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 <libcfs/libcfs.h>
42 #include <obd_support.h>
43 #include <lustre_net.h>
45 #include <obd_class.h>
46 #include <lustre_disk.h>
47 #include <lustre_fid.h>
49 #include "osd_internal.h"
51 #include <sys/dnode.h>
56 #include <sys/spa_impl.h>
57 #include <sys/zfs_znode.h>
58 #include <sys/dmu_tx.h>
59 #include <sys/dmu_objset.h>
60 #include <sys/dsl_prop.h>
61 #include <sys/sa_impl.h>
64 char *osd_obj_tag = "osd_object";
66 static struct dt_object_operations osd_obj_ops;
67 static struct lu_object_operations osd_lu_obj_ops;
68 extern struct dt_body_operations osd_body_ops;
69 static struct dt_object_operations osd_obj_otable_it_ops;
71 extern struct kmem_cache *osd_object_kmem;
74 osd_object_sa_fini(struct osd_object *obj)
77 sa_handle_destroy(obj->oo_sa_hdl);
78 obj->oo_sa_hdl = NULL;
83 osd_object_sa_init(struct osd_object *obj, struct osd_device *o)
87 LASSERT(obj->oo_sa_hdl == NULL);
88 LASSERT(obj->oo_dn != NULL);
90 rc = osd_sa_handle_get(obj);
94 /* Cache the xattr object id, valid for the life of the object */
95 rc = -sa_lookup(obj->oo_sa_hdl, SA_ZPL_XATTR(o), &obj->oo_xattr, 8);
97 obj->oo_xattr = ZFS_NO_OBJECT;
100 osd_object_sa_fini(obj);
107 * Add object to list of dirty objects in tx handle.
109 void osd_object_sa_dirty_add(struct osd_object *obj, struct osd_thandle *oh)
111 if (!list_empty(&obj->oo_sa_linkage))
114 write_lock(&obj->oo_attr_lock);
115 if (likely(list_empty(&obj->oo_sa_linkage)))
116 list_add(&obj->oo_sa_linkage, &oh->ot_sa_list);
117 write_unlock(&obj->oo_attr_lock);
121 * Release spill block dbuf hold for all dirty SAs.
123 void osd_object_sa_dirty_rele(const struct lu_env *env, struct osd_thandle *oh)
125 struct osd_object *obj;
127 while (!list_empty(&oh->ot_sa_list)) {
128 obj = list_entry(oh->ot_sa_list.next,
129 struct osd_object, oo_sa_linkage);
130 write_lock(&obj->oo_attr_lock);
131 list_del_init(&obj->oo_sa_linkage);
132 write_unlock(&obj->oo_attr_lock);
133 if (obj->oo_late_xattr) {
135 * take oo_guard to protect oo_sa_xattr buffer
136 * from concurrent update by osd_xattr_set()
138 LASSERT(oh->ot_assigned != 0);
139 down_write(&obj->oo_guard);
140 if (obj->oo_late_attr_set)
141 __osd_sa_attr_init(env, obj, oh);
142 else if (obj->oo_late_xattr)
143 __osd_sa_xattr_update(env, obj, oh);
144 up_write(&obj->oo_guard);
146 sa_spill_rele(obj->oo_sa_hdl);
151 * Update the SA and add the object to the dirty list.
153 int osd_object_sa_update(struct osd_object *obj, sa_attr_type_t type,
154 void *buf, uint32_t buflen, struct osd_thandle *oh)
158 LASSERT(obj->oo_sa_hdl != NULL);
159 LASSERT(oh->ot_tx != NULL);
161 rc = -sa_update(obj->oo_sa_hdl, type, buf, buflen, oh->ot_tx);
162 osd_object_sa_dirty_add(obj, oh);
168 * Bulk update the SA and add the object to the dirty list.
171 osd_object_sa_bulk_update(struct osd_object *obj, sa_bulk_attr_t *attrs,
172 int count, struct osd_thandle *oh)
176 LASSERT(obj->oo_sa_hdl != NULL);
177 LASSERT(oh->ot_tx != NULL);
179 rc = -sa_bulk_update(obj->oo_sa_hdl, attrs, count, oh->ot_tx);
180 osd_object_sa_dirty_add(obj, oh);
186 * Retrieve the attributes of a DMU object
188 int __osd_object_attr_get(const struct lu_env *env, struct osd_device *o,
189 struct osd_object *obj, struct lu_attr *la)
191 struct osa_attr *osa = &osd_oti_get(env)->oti_osa;
192 sa_bulk_attr_t *bulk = osd_oti_get(env)->oti_attr_bulk;
197 LASSERT(obj->oo_dn != NULL);
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(obj->oo_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(obj->oo_sa_hdl, SA_ZPL_RDEV(o), &osa->rdev, 8);
253 la->la_rdev = osa->rdev;
254 la->la_valid |= LA_RDEV;
261 int __osd_obj2dnode(objset_t *os, uint64_t oid, dnode_t **dnp)
267 rc = -dmu_bonus_hold(os, oid, osd_obj_tag, &db);
271 dbi = (dmu_buf_impl_t *)db;
273 *dnp = DB_DNODE(dbi);
274 LASSERT(*dnp != NULL);
280 * Concurrency: no concurrent access is possible that early in object
283 struct lu_object *osd_object_alloc(const struct lu_env *env,
284 const struct lu_object_header *hdr,
287 struct osd_object *mo;
289 OBD_SLAB_ALLOC_PTR_GFP(mo, osd_object_kmem, GFP_NOFS);
293 l = &mo->oo_dt.do_lu;
294 dt_object_init(&mo->oo_dt, NULL, d);
295 mo->oo_dt.do_ops = &osd_obj_ops;
296 l->lo_ops = &osd_lu_obj_ops;
297 INIT_LIST_HEAD(&mo->oo_sa_linkage);
298 INIT_LIST_HEAD(&mo->oo_unlinked_linkage);
299 init_rwsem(&mo->oo_sem);
300 init_rwsem(&mo->oo_guard);
301 rwlock_init(&mo->oo_attr_lock);
302 mo->oo_destroy = OSD_DESTROY_NONE;
310 * Concurrency: shouldn't matter.
