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 = -sa_handle_get(o->od_os, obj->oo_dn->dn_object, obj,
92 SA_HDL_PRIVATE, &obj->oo_sa_hdl);
96 /* Cache the xattr object id, valid for the life of the object */
97 rc = -sa_lookup(obj->oo_sa_hdl, SA_ZPL_XATTR(o), &obj->oo_xattr, 8);
99 obj->oo_xattr = ZFS_NO_OBJECT;
102 osd_object_sa_fini(obj);
109 * Add object to list of dirty objects in tx handle.
112 osd_object_sa_dirty_add(struct osd_object *obj, struct osd_thandle *oh)
114 if (!list_empty(&obj->oo_sa_linkage))
117 down(&oh->ot_sa_lock);
118 write_lock(&obj->oo_attr_lock);
119 if (likely(list_empty(&obj->oo_sa_linkage)))
120 list_add(&obj->oo_sa_linkage, &oh->ot_sa_list);
121 write_unlock(&obj->oo_attr_lock);
126 * Release spill block dbuf hold for all dirty SAs.
128 void osd_object_sa_dirty_rele(struct osd_thandle *oh)
130 struct osd_object *obj;
132 down(&oh->ot_sa_lock);
133 while (!list_empty(&oh->ot_sa_list)) {
134 obj = list_entry(oh->ot_sa_list.next,
135 struct osd_object, oo_sa_linkage);
136 sa_spill_rele(obj->oo_sa_hdl);
137 write_lock(&obj->oo_attr_lock);
138 list_del_init(&obj->oo_sa_linkage);
139 write_unlock(&obj->oo_attr_lock);
145 * Update the SA and add the object to the dirty list.
147 int osd_object_sa_update(struct osd_object *obj, sa_attr_type_t type,
148 void *buf, uint32_t buflen, struct osd_thandle *oh)
152 LASSERT(obj->oo_sa_hdl != NULL);
153 LASSERT(oh->ot_tx != NULL);
155 rc = -sa_update(obj->oo_sa_hdl, type, buf, buflen, oh->ot_tx);
156 osd_object_sa_dirty_add(obj, oh);
162 * Bulk update the SA and add the object to the dirty list.
165 osd_object_sa_bulk_update(struct osd_object *obj, sa_bulk_attr_t *attrs,
166 int count, struct osd_thandle *oh)
170 LASSERT(obj->oo_sa_hdl != NULL);
171 LASSERT(oh->ot_tx != NULL);
173 rc = -sa_bulk_update(obj->oo_sa_hdl, attrs, count, oh->ot_tx);
174 osd_object_sa_dirty_add(obj, oh);
180 * Retrieve the attributes of a DMU object
182 int __osd_object_attr_get(const struct lu_env *env, struct osd_device *o,
183 struct osd_object *obj, struct lu_attr *la)
185 struct osa_attr *osa = &osd_oti_get(env)->oti_osa;
186 sa_bulk_attr_t *bulk = osd_oti_get(env)->oti_attr_bulk;
192 LASSERT(obj->oo_dn != NULL);
194 rc = -sa_handle_get(o->od_os, obj->oo_dn->dn_object, NULL,
195 SA_HDL_PRIVATE, &sa_hdl);
199 la->la_valid |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE | LA_TYPE |
200 LA_SIZE | LA_UID | LA_GID | LA_FLAGS | LA_NLINK;
202 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(o), NULL, osa->atime, 16);
203 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(o), NULL, osa->mtime, 16);
204 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(o), NULL, osa->ctime, 16);
205 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(o), NULL, &osa->mode, 8);
206 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(o), NULL, &osa->size, 8);
207 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(o), NULL, &osa->nlink, 8);
208 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(o), NULL, &osa->uid, 8);
209 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(o), NULL, &osa->gid, 8);
210 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(o), NULL, &osa->flags, 8);
211 LASSERT(cnt <= ARRAY_SIZE(osd_oti_get(env)->oti_attr_bulk));
213 rc = -sa_bulk_lookup(sa_hdl, bulk, cnt);
217 la->la_atime = osa->atime[0];
218 la->la_mtime = osa->mtime[0];
219 la->la_ctime = osa->ctime[0];
220 la->la_mode = osa->mode;
221 la->la_uid = osa->uid;
222 la->la_gid = osa->gid;
223 la->la_nlink = osa->nlink;
224 la->la_flags = attrs_zfs2fs(osa->flags);
225 la->la_size = osa->size;
227 /* Try to get extra flag from LMA. Right now, only LMAI_ORPHAN
228 * flags is stored in LMA, and it is only for orphan directory */
229 if (S_ISDIR(la->la_mode) && dt_object_exists(&obj->oo_dt)) {
230 struct osd_thread_info *info = osd_oti_get(env);
231 struct lustre_mdt_attrs *lma;
234 lma = (struct lustre_mdt_attrs *)info->oti_buf;
236 buf.lb_len = sizeof(info->oti_buf);
237 rc = osd_xattr_get(env, &obj->oo_dt, &buf, XATTR_NAME_LMA);
240 lma->lma_incompat = le32_to_cpu(lma->lma_incompat);
242 lma_to_lustre_flags(lma->lma_incompat);
244 } else if (rc == -ENODATA) {
249 if (S_ISCHR(la->la_mode) || S_ISBLK(la->la_mode)) {
250 rc = -sa_lookup(sa_hdl, SA_ZPL_RDEV(o), &osa->rdev, 8);
253 la->la_rdev = osa->rdev;
254 la->la_valid |= LA_RDEV;
257 sa_handle_destroy(sa_hdl);
262 int __osd_obj2dnode(const struct lu_env *env, objset_t *os,
263 uint64_t oid, dnode_t **dnp)
265 dmu_object_info_t *doi = &osd_oti_get(env)->oti_doi;
270 rc = dmu_bonus_hold(os, oid, osd_obj_tag, &db);
274 dbi = (dmu_buf_impl_t *)db;
276 *dnp = DB_DNODE(dbi);
278 LASSERT(*dnp != NULL);
279 dmu_object_info_from_dnode(*dnp, doi);
280 if (unlikely (oid != DMU_USERUSED_OBJECT &&
281 oid != DMU_GROUPUSED_OBJECT && doi->doi_bonus_type != DMU_OT_SA)) {
282 osd_dnode_rele(*dnp);
291 * Concurrency: no concurrent access is possible that early in object
294 struct lu_object *osd_object_alloc(const struct lu_env *env,
295 const struct lu_object_header *hdr,
298 struct osd_object *mo;
300 OBD_SLAB_ALLOC_PTR_GFP(mo, osd_object_kmem, GFP_NOFS);
304 l = &mo->oo_dt.do_lu;
305 dt_object_init(&mo->oo_dt, NULL, d);
306 mo->oo_dt.do_ops = &osd_obj_ops;
307 l->lo_ops = &osd_lu_obj_ops;
308 INIT_LIST_HEAD(&mo->oo_sa_linkage);
309 INIT_LIST_HEAD(&mo->oo_unlinked_linkage);
310 init_rwsem(&mo->oo_sem);
311 init_rwsem(&mo->oo_guard);
312 rwlock_init(&mo->oo_attr_lock);
313 mo->oo_destroy = OSD_DESTROY_NONE;
321 * Concurrency: shouldn't matter.
