)) {
1251 dnode_t *dn = txh->txh_dnode;
1256 if (dn->dn_object != oid)
1260 rw_enter(&dn->dn_struct_rwlock, RW_WRITER);
1261 if (dn->dn_bonus == NULL)
1262 dbuf_create_bonus(dn);
1263 rw_exit(&dn->dn_struct_rwlock);
1267 LASSERT(dn->dn_handle);
1269 if (refcount_add(&db->db_holds, osd_obj_tag) == 1) {
1270 refcount_add(&dn->dn_holds, tag);
1271 atomic_inc_32(&dn->dn_dbufs_count);
1274 dbuf_read(db, NULL, DB_RF_MUST_SUCCEED | DB_RF_NOPREFETCH);
1278 if (unlikely(*dnp == NULL))
1279 rc = __osd_obj2dnode(tx->tx_objset, oid, dnp);
1285 * The transaction passed to this routine must have
1286 * dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT) called and then assigned
1287 * to a transaction group.
1289 int __osd_object_create(const struct lu_env *env, struct osd_object *obj,
1290 dnode_t **dnp, dmu_tx_t *tx, struct lu_attr *la)
1292 struct osd_device *osd = osd_obj2dev(obj);
1293 const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
1294 dmu_object_type_t type = DMU_OT_PLAIN_FILE_CONTENTS;
1297 /* Use DMU_OTN_UINT8_METADATA for local objects so their data blocks
1298 * would get an additional ditto copy */
1299 if (unlikely(S_ISREG(la->la_mode) &&
1300 fid_seq_is_local_file(fid_seq(fid))))
1301 type = DMU_OTN_UINT8_METADATA;
1303 /* Create a new DMU object using the default dnode size. */
1304 oid = osd_dmu_object_alloc(osd->od_os, type, 0, 0, tx);
1306 LASSERT(la->la_valid & LA_MODE);
1310 return osd_find_new_dnode(env, tx, oid, dnp);
1314 * The transaction passed to this routine must have
1315 * dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, ...) called and then assigned
1316 * to a transaction group.
1318 * Using ZAP_FLAG_HASH64 will force the ZAP to always be a FAT ZAP.
1319 * This is fine for directories today, because storing the FID in the dirent
1320 * will also require a FAT ZAP. If there is a new type of micro ZAP created
1321 * then we might need to re-evaluate the use of this flag and instead do
1322 * a conversion from the different internal ZAP hash formats being used. */
1323 int __osd_zap_create(const struct lu_env *env, struct osd_device *osd,
1324 dnode_t **dnp, dmu_tx_t *tx, struct lu_attr *la,
1329 /* Assert that the transaction has been assigned to a
1330 transaction group. */
1331 LASSERT(tx->tx_txg != 0);
1334 oid = osd_zap_create_flags(osd->od_os, 0, flags | ZAP_FLAG_HASH64,
1335 DMU_OT_DIRECTORY_CONTENTS,
1336 14, /* == ZFS fzap_default_blockshift */
1337 DN_MAX_INDBLKSHIFT, /* indirect blockshift */
1343 return osd_find_new_dnode(env, tx, oid, dnp);
1346 static dnode_t *osd_mkidx(const struct lu_env *env, struct osd_object *obj,
1347 struct lu_attr *la, struct osd_thandle *oh)
1352 /* Index file should be created as regular file in order not to confuse
1353 * ZPL which could interpret them as directory.
1354 * We set ZAP_FLAG_UINT64_KEY to let ZFS know than we are going to use
1356 LASSERT(S_ISREG(la->la_mode));
1357 rc = __osd_zap_create(env, osd_obj2dev(obj), &dn, oh->ot_tx, la,
1358 ZAP_FLAG_UINT64_KEY);
1364 static dnode_t *osd_mkdir(const struct lu_env *env, struct osd_object *obj,
1365 struct lu_attr *la, struct osd_thandle *oh)
1370 LASSERT(S_ISDIR(la->la_mode));
1371 rc = __osd_zap_create(env, osd_obj2dev(obj), &dn, oh->ot_tx, la, 0);
1377 static dnode_t *osd_mkreg(const struct lu_env *env, struct osd_object *obj,
1378 struct lu_attr *la, struct osd_thandle *oh)
1380 const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
1381 struct osd_device *osd = osd_obj2dev(obj);
1385 LASSERT(S_ISREG(la->la_mode));
1386 rc = __osd_object_create(env, obj, &dn, oh->ot_tx, la);
1390 if ((fid_is_idif(fid) || fid_is_norm(fid) || fid_is_echo(fid)) &&
1392 /* The minimum block size must be at least page size otherwise
1393 * it will break the assumption in tgt_thread_big_cache where
1394 * the array size is PTLRPC_MAX_BRW_PAGES. It will also affect
1395 * RDMA due to subpage transfer size */
1396 rc = -dmu_object_set_blocksize(osd->od_os, dn->dn_object,
1397 PAGE_SIZE, 0, oh->ot_tx);
1399 CERROR("%s: can't change blocksize: %d\n",
1400 osd->od_svname, rc);
1408 static dnode_t *osd_mksym(const struct lu_env *env, struct osd_object *obj,
1409 struct lu_attr *la, struct osd_thandle *oh)
1414 LASSERT(S_ISLNK(la->la_mode));
1415 rc = __osd_object_create(env, obj, &dn, oh->ot_tx, la);
1421 static dnode_t *osd_mknod(const struct lu_env *env, struct osd_object *obj,
1422 struct lu_attr *la, struct osd_thandle *oh)
1427 if (S_ISCHR(la->la_mode) || S_ISBLK(la->la_mode))
1428 la->la_valid |= LA_RDEV;
1430 rc = __osd_object_create(env, obj, &dn, oh->ot_tx, la);
1436 typedef dnode_t *(*osd_obj_type_f)(const struct lu_env *env,
1437 struct osd_object *obj,
1439 struct osd_thandle *oh);
1441 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
1443 osd_obj_type_f result;
1469 * Concurrency: @dt is write locked.