312 int osd_object_init0(const struct lu_env *env, struct osd_object *obj)
314 struct osd_device *osd = osd_obj2dev(obj);
315 const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
319 if (obj->oo_dn == NULL)
324 rc = osd_object_sa_init(obj, osd);
328 /* cache attrs in object */
329 rc = __osd_object_attr_get(env, osd, obj, &obj->oo_attr);
333 if (likely(!fid_is_acct(fid)))
334 /* no body operations for accounting objects */
335 obj->oo_dt.do_body_ops = &osd_body_ops;
338 * initialize object before marking it existing
340 obj->oo_dt.do_lu.lo_header->loh_attr |= obj->oo_attr.la_mode & S_IFMT;
343 obj->oo_dt.do_lu.lo_header->loh_attr |= LOHA_EXISTS;
348 static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
350 struct osd_thread_info *info = osd_oti_get(env);
353 struct lustre_mdt_attrs *lma;
356 CLASSERT(sizeof(info->oti_buf) >= sizeof(*lma));
357 lma = (struct lustre_mdt_attrs *)info->oti_buf;
359 buf.lb_len = sizeof(info->oti_buf);
361 rc = osd_xattr_get(env, &obj->oo_dt, &buf, XATTR_NAME_LMA);
364 lustre_lma_swab(lma);
365 if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
366 CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT))) {
367 CWARN("%s: unsupported incompat LMA feature(s) %#x for "
368 "fid = "DFID"\n", osd_obj2dev(obj)->od_svname,
369 lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
370 PFID(lu_object_fid(&obj->oo_dt.do_lu)));
373 } else if (rc == -ENODATA) {
374 /* haven't initialize LMA xattr */
382 * Helper function to retrieve DMU object id from fid for accounting object
384 static dnode_t *osd_quota_fid2dmu(const struct osd_device *osd,
385 const struct lu_fid *fid)
389 LASSERT(fid_is_acct(fid));
391 switch (fid_oid(fid)) {
393 dn = osd->od_userused_dn;
396 dn = osd->od_groupused_dn;
406 * Concurrency: no concurrent access is possible that early in object
409 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
410 const struct lu_object_conf *conf)
412 struct osd_object *obj = osd_obj(l);
413 struct osd_device *osd = osd_obj2dev(obj);
414 const struct lu_fid *fid = lu_object_fid(l);
419 LASSERT(osd_invariant(obj));
421 if (fid_is_otable_it(&l->lo_header->loh_fid)) {
422 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
423 l->lo_header->loh_attr |= LOHA_EXISTS;
427 if (conf != NULL && conf->loc_flags & LOC_F_NEW)
430 if (unlikely(fid_is_acct(fid))) {
431 obj->oo_dn = osd_quota_fid2dmu(osd, fid);
433 obj->oo_dt.do_index_ops = &osd_acct_index_ops;
434 l->lo_header->loh_attr |= LOHA_EXISTS;
440 rc = osd_fid_lookup(env, osd, fid, &oid);
442 LASSERT(obj->oo_dn == NULL);
443 rc = __osd_obj2dnode(osd->od_os, oid, &obj->oo_dn);
444 /* EEXIST will be returned if object is being deleted in ZFS */
450 CERROR("%s: lookup "DFID"/%#llx failed: rc = %d\n",
451 osd->od_svname, PFID(lu_object_fid(l)), oid, rc);
455 rc = osd_object_init0(env, obj);
459 rc = osd_check_lma(env, obj);
462 } else if (rc == -ENOENT) {
465 LASSERT(osd_invariant(obj));
471 * Concurrency: no concurrent access is possible that late in object
474 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
476 struct osd_object *obj = osd_obj(l);
478 LASSERT(osd_invariant(obj));
480 dt_object_fini(&obj->oo_dt);
481 OBD_SLAB_FREE_PTR(obj, osd_object_kmem);
485 osd_object_unlinked_add(struct osd_object *obj, struct osd_thandle *oh)
489 LASSERT(obj->oo_destroy == OSD_DESTROY_ASYNC);
491 /* the object is supposed to be exclusively locked by
492 * the caller (osd_destroy()), while the transaction
493 * (oh) is per-thread and not shared */
494 if (likely(list_empty(&obj->oo_unlinked_linkage))) {
495 list_add(&obj->oo_unlinked_linkage, &oh->ot_unlinked_list);
502 /* Default to max data size covered by a level-1 indirect block */
503 static unsigned long osd_sync_destroy_max_size =
504 1UL << (DN_MAX_INDBLKSHIFT - SPA_BLKPTRSHIFT + SPA_MAXBLOCKSHIFT);
505 module_param(osd_sync_destroy_max_size, ulong, 0444);
506 MODULE_PARM_DESC(osd_sync_destroy_max_size, "Maximum object size to use synchronous destroy.");
509 osd_object_set_destroy_type(struct osd_object *obj)
512 * Lock-less OST_WRITE can race with OST_DESTROY, so set destroy type
513 * only once and use it consistently thereafter.