323 int osd_object_init0(const struct lu_env *env, struct osd_object *obj)
325 struct osd_device *osd = osd_obj2dev(obj);
326 const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
330 if (obj->oo_dn == NULL)
335 rc = osd_object_sa_init(obj, osd);
339 /* cache attrs in object */
340 rc = __osd_object_attr_get(env, osd, obj, &obj->oo_attr);
344 if (likely(!fid_is_acct(fid)))
345 /* no body operations for accounting objects */
346 obj->oo_dt.do_body_ops = &osd_body_ops;
349 * initialize object before marking it existing
351 obj->oo_dt.do_lu.lo_header->loh_attr |= obj->oo_attr.la_mode & S_IFMT;
354 obj->oo_dt.do_lu.lo_header->loh_attr |= LOHA_EXISTS;
359 static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
361 struct osd_thread_info *info = osd_oti_get(env);
364 struct lustre_mdt_attrs *lma;
367 CLASSERT(sizeof(info->oti_buf) >= sizeof(*lma));
368 lma = (struct lustre_mdt_attrs *)info->oti_buf;
370 buf.lb_len = sizeof(info->oti_buf);
372 rc = osd_xattr_get(env, &obj->oo_dt, &buf, XATTR_NAME_LMA);
375 lustre_lma_swab(lma);
376 if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
377 CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT))) {
378 CWARN("%s: unsupported incompat LMA feature(s) %#x for "
379 "fid = "DFID"\n", osd_obj2dev(obj)->od_svname,
380 lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
381 PFID(lu_object_fid(&obj->oo_dt.do_lu)));
384 } else if (rc == -ENODATA) {
385 /* haven't initialize LMA xattr */
393 * Concurrency: no concurrent access is possible that early in object
396 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
397 const struct lu_object_conf *conf)
399 struct osd_object *obj = osd_obj(l);
400 struct osd_device *osd = osd_obj2dev(obj);
405 LASSERT(osd_invariant(obj));
407 if (fid_is_otable_it(&l->lo_header->loh_fid)) {
408 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
409 l->lo_header->loh_attr |= LOHA_EXISTS;
413 if (conf != NULL && conf->loc_flags & LOC_F_NEW)
416 rc = osd_fid_lookup(env, osd, lu_object_fid(l), &oid);
418 LASSERT(obj->oo_dn == NULL);
419 rc = __osd_obj2dnode(env, osd->od_os, oid, &obj->oo_dn);
420 /* EEXIST will be returned if object is being deleted in ZFS */
426 CERROR("%s: lookup "DFID"/%#llx failed: rc = %d\n",
427 osd->od_svname, PFID(lu_object_fid(l)), oid, rc);
431 rc = osd_object_init0(env, obj);
435 rc = osd_check_lma(env, obj);
438 } else if (rc == -ENOENT) {
441 LASSERT(osd_invariant(obj));
447 * Concurrency: no concurrent access is possible that late in object
450 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
452 struct osd_object *obj = osd_obj(l);
454 LASSERT(osd_invariant(obj));
456 dt_object_fini(&obj->oo_dt);
457 OBD_SLAB_FREE_PTR(obj, osd_object_kmem);
461 osd_object_unlinked_add(struct osd_object *obj, struct osd_thandle *oh)
465 LASSERT(obj->oo_destroy == OSD_DESTROY_ASYNC);
467 /* the object is supposed to be exclusively locked by
468 * the caller (osd_object_destroy()), while the transaction
469 * (oh) is per-thread and not shared */
470 if (likely(list_empty(&obj->oo_unlinked_linkage))) {
471 list_add(&obj->oo_unlinked_linkage, &oh->ot_unlinked_list);
478 /* Default to max data size covered by a level-1 indirect block */
479 static unsigned long osd_sync_destroy_max_size =
480 1UL << (DN_MAX_INDBLKSHIFT - SPA_BLKPTRSHIFT + SPA_MAXBLOCKSHIFT);
481 module_param(osd_sync_destroy_max_size, ulong, 0444);
482 MODULE_PARM_DESC(osd_sync_destroy_max_size, "Maximum object size to use synchronous destroy.");
485 osd_object_set_destroy_type(struct osd_object *obj)
488 * Lock-less OST_WRITE can race with OST_DESTROY, so set destroy type
489 * only once and use it consistently thereafter.
