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 #ifdef ZFS_PROJINHERIT
218 if (o->od_projectused_dn && osa->flags & ZFS_PROJID) {
219 rc = -sa_lookup(obj->oo_sa_hdl, SA_ZPL_PROJID(o),
224 la->la_projid = osa->projid;
225 la->la_valid |= LA_PROJID;
226 obj->oo_with_projid = 1;
228 la->la_projid = ZFS_DEFAULT_PROJID;
229 la->la_valid &= ~LA_PROJID;
233 la->la_valid &= ~LA_PROJID;
236 la->la_atime = osa->atime[0];
237 la->la_mtime = osa->mtime[0];
238 la->la_ctime = osa->ctime[0];
239 la->la_mode = osa->mode;
240 la->la_uid = osa->uid;
241 la->la_gid = osa->gid;
242 la->la_nlink = osa->nlink;
243 la->la_flags = attrs_zfs2fs(osa->flags);
244 la->la_size = osa->size;
246 /* Try to get extra flag from LMA. Right now, only LMAI_ORPHAN
247 * flags is stored in LMA, and it is only for orphan directory */
248 if (S_ISDIR(la->la_mode) && dt_object_exists(&obj->oo_dt)) {
249 struct osd_thread_info *info = osd_oti_get(env);
250 struct lustre_mdt_attrs *lma;
253 lma = (struct lustre_mdt_attrs *)info->oti_buf;
255 buf.lb_len = sizeof(info->oti_buf);
256 rc = osd_xattr_get(env, &obj->oo_dt, &buf, XATTR_NAME_LMA);
259 lma->lma_incompat = le32_to_cpu(lma->lma_incompat);
261 lma_to_lustre_flags(lma->lma_incompat);
263 } else if (rc == -ENODATA) {
268 if (S_ISCHR(la->la_mode) || S_ISBLK(la->la_mode)) {
269 rc = -sa_lookup(obj->oo_sa_hdl, SA_ZPL_RDEV(o), &osa->rdev, 8);
272 la->la_rdev = osa->rdev;
273 la->la_valid |= LA_RDEV;
280 int __osd_obj2dnode(objset_t *os, uint64_t oid, dnode_t **dnp)
286 rc = -dmu_bonus_hold(os, oid, osd_obj_tag, &db);
290 dbi = (dmu_buf_impl_t *)db;
292 *dnp = DB_DNODE(dbi);
293 LASSERT(*dnp != NULL);
299 * Concurrency: no concurrent access is possible that early in object
302 struct lu_object *osd_object_alloc(const struct lu_env *env,
303 const struct lu_object_header *hdr,
306 struct osd_object *mo;
308 OBD_SLAB_ALLOC_PTR_GFP(mo, osd_object_kmem, GFP_NOFS);
312 l = &mo->oo_dt.do_lu;
313 dt_object_init(&mo->oo_dt, NULL, d);
314 mo->oo_dt.do_ops = &osd_obj_ops;
315 l->lo_ops = &osd_lu_obj_ops;
316 INIT_LIST_HEAD(&mo->oo_sa_linkage);
317 INIT_LIST_HEAD(&mo->oo_unlinked_linkage);
318 init_rwsem(&mo->oo_sem);
319 init_rwsem(&mo->oo_guard);
320 rwlock_init(&mo->oo_attr_lock);
321 mo->oo_destroy = OSD_DESTROY_NONE;
328 static void osd_obj_set_blksize(const struct lu_env *env,
329 struct osd_device *osd, struct osd_object *obj)
331 const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
333 dnode_t *dn = obj->oo_dn;
338 LASSERT(!osd_oti_get(env)->oti_in_trans);
340 tx = dmu_tx_create(osd->od_os);
342 CERROR("%s: fail to create tx to set blksize for "DFID"\n",
343 osd->od_svname, PFID(fid));
347 dmu_tx_hold_bonus(tx, dn->dn_object);
348 rc = -dmu_tx_assign(tx, TXG_WAIT);
351 CERROR("%s: fail to assign tx to set blksize for "DFID
352 ": rc = %d\n", osd->od_svname, PFID(fid), rc);
356 down_write(&obj->oo_guard);
357 if (unlikely((1 << dn->dn_datablkshift) >= PAGE_SIZE))
360 blksz = dn->dn_datablksz;
361 if (!is_power_of_2(blksz))
362 blksz = size_roundup_power2(blksz);
364 if (blksz > osd->od_max_blksz)
365 blksz = osd->od_max_blksz;
366 else if (blksz < PAGE_SIZE)
368 rc = -dmu_object_set_blocksize(osd->od_os, dn->dn_object, blksz, 0, tx);
373 up_write(&obj->oo_guard);
376 if (unlikely(obj->oo_dn->dn_maxblkid > 0))
379 CERROR("%s: fail to set blksize for "DFID": rc = %d\n",
380 osd->od_svname, PFID(fid), rc);
383 CDEBUG(D_INODE, "%s: set blksize as %u for "DFID"\n",
384 osd->od_svname, blksz, PFID(fid));
389 * Concurrency: shouldn't matter.