515 down_write(&obj->oo_guard);
516 if (obj->oo_destroy == OSD_DESTROY_NONE) {
517 if (obj->oo_attr.la_size <= osd_sync_destroy_max_size)
518 obj->oo_destroy = OSD_DESTROY_SYNC;
519 else /* Larger objects are destroyed asynchronously */
520 obj->oo_destroy = OSD_DESTROY_ASYNC;
522 up_write(&obj->oo_guard);
525 static int osd_declare_destroy(const struct lu_env *env, struct dt_object *dt,
528 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
529 struct osd_object *obj = osd_dt_obj(dt);
530 struct osd_device *osd = osd_obj2dev(obj);
531 struct osd_thandle *oh;
538 LASSERT(dt_object_exists(dt));
540 oh = container_of0(th, struct osd_thandle, ot_super);
541 LASSERT(oh->ot_tx != NULL);
543 /* declare that we'll remove object from fid-dnode mapping */
544 zapid = osd_get_name_n_idx(env, osd, fid, NULL, 0, &dn);
545 osd_tx_hold_zap(oh->ot_tx, zapid, dn, FALSE, NULL);
547 osd_declare_xattrs_destroy(env, obj, oh);
550 rc = osd_declare_quota(env, osd, obj->oo_attr.la_uid,
551 obj->oo_attr.la_gid, -1, oh, false, NULL, false);
555 /* data to be truncated */
556 rc = osd_declare_quota(env, osd, obj->oo_attr.la_uid,
557 obj->oo_attr.la_gid, 0, oh, true, NULL, false);
561 osd_object_set_destroy_type(obj);
562 if (obj->oo_destroy == OSD_DESTROY_SYNC)
563 dmu_tx_hold_free(oh->ot_tx, obj->oo_dn->dn_object,
566 osd_tx_hold_zap(oh->ot_tx, osd->od_unlinked->dn_object,
567 osd->od_unlinked, TRUE, NULL);
569 /* will help to find FID->ino when this object is being
570 * added to PENDING/ */
571 osd_idc_find_and_init(env, osd, obj);
576 static int osd_destroy(const struct lu_env *env, struct dt_object *dt,
579 struct osd_thread_info *info = osd_oti_get(env);
580 char *buf = info->oti_str;
581 struct osd_object *obj = osd_dt_obj(dt);
582 struct osd_device *osd = osd_obj2dev(obj);
583 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
584 struct osd_thandle *oh;
590 down_write(&obj->oo_guard);
592 if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
593 GOTO(out, rc = -ENOENT);
595 LASSERT(obj->oo_dn != NULL);
597 oh = container_of0(th, struct osd_thandle, ot_super);
599 LASSERT(oh->ot_tx != NULL);
601 /* remove obj ref from index dir (it depends) */
602 zapid = osd_get_name_n_idx(env, osd, fid, buf,
603 sizeof(info->oti_str), &zdn);
604 rc = osd_zap_remove(osd, zapid, zdn, buf, oh->ot_tx);
606 CERROR("%s: zap_remove(%s) failed: rc = %d\n",
607 osd->od_svname, buf, rc);
611 rc = osd_xattrs_destroy(env, obj, oh);
613 CERROR("%s: cannot destroy xattrs for %s: rc = %d\n",
614 osd->od_svname, buf, rc);
618 oid = obj->oo_dn->dn_object;
619 if (unlikely(obj->oo_destroy == OSD_DESTROY_NONE)) {
620 /* this may happen if the destroy wasn't declared
621 * e.g. when the object is created and then destroyed
622 * in the same transaction - we don't need additional
623 * space for destroy specifically */
624 LASSERT(obj->oo_attr.la_size <= osd_sync_destroy_max_size);
625 rc = -dmu_object_free(osd->od_os, oid, oh->ot_tx);
627 CERROR("%s: failed to free %s %llu: rc = %d\n",
628 osd->od_svname, buf, oid, rc);
629 } else if (obj->oo_destroy == OSD_DESTROY_SYNC) {
630 rc = -dmu_object_free(osd->od_os, oid, oh->ot_tx);
632 CERROR("%s: failed to free %s %llu: rc = %d\n",
633 osd->od_svname, buf, oid, rc);
634 } else { /* asynchronous destroy */
635 char *key = info->oti_key;
637 rc = osd_object_unlinked_add(obj, oh);
641 snprintf(key, sizeof(info->oti_key), "%llx", oid);
642 rc = osd_zap_add(osd, osd->od_unlinked->dn_object,
643 osd->od_unlinked, key, 8, 1, &oid, oh->ot_tx);
645 CERROR("%s: zap_add_int() failed %s %llu: rc = %d\n",
646 osd->od_svname, buf, oid, rc);
650 /* not needed in the cache anymore */
651 set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
653 obj->oo_destroyed = 1;
654 up_write(&obj->oo_guard);
658 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
660 struct osd_object *obj = osd_obj(l);
661 const struct lu_fid *fid = lu_object_fid(l);
664 if (likely(!fid_is_acct(fid))) {
665 osd_object_sa_fini(obj);
666 if (obj->oo_sa_xattr) {
667 nvlist_free(obj->oo_sa_xattr);
668 obj->oo_sa_xattr = NULL;
670 osd_dnode_rele(obj->oo_dn);
671 list_del(&obj->oo_sa_linkage);
678 * Concurrency: ->loo_object_release() is called under site spin-lock.
680 static void osd_object_release(const struct lu_env *env,
686 * Concurrency: shouldn't matter.
688 static int osd_object_print(const struct lu_env *env, void *cookie,
689 lu_printer_t p, const struct lu_object *l)
691 struct osd_object *o = osd_obj(l);
693 return (*p)(env, cookie, LUSTRE_OSD_ZFS_NAME"-object@%p", o);
696 static void osd_read_lock(const struct lu_env *env, struct dt_object *dt,
699 struct osd_object *obj = osd_dt_obj(dt);
701 LASSERT(osd_invariant(obj));
703 down_read_nested(&obj->oo_sem, role);
706 static void osd_write_lock(const struct lu_env *env, struct dt_object *dt,
709 struct osd_object *obj = osd_dt_obj(dt);
711 LASSERT(osd_invariant(obj));
713 down_write_nested(&obj->oo_sem, role);
716 static void osd_read_unlock(const struct lu_env *env, struct dt_object *dt)
718 struct osd_object *obj = osd_dt_obj(dt);
720 LASSERT(osd_invariant(obj));
721 up_read(&obj->oo_sem);
724 static void osd_write_unlock(const struct lu_env *env, struct dt_object *dt)