491 down_write(&obj->oo_guard);
492 if (obj->oo_destroy == OSD_DESTROY_NONE) {
493 if (obj->oo_attr.la_size <= osd_sync_destroy_max_size)
494 obj->oo_destroy = OSD_DESTROY_SYNC;
495 else /* Larger objects are destroyed asynchronously */
496 obj->oo_destroy = OSD_DESTROY_ASYNC;
498 up_write(&obj->oo_guard);
501 static int osd_declare_object_destroy(const struct lu_env *env,
502 struct dt_object *dt,
505 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
506 struct osd_object *obj = osd_dt_obj(dt);
507 struct osd_device *osd = osd_obj2dev(obj);
508 struct osd_thandle *oh;
514 LASSERT(dt_object_exists(dt));
516 oh = container_of0(th, struct osd_thandle, ot_super);
517 LASSERT(oh->ot_tx != NULL);
519 /* declare that we'll remove object from fid-dnode mapping */
520 zapid = osd_get_name_n_idx(env, osd, fid, NULL, 0);
521 dmu_tx_hold_zap(oh->ot_tx, zapid, FALSE, NULL);
523 osd_declare_xattrs_destroy(env, obj, oh);
526 rc = osd_declare_quota(env, osd, obj->oo_attr.la_uid,
527 obj->oo_attr.la_gid, -1, oh, false, NULL, false);
531 /* data to be truncated */
532 rc = osd_declare_quota(env, osd, obj->oo_attr.la_uid,
533 obj->oo_attr.la_gid, 0, oh, true, NULL, false);
537 osd_object_set_destroy_type(obj);
538 if (obj->oo_destroy == OSD_DESTROY_SYNC)
539 dmu_tx_hold_free(oh->ot_tx, obj->oo_dn->dn_object,
542 dmu_tx_hold_zap(oh->ot_tx, osd->od_unlinkedid, TRUE, NULL);
544 /* will help to find FID->ino when this object is being
545 * added to PENDING/ */
546 osd_idc_find_and_init(env, osd, obj);
551 static int osd_object_destroy(const struct lu_env *env,
552 struct dt_object *dt, struct thandle *th)
554 struct osd_thread_info *info = osd_oti_get(env);
555 char *buf = info->oti_str;
556 struct osd_object *obj = osd_dt_obj(dt);
557 struct osd_device *osd = osd_obj2dev(obj);
558 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
559 struct osd_thandle *oh;
564 down_write(&obj->oo_guard);
566 if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
567 GOTO(out, rc = -ENOENT);
569 LASSERT(obj->oo_dn != NULL);
571 oh = container_of0(th, struct osd_thandle, ot_super);
573 LASSERT(oh->ot_tx != NULL);
575 /* remove obj ref from index dir (it depends) */
576 zapid = osd_get_name_n_idx(env, osd, fid, buf, sizeof(info->oti_str));
577 rc = -zap_remove(osd->od_os, zapid, buf, oh->ot_tx);
579 CERROR("%s: zap_remove(%s) failed: rc = %d\n",
580 osd->od_svname, buf, rc);
584 rc = osd_xattrs_destroy(env, obj, oh);
586 CERROR("%s: cannot destroy xattrs for %s: rc = %d\n",
587 osd->od_svname, buf, rc);
591 oid = obj->oo_dn->dn_object;
592 if (unlikely(obj->oo_destroy == OSD_DESTROY_NONE)) {
593 /* this may happen if the destroy wasn't declared
594 * e.g. when the object is created and then destroyed
595 * in the same transaction - we don't need additional
596 * space for destroy specifically */
597 LASSERT(obj->oo_attr.la_size <= osd_sync_destroy_max_size);
598 rc = -dmu_object_free(osd->od_os, oid, oh->ot_tx);
600 CERROR("%s: failed to free %s %llu: rc = %d\n",
601 osd->od_svname, buf, oid, rc);
602 } else if (obj->oo_destroy == OSD_DESTROY_SYNC) {
603 rc = -dmu_object_free(osd->od_os, oid, oh->ot_tx);
605 CERROR("%s: failed to free %s %llu: rc = %d\n",
606 osd->od_svname, buf, oid, rc);
607 } else { /* asynchronous destroy */
608 rc = osd_object_unlinked_add(obj, oh);
612 rc = -zap_add_int(osd->od_os, osd->od_unlinkedid,
615 CERROR("%s: zap_add_int() failed %s %llu: rc = %d\n",
616 osd->od_svname, buf, oid, rc);
620 /* not needed in the cache anymore */
621 set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
623 obj->oo_destroyed = 1;
624 up_write(&obj->oo_guard);
628 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
630 struct osd_object *obj = osd_obj(l);
632 if (obj->oo_dn != NULL) {
633 osd_object_sa_fini(obj);
634 if (obj->oo_sa_xattr) {
635 nvlist_free(obj->oo_sa_xattr);
636 obj->oo_sa_xattr = NULL;
638 osd_dnode_rele(obj->oo_dn);
639 list_del(&obj->oo_sa_linkage);
645 * Concurrency: ->loo_object_release() is called under site spin-lock.
647 static void osd_object_release(const struct lu_env *env,
653 * Concurrency: shouldn't matter.