391 int osd_object_init0(const struct lu_env *env, struct osd_object *obj)
393 struct osd_device *osd = osd_obj2dev(obj);
394 const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
400 rc = osd_object_sa_init(obj, osd);
404 /* cache attrs in object */
405 rc = __osd_object_attr_get(env, osd, obj, &obj->oo_attr);
409 if (likely(!fid_is_acct(fid))) {
410 /* no body operations for accounting objects */
411 obj->oo_dt.do_body_ops = &osd_body_ops;
413 if (S_ISREG(obj->oo_attr.la_mode) &&
414 obj->oo_dn->dn_maxblkid == 0 &&
415 (1 << obj->oo_dn->dn_datablkshift) < PAGE_SIZE &&
416 (fid_is_idif(fid) || fid_is_norm(fid) ||
418 osd->od_is_ost && !osd->od_dt_dev.dd_rdonly)
419 osd_obj_set_blksize(env, osd, obj);
423 * initialize object before marking it existing
425 obj->oo_dt.do_lu.lo_header->loh_attr |= obj->oo_attr.la_mode & S_IFMT;
428 obj->oo_dt.do_lu.lo_header->loh_attr |= LOHA_EXISTS;
433 static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
435 struct osd_thread_info *info = osd_oti_get(env);
438 struct lustre_mdt_attrs *lma;
441 CLASSERT(sizeof(info->oti_buf) >= sizeof(*lma));
442 lma = (struct lustre_mdt_attrs *)info->oti_buf;
444 buf.lb_len = sizeof(info->oti_buf);
446 rc = osd_xattr_get(env, &obj->oo_dt, &buf, XATTR_NAME_LMA);
449 lustre_lma_swab(lma);
450 if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
451 CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT))) {
452 CWARN("%s: unsupported incompat LMA feature(s) %#x for "
453 "fid = "DFID"\n", osd_obj2dev(obj)->od_svname,
454 lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
455 PFID(lu_object_fid(&obj->oo_dt.do_lu)));
458 struct osd_device *osd = osd_obj2dev(obj);
460 if (lma->lma_compat & LMAC_STRIPE_INFO &&
462 obj->oo_pfid_in_lma = 1;
463 if (unlikely(lma->lma_incompat & LMAI_REMOTE_PARENT) &&
464 osd->od_remote_parent_dir != ZFS_NO_OBJECT)
465 lu_object_set_agent_entry(&obj->oo_dt.do_lu);
467 } else if (rc == -ENODATA) {
468 /* haven't initialize LMA xattr */
476 * Helper function to retrieve DMU object id from fid for accounting object
478 static dnode_t *osd_quota_fid2dmu(const struct osd_device *osd,
479 const struct lu_fid *fid)
483 LASSERT(fid_is_acct(fid));
485 switch (fid_oid(fid)) {
487 dn = osd->od_userused_dn;
490 dn = osd->od_groupused_dn;
492 #ifdef ZFS_PROJINHERIT
493 case ACCT_PROJECT_OID:
494 dn = osd->od_projectused_dn;
505 * Concurrency: no concurrent access is possible that early in object
508 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
509 const struct lu_object_conf *conf)
511 struct osd_object *obj = osd_obj(l);
512 struct osd_device *osd = osd_obj2dev(obj);
513 const struct lu_fid *fid = lu_object_fid(l);
518 LASSERT(osd_invariant(obj));
520 if (fid_is_otable_it(&l->lo_header->loh_fid)) {
521 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
522 l->lo_header->loh_attr |= LOHA_EXISTS;
526 if (conf != NULL && conf->loc_flags & LOC_F_NEW)
529 if (unlikely(fid_is_acct(fid))) {
530 obj->oo_dn = osd_quota_fid2dmu(osd, fid);
532 obj->oo_dt.do_index_ops = &osd_acct_index_ops;
533 l->lo_header->loh_attr |= LOHA_EXISTS;
539 rc = osd_fid_lookup(env, osd, fid, &oid);
541 LASSERT(obj->oo_dn == NULL);
542 rc = __osd_obj2dnode(osd->od_os, oid, &obj->oo_dn);
543 /* EEXIST will be returned if object is being deleted in ZFS */
549 CERROR("%s: lookup "DFID"/%#llx failed: rc = %d\n",
550 osd->od_svname, PFID(lu_object_fid(l)), oid, rc);
553 rc = osd_object_init0(env, obj);
557 rc = osd_check_lma(env, obj);
560 } else if (rc == -ENOENT) {
563 LASSERT(osd_invariant(obj));
569 * Concurrency: no concurrent access is possible that late in object
572 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
574 struct osd_object *obj = osd_obj(l);
576 LASSERT(osd_invariant(obj));
578 dt_object_fini(&obj->oo_dt);
579 OBD_SLAB_FREE_PTR(obj, osd_object_kmem);
583 osd_object_unlinked_add(struct osd_object *obj, struct osd_thandle *oh)
587 LASSERT(obj->oo_destroy == OSD_DESTROY_ASYNC);
589 /* the object is supposed to be exclusively locked by
590 * the caller (osd_destroy()), while the transaction
591 * (oh) is per-thread and not shared */
592 if (likely(list_empty(&obj->oo_unlinked_linkage))) {
593 list_add(&obj->oo_unlinked_linkage, &oh->ot_unlinked_list);
600 /* Default to max data size covered by a level-1 indirect block */
601 static unsigned long osd_sync_destroy_max_size =
602 1UL << (DN_MAX_INDBLKSHIFT - SPA_BLKPTRSHIFT + SPA_MAXBLOCKSHIFT);
603 module_param(osd_sync_destroy_max_size, ulong, 0444);
604 MODULE_PARM_DESC(osd_sync_destroy_max_size, "Maximum object size to use synchronous destroy.");
607 osd_object_set_destroy_type(struct osd_object *obj)
610 * Lock-less OST_WRITE can race with OST_DESTROY, so set destroy type
611 * only once and use it consistently thereafter.