726 struct osd_object *obj = osd_dt_obj(dt);
728 LASSERT(osd_invariant(obj));
729 up_write(&obj->oo_sem);
732 static int osd_write_locked(const struct lu_env *env, struct dt_object *dt)
734 struct osd_object *obj = osd_dt_obj(dt);
737 LASSERT(osd_invariant(obj));
739 if (down_write_trylock(&obj->oo_sem)) {
741 up_write(&obj->oo_sem);
746 static int osd_attr_get(const struct lu_env *env,
747 struct dt_object *dt,
748 struct lu_attr *attr)
750 struct osd_object *obj = osd_dt_obj(dt);
755 down_read(&obj->oo_guard);
757 if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
758 GOTO(out, rc = -ENOENT);
760 if (unlikely(fid_is_acct(lu_object_fid(&dt->do_lu))))
763 LASSERT(osd_invariant(obj));
766 read_lock(&obj->oo_attr_lock);
767 *attr = obj->oo_attr;
768 if (obj->oo_lma_flags & LUSTRE_ORPHAN_FL)
769 attr->la_flags |= LUSTRE_ORPHAN_FL;
770 read_unlock(&obj->oo_attr_lock);
772 /* with ZFS_DEBUG zrl_add_debug() called by DB_DNODE_ENTER()
773 * from within sa_object_size() can block on a mutex, so
774 * we can't call sa_object_size() holding rwlock */
775 sa_object_size(obj->oo_sa_hdl, &blksize, &blocks);
776 /* we do not control size of indices, so always calculate
777 * it from number of blocks reported by DMU */
778 if (S_ISDIR(attr->la_mode))
779 attr->la_size = 512 * blocks;
780 /* Block size may be not set; suggest maximal I/O transfers. */
782 blksize = osd_spa_maxblocksize(
783 dmu_objset_spa(osd_obj2dev(obj)->od_os));
785 attr->la_blksize = blksize;
786 attr->la_blocks = blocks;
787 attr->la_valid |= LA_BLOCKS | LA_BLKSIZE;
790 up_read(&obj->oo_guard);
794 /* Simple wrapper on top of qsd API which implement quota transfer for osd
795 * setattr needs. As a reminder, only the root user can change ownership of
796 * a file, that's why EDQUOT & EINPROGRESS errors are discarded */
797 static inline int qsd_transfer(const struct lu_env *env,
798 struct qsd_instance *qsd,
799 struct lquota_trans *trans, int qtype,
800 __u64 orig_id, __u64 new_id, __u64 bspace,
801 struct lquota_id_info *qi)
805 if (unlikely(qsd == NULL))
808 LASSERT(qtype >= 0 && qtype < LL_MAXQUOTAS);
809 qi->lqi_type = qtype;
811 /* inode accounting */
812 qi->lqi_is_blk = false;
814 /* one more inode for the new owner ... */
815 qi->lqi_id.qid_uid = new_id;
817 rc = qsd_op_begin(env, qsd, trans, qi, NULL);
818 if (rc == -EDQUOT || rc == -EINPROGRESS)
823 /* and one less inode for the current id */
824 qi->lqi_id.qid_uid = orig_id;;
826 /* can't get EDQUOT when reducing usage */
827 rc = qsd_op_begin(env, qsd, trans, qi, NULL);
828 if (rc == -EINPROGRESS)
833 /* block accounting */
834 qi->lqi_is_blk = true;
836 /* more blocks for the new owner ... */
837 qi->lqi_id.qid_uid = new_id;
838 qi->lqi_space = bspace;
839 rc = qsd_op_begin(env, qsd, trans, qi, NULL);
840 if (rc == -EDQUOT || rc == -EINPROGRESS)
845 /* and finally less blocks for the current owner */
846 qi->lqi_id.qid_uid = orig_id;
847 qi->lqi_space = -bspace;
848 rc = qsd_op_begin(env, qsd, trans, qi, NULL);
849 /* can't get EDQUOT when reducing usage */
850 if (rc == -EINPROGRESS)
855 static int osd_declare_attr_set(const struct lu_env *env,
856 struct dt_object *dt,
857 const struct lu_attr *attr,
858 struct thandle *handle)
860 struct osd_thread_info *info = osd_oti_get(env);
861 struct osd_object *obj = osd_dt_obj(dt);
862 struct osd_device *osd = osd_obj2dev(obj);
864 struct osd_thandle *oh;
872 LASSERT(handle != NULL);
873 LASSERT(osd_invariant(obj));
875 oh = container_of0(handle, struct osd_thandle, ot_super);
877 down_read(&obj->oo_guard);
878 if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
881 LASSERT(obj->oo_sa_hdl != NULL);
882 LASSERT(oh->ot_tx != NULL);
883 /* regular attributes are part of the bonus buffer */
884 /* let's check whether this object is already part of
887 for (txh = list_head(&oh->ot_tx->tx_holds); txh;
888 txh = list_next(&oh->ot_tx->tx_holds, txh)) {
889 if (txh->txh_dnode == NULL)
891 if (txh->txh_dnode->dn_object != obj->oo_dn->dn_object)
893 /* this object is part of the transaction already
894 * we don't need to declare bonus again */
899 dmu_tx_hold_bonus(oh->ot_tx, obj->oo_dn->dn_object);
900 if (oh->ot_tx->tx_err != 0)
901 GOTO(out, rc = -oh->ot_tx->tx_err);
903 if (attr && attr->la_valid & LA_FLAGS) {
904 /* LMA is usually a part of bonus, no need to declare
908 if (attr && (attr->la_valid & (LA_UID | LA_GID))) {
909 sa_object_size(obj->oo_sa_hdl, &blksize, &bspace);
910 bspace = toqb(bspace * blksize);
913 if (attr && attr->la_valid & LA_UID) {
914 /* quota enforcement for user */
915 if (attr->la_uid != obj->oo_attr.la_uid) {
916 rc = qsd_transfer(env, osd->od_quota_slave,
917 &oh->ot_quota_trans, USRQUOTA,
918 obj->oo_attr.la_uid, attr->la_uid,
919 bspace, &info->oti_qi);
924 if (attr && attr->la_valid & LA_GID) {
925 /* quota enforcement for group */
926 if (attr->la_gid != obj->oo_attr.la_gid) {
927 rc = qsd_transfer(env, osd->od_quota_slave,
928 &oh->ot_quota_trans, GRPQUOTA,
929 obj->oo_attr.la_gid, attr->la_gid,
930 bspace, &info->oti_qi);
937 up_read(&obj->oo_guard);
942 * Set the attributes of an object
944 * The transaction passed to this routine must have
945 * dmu_tx_hold_bonus(tx, oid) called and then assigned
946 * to a transaction group.