655 static int osd_object_print(const struct lu_env *env, void *cookie,
656 lu_printer_t p, const struct lu_object *l)
658 struct osd_object *o = osd_obj(l);
660 return (*p)(env, cookie, LUSTRE_OSD_ZFS_NAME"-object@%p", o);
663 static void osd_object_read_lock(const struct lu_env *env,
664 struct dt_object *dt, unsigned role)
666 struct osd_object *obj = osd_dt_obj(dt);
668 LASSERT(osd_invariant(obj));
670 down_read_nested(&obj->oo_sem, role);
673 static void osd_object_write_lock(const struct lu_env *env,
674 struct dt_object *dt, unsigned role)
676 struct osd_object *obj = osd_dt_obj(dt);
678 LASSERT(osd_invariant(obj));
680 down_write_nested(&obj->oo_sem, role);
683 static void osd_object_read_unlock(const struct lu_env *env,
684 struct dt_object *dt)
686 struct osd_object *obj = osd_dt_obj(dt);
688 LASSERT(osd_invariant(obj));
689 up_read(&obj->oo_sem);
692 static void osd_object_write_unlock(const struct lu_env *env,
693 struct dt_object *dt)
695 struct osd_object *obj = osd_dt_obj(dt);
697 LASSERT(osd_invariant(obj));
698 up_write(&obj->oo_sem);
701 static int osd_object_write_locked(const struct lu_env *env,
702 struct dt_object *dt)
704 struct osd_object *obj = osd_dt_obj(dt);
707 LASSERT(osd_invariant(obj));
709 if (down_write_trylock(&obj->oo_sem)) {
711 up_write(&obj->oo_sem);
716 static int osd_attr_get(const struct lu_env *env,
717 struct dt_object *dt,
718 struct lu_attr *attr)
720 struct osd_object *obj = osd_dt_obj(dt);
725 down_read(&obj->oo_guard);
727 if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
728 GOTO(out, rc = -ENOENT);
730 LASSERT(osd_invariant(obj));
733 read_lock(&obj->oo_attr_lock);
734 *attr = obj->oo_attr;
735 if (obj->oo_lma_flags & LUSTRE_ORPHAN_FL)
736 attr->la_flags |= LUSTRE_ORPHAN_FL;
737 read_unlock(&obj->oo_attr_lock);
739 /* with ZFS_DEBUG zrl_add_debug() called by DB_DNODE_ENTER()
740 * from within sa_object_size() can block on a mutex, so
741 * we can't call sa_object_size() holding rwlock */
742 sa_object_size(obj->oo_sa_hdl, &blksize, &blocks);
743 /* we do not control size of indices, so always calculate
744 * it from number of blocks reported by DMU */
745 if (S_ISDIR(attr->la_mode))
746 attr->la_size = 512 * blocks;
747 /* Block size may be not set; suggest maximal I/O transfers. */
749 blksize = osd_spa_maxblocksize(
750 dmu_objset_spa(osd_obj2dev(obj)->od_os));
752 attr->la_blksize = blksize;
753 attr->la_blocks = blocks;
754 attr->la_valid |= LA_BLOCKS | LA_BLKSIZE;
757 up_read(&obj->oo_guard);
761 /* Simple wrapper on top of qsd API which implement quota transfer for osd
762 * setattr needs. As a reminder, only the root user can change ownership of
763 * a file, that's why EDQUOT & EINPROGRESS errors are discarded */
764 static inline int qsd_transfer(const struct lu_env *env,
765 struct qsd_instance *qsd,
766 struct lquota_trans *trans, int qtype,
767 __u64 orig_id, __u64 new_id, __u64 bspace,
768 struct lquota_id_info *qi)
772 if (unlikely(qsd == NULL))
775 LASSERT(qtype >= 0 && qtype < LL_MAXQUOTAS);
776 qi->lqi_type = qtype;
778 /* inode accounting */
779 qi->lqi_is_blk = false;
781 /* one more inode for the new owner ... */
782 qi->lqi_id.qid_uid = new_id;
784 rc = qsd_op_begin(env, qsd, trans, qi, NULL);
785 if (rc == -EDQUOT || rc == -EINPROGRESS)
790 /* and one less inode for the current id */
791 qi->lqi_id.qid_uid = orig_id;;
793 /* can't get EDQUOT when reducing usage */
794 rc = qsd_op_begin(env, qsd, trans, qi, NULL);
795 if (rc == -EINPROGRESS)
800 /* block accounting */
801 qi->lqi_is_blk = true;
803 /* more blocks for the new owner ... */
804 qi->lqi_id.qid_uid = new_id;
805 qi->lqi_space = bspace;
806 rc = qsd_op_begin(env, qsd, trans, qi, NULL);
807 if (rc == -EDQUOT || rc == -EINPROGRESS)
812 /* and finally less blocks for the current owner */
813 qi->lqi_id.qid_uid = orig_id;
814 qi->lqi_space = -bspace;
815 rc = qsd_op_begin(env, qsd, trans, qi, NULL);
816 /* can't get EDQUOT when reducing usage */
817 if (rc == -EINPROGRESS)
822 static int osd_declare_attr_set(const struct lu_env *env,
823 struct dt_object *dt,
824 const struct lu_attr *attr,
825 struct thandle *handle)
827 struct osd_thread_info *info = osd_oti_get(env);
828 struct osd_object *obj = osd_dt_obj(dt);
829 struct osd_device *osd = osd_obj2dev(obj);
831 struct osd_thandle *oh;
839 LASSERT(handle != NULL);
840 LASSERT(osd_invariant(obj));
842 oh = container_of0(handle, struct osd_thandle, ot_super);
844 down_read(&obj->oo_guard);
845 if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
848 LASSERT(obj->oo_sa_hdl != NULL);
849 LASSERT(oh->ot_tx != NULL);
850 /* regular attributes are part of the bonus buffer */
851 /* let's check whether this object is already part of
854 for (txh = list_head(&oh->ot_tx->tx_holds); txh;
855 txh = list_next(&oh->ot_tx->tx_holds, txh)) {
856 if (txh->txh_dnode == NULL)
858 if (txh->txh_dnode->dn_object != obj->oo_dn->dn_object)
860 /* this object is part of the transaction already
861 * we don't need to declare bonus again */
866 dmu_tx_hold_bonus(oh->ot_tx, obj->oo_dn->dn_object);
867 if (oh->ot_tx->tx_err != 0)
868 GOTO(out, rc = -oh->ot_tx->tx_err);
870 if (attr && attr->la_valid & LA_FLAGS) {
871 /* LMA is usually a part of bonus, no need to declare
875 if (attr && (attr->la_valid & (LA_UID | LA_GID))) {
876 sa_object_size(obj->oo_sa_hdl, &blksize, &bspace);
877 bspace = toqb(bspace * blksize);
880 if (attr && attr->la_valid & LA_UID) {
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 /* quota enforcement for group */
893 if (attr->la_gid != obj->oo_attr.la_gid) {
894 rc = qsd_transfer(env, osd->od_quota_slave,
895 &oh->ot_quota_trans, GRPQUOTA,
896 obj->oo_attr.la_gid, attr->la_gid,
897 bspace, &info->oti_qi);
904 up_read(&obj->oo_guard);
909 * Set the attributes of an object
911 * The transaction passed to this routine must have
912 * dmu_tx_hold_bonus(tx, oid) called and then assigned
913 * to a transaction group.