613 down_write(&obj->oo_guard);
614 if (obj->oo_destroy == OSD_DESTROY_NONE) {
615 if (obj->oo_attr.la_size <= osd_sync_destroy_max_size)
616 obj->oo_destroy = OSD_DESTROY_SYNC;
617 else /* Larger objects are destroyed asynchronously */
618 obj->oo_destroy = OSD_DESTROY_ASYNC;
620 up_write(&obj->oo_guard);
623 static int osd_declare_destroy(const struct lu_env *env, struct dt_object *dt,
626 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
627 struct osd_object *obj = osd_dt_obj(dt);
628 struct osd_device *osd = osd_obj2dev(obj);
629 struct osd_thandle *oh;
636 LASSERT(dt_object_exists(dt));
638 oh = container_of0(th, struct osd_thandle, ot_super);
639 LASSERT(oh->ot_tx != NULL);
641 /* declare that we'll remove object from fid-dnode mapping */
642 zapid = osd_get_name_n_idx(env, osd, fid, NULL, 0, &dn);
643 osd_tx_hold_zap(oh->ot_tx, zapid, dn, FALSE, NULL);
645 osd_declare_xattrs_destroy(env, obj, oh);
648 rc = osd_declare_quota(env, osd, obj->oo_attr.la_uid,
649 obj->oo_attr.la_gid, obj->oo_attr.la_projid,
650 -1, oh, NULL, OSD_QID_INODE);
654 /* data to be truncated */
655 rc = osd_declare_quota(env, osd, obj->oo_attr.la_uid,
656 obj->oo_attr.la_gid, obj->oo_attr.la_projid,
657 0, oh, NULL, OSD_QID_BLK);
661 osd_object_set_destroy_type(obj);
662 if (obj->oo_destroy == OSD_DESTROY_SYNC)
663 dmu_tx_hold_free(oh->ot_tx, obj->oo_dn->dn_object,
666 osd_tx_hold_zap(oh->ot_tx, osd->od_unlinked->dn_object,
667 osd->od_unlinked, TRUE, NULL);
669 /* remove agent entry (if have) from remote parent */
670 if (lu_object_has_agent_entry(&obj->oo_dt.do_lu))
671 osd_tx_hold_zap(oh->ot_tx, osd->od_remote_parent_dir,
674 /* will help to find FID->ino when this object is being
675 * added to PENDING/ */
676 osd_idc_find_and_init(env, osd, obj);
681 static int osd_destroy(const struct lu_env *env, struct dt_object *dt,
684 struct osd_thread_info *info = osd_oti_get(env);
685 char *buf = info->oti_str;
686 struct osd_object *obj = osd_dt_obj(dt);
687 struct osd_device *osd = osd_obj2dev(obj);
688 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
689 struct osd_thandle *oh;
695 down_write(&obj->oo_guard);
697 if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
698 GOTO(out, rc = -ENOENT);
700 LASSERT(obj->oo_dn != NULL);
702 oh = container_of0(th, struct osd_thandle, ot_super);
704 LASSERT(oh->ot_tx != NULL);
706 /* remove obj ref from index dir (it depends) */
707 zapid = osd_get_name_n_idx(env, osd, fid, buf,
708 sizeof(info->oti_str), &zdn);
709 rc = osd_zap_remove(osd, zapid, zdn, buf, oh->ot_tx);
711 CERROR("%s: zap_remove(%s) failed: rc = %d\n",
712 osd->od_svname, buf, rc);
716 rc = osd_xattrs_destroy(env, obj, oh);
718 CERROR("%s: cannot destroy xattrs for %s: rc = %d\n",
719 osd->od_svname, buf, rc);
723 if (lu_object_has_agent_entry(&obj->oo_dt.do_lu)) {
724 rc = osd_delete_from_remote_parent(env, osd, obj, oh, true);
729 oid = obj->oo_dn->dn_object;
730 if (unlikely(obj->oo_destroy == OSD_DESTROY_NONE)) {
731 /* this may happen if the destroy wasn't declared
732 * e.g. when the object is created and then destroyed
733 * in the same transaction - we don't need additional
734 * space for destroy specifically */
735 LASSERT(obj->oo_attr.la_size <= osd_sync_destroy_max_size);
736 rc = -dmu_object_free(osd->od_os, oid, oh->ot_tx);
738 CERROR("%s: failed to free %s %llu: rc = %d\n",
739 osd->od_svname, buf, oid, rc);
740 } else if (obj->oo_destroy == OSD_DESTROY_SYNC) {
741 rc = -dmu_object_free(osd->od_os, oid, oh->ot_tx);
743 CERROR("%s: failed to free %s %llu: rc = %d\n",
744 osd->od_svname, buf, oid, rc);
745 } else { /* asynchronous destroy */
746 char *key = info->oti_key;
748 rc = osd_object_unlinked_add(obj, oh);
752 snprintf(key, sizeof(info->oti_key), "%llx", oid);
753 rc = osd_zap_add(osd, osd->od_unlinked->dn_object,
754 osd->od_unlinked, key, 8, 1, &oid, oh->ot_tx);
756 CERROR("%s: zap_add_int() failed %s %llu: rc = %d\n",
757 osd->od_svname, buf, oid, rc);
761 /* not needed in the cache anymore */
762 set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
764 obj->oo_destroyed = 1;
765 up_write(&obj->oo_guard);
769 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
771 struct osd_object *obj = osd_obj(l);
772 const struct lu_fid *fid = lu_object_fid(l);
775 if (likely(!fid_is_acct(fid))) {
776 osd_object_sa_fini(obj);
777 if (obj->oo_sa_xattr) {
778 nvlist_free(obj->oo_sa_xattr);
779 obj->oo_sa_xattr = NULL;
781 osd_dnode_rele(obj->oo_dn);
782 list_del(&obj->oo_sa_linkage);
789 * Concurrency: ->loo_object_release() is called under site spin-lock.
791 static void osd_object_release(const struct lu_env *env,
797 * Concurrency: shouldn't matter.