948 static int osd_attr_set(const struct lu_env *env, struct dt_object *dt,
949 const struct lu_attr *la, struct thandle *handle)
951 struct osd_thread_info *info = osd_oti_get(env);
952 sa_bulk_attr_t *bulk = osd_oti_get(env)->oti_attr_bulk;
953 struct osd_object *obj = osd_dt_obj(dt);
954 struct osd_device *osd = osd_obj2dev(obj);
955 struct osd_thandle *oh;
956 struct osa_attr *osa = &info->oti_osa;
957 __u64 valid = la->la_valid;
963 down_read(&obj->oo_guard);
964 if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
965 GOTO(out, rc = -ENOENT);
967 LASSERT(handle != NULL);
968 LASSERT(osd_invariant(obj));
969 LASSERT(obj->oo_sa_hdl);
971 oh = container_of0(handle, struct osd_thandle, ot_super);
972 /* Assert that the transaction has been assigned to a
973 transaction group. */
974 LASSERT(oh->ot_tx->tx_txg != 0);
976 /* Only allow set size for regular file */
977 if (!S_ISREG(dt->do_lu.lo_header->loh_attr))
978 valid &= ~(LA_SIZE | LA_BLOCKS);
980 if (valid & LA_CTIME && la->la_ctime == obj->oo_attr.la_ctime)
983 if (valid & LA_MTIME && la->la_mtime == obj->oo_attr.la_mtime)
986 if (valid & LA_ATIME && la->la_atime == obj->oo_attr.la_atime)
992 if (valid & LA_FLAGS) {
993 struct lustre_mdt_attrs *lma;
996 if (la->la_flags & LUSTRE_LMA_FL_MASKS) {
997 CLASSERT(sizeof(info->oti_buf) >= sizeof(*lma));
998 lma = (struct lustre_mdt_attrs *)&info->oti_buf;
1000 buf.lb_len = sizeof(info->oti_buf);
1001 rc = osd_xattr_get(env, &obj->oo_dt, &buf,
1005 le32_to_cpu(lma->lma_incompat);
1006 lma->lma_incompat |=
1007 lustre_to_lma_flags(la->la_flags);
1009 cpu_to_le32(lma->lma_incompat);
1011 buf.lb_len = sizeof(*lma);
1012 rc = osd_xattr_set_internal(env, obj, &buf,
1018 CWARN("%s: failed to set LMA flags: rc = %d\n",
1019 osd->od_svname, rc);
1025 write_lock(&obj->oo_attr_lock);
1027 if (valid & LA_ATIME) {
1028 osa->atime[0] = obj->oo_attr.la_atime = la->la_atime;
1029 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(osd), NULL,
1032 if (valid & LA_MTIME) {
1033 osa->mtime[0] = obj->oo_attr.la_mtime = la->la_mtime;
1034 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(osd), NULL,
1037 if (valid & LA_CTIME) {
1038 osa->ctime[0] = obj->oo_attr.la_ctime = la->la_ctime;
1039 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(osd), NULL,
1042 if (valid & LA_MODE) {
1043 /* mode is stored along with type, so read it first */
1044 obj->oo_attr.la_mode = (obj->oo_attr.la_mode & S_IFMT) |
1045 (la->la_mode & ~S_IFMT);
1046 osa->mode = obj->oo_attr.la_mode;
1047 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(osd), NULL,
1050 if (valid & LA_SIZE) {
1051 osa->size = obj->oo_attr.la_size = la->la_size;
1052 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(osd), NULL,
1055 if (valid & LA_NLINK) {
1056 osa->nlink = obj->oo_attr.la_nlink = la->la_nlink;
1057 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(osd), NULL,
1060 if (valid & LA_RDEV) {
1061 osa->rdev = obj->oo_attr.la_rdev = la->la_rdev;
1062 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_RDEV(osd), NULL,
1065 if (valid & LA_FLAGS) {
1066 osa->flags = attrs_fs2zfs(la->la_flags);
1067 /* many flags are not supported by zfs, so ensure a good cached
1069 obj->oo_attr.la_flags = attrs_zfs2fs(osa->flags);
1070 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(osd), NULL,
1073 if (valid & LA_UID) {
1074 osa->uid = obj->oo_attr.la_uid = la->la_uid;
1075 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(osd), NULL,
1078 if (valid & LA_GID) {
1079 osa->gid = obj->oo_attr.la_gid = la->la_gid;
1080 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(osd), NULL,
1083 obj->oo_attr.la_valid |= valid;
1084 write_unlock(&obj->oo_attr_lock);
1086 LASSERT(cnt <= ARRAY_SIZE(osd_oti_get(env)->oti_attr_bulk));
1087 rc = osd_object_sa_bulk_update(obj, bulk, cnt, oh);
1090 up_read(&obj->oo_guard);
1097 * XXX temporary solution.
1100 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
1101 struct dt_object *parent, struct dt_object *child,
1106 ah->dah_parent = parent;
1107 ah->dah_mode = child_mode;
1109 if (parent != NULL && !dt_object_remote(parent)) {
1110 /* will help to find FID->ino at dt_insert("..") */
1111 struct osd_object *pobj = osd_dt_obj(parent);
1113 osd_idc_find_and_init(env, osd_obj2dev(pobj), pobj);
1117 static int osd_declare_create(const struct lu_env *env, struct dt_object *dt,
1118 struct lu_attr *attr,
1119 struct dt_allocation_hint *hint,
1120 struct dt_object_format *dof,
1121 struct thandle *handle)
1123 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1124 struct osd_object *obj = osd_dt_obj(dt);
1125 struct osd_device *osd = osd_obj2dev(obj);
1126 struct osd_thandle *oh;
1134 switch (dof->dof_type) {
1138 if (obj->oo_dt.do_body_ops == NULL)
1139 obj->oo_dt.do_body_ops = &osd_body_ops;
1145 LASSERT(handle != NULL);
1146 oh = container_of0(handle, struct osd_thandle, ot_super);
1147 LASSERT(oh->ot_tx != NULL);
1149 /* this is the minimum set of EAs on every Lustre object */
1150 obj->oo_ea_in_bonus = ZFS_SA_BASE_ATTR_SIZE +
1151 sizeof(__u64) + /* VBR VERSION */
1152 sizeof(struct lustre_mdt_attrs); /* LMA */
1153 /* reserve 32 bytes for extra stuff like ACLs */
1154 dnode_size = size_roundup_power2(obj->oo_ea_in_bonus + 32);
1156 switch (dof->dof_type) {
1158 dt->do_index_ops = &osd_dir_ops;
1160 /* for zap create */
1161 dmu_tx_hold_zap(oh->ot_tx, DMU_NEW_OBJECT, FALSE, NULL);
1162 dmu_tx_hold_sa_create(oh->ot_tx, dnode_size);
1167 /* first, we'll create new object */
1168 dmu_tx_hold_sa_create(oh->ot_tx, dnode_size);
1176 /* and we'll add it to some mapping */
1177 zapid = osd_get_name_n_idx(env, osd, fid, NULL, 0, &dn);
1178 osd_tx_hold_zap(oh->ot_tx, zapid, dn, TRUE, NULL);
1180 /* will help to find FID->ino mapping at dt_insert() */
1181 osd_idc_find_and_init(env, osd, obj);
1183 rc = osd_declare_quota(env, osd, attr->la_uid, attr->la_gid, 1, oh,
1184 false, NULL, false);
1189 int __osd_attr_init(const struct lu_env *env, struct osd_device *osd,
1190 sa_handle_t *sa_hdl, dmu_tx_t *tx,
1191 struct lu_attr *la, uint64_t parent,
1194 sa_bulk_attr_t *bulk = osd_oti_get(env)->oti_attr_bulk;
1195 struct osa_attr *osa = &osd_oti_get(env)->oti_osa;
1201 char *dxattr = NULL;
1207 gen = dmu_tx_get_txg(tx);
1209 ZFS_TIME_ENCODE(&now, crtime);
1211 osa->atime[0] = la->la_atime;
1212 osa->ctime[0] = la->la_ctime;
1213 osa->mtime[0] = la->la_mtime;
1214 osa->mode = la->la_mode;
1215 osa->uid = la->la_uid;
1216 osa->gid = la->la_gid;
1217 osa->rdev = la->la_rdev;
1218 osa->nlink = la->la_nlink;
1219 osa->flags = attrs_fs2zfs(la->la_flags);
1220 osa->size = la->la_size;
1223 * we need to create all SA below upon object create.