915 static int osd_attr_set(const struct lu_env *env, struct dt_object *dt,
916 const struct lu_attr *la, struct thandle *handle)
918 struct osd_thread_info *info = osd_oti_get(env);
919 sa_bulk_attr_t *bulk = osd_oti_get(env)->oti_attr_bulk;
920 struct osd_object *obj = osd_dt_obj(dt);
921 struct osd_device *osd = osd_obj2dev(obj);
922 struct osd_thandle *oh;
923 struct osa_attr *osa = &info->oti_osa;
924 __u64 valid = la->la_valid;
930 down_read(&obj->oo_guard);
931 if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
932 GOTO(out, rc = -ENOENT);
934 LASSERT(handle != NULL);
935 LASSERT(osd_invariant(obj));
936 LASSERT(obj->oo_sa_hdl);
938 oh = container_of0(handle, struct osd_thandle, ot_super);
939 /* Assert that the transaction has been assigned to a
940 transaction group. */
941 LASSERT(oh->ot_tx->tx_txg != 0);
943 /* Only allow set size for regular file */
944 if (!S_ISREG(dt->do_lu.lo_header->loh_attr))
945 valid &= ~(LA_SIZE | LA_BLOCKS);
947 if (valid & LA_CTIME && la->la_ctime == obj->oo_attr.la_ctime)
950 if (valid & LA_MTIME && la->la_mtime == obj->oo_attr.la_mtime)
953 if (valid & LA_ATIME && la->la_atime == obj->oo_attr.la_atime)
959 if (valid & LA_FLAGS) {
960 struct lustre_mdt_attrs *lma;
963 if (la->la_flags & LUSTRE_LMA_FL_MASKS) {
964 CLASSERT(sizeof(info->oti_buf) >= sizeof(*lma));
965 lma = (struct lustre_mdt_attrs *)&info->oti_buf;
967 buf.lb_len = sizeof(info->oti_buf);
968 rc = osd_xattr_get(env, &obj->oo_dt, &buf,
972 le32_to_cpu(lma->lma_incompat);
974 lustre_to_lma_flags(la->la_flags);
976 cpu_to_le32(lma->lma_incompat);
978 buf.lb_len = sizeof(*lma);
979 rc = osd_xattr_set_internal(env, obj, &buf,
985 CWARN("%s: failed to set LMA flags: rc = %d\n",
992 write_lock(&obj->oo_attr_lock);
994 if (valid & LA_ATIME) {
995 osa->atime[0] = obj->oo_attr.la_atime = la->la_atime;
996 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(osd), NULL,
999 if (valid & LA_MTIME) {
1000 osa->mtime[0] = obj->oo_attr.la_mtime = la->la_mtime;
1001 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(osd), NULL,
1004 if (valid & LA_CTIME) {
1005 osa->ctime[0] = obj->oo_attr.la_ctime = la->la_ctime;
1006 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(osd), NULL,
1009 if (valid & LA_MODE) {
1010 /* mode is stored along with type, so read it first */
1011 obj->oo_attr.la_mode = (obj->oo_attr.la_mode & S_IFMT) |
1012 (la->la_mode & ~S_IFMT);
1013 osa->mode = obj->oo_attr.la_mode;
1014 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(osd), NULL,
1017 if (valid & LA_SIZE) {
1018 osa->size = obj->oo_attr.la_size = la->la_size;
1019 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(osd), NULL,
1022 if (valid & LA_NLINK) {
1023 osa->nlink = obj->oo_attr.la_nlink = la->la_nlink;
1024 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(osd), NULL,
1027 if (valid & LA_RDEV) {
1028 osa->rdev = obj->oo_attr.la_rdev = la->la_rdev;
1029 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_RDEV(osd), NULL,
1032 if (valid & LA_FLAGS) {
1033 osa->flags = attrs_fs2zfs(la->la_flags);
1034 /* many flags are not supported by zfs, so ensure a good cached
1036 obj->oo_attr.la_flags = attrs_zfs2fs(osa->flags);
1037 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(osd), NULL,
1040 if (valid & LA_UID) {
1041 osa->uid = obj->oo_attr.la_uid = la->la_uid;
1042 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(osd), NULL,
1045 if (valid & LA_GID) {
1046 osa->gid = obj->oo_attr.la_gid = la->la_gid;
1047 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(osd), NULL,
1050 obj->oo_attr.la_valid |= valid;
1051 write_unlock(&obj->oo_attr_lock);
1053 LASSERT(cnt <= ARRAY_SIZE(osd_oti_get(env)->oti_attr_bulk));
1054 rc = osd_object_sa_bulk_update(obj, bulk, cnt, oh);
1057 up_read(&obj->oo_guard);
1064 * XXX temporary solution.
1067 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
1068 struct dt_object *parent, struct dt_object *child,
1073 ah->dah_parent = parent;
1074 ah->dah_mode = child_mode;
1076 if (parent != NULL && !dt_object_remote(parent)) {
1077 /* will help to find FID->ino at dt_insert("..") */
1078 struct osd_object *pobj = osd_dt_obj(parent);
1080 osd_idc_find_and_init(env, osd_obj2dev(pobj), pobj);
1084 static int osd_declare_object_create(const struct lu_env *env,
1085 struct dt_object *dt,
1086 struct lu_attr *attr,
1087 struct dt_allocation_hint *hint,
1088 struct dt_object_format *dof,
1089 struct thandle *handle)
1091 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1092 struct osd_object *obj = osd_dt_obj(dt);
1093 struct osd_device *osd = osd_obj2dev(obj);
1094 struct osd_thandle *oh;
1101 switch (dof->dof_type) {
1105 if (obj->oo_dt.do_body_ops == NULL)
1106 obj->oo_dt.do_body_ops = &osd_body_ops;
1112 LASSERT(handle != NULL);
1113 oh = container_of0(handle, struct osd_thandle, ot_super);
1114 LASSERT(oh->ot_tx != NULL);
1116 /* this is the minimum set of EAs on every Lustre object */
1117 obj->oo_ea_in_bonus = ZFS_SA_BASE_ATTR_SIZE +
1118 sizeof(__u64) + /* VBR VERSION */
1119 sizeof(struct lustre_mdt_attrs); /* LMA */
1120 /* reserve 32 bytes for extra stuff like ACLs */
1121 dnode_size = size_roundup_power2(obj->oo_ea_in_bonus + 32);
1123 switch (dof->dof_type) {
1125 dt->do_index_ops = &osd_dir_ops;
1127 /* for zap create */
1128 dmu_tx_hold_zap(oh->ot_tx, DMU_NEW_OBJECT, FALSE, NULL);
1129 dmu_tx_hold_sa_create(oh->ot_tx, dnode_size);
1134 /* first, we'll create new object */
1135 dmu_tx_hold_sa_create(oh->ot_tx, dnode_size);
1143 /* and we'll add it to some mapping */
1144 zapid = osd_get_name_n_idx(env, osd, fid, NULL, 0);
1145 dmu_tx_hold_zap(oh->ot_tx, zapid, TRUE, NULL);
1147 /* will help to find FID->ino mapping at dt_insert() */
1148 osd_idc_find_and_init(env, osd, obj);
1150 rc = osd_declare_quota(env, osd, attr->la_uid, attr->la_gid, 1, oh,
1151 false, NULL, false);
1156 int __osd_attr_init(const struct lu_env *env, struct osd_device *osd,
1157 sa_handle_t *sa_hdl, dmu_tx_t *tx,
1158 struct lu_attr *la, uint64_t parent)
1160 sa_bulk_attr_t *bulk = osd_oti_get(env)->oti_attr_bulk;
1161 struct osa_attr *osa = &osd_oti_get(env)->oti_osa;
1170 gen = dmu_tx_get_txg(tx);
1172 ZFS_TIME_ENCODE(&now, crtime);
1174 osa->atime[0] = la->la_atime;
1175 osa->ctime[0] = la->la_ctime;
1176 osa->mtime[0] = la->la_mtime;
1177 osa->mode = la->la_mode;
1178 osa->uid = la->la_uid;
1179 osa->gid = la->la_gid;
1180 osa->rdev = la->la_rdev;
1181 osa->nlink = la->la_nlink;
1182 osa->flags = attrs_fs2zfs(la->la_flags);
1183 osa->size = la->la_size;
1186 * we need to create all SA below upon object create.