799 static int osd_object_print(const struct lu_env *env, void *cookie,
800 lu_printer_t p, const struct lu_object *l)
802 struct osd_object *o = osd_obj(l);
804 return (*p)(env, cookie, LUSTRE_OSD_ZFS_NAME"-object@%p", o);
807 static void osd_read_lock(const struct lu_env *env, struct dt_object *dt,
810 struct osd_object *obj = osd_dt_obj(dt);
812 LASSERT(osd_invariant(obj));
814 down_read_nested(&obj->oo_sem, role);
817 static void osd_write_lock(const struct lu_env *env, struct dt_object *dt,
820 struct osd_object *obj = osd_dt_obj(dt);
822 LASSERT(osd_invariant(obj));
824 down_write_nested(&obj->oo_sem, role);
827 static void osd_read_unlock(const struct lu_env *env, struct dt_object *dt)
829 struct osd_object *obj = osd_dt_obj(dt);
831 LASSERT(osd_invariant(obj));
832 up_read(&obj->oo_sem);
835 static void osd_write_unlock(const struct lu_env *env, struct dt_object *dt)
837 struct osd_object *obj = osd_dt_obj(dt);
839 LASSERT(osd_invariant(obj));
840 up_write(&obj->oo_sem);
843 static int osd_write_locked(const struct lu_env *env, struct dt_object *dt)
845 struct osd_object *obj = osd_dt_obj(dt);
848 LASSERT(osd_invariant(obj));
850 if (down_write_trylock(&obj->oo_sem)) {
852 up_write(&obj->oo_sem);
857 static int osd_attr_get(const struct lu_env *env,
858 struct dt_object *dt,
859 struct lu_attr *attr)
861 struct osd_object *obj = osd_dt_obj(dt);
866 down_read(&obj->oo_guard);
868 if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
869 GOTO(out, rc = -ENOENT);
871 if (unlikely(fid_is_acct(lu_object_fid(&dt->do_lu))))
874 LASSERT(osd_invariant(obj));
877 read_lock(&obj->oo_attr_lock);
878 *attr = obj->oo_attr;
879 if (obj->oo_lma_flags & LUSTRE_ORPHAN_FL)
880 attr->la_flags |= LUSTRE_ORPHAN_FL;
881 read_unlock(&obj->oo_attr_lock);
883 /* with ZFS_DEBUG zrl_add_debug() called by DB_DNODE_ENTER()
884 * from within sa_object_size() can block on a mutex, so
885 * we can't call sa_object_size() holding rwlock */
886 sa_object_size(obj->oo_sa_hdl, &blksize, &blocks);
887 /* we do not control size of indices, so always calculate
888 * it from number of blocks reported by DMU */
889 if (S_ISDIR(attr->la_mode))
890 attr->la_size = 512 * blocks;
891 /* Block size may be not set; suggest maximal I/O transfers. */
893 blksize = osd_spa_maxblocksize(
894 dmu_objset_spa(osd_obj2dev(obj)->od_os));
896 attr->la_blksize = blksize;
897 attr->la_blocks = blocks;
898 attr->la_valid |= LA_BLOCKS | LA_BLKSIZE;
901 up_read(&obj->oo_guard);
905 /* Simple wrapper on top of qsd API which implement quota transfer for osd
906 * setattr needs. As a reminder, only the root user can change ownership of
907 * a file, that's why EDQUOT & EINPROGRESS errors are discarded */
908 static inline int qsd_transfer(const struct lu_env *env,
909 struct qsd_instance *qsd,
910 struct lquota_trans *trans, int qtype,
911 __u64 orig_id, __u64 new_id, __u64 bspace,
912 struct lquota_id_info *qi)
916 if (unlikely(qsd == NULL))
919 LASSERT(qtype >= 0 && qtype < LL_MAXQUOTAS);
920 qi->lqi_type = qtype;
922 /* inode accounting */
923 qi->lqi_is_blk = false;
925 /* one more inode for the new owner ... */
926 qi->lqi_id.qid_uid = new_id;
928 rc = qsd_op_begin(env, qsd, trans, qi, NULL);
929 if (rc == -EDQUOT || rc == -EINPROGRESS)
934 /* and one less inode for the current id */
935 qi->lqi_id.qid_uid = orig_id;;
937 /* can't get EDQUOT when reducing usage */
938 rc = qsd_op_begin(env, qsd, trans, qi, NULL);
939 if (rc == -EINPROGRESS)
944 /* block accounting */
945 qi->lqi_is_blk = true;
947 /* more blocks for the new owner ... */
948 qi->lqi_id.qid_uid = new_id;
949 qi->lqi_space = bspace;
950 rc = qsd_op_begin(env, qsd, trans, qi, NULL);
951 if (rc == -EDQUOT || rc == -EINPROGRESS)
956 /* and finally less blocks for the current owner */
957 qi->lqi_id.qid_uid = orig_id;
958 qi->lqi_space = -bspace;
959 rc = qsd_op_begin(env, qsd, trans, qi, NULL);
960 /* can't get EDQUOT when reducing usage */
961 if (rc == -EINPROGRESS)
966 static int osd_declare_attr_set(const struct lu_env *env,
967 struct dt_object *dt,
968 const struct lu_attr *attr,
969 struct thandle *handle)
971 struct osd_thread_info *info = osd_oti_get(env);
972 struct osd_object *obj = osd_dt_obj(dt);
973 struct osd_device *osd = osd_obj2dev(obj);
975 struct osd_thandle *oh;
983 LASSERT(handle != NULL);
984 LASSERT(osd_invariant(obj));
986 oh = container_of0(handle, struct osd_thandle, ot_super);
988 down_read(&obj->oo_guard);
989 if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
992 LASSERT(obj->oo_sa_hdl != NULL);
993 LASSERT(oh->ot_tx != NULL);
994 /* regular attributes are part of the bonus buffer */
995 /* let's check whether this object is already part of
998 for (txh = list_head(&oh->ot_tx->tx_holds); txh;
999 txh = list_next(&oh->ot_tx->tx_holds, txh)) {
1000 if (txh->txh_dnode == NULL)
1002 if (txh->txh_dnode->dn_object != obj->oo_dn->dn_object)
1004 /* this object is part of the transaction already
1005 * we don't need to declare bonus again */
1010 dmu_tx_hold_bonus(oh->ot_tx, obj->oo_dn->dn_object);
1011 if (oh->ot_tx->tx_err != 0)
1012 GOTO(out, rc = -oh->ot_tx->tx_err);
1014 if (attr && attr->la_valid & LA_FLAGS) {
1015 /* LMA is usually a part of bonus, no need to declare
1019 if (attr && (attr->la_valid & (LA_UID | LA_GID | LA_PROJID))) {
1020 sa_object_size(obj->oo_sa_hdl, &blksize, &bspace);
1021 bspace = toqb(bspace * blksize);
1024 if (attr && attr->la_valid & LA_UID) {
1025 /* quota enforcement for user */
1026 if (attr->la_uid != obj->oo_attr.la_uid) {
1027 rc = qsd_transfer(env, osd->od_quota_slave,
1028 &oh->ot_quota_trans, USRQUOTA,
1029 obj->oo_attr.la_uid, attr->la_uid,
1030 bspace, &info->oti_qi);
1035 if (attr && attr->la_valid & LA_GID) {
1036 /* quota enforcement for group */
1037 if (attr->la_gid != obj->oo_attr.la_gid) {
1038 rc = qsd_transfer(env, osd->od_quota_slave,
1039 &oh->ot_quota_trans, GRPQUOTA,
1040 obj->oo_attr.la_gid, attr->la_gid,
1041 bspace, &info->oti_qi);
1046 #ifdef ZFS_PROJINHERIT
1047 if (attr && attr->la_valid & LA_PROJID) {
1048 if (!osd->od_projectused_dn)
1049 GOTO(out, rc = -EOPNOTSUPP);
1051 /* Usually, if project quota is upgradable for the device,
1052 * then the upgrade will be done before or when mount the
1053 * device. So when we come here, this project should have
1054 * project ID attribute already (that is zero by default).