1471 static int osd_create(const struct lu_env *env, struct dt_object *dt,
1472 struct lu_attr *attr, struct dt_allocation_hint *hint,
1473 struct dt_object_format *dof, struct thandle *th)
1475 struct osd_thread_info *info = osd_oti_get(env);
1476 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
1477 struct zpl_direntry *zde = &info->oti_zde.lzd_reg;
1478 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1479 struct osd_object *obj = osd_dt_obj(dt);
1480 struct osd_device *osd = osd_obj2dev(obj);
1481 char *buf = info->oti_str;
1482 struct osd_thandle *oh;
1483 dnode_t *dn = NULL, *zdn = NULL;
1484 uint64_t zapid, parent = 0;
1489 LASSERT(!fid_is_acct(fid));
1491 /* concurrent create declarations should not see
1492 * the object inconsistent (db, attr, etc).
1493 * in regular cases acquisition should be cheap */
1494 down_write(&obj->oo_guard);
1496 if (unlikely(dt_object_exists(dt)))
1497 GOTO(out, rc = -EEXIST);
1499 LASSERT(osd_invariant(obj));
1500 LASSERT(dof != NULL);
1502 LASSERT(th != NULL);
1503 oh = container_of0(th, struct osd_thandle, ot_super);
1505 LASSERT(obj->oo_dn == NULL);
1507 /* to follow ZFS on-disk format we need
1508 * to initialize parent dnode properly */
1509 if (hint != NULL && hint->dah_parent != NULL &&
1510 !dt_object_remote(hint->dah_parent))
1511 parent = osd_dt_obj(hint->dah_parent)->oo_dn->dn_object;
1513 /* we may fix some attributes, better do not change the source */
1514 obj->oo_attr = *attr;
1515 obj->oo_attr.la_valid |= LA_SIZE | LA_NLINK | LA_TYPE;
1517 dn = osd_create_type_f(dof->dof_type)(env, obj, &obj->oo_attr, oh);
1525 zde->zde_dnode = dn->dn_object;
1526 zde->zde_type = IFTODT(attr->la_mode & S_IFMT);
1528 zapid = osd_get_name_n_idx(env, osd, fid, buf,
1529 sizeof(info->oti_str), &zdn);
1530 rc = osd_zap_add(osd, zapid, zdn, buf, 8, 1, zde, oh->ot_tx);
1534 /* Now add in all of the "SA" attributes */
1535 rc = osd_sa_handle_get(obj);
1539 /* configure new osd object */
1540 parent = parent != 0 ? parent : zapid;
1541 rc = __osd_attr_init(env, osd, obj->oo_sa_hdl, oh->ot_tx,
1542 &obj->oo_attr, parent);
1546 /* XXX: oo_lma_flags */
1547 obj->oo_dt.do_lu.lo_header->loh_attr |= obj->oo_attr.la_mode & S_IFMT;
1548 obj->oo_dt.do_body_ops = &osd_body_ops;
1550 rc = -nvlist_alloc(&obj->oo_sa_xattr, NV_UNIQUE_NAME, KM_SLEEP);
1554 /* initialize LMA */
1555 lustre_lma_init(lma, fid, 0, 0);
1556 lustre_lma_swab(lma);
1557 rc = -nvlist_add_byte_array(obj->oo_sa_xattr, XATTR_NAME_LMA,
1558 (uchar_t *)lma, sizeof(*lma));
1561 rc = __osd_sa_xattr_update(env, obj, oh);
1564 osd_idc_find_and_init(env, osd, obj);
1567 if (unlikely(rc && dn)) {
1568 dmu_object_free(osd->od_os, dn->dn_object, oh->ot_tx);
1572 obj->oo_dt.do_lu.lo_header->loh_attr |= LOHA_EXISTS;
1574 up_write(&obj->oo_guard);
1578 static int osd_declare_ref_add(const struct lu_env *env, struct dt_object *dt,
1581 return osd_declare_attr_set(env, dt, NULL, th);
1585 * Concurrency: @dt is write locked.