1188 * XXX The attribute order matters since the accounting callback relies
1189 * on static offsets (i.e. SA_*_OFFSET, see zfs_space_delta_cb()) to
1190 * look up the UID/GID attributes. Moreover, the callback does not seem
1191 * to support the spill block.
1192 * We define attributes in the same order as SA_*_OFFSET in order to
1193 * work around the problem. See ORI-610.
1196 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(osd), NULL, &osa->mode, 8);
1197 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(osd), NULL, &osa->size, 8);
1198 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GEN(osd), NULL, &gen, 8);
1199 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(osd), NULL, &osa->uid, 8);
1200 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(osd), NULL, &osa->gid, 8);
1201 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_PARENT(osd), NULL, &parent, 8);
1202 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(osd), NULL, &osa->flags, 8);
1203 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(osd), NULL, osa->atime, 16);
1204 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(osd), NULL, osa->mtime, 16);
1205 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(osd), NULL, osa->ctime, 16);
1206 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CRTIME(osd), NULL, crtime, 16);
1207 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(osd), NULL, &osa->nlink, 8);
1208 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_RDEV(osd), NULL, &osa->rdev, 8);
1209 LASSERT(cnt <= ARRAY_SIZE(osd_oti_get(env)->oti_attr_bulk));
1211 rc = -sa_replace_all_by_template(sa_hdl, bulk, cnt, tx);
1216 static int osd_find_new_dnode(const struct lu_env *env, dmu_tx_t *tx,
1217 uint64_t oid, dnode_t **dnp)
1222 /* take dnode_t from tx to save on dnode#->dnode_t lookup */
1223 for (txh = list_tail(&tx->tx_holds); txh;
1224 txh = list_prev(&tx->tx_holds, txh)) {
1225 dnode_t *dn = txh->txh_dnode;
1230 if (dn->dn_object != oid)
1234 rw_enter(&dn->dn_struct_rwlock, RW_WRITER);
1235 if (dn->dn_bonus == NULL)
1236 dbuf_create_bonus(dn);
1237 rw_exit(&dn->dn_struct_rwlock);
1241 LASSERT(dn->dn_handle);
1243 if (refcount_add(&db->db_holds, osd_obj_tag) == 1) {
1244 refcount_add(&dn->dn_holds, tag);
1245 atomic_inc_32(&dn->dn_dbufs_count);
1251 if (unlikely(*dnp == NULL))
1252 rc = __osd_obj2dnode(env, tx->tx_objset, oid, dnp);
1258 * The transaction passed to this routine must have
1259 * dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT) called and then assigned
1260 * to a transaction group.
1262 int __osd_object_create(const struct lu_env *env, struct osd_object *obj,
1263 dnode_t **dnp, dmu_tx_t *tx, struct lu_attr *la)
1265 struct osd_device *osd = osd_obj2dev(obj);
1266 const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
1267 dmu_object_type_t type = DMU_OT_PLAIN_FILE_CONTENTS;
1270 /* Use DMU_OTN_UINT8_METADATA for local objects so their data blocks
1271 * would get an additional ditto copy */
1272 if (unlikely(S_ISREG(la->la_mode) &&
1273 fid_seq_is_local_file(fid_seq(fid))))
1274 type = DMU_OTN_UINT8_METADATA;
1276 /* Create a new DMU object using the default dnode size. */
1277 oid = osd_dmu_object_alloc(osd->od_os, type, 0, 0, tx);
1279 LASSERT(la->la_valid & LA_MODE);
1283 return osd_find_new_dnode(env, tx, oid, dnp);
1287 * The transaction passed to this routine must have
1288 * dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, ...) called and then assigned
1289 * to a transaction group.
1291 * Using ZAP_FLAG_HASH64 will force the ZAP to always be a FAT ZAP.
1292 * This is fine for directories today, because storing the FID in the dirent
1293 * will also require a FAT ZAP. If there is a new type of micro ZAP created
1294 * then we might need to re-evaluate the use of this flag and instead do
1295 * a conversion from the different internal ZAP hash formats being used. */
1296 int __osd_zap_create(const struct lu_env *env, struct osd_device *osd,
1297 dnode_t **dnp, dmu_tx_t *tx, struct lu_attr *la,
1302 /* Assert that the transaction has been assigned to a
1303 transaction group. */
1304 LASSERT(tx->tx_txg != 0);
1307 oid = osd_zap_create_flags(osd->od_os, 0, flags | ZAP_FLAG_HASH64,
1308 DMU_OT_DIRECTORY_CONTENTS,
1309 14, /* == ZFS fzap_default_blockshift */
1310 DN_MAX_INDBLKSHIFT, /* indirect blockshift */
1316 return osd_find_new_dnode(env, tx, oid, dnp);
1319 static dnode_t *osd_mkidx(const struct lu_env *env, struct osd_object *obj,
1320 struct lu_attr *la, struct osd_thandle *oh)
1325 /* Index file should be created as regular file in order not to confuse
1326 * ZPL which could interpret them as directory.