1225 * XXX The attribute order matters since the accounting callback relies
1226 * on static offsets (i.e. SA_*_OFFSET, see zfs_space_delta_cb()) to
1227 * look up the UID/GID attributes. Moreover, the callback does not seem
1228 * to support the spill block.
1229 * We define attributes in the same order as SA_*_OFFSET in order to
1230 * work around the problem. See ORI-610.
1233 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(osd), NULL, &osa->mode, 8);
1234 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(osd), NULL, &osa->size, 8);
1235 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GEN(osd), NULL, &gen, 8);
1236 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(osd), NULL, &osa->uid, 8);
1237 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(osd), NULL, &osa->gid, 8);
1238 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_PARENT(osd), NULL, &parent, 8);
1239 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(osd), NULL, &osa->flags, 8);
1240 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(osd), NULL, osa->atime, 16);
1241 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(osd), NULL, osa->mtime, 16);
1242 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(osd), NULL, osa->ctime, 16);
1243 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CRTIME(osd), NULL, crtime, 16);
1244 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(osd), NULL, &osa->nlink, 8);
1245 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_RDEV(osd), NULL, &osa->rdev, 8);
1246 LASSERT(cnt <= ARRAY_SIZE(osd_oti_get(env)->oti_attr_bulk));
1249 rc = -nvlist_size(xattr, &sa_size, NV_ENCODE_XDR);
1252 dxattr = osd_zio_buf_alloc(sa_size);
1255 rc = -nvlist_pack(xattr, &dxattr, &sa_size,
1256 NV_ENCODE_XDR, KM_SLEEP);
1259 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_DXATTR(osd),
1260 NULL, dxattr, sa_size);
1263 rc = -sa_replace_all_by_template(sa_hdl, bulk, cnt, tx);
1265 osd_zio_buf_free(dxattr, sa_size);
1270 static int osd_find_new_dnode(const struct lu_env *env, dmu_tx_t *tx,
1271 uint64_t oid, dnode_t **dnp)
1276 /* take dnode_t from tx to save on dnode#->dnode_t lookup */
1277 for (txh = list_tail(&tx->tx_holds); txh;
1278 txh = list_prev(&tx->tx_holds, txh)) {
1279 dnode_t *dn = txh->txh_dnode;
1284 if (dn->dn_object != oid)
1288 rw_enter(&dn->dn_struct_rwlock, RW_WRITER);
1289 if (dn->dn_bonus == NULL)
1290 dbuf_create_bonus(dn);
1291 rw_exit(&dn->dn_struct_rwlock);
1295 LASSERT(dn->dn_handle);
1297 if (refcount_add(&db->db_holds, osd_obj_tag) == 1) {
1298 refcount_add(&dn->dn_holds, tag);
1299 atomic_inc_32(&dn->dn_dbufs_count);
1302 dbuf_read(db, NULL, DB_RF_MUST_SUCCEED | DB_RF_NOPREFETCH);
1306 if (unlikely(*dnp == NULL))
1307 rc = __osd_obj2dnode(tx->tx_objset, oid, dnp);
1313 * The transaction passed to this routine must have
1314 * dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT) called and then assigned
1315 * to a transaction group.
1317 int __osd_object_create(const struct lu_env *env, struct osd_object *obj,
1318 dnode_t **dnp, dmu_tx_t *tx, struct lu_attr *la)
1320 struct osd_device *osd = osd_obj2dev(obj);
1321 const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
1322 dmu_object_type_t type = DMU_OT_PLAIN_FILE_CONTENTS;
1325 /* Use DMU_OTN_UINT8_METADATA for local objects so their data blocks
1326 * would get an additional ditto copy */
1327 if (unlikely(S_ISREG(la->la_mode) &&
1328 fid_seq_is_local_file(fid_seq(fid))))
1329 type = DMU_OTN_UINT8_METADATA;
1331 /* Create a new DMU object using the default dnode size. */
1332 oid = osd_dmu_object_alloc(osd->od_os, type, 0, 0, tx);
1334 LASSERT(la->la_valid & LA_MODE);
1338 return osd_find_new_dnode(env, tx, oid, dnp);
1342 * The transaction passed to this routine must have
1343 * dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, ...) called and then assigned
1344 * to a transaction group.
1346 * Using ZAP_FLAG_HASH64 will force the ZAP to always be a FAT ZAP.