1055 * Otherwise, there was something wrong during the former
1056 * upgrade, let's return failure to report that.
1058 * Please note that, different from other attributes, you
1059 * can NOT simply set the project ID attribute under such
1060 * case, because adding (NOT change) project ID attribute
1061 * needs to change the object's attribute layout to match
1062 * zfs backend quota accounting requirement. */
1063 if (unlikely(!obj->oo_with_projid))
1064 GOTO(out, rc = -ENXIO);
1066 /* quota enforcement for project */
1067 if (attr->la_projid != obj->oo_attr.la_projid) {
1068 rc = qsd_transfer(env, osd->od_quota_slave,
1069 &oh->ot_quota_trans, PRJQUOTA,
1070 obj->oo_attr.la_projid,
1071 attr->la_projid, bspace,
1079 up_read(&obj->oo_guard);
1084 * Set the attributes of an object
1086 * The transaction passed to this routine must have
1087 * dmu_tx_hold_bonus(tx, oid) called and then assigned
1088 * to a transaction group.
1090 static int osd_attr_set(const struct lu_env *env, struct dt_object *dt,
1091 const struct lu_attr *la, struct thandle *handle)
1093 struct osd_thread_info *info = osd_oti_get(env);
1094 sa_bulk_attr_t *bulk = osd_oti_get(env)->oti_attr_bulk;
1095 struct osd_object *obj = osd_dt_obj(dt);
1096 struct osd_device *osd = osd_obj2dev(obj);
1097 struct osd_thandle *oh;
1098 struct osa_attr *osa = &info->oti_osa;
1099 __u64 valid = la->la_valid;
1105 down_read(&obj->oo_guard);
1106 if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
1107 GOTO(out, rc = -ENOENT);
1109 LASSERT(handle != NULL);
1110 LASSERT(osd_invariant(obj));
1111 LASSERT(obj->oo_sa_hdl);
1113 oh = container_of0(handle, struct osd_thandle, ot_super);
1114 /* Assert that the transaction has been assigned to a
1115 transaction group. */
1116 LASSERT(oh->ot_tx->tx_txg != 0);
1118 /* Only allow set size for regular file */
1119 if (!S_ISREG(dt->do_lu.lo_header->loh_attr))
1120 valid &= ~(LA_SIZE | LA_BLOCKS);
1122 if (valid & LA_CTIME && la->la_ctime == obj->oo_attr.la_ctime)
1125 if (valid & LA_MTIME && la->la_mtime == obj->oo_attr.la_mtime)
1128 if (valid & LA_ATIME && la->la_atime == obj->oo_attr.la_atime)
1134 if (valid & LA_FLAGS) {
1135 struct lustre_mdt_attrs *lma;
1138 if (la->la_flags & LUSTRE_LMA_FL_MASKS) {
1139 LASSERT(!obj->oo_pfid_in_lma);
1140 CLASSERT(sizeof(info->oti_buf) >= sizeof(*lma));
1141 lma = (struct lustre_mdt_attrs *)&info->oti_buf;
1143 buf.lb_len = sizeof(info->oti_buf);
1144 rc = osd_xattr_get(env, &obj->oo_dt, &buf,
1148 le32_to_cpu(lma->lma_incompat);
1149 lma->lma_incompat |=
1150 lustre_to_lma_flags(la->la_flags);
1152 cpu_to_le32(lma->lma_incompat);
1154 buf.lb_len = sizeof(*lma);
1155 rc = osd_xattr_set_internal(env, obj, &buf,
1161 CWARN("%s: failed to set LMA flags: rc = %d\n",
1162 osd->od_svname, rc);
1168 write_lock(&obj->oo_attr_lock);
1171 if (valid & LA_PROJID) {
1172 #ifdef ZFS_PROJINHERIT
1173 /* osd_declare_attr_set() must be called firstly.
1174 * If osd::od_projectused_dn is not set, then we
1175 * can not arrive at here. */
1176 LASSERT(osd->od_projectused_dn);
1177 LASSERT(obj->oo_with_projid);
1179 osa->projid = obj->oo_attr.la_projid = la->la_projid;
1180 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_PROJID(osd), NULL,
1183 valid &= ~LA_PROJID;
1187 if (valid & LA_ATIME) {
1188 osa->atime[0] = obj->oo_attr.la_atime = la->la_atime;
1189 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(osd), NULL,
1192 if (valid & LA_MTIME) {
1193 osa->mtime[0] = obj->oo_attr.la_mtime = la->la_mtime;
1194 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(osd), NULL,
1197 if (valid & LA_CTIME) {
1198 osa->ctime[0] = obj->oo_attr.la_ctime = la->la_ctime;
1199 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(osd), NULL,
1202 if (valid & LA_MODE) {
1203 /* mode is stored along with type, so read it first */
1204 obj->oo_attr.la_mode = (obj->oo_attr.la_mode & S_IFMT) |
1205 (la->la_mode & ~S_IFMT);
1206 osa->mode = obj->oo_attr.la_mode;
1207 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(osd), NULL,
1210 if (valid & LA_SIZE) {
1211 osa->size = obj->oo_attr.la_size = la->la_size;
1212 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(osd), NULL,
1215 if (valid & LA_NLINK) {
1216 osa->nlink = obj->oo_attr.la_nlink = la->la_nlink;
1217 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(osd), NULL,
1220 if (valid & LA_RDEV) {
1221 osa->rdev = obj->oo_attr.la_rdev = la->la_rdev;
1222 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_RDEV(osd), NULL,
1225 if (valid & LA_FLAGS) {
1226 osa->flags = attrs_fs2zfs(la->la_flags);
1227 /* many flags are not supported by zfs, so ensure a good cached
1229 obj->oo_attr.la_flags = attrs_zfs2fs(osa->flags);
1230 #ifdef ZFS_PROJINHERIT
1231 if (obj->oo_with_projid)
1232 osa->flags |= ZFS_PROJID;
1234 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(osd), NULL,
1237 if (valid & LA_UID) {
1238 osa->uid = obj->oo_attr.la_uid = la->la_uid;
1239 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(osd), NULL,
1242 if (valid & LA_GID) {
1243 osa->gid = obj->oo_attr.la_gid = la->la_gid;
1244 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(osd), NULL,
1247 obj->oo_attr.la_valid |= valid;
1248 write_unlock(&obj->oo_attr_lock);
1250 LASSERT(cnt <= ARRAY_SIZE(osd_oti_get(env)->oti_attr_bulk));
1251 rc = osd_object_sa_bulk_update(obj, bulk, cnt, oh);
1254 up_read(&obj->oo_guard);
1261 * XXX temporary solution.