1327 * We set ZAP_FLAG_UINT64_KEY to let ZFS know than we are going to use
1329 LASSERT(S_ISREG(la->la_mode));
1330 rc = __osd_zap_create(env, osd_obj2dev(obj), &dn, oh->ot_tx, la,
1331 ZAP_FLAG_UINT64_KEY);
1337 static dnode_t *osd_mkdir(const struct lu_env *env, struct osd_object *obj,
1338 struct lu_attr *la, struct osd_thandle *oh)
1343 LASSERT(S_ISDIR(la->la_mode));
1344 rc = __osd_zap_create(env, osd_obj2dev(obj), &dn, oh->ot_tx, la, 0);
1350 static dnode_t *osd_mkreg(const struct lu_env *env, struct osd_object *obj,
1351 struct lu_attr *la, struct osd_thandle *oh)
1353 const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
1354 struct osd_device *osd = osd_obj2dev(obj);
1358 LASSERT(S_ISREG(la->la_mode));
1359 rc = __osd_object_create(env, obj, &dn, oh->ot_tx, la);
1363 if ((fid_is_idif(fid) || fid_is_norm(fid) || fid_is_echo(fid)) &&
1365 /* The minimum block size must be at least page size otherwise
1366 * it will break the assumption in tgt_thread_big_cache where
1367 * the array size is PTLRPC_MAX_BRW_PAGES. It will also affect
1368 * RDMA due to subpage transfer size */
1369 rc = -dmu_object_set_blocksize(osd->od_os, dn->dn_object,
1370 PAGE_SIZE, 0, oh->ot_tx);
1372 CERROR("%s: can't change blocksize: %d\n",
1373 osd->od_svname, rc);
1381 static dnode_t *osd_mksym(const struct lu_env *env, struct osd_object *obj,
1382 struct lu_attr *la, struct osd_thandle *oh)
1387 LASSERT(S_ISLNK(la->la_mode));
1388 rc = __osd_object_create(env, obj, &dn, oh->ot_tx, la);
1394 static dnode_t *osd_mknod(const struct lu_env *env, struct osd_object *obj,
1395 struct lu_attr *la, struct osd_thandle *oh)
1400 if (S_ISCHR(la->la_mode) || S_ISBLK(la->la_mode))
1401 la->la_valid |= LA_RDEV;
1403 rc = __osd_object_create(env, obj, &dn, oh->ot_tx, la);
1409 typedef dnode_t *(*osd_obj_type_f)(const struct lu_env *env,
1410 struct osd_object *obj,
1412 struct osd_thandle *oh);
1414 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
1416 osd_obj_type_f result;
1442 * Concurrency: @dt is write locked.
1444 static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
1445 struct lu_attr *attr,
1446 struct dt_allocation_hint *hint,
1447 struct dt_object_format *dof,
1450 struct osd_thread_info *info = osd_oti_get(env);
1451 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
1452 struct zpl_direntry *zde = &info->oti_zde.lzd_reg;
1453 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1454 struct osd_object *obj = osd_dt_obj(dt);
1455 struct osd_device *osd = osd_obj2dev(obj);
1456 char *buf = info->oti_str;
1457 struct osd_thandle *oh;
1459 uint64_t zapid, parent = 0;
1464 /* concurrent create declarations should not see
1465 * the object inconsistent (db, attr, etc).
1466 * in regular cases acquisition should be cheap */
1467 down_write(&obj->oo_guard);
1469 if (unlikely(dt_object_exists(dt)))
1470 GOTO(out, rc = -EEXIST);
1472 LASSERT(osd_invariant(obj));
1473 LASSERT(dof != NULL);
1475 LASSERT(th != NULL);
1476 oh = container_of0(th, struct osd_thandle, ot_super);
1478 LASSERT(obj->oo_dn == NULL);
1480 /* to follow ZFS on-disk format we need
1481 * to initialize parent dnode properly */
1482 if (hint != NULL && hint->dah_parent != NULL &&
1483 !dt_object_remote(hint->dah_parent))
1484 parent = osd_dt_obj(hint->dah_parent)->oo_dn->dn_object;
1486 /* we may fix some attributes, better do not change the source */
1487 obj->oo_attr = *attr;
1488 obj->oo_attr.la_valid |= LA_SIZE | LA_NLINK | LA_TYPE;
1490 dn = osd_create_type_f(dof->dof_type)(env, obj, &obj->oo_attr, oh);
1498 zde->zde_dnode = dn->dn_object;
1499 zde->zde_type = IFTODT(attr->la_mode & S_IFMT);
1501 zapid = osd_get_name_n_idx(env, osd, fid, buf, sizeof(info->oti_str));
1503 rc = -zap_add(osd->od_os, zapid, buf, 8, 1, zde, oh->ot_tx);
1507 /* Now add in all of the "SA" attributes */
1508 rc = -sa_handle_get(osd->od_os, dn->dn_object, NULL,
1509 SA_HDL_PRIVATE, &obj->oo_sa_hdl);
1513 /* configure new osd object */
1515 parent = parent != 0 ? parent : zapid;
1516 rc = __osd_attr_init(env, osd, obj->oo_sa_hdl, oh->ot_tx,
1517 &obj->oo_attr, parent);
1521 /* XXX: oo_lma_flags */
1522 obj->oo_dt.do_lu.lo_header->loh_attr |= obj->oo_attr.la_mode & S_IFMT;
1524 obj->oo_dt.do_lu.lo_header->loh_attr |= LOHA_EXISTS;
1525 if (likely(!fid_is_acct(lu_object_fid(&obj->oo_dt.do_lu))))
1526 /* no body operations for accounting objects */
1527 obj->oo_dt.do_body_ops = &osd_body_ops;
1529 rc = -nvlist_alloc(&obj->oo_sa_xattr, NV_UNIQUE_NAME, KM_SLEEP);
1533 /* initialize LMA */
1534 lustre_lma_init(lma, lu_object_fid(&obj->oo_dt.do_lu), 0, 0);
1535 lustre_lma_swab(lma);
1536 rc = -nvlist_add_byte_array(obj->oo_sa_xattr, XATTR_NAME_LMA,
1537 (uchar_t *)lma, sizeof(*lma));
1540 rc = __osd_sa_xattr_update(env, obj, oh);
1543 osd_idc_find_and_init(env, osd, obj);
1546 if (unlikely(rc && dn)) {
1547 dmu_object_free(osd->od_os, dn->dn_object, oh->ot_tx);
1551 up_write(&obj->oo_guard);
1555 static int osd_declare_object_ref_add(const struct lu_env *env,
1556 struct dt_object *dt,
1559 return osd_declare_attr_set(env, dt, NULL, th);
1563 * Concurrency: @dt is write locked.