1347 * This is fine for directories today, because storing the FID in the dirent
1348 * will also require a FAT ZAP. If there is a new type of micro ZAP created
1349 * then we might need to re-evaluate the use of this flag and instead do
1350 * a conversion from the different internal ZAP hash formats being used. */
1351 int __osd_zap_create(const struct lu_env *env, struct osd_device *osd,
1352 dnode_t **dnp, dmu_tx_t *tx, struct lu_attr *la,
1357 /* Assert that the transaction has been assigned to a
1358 transaction group. */
1359 LASSERT(tx->tx_txg != 0);
1362 oid = osd_zap_create_flags(osd->od_os, 0, flags | ZAP_FLAG_HASH64,
1363 DMU_OT_DIRECTORY_CONTENTS,
1364 14, /* == ZFS fzap_default_blockshift */
1365 DN_MAX_INDBLKSHIFT, /* indirect blockshift */
1371 return osd_find_new_dnode(env, tx, oid, dnp);
1374 static dnode_t *osd_mkidx(const struct lu_env *env, struct osd_object *obj,
1375 struct lu_attr *la, struct osd_thandle *oh)
1380 /* Index file should be created as regular file in order not to confuse
1381 * ZPL which could interpret them as directory.
1382 * We set ZAP_FLAG_UINT64_KEY to let ZFS know than we are going to use
1384 LASSERT(S_ISREG(la->la_mode));
1385 rc = __osd_zap_create(env, osd_obj2dev(obj), &dn, oh->ot_tx, la,
1386 ZAP_FLAG_UINT64_KEY);
1392 static dnode_t *osd_mkdir(const struct lu_env *env, struct osd_object *obj,
1393 struct lu_attr *la, struct osd_thandle *oh)
1398 LASSERT(S_ISDIR(la->la_mode));
1399 rc = __osd_zap_create(env, osd_obj2dev(obj), &dn, oh->ot_tx, la, 0);
1405 static dnode_t *osd_mkreg(const struct lu_env *env, struct osd_object *obj,
1406 struct lu_attr *la, struct osd_thandle *oh)
1408 const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
1409 struct osd_device *osd = osd_obj2dev(obj);
1413 LASSERT(S_ISREG(la->la_mode));
1414 rc = __osd_object_create(env, obj, &dn, oh->ot_tx, la);
1418 if ((fid_is_idif(fid) || fid_is_norm(fid) || fid_is_echo(fid)) &&
1420 /* The minimum block size must be at least page size otherwise
1421 * it will break the assumption in tgt_thread_big_cache where
1422 * the array size is PTLRPC_MAX_BRW_PAGES. It will also affect
1423 * RDMA due to subpage transfer size */
1424 rc = -dmu_object_set_blocksize(osd->od_os, dn->dn_object,
1425 PAGE_SIZE, 0, oh->ot_tx);
1427 CERROR("%s: can't change blocksize: %d\n",
1428 osd->od_svname, rc);
1436 static dnode_t *osd_mksym(const struct lu_env *env, struct osd_object *obj,
1437 struct lu_attr *la, struct osd_thandle *oh)
1442 LASSERT(S_ISLNK(la->la_mode));
1443 rc = __osd_object_create(env, obj, &dn, oh->ot_tx, la);
1449 static dnode_t *osd_mknod(const struct lu_env *env, struct osd_object *obj,
1450 struct lu_attr *la, struct osd_thandle *oh)
1455 if (S_ISCHR(la->la_mode) || S_ISBLK(la->la_mode))
1456 la->la_valid |= LA_RDEV;
1458 rc = __osd_object_create(env, obj, &dn, oh->ot_tx, la);
1464 typedef dnode_t *(*osd_obj_type_f)(const struct lu_env *env,
1465 struct osd_object *obj,
1467 struct osd_thandle *oh);
1469 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
1471 osd_obj_type_f result;
1497 * Concurrency: @dt is write locked.
1499 static int osd_create(const struct lu_env *env, struct dt_object *dt,
1500 struct lu_attr *attr, struct dt_allocation_hint *hint,
1501 struct dt_object_format *dof, struct thandle *th)
1503 struct osd_thread_info *info = osd_oti_get(env);
1504 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
1505 struct zpl_direntry *zde = &info->oti_zde.lzd_reg;
1506 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1507 struct osd_object *obj = osd_dt_obj(dt);
1508 struct osd_device *osd = osd_obj2dev(obj);
1509 char *buf = info->oti_str;
1510 struct osd_thandle *oh;
1511 dnode_t *dn = NULL, *zdn = NULL;
1512 uint64_t zapid, parent = 0;
1517 LASSERT(!fid_is_acct(fid));
1519 /* concurrent create declarations should not see
1520 * the object inconsistent (db, attr, etc).
1521 * in regular cases acquisition should be cheap */
1522 down_write(&obj->oo_guard);
1524 if (unlikely(dt_object_exists(dt)))
1525 GOTO(out, rc = -EEXIST);
1527 LASSERT(osd_invariant(obj));
1528 LASSERT(dof != NULL);
1530 LASSERT(th != NULL);
1531 oh = container_of0(th, struct osd_thandle, ot_super);
1533 LASSERT(obj->oo_dn == NULL);
1535 /* to follow ZFS on-disk format we need
1536 * to initialize parent dnode properly */
1537 if (hint != NULL && hint->dah_parent != NULL &&
1538 !dt_object_remote(hint->dah_parent))
1539 parent = osd_dt_obj(hint->dah_parent)->oo_dn->dn_object;
1541 /* we may fix some attributes, better do not change the source */
1542 obj->oo_attr = *attr;
1543 obj->oo_attr.la_valid |= LA_SIZE | LA_NLINK | LA_TYPE;
1545 dn = osd_create_type_f(dof->dof_type)(env, obj, &obj->oo_attr, oh);
1553 zde->zde_dnode = dn->dn_object;
1554 zde->zde_type = IFTODT(attr->la_mode & S_IFMT);
1556 zapid = osd_get_name_n_idx(env, osd, fid, buf,
1557 sizeof(info->oti_str), &zdn);
1558 rc = osd_zap_add(osd, zapid, zdn, buf, 8, 1, zde, oh->ot_tx);
1562 /* Now add in all of the "SA" attributes */
1563 rc = osd_sa_handle_get(obj);
1567 rc = -nvlist_alloc(&obj->oo_sa_xattr, NV_UNIQUE_NAME, KM_SLEEP);
1571 /* initialize LMA */
1572 lustre_lma_init(lma, fid, 0, 0);
1573 lustre_lma_swab(lma);
1574 rc = -nvlist_add_byte_array(obj->oo_sa_xattr, XATTR_NAME_LMA,
1575 (uchar_t *)lma, sizeof(*lma));
1579 /* configure new osd object */
1580 obj->oo_parent = parent != 0 ? parent : zapid;
1581 obj->oo_late_attr_set = 1;
1582 rc = __osd_sa_xattr_schedule_update(env, obj, oh);
1586 /* XXX: oo_lma_flags */
1587 obj->oo_dt.do_lu.lo_header->loh_attr |= obj->oo_attr.la_mode & S_IFMT;
1588 if (likely(!fid_is_acct(lu_object_fid(&obj->oo_dt.do_lu))))
1589 /* no body operations for accounting objects */
1590 obj->oo_dt.do_body_ops = &osd_body_ops;
1592 osd_idc_find_and_init(env, osd, obj);
1595 if (unlikely(rc && dn)) {
1596 dmu_object_free(osd->od_os, dn->dn_object, oh->ot_tx);
1600 obj->oo_dt.do_lu.lo_header->loh_attr |= LOHA_EXISTS;
1602 up_write(&obj->oo_guard);
1606 static int osd_declare_ref_add(const struct lu_env *env, struct dt_object *dt,
1609 return osd_declare_attr_set(env, dt, NULL, th);
1613 * Concurrency: @dt is write locked.