1264 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
1265 struct dt_object *parent, struct dt_object *child,
1270 ah->dah_parent = parent;
1271 ah->dah_mode = child_mode;
1273 if (parent != NULL && !dt_object_remote(parent)) {
1274 /* will help to find FID->ino at dt_insert("..") */
1275 struct osd_object *pobj = osd_dt_obj(parent);
1277 osd_idc_find_and_init(env, osd_obj2dev(pobj), pobj);
1281 static int osd_declare_create(const struct lu_env *env, struct dt_object *dt,
1282 struct lu_attr *attr,
1283 struct dt_allocation_hint *hint,
1284 struct dt_object_format *dof,
1285 struct thandle *handle)
1287 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1288 struct osd_object *obj = osd_dt_obj(dt);
1289 struct osd_device *osd = osd_obj2dev(obj);
1290 struct osd_thandle *oh;
1298 switch (dof->dof_type) {
1302 if (obj->oo_dt.do_body_ops == NULL)
1303 obj->oo_dt.do_body_ops = &osd_body_ops;
1309 LASSERT(handle != NULL);
1310 oh = container_of0(handle, struct osd_thandle, ot_super);
1311 LASSERT(oh->ot_tx != NULL);
1313 /* this is the minimum set of EAs on every Lustre object */
1314 obj->oo_ea_in_bonus = OSD_BASE_EA_IN_BONUS;
1315 /* reserve 32 bytes for extra stuff like ACLs */
1316 dnode_size = size_roundup_power2(obj->oo_ea_in_bonus + 32);
1318 switch (dof->dof_type) {
1320 dt->do_index_ops = &osd_dir_ops;
1322 /* for zap create */
1323 dmu_tx_hold_zap(oh->ot_tx, DMU_NEW_OBJECT, FALSE, NULL);
1324 dmu_tx_hold_sa_create(oh->ot_tx, dnode_size);
1329 /* first, we'll create new object */
1330 dmu_tx_hold_sa_create(oh->ot_tx, dnode_size);
1338 /* and we'll add it to some mapping */
1339 zapid = osd_get_name_n_idx(env, osd, fid, NULL, 0, &dn);
1340 osd_tx_hold_zap(oh->ot_tx, zapid, dn, TRUE, NULL);
1342 /* will help to find FID->ino mapping at dt_insert() */
1343 osd_idc_find_and_init(env, osd, obj);
1345 rc = osd_declare_quota(env, osd, attr->la_uid, attr->la_gid,
1346 attr->la_projid, 1, oh, NULL, OSD_QID_INODE);
1351 int __osd_attr_init(const struct lu_env *env, struct osd_device *osd,
1352 struct osd_object *obj, sa_handle_t *sa_hdl, dmu_tx_t *tx,
1353 struct lu_attr *la, uint64_t parent,
1356 sa_bulk_attr_t *bulk = osd_oti_get(env)->oti_attr_bulk;
1357 struct osa_attr *osa = &osd_oti_get(env)->oti_osa;
1363 char *dxattr = NULL;
1369 gen = dmu_tx_get_txg(tx);
1371 ZFS_TIME_ENCODE(&now, crtime);
1373 osa->atime[0] = la->la_atime;
1374 osa->ctime[0] = la->la_ctime;
1375 osa->mtime[0] = la->la_mtime;
1376 osa->mode = la->la_mode;
1377 osa->uid = la->la_uid;
1378 osa->gid = la->la_gid;
1379 osa->rdev = la->la_rdev;
1380 osa->nlink = la->la_nlink;
1381 if (la->la_valid & LA_FLAGS)
1382 osa->flags = attrs_fs2zfs(la->la_flags);
1385 osa->size = la->la_size;
1386 #ifdef ZFS_PROJINHERIT
1387 if (osd->od_projectused_dn) {
1388 if (la->la_valid & LA_PROJID)
1389 osa->projid = la->la_projid;
1391 osa->projid = ZFS_DEFAULT_PROJID;
1392 osa->flags |= ZFS_PROJID;
1394 obj->oo_with_projid = 1;
1396 osa->flags &= ~ZFS_PROJID;
1401 * we need to create all SA below upon object create.
1403 * XXX The attribute order matters since the accounting callback relies
1404 * on static offsets (i.e. SA_*_OFFSET, see zfs_space_delta_cb()) to
1405 * look up the UID/GID/PROJID attributes. Moreover, the callback does
1406 * not seem to support the spill block.
1407 * We define attributes in the same order as SA_*_OFFSET in order to
1408 * work around the problem. See ORI-610.