1565 static int osd_object_ref_add(const struct lu_env *env,
1566 struct dt_object *dt,
1567 struct thandle *handle)
1569 struct osd_object *obj = osd_dt_obj(dt);
1570 struct osd_thandle *oh;
1571 struct osd_device *osd = osd_obj2dev(obj);
1577 down_read(&obj->oo_guard);
1578 if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
1579 GOTO(out, rc = -ENOENT);
1581 LASSERT(osd_invariant(obj));
1582 LASSERT(obj->oo_sa_hdl != NULL);
1584 oh = container_of0(handle, struct osd_thandle, ot_super);
1586 write_lock(&obj->oo_attr_lock);
1587 nlink = ++obj->oo_attr.la_nlink;
1588 write_unlock(&obj->oo_attr_lock);
1590 rc = osd_object_sa_update(obj, SA_ZPL_LINKS(osd), &nlink, 8, oh);
1593 up_read(&obj->oo_guard);
1597 static int osd_declare_object_ref_del(const struct lu_env *env,
1598 struct dt_object *dt,
1599 struct thandle *handle)
1601 return osd_declare_attr_set(env, dt, NULL, handle);
1605 * Concurrency: @dt is write locked.
1607 static int osd_object_ref_del(const struct lu_env *env,
1608 struct dt_object *dt,
1609 struct thandle *handle)
1611 struct osd_object *obj = osd_dt_obj(dt);
1612 struct osd_thandle *oh;
1613 struct osd_device *osd = osd_obj2dev(obj);
1619 down_read(&obj->oo_guard);
1621 if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
1622 GOTO(out, rc = -ENOENT);
1624 LASSERT(osd_invariant(obj));
1625 LASSERT(obj->oo_sa_hdl != NULL);
1627 oh = container_of0(handle, struct osd_thandle, ot_super);
1628 LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
1630 write_lock(&obj->oo_attr_lock);
1631 nlink = --obj->oo_attr.la_nlink;
1632 write_unlock(&obj->oo_attr_lock);
1634 rc = osd_object_sa_update(obj, SA_ZPL_LINKS(osd), &nlink, 8, oh);
1637 up_read(&obj->oo_guard);
1641 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
1642 __u64 start, __u64 end)
1644 struct osd_device *osd = osd_obj2dev(osd_dt_obj(dt));
1647 /* XXX: no other option than syncing the whole filesystem until we
1648 * support ZIL. If the object tracked the txg that it was last
1649 * modified in, it could pass that txg here instead of "0". Maybe
1650 * the changes are already committed, so no wait is needed at all? */
1651 if (!osd->od_dt_dev.dd_rdonly)
1652 txg_wait_synced(dmu_objset_pool(osd->od_os), 0ULL);
1657 static int osd_invalidate(const struct lu_env *env, struct dt_object *dt)
1662 static struct dt_object_operations osd_obj_ops = {
1663 .do_read_lock = osd_object_read_lock,
1664 .do_write_lock = osd_object_write_lock,
1665 .do_read_unlock = osd_object_read_unlock,
1666 .do_write_unlock = osd_object_write_unlock,
1667 .do_write_locked = osd_object_write_locked,
1668 .do_attr_get = osd_attr_get,
1669 .do_declare_attr_set = osd_declare_attr_set,
1670 .do_attr_set = osd_attr_set,
1671 .do_ah_init = osd_ah_init,
1672 .do_declare_create = osd_declare_object_create,
1673 .do_create = osd_object_create,
1674 .do_declare_destroy = osd_declare_object_destroy,
1675 .do_destroy = osd_object_destroy,
1676 .do_index_try = osd_index_try,
1677 .do_declare_ref_add = osd_declare_object_ref_add,
1678 .do_ref_add = osd_object_ref_add,
1679 .do_declare_ref_del = osd_declare_object_ref_del,
1680 .do_ref_del = osd_object_ref_del,
1681 .do_xattr_get = osd_xattr_get,
1682 .do_declare_xattr_set = osd_declare_xattr_set,
1683 .do_xattr_set = osd_xattr_set,
1684 .do_declare_xattr_del = osd_declare_xattr_del,
1685 .do_xattr_del = osd_xattr_del,
1686 .do_xattr_list = osd_xattr_list,
1687 .do_object_sync = osd_object_sync,
1688 .do_invalidate = osd_invalidate,
1691 static struct lu_object_operations osd_lu_obj_ops = {
1692 .loo_object_init = osd_object_init,
1693 .loo_object_delete = osd_object_delete,
1694 .loo_object_release = osd_object_release,
1695 .loo_object_free = osd_object_free,
1696 .loo_object_print = osd_object_print,
1697 .loo_object_invariant = osd_object_invariant,
1700 static int osd_otable_it_attr_get(const struct lu_env *env,
1701 struct dt_object *dt,
1702 struct lu_attr *attr)
1708 static struct dt_object_operations osd_obj_otable_it_ops = {
1709 .do_attr_get = osd_otable_it_attr_get,
1710 .do_index_try = osd_index_try,