1615 static int osd_ref_add(const struct lu_env *env, struct dt_object *dt,
1616 struct thandle *handle)
1618 struct osd_object *obj = osd_dt_obj(dt);
1619 struct osd_thandle *oh;
1620 struct osd_device *osd = osd_obj2dev(obj);
1626 down_read(&obj->oo_guard);
1627 if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
1628 GOTO(out, rc = -ENOENT);
1630 LASSERT(osd_invariant(obj));
1631 LASSERT(obj->oo_sa_hdl != NULL);
1633 oh = container_of0(handle, struct osd_thandle, ot_super);
1635 write_lock(&obj->oo_attr_lock);
1636 nlink = ++obj->oo_attr.la_nlink;
1637 write_unlock(&obj->oo_attr_lock);
1639 rc = osd_object_sa_update(obj, SA_ZPL_LINKS(osd), &nlink, 8, oh);
1642 up_read(&obj->oo_guard);
1646 static int osd_declare_ref_del(const struct lu_env *env, struct dt_object *dt,
1647 struct thandle *handle)
1649 return osd_declare_attr_set(env, dt, NULL, handle);
1653 * Concurrency: @dt is write locked.
1655 static int osd_ref_del(const struct lu_env *env, struct dt_object *dt,
1656 struct thandle *handle)
1658 struct osd_object *obj = osd_dt_obj(dt);
1659 struct osd_thandle *oh;
1660 struct osd_device *osd = osd_obj2dev(obj);
1666 down_read(&obj->oo_guard);
1668 if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
1669 GOTO(out, rc = -ENOENT);
1671 LASSERT(osd_invariant(obj));
1672 LASSERT(obj->oo_sa_hdl != NULL);
1674 oh = container_of0(handle, struct osd_thandle, ot_super);
1675 LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
1677 write_lock(&obj->oo_attr_lock);
1678 nlink = --obj->oo_attr.la_nlink;
1679 write_unlock(&obj->oo_attr_lock);
1681 rc = osd_object_sa_update(obj, SA_ZPL_LINKS(osd), &nlink, 8, oh);
1684 up_read(&obj->oo_guard);
1688 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
1689 __u64 start, __u64 end)
1691 struct osd_device *osd = osd_obj2dev(osd_dt_obj(dt));
1694 /* XXX: no other option than syncing the whole filesystem until we
1695 * support ZIL. If the object tracked the txg that it was last
1696 * modified in, it could pass that txg here instead of "0". Maybe
1697 * the changes are already committed, so no wait is needed at all? */
1698 if (!osd->od_dt_dev.dd_rdonly)
1699 txg_wait_synced(dmu_objset_pool(osd->od_os), 0ULL);
1704 static int osd_invalidate(const struct lu_env *env, struct dt_object *dt)
1709 static struct dt_object_operations osd_obj_ops = {
1710 .do_read_lock = osd_read_lock,
1711 .do_write_lock = osd_write_lock,
1712 .do_read_unlock = osd_read_unlock,
1713 .do_write_unlock = osd_write_unlock,
1714 .do_write_locked = osd_write_locked,
1715 .do_attr_get = osd_attr_get,
1716 .do_declare_attr_set = osd_declare_attr_set,
1717 .do_attr_set = osd_attr_set,
1718 .do_ah_init = osd_ah_init,
1719 .do_declare_create = osd_declare_create,
1720 .do_create = osd_create,
1721 .do_declare_destroy = osd_declare_destroy,
1722 .do_destroy = osd_destroy,
1723 .do_index_try = osd_index_try,
1724 .do_declare_ref_add = osd_declare_ref_add,
1725 .do_ref_add = osd_ref_add,
1726 .do_declare_ref_del = osd_declare_ref_del,
1727 .do_ref_del = osd_ref_del,
1728 .do_xattr_get = osd_xattr_get,
1729 .do_declare_xattr_set = osd_declare_xattr_set,
1730 .do_xattr_set = osd_xattr_set,
1731 .do_declare_xattr_del = osd_declare_xattr_del,
1732 .do_xattr_del = osd_xattr_del,
1733 .do_xattr_list = osd_xattr_list,
1734 .do_object_sync = osd_object_sync,
1735 .do_invalidate = osd_invalidate,
1738 static struct lu_object_operations osd_lu_obj_ops = {
1739 .loo_object_init = osd_object_init,
1740 .loo_object_delete = osd_object_delete,
1741 .loo_object_release = osd_object_release,
1742 .loo_object_free = osd_object_free,
1743 .loo_object_print = osd_object_print,
1744 .loo_object_invariant = osd_object_invariant,
1747 static int osd_otable_it_attr_get(const struct lu_env *env,
1748 struct dt_object *dt,
1749 struct lu_attr *attr)
1755 static struct dt_object_operations osd_obj_otable_it_ops = {
1756 .do_attr_get = osd_otable_it_attr_get,
1757 .do_index_try = osd_index_try,