1411 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(osd), NULL, &osa->mode, 8);
1412 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(osd), NULL, &osa->size, 8);
1413 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GEN(osd), NULL, &gen, 8);
1414 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(osd), NULL, &osa->uid, 8);
1415 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(osd), NULL, &osa->gid, 8);
1416 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_PARENT(osd), NULL, &parent, 8);
1417 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(osd), NULL, &osa->flags, 8);
1418 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(osd), NULL, osa->atime, 16);
1419 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(osd), NULL, osa->mtime, 16);
1420 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(osd), NULL, osa->ctime, 16);
1421 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CRTIME(osd), NULL, crtime, 16);
1422 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(osd), NULL, &osa->nlink, 8);
1423 #ifdef ZFS_PROJINHERIT
1424 if (osd->od_projectused_dn)
1425 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_PROJID(osd), NULL,
1428 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_RDEV(osd), NULL, &osa->rdev, 8);
1429 LASSERT(cnt <= ARRAY_SIZE(osd_oti_get(env)->oti_attr_bulk));
1432 rc = -nvlist_size(xattr, &sa_size, NV_ENCODE_XDR);
1435 dxattr = osd_zio_buf_alloc(sa_size);
1438 rc = -nvlist_pack(xattr, &dxattr, &sa_size,
1439 NV_ENCODE_XDR, KM_SLEEP);
1442 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_DXATTR(osd),
1443 NULL, dxattr, sa_size);
1446 rc = -sa_replace_all_by_template(sa_hdl, bulk, cnt, tx);
1448 osd_zio_buf_free(dxattr, sa_size);
1453 static int osd_find_new_dnode(const struct lu_env *env, dmu_tx_t *tx,
1454 uint64_t oid, dnode_t **dnp)
1459 /* take dnode_t from tx to save on dnode#->dnode_t lookup */
1460 for (txh = list_tail(&tx->tx_holds); txh;
1461 txh = list_prev(&tx->tx_holds, txh)) {
1462 dnode_t *dn = txh->txh_dnode;
1467 if (dn->dn_object != oid)
1471 rw_enter(&dn->dn_struct_rwlock, RW_WRITER);
1472 if (dn->dn_bonus == NULL)
1473 dbuf_create_bonus(dn);
1474 rw_exit(&dn->dn_struct_rwlock);
1478 LASSERT(dn->dn_handle);
1480 if (refcount_add(&db->db_holds, osd_obj_tag) == 1) {
1481 refcount_add(&dn->dn_holds, tag);
1482 atomic_inc_32(&dn->dn_dbufs_count);
1485 dbuf_read(db, NULL, DB_RF_MUST_SUCCEED | DB_RF_NOPREFETCH);
1489 if (unlikely(*dnp == NULL))
1490 rc = __osd_obj2dnode(tx->tx_objset, oid, dnp);
1495 #ifdef HAVE_DMU_OBJECT_ALLOC_DNSIZE
1496 int osd_find_dnsize(struct osd_device *osd, int ea_in_bonus)
1500 if (osd->od_dnsize == ZFS_DNSIZE_AUTO) {
1501 dnsize = DNODE_MIN_SIZE;
1503 if (DN_BONUS_SIZE(dnsize) >= ea_in_bonus + 32)
1506 } while (dnsize < DNODE_MAX_SIZE);
1507 if (dnsize > DNODE_MAX_SIZE)
1508 dnsize = DNODE_MAX_SIZE;
1509 } else if (osd->od_dnsize == ZFS_DNSIZE_1K) {
1511 } else if (osd->od_dnsize == ZFS_DNSIZE_2K) {
1513 } else if (osd->od_dnsize == ZFS_DNSIZE_4K) {
1515 } else if (osd->od_dnsize == ZFS_DNSIZE_8K) {
1517 } else if (osd->od_dnsize == ZFS_DNSIZE_16K) {
1520 dnsize = DNODE_MIN_SIZE;
1527 * The transaction passed to this routine must have
1528 * dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT) called and then assigned
1529 * to a transaction group.
1531 int __osd_object_create(const struct lu_env *env, struct osd_device *osd,
1532 struct osd_object *obj, const struct lu_fid *fid,
1533 dnode_t **dnp, dmu_tx_t *tx, struct lu_attr *la)
1535 dmu_object_type_t type = DMU_OT_PLAIN_FILE_CONTENTS;
1539 /* Use DMU_OTN_UINT8_METADATA for local objects so their data blocks
1540 * would get an additional ditto copy */
1541 if (unlikely(S_ISREG(la->la_mode) &&
1542 fid_seq_is_local_file(fid_seq(fid))))
1543 type = DMU_OTN_UINT8_METADATA;
1545 /* Create a new DMU object using the default dnode size. */
1547 size = obj->oo_ea_in_bonus;
1549 size = OSD_BASE_EA_IN_BONUS;
1550 oid = osd_dmu_object_alloc(osd->od_os, type, 0,
1551 osd_find_dnsize(osd, size), tx);
1553 LASSERT(la->la_valid & LA_MODE);
1557 return osd_find_new_dnode(env, tx, oid, dnp);
1561 * The transaction passed to this routine must have
1562 * dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, ...) called and then assigned
1563 * to a transaction group.
1565 * Using ZAP_FLAG_HASH64 will force the ZAP to always be a FAT ZAP.
1566 * This is fine for directories today, because storing the FID in the dirent
1567 * will also require a FAT ZAP. If there is a new type of micro ZAP created
1568 * then we might need to re-evaluate the use of this flag and instead do
1569 * a conversion from the different internal ZAP hash formats being used. */
1570 int __osd_zap_create(const struct lu_env *env, struct osd_device *osd,
1571 dnode_t **dnp, dmu_tx_t *tx, struct lu_attr *la,
1572 unsigned dnsize, zap_flags_t flags)
1576 /* Assert that the transaction has been assigned to a
1577 transaction group. */
1578 LASSERT(tx->tx_txg != 0);
1581 oid = osd_zap_create_flags(osd->od_os, 0, flags | ZAP_FLAG_HASH64,
1582 DMU_OT_DIRECTORY_CONTENTS,
1583 14, /* == ZFS fzap_default_blockshift */
1584 DN_MAX_INDBLKSHIFT, /* indirect blockshift */
1590 return osd_find_new_dnode(env, tx, oid, dnp);
1593 static dnode_t *osd_mkidx(const struct lu_env *env, struct osd_object *obj,
1594 struct lu_attr *la, struct osd_thandle *oh)
1596 struct osd_device *osd = osd_obj2dev(obj);
1600 /* Index file should be created as regular file in order not to confuse
1601 * ZPL which could interpret them as directory.