1587 static int osd_ref_add(const struct lu_env *env, 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 down_read(&obj->oo_guard);
1599 if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
1600 GOTO(out, rc = -ENOENT);
1602 LASSERT(osd_invariant(obj));
1603 LASSERT(obj->oo_sa_hdl != NULL);
1605 oh = container_of0(handle, struct osd_thandle, ot_super);
1607 write_lock(&obj->oo_attr_lock);
1608 nlink = ++obj->oo_attr.la_nlink;
1609 write_unlock(&obj->oo_attr_lock);
1611 rc = osd_object_sa_update(obj, SA_ZPL_LINKS(osd), &nlink, 8, oh);
1614 up_read(&obj->oo_guard);
1618 static int osd_declare_ref_del(const struct lu_env *env, struct dt_object *dt,
1619 struct thandle *handle)
1621 return osd_declare_attr_set(env, dt, NULL, handle);
1625 * Concurrency: @dt is write locked.
1627 static int osd_ref_del(const struct lu_env *env, struct dt_object *dt,
1628 struct thandle *handle)
1630 struct osd_object *obj = osd_dt_obj(dt);
1631 struct osd_thandle *oh;
1632 struct osd_device *osd = osd_obj2dev(obj);
1638 down_read(&obj->oo_guard);
1640 if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
1641 GOTO(out, rc = -ENOENT);
1643 LASSERT(osd_invariant(obj));
1644 LASSERT(obj->oo_sa_hdl != NULL);
1646 oh = container_of0(handle, struct osd_thandle, ot_super);
1647 LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
1649 write_lock(&obj->oo_attr_lock);
1650 nlink = --obj->oo_attr.la_nlink;
1651 write_unlock(&obj->oo_attr_lock);
1653 rc = osd_object_sa_update(obj, SA_ZPL_LINKS(osd), &nlink, 8, oh);
1656 up_read(&obj->oo_guard);
1660 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
1661 __u64 start, __u64 end)
1663 struct osd_device *osd = osd_obj2dev(osd_dt_obj(dt));
1666 /* XXX: no other option than syncing the whole filesystem until we
1667 * support ZIL. If the object tracked the txg that it was last
1668 * modified in, it could pass that txg here instead of "0". Maybe
1669 * the changes are already committed, so no wait is needed at all? */
1670 if (!osd->od_dt_dev.dd_rdonly)
1671 txg_wait_synced(dmu_objset_pool(osd->od_os), 0ULL);
1676 static int osd_invalidate(const struct lu_env *env, struct dt_object *dt)
1681 static struct dt_object_operations osd_obj_ops = {
1682 .do_read_lock = osd_read_lock,
1683 .do_write_lock = osd_write_lock,
1684 .do_read_unlock = osd_read_unlock,
1685 .do_write_unlock = osd_write_unlock,
1686 .do_write_locked = osd_write_locked,
1687 .do_attr_get = osd_attr_get,
1688 .do_declare_attr_set = osd_declare_attr_set,
1689 .do_attr_set = osd_attr_set,
1690 .do_ah_init = osd_ah_init,
1691 .do_declare_create = osd_declare_create,
1692 .do_create = osd_create,
1693 .do_declare_destroy = osd_declare_destroy,
1694 .do_destroy = osd_destroy,
1695 .do_index_try = osd_index_try,
1696 .do_declare_ref_add = osd_declare_ref_add,
1697 .do_ref_add = osd_ref_add,
1698 .do_declare_ref_del = osd_declare_ref_del,
1699 .do_ref_del = osd_ref_del,
1700 .do_xattr_get = osd_xattr_get,
1701 .do_declare_xattr_set = osd_declare_xattr_set,
1702 .do_xattr_set = osd_xattr_set,
1703 .do_declare_xattr_del = osd_declare_xattr_del,
1704 .do_xattr_del = osd_xattr_del,
1705 .do_xattr_list = osd_xattr_list,
1706 .do_object_sync = osd_object_sync,
1707 .do_invalidate = osd_invalidate,
1710 static struct lu_object_operations osd_lu_obj_ops = {
1711 .loo_object_init = osd_object_init,
1712 .loo_object_delete = osd_object_delete,
1713 .loo_object_release = osd_object_release,
1714 .loo_object_free = osd_object_free,
1715 .loo_object_print = osd_object_print,
1716 .loo_object_invariant = osd_object_invariant,
1719 static int osd_otable_it_attr_get(const struct lu_env *env,
1720 struct dt_object *dt,
1721 struct lu_attr *attr)
1727 static struct dt_object_operations osd_obj_otable_it_ops = {
1728 .do_attr_get = osd_otable_it_attr_get,
1729 .do_index_try = osd_index_try,