1602 * We set ZAP_FLAG_UINT64_KEY to let ZFS know than we are going to use
1604 LASSERT(S_ISREG(la->la_mode));
1605 rc = __osd_zap_create(env, osd, &dn, oh->ot_tx, la,
1606 osd_find_dnsize(osd, obj->oo_ea_in_bonus), ZAP_FLAG_UINT64_KEY);
1612 static dnode_t *osd_mkdir(const struct lu_env *env, struct osd_object *obj,
1613 struct lu_attr *la, struct osd_thandle *oh)
1615 struct osd_device *osd = osd_obj2dev(obj);
1619 LASSERT(S_ISDIR(la->la_mode));
1620 rc = __osd_zap_create(env, osd, &dn, oh->ot_tx, la,
1621 osd_find_dnsize(osd, obj->oo_ea_in_bonus), 0);
1627 static dnode_t *osd_mkreg(const struct lu_env *env, struct osd_object *obj,
1628 struct lu_attr *la, struct osd_thandle *oh)
1630 const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
1631 struct osd_device *osd = osd_obj2dev(obj);
1635 LASSERT(S_ISREG(la->la_mode));
1636 rc = __osd_object_create(env, osd, obj, fid, &dn, oh->ot_tx, la);
1640 if ((fid_is_idif(fid) || fid_is_norm(fid) || fid_is_echo(fid))) {
1641 /* The minimum block size must be at least page size otherwise
1642 * it will break the assumption in tgt_thread_big_cache where
1643 * the array size is PTLRPC_MAX_BRW_PAGES. It will also affect
1644 * RDMA due to subpage transfer size */
1645 rc = -dmu_object_set_blocksize(osd->od_os, dn->dn_object,
1646 PAGE_SIZE, 0, oh->ot_tx);
1648 CERROR("%s: can't change blocksize: %d\n",
1649 osd->od_svname, rc);
1657 static dnode_t *osd_mksym(const struct lu_env *env, struct osd_object *obj,
1658 struct lu_attr *la, struct osd_thandle *oh)
1663 LASSERT(S_ISLNK(la->la_mode));
1664 rc = __osd_object_create(env, osd_obj2dev(obj), obj,
1665 lu_object_fid(&obj->oo_dt.do_lu),
1666 &dn, oh->ot_tx, la);
1672 static dnode_t *osd_mknod(const struct lu_env *env, struct osd_object *obj,
1673 struct lu_attr *la, struct osd_thandle *oh)
1678 if (S_ISCHR(la->la_mode) || S_ISBLK(la->la_mode))
1679 la->la_valid |= LA_RDEV;
1681 rc = __osd_object_create(env, osd_obj2dev(obj), obj,
1682 lu_object_fid(&obj->oo_dt.do_lu),
1683 &dn, oh->ot_tx, la);
1689 typedef dnode_t *(*osd_obj_type_f)(const struct lu_env *env,
1690 struct osd_object *obj,
1692 struct osd_thandle *oh);
1694 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
1696 osd_obj_type_f result;
1722 * Concurrency: @dt is write locked.
1724 static int osd_create(const struct lu_env *env, struct dt_object *dt,
1725 struct lu_attr *attr, struct dt_allocation_hint *hint,
1726 struct dt_object_format *dof, struct thandle *th)
1728 struct osd_thread_info *info = osd_oti_get(env);
1729 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
1730 struct zpl_direntry *zde = &info->oti_zde.lzd_reg;
1731 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1732 struct osd_object *obj = osd_dt_obj(dt);
1733 struct osd_device *osd = osd_obj2dev(obj);
1734 char *buf = info->oti_str;
1735 struct osd_thandle *oh;
1736 dnode_t *dn = NULL, *zdn = NULL;
1737 uint64_t zapid, parent = 0;
1742 LASSERT(!fid_is_acct(fid));
1744 /* concurrent create declarations should not see
1745 * the object inconsistent (db, attr, etc).
1746 * in regular cases acquisition should be cheap */
1747 down_write(&obj->oo_guard);
1749 if (unlikely(dt_object_exists(dt)))
1750 GOTO(out, rc = -EEXIST);
1752 LASSERT(osd_invariant(obj));
1753 LASSERT(dof != NULL);
1755 LASSERT(th != NULL);
1756 oh = container_of0(th, struct osd_thandle, ot_super);
1758 LASSERT(obj->oo_dn == NULL);
1760 /* to follow ZFS on-disk format we need
1761 * to initialize parent dnode properly */
1762 if (hint != NULL && hint->dah_parent != NULL &&
1763 !dt_object_remote(hint->dah_parent))
1764 parent = osd_dt_obj(hint->dah_parent)->oo_dn->dn_object;
1766 /* we may fix some attributes, better do not change the source */
1767 obj->oo_attr = *attr;
1768 obj->oo_attr.la_valid |= LA_SIZE | LA_NLINK | LA_TYPE;
1770 #ifdef ZFS_PROJINHERIT
1771 if (osd->od_projectused_dn) {
1772 if (!(obj->oo_attr.la_valid & LA_PROJID))
1773 obj->oo_attr.la_projid = ZFS_DEFAULT_PROJID;
1774 obj->oo_with_projid = 1;
1778 dn = osd_create_type_f(dof->dof_type)(env, obj, &obj->oo_attr, oh);
1786 zde->zde_dnode = dn->dn_object;
1787 zde->zde_type = IFTODT(attr->la_mode & S_IFMT);
1789 zapid = osd_get_name_n_idx(env, osd, fid, buf,
1790 sizeof(info->oti_str), &zdn);
1791 rc = osd_zap_add(osd, zapid, zdn, buf, 8, 1, zde, oh->ot_tx);
1795 /* Now add in all of the "SA" attributes */
1796 rc = osd_sa_handle_get(obj);
1800 rc = -nvlist_alloc(&obj->oo_sa_xattr, NV_UNIQUE_NAME, KM_SLEEP);
1804 /* initialize LMA */
1805 lustre_lma_init(lma, fid, 0, 0);
1806 lustre_lma_swab(lma);
1807 rc = -nvlist_add_byte_array(obj->oo_sa_xattr, XATTR_NAME_LMA,
1808 (uchar_t *)lma, sizeof(*lma));
1812 /* configure new osd object */
1813 obj->oo_parent = parent != 0 ? parent : zapid;
1814 obj->oo_late_attr_set = 1;
1815 rc = __osd_sa_xattr_schedule_update(env, obj, oh);
1819 /* XXX: oo_lma_flags */
1820 obj->oo_dt.do_lu.lo_header->loh_attr |= obj->oo_attr.la_mode & S_IFMT;
1821 if (likely(!fid_is_acct(lu_object_fid(&obj->oo_dt.do_lu))))
1822 /* no body operations for accounting objects */
1823 obj->oo_dt.do_body_ops = &osd_body_ops;
1825 osd_idc_find_and_init(env, osd, obj);
1828 if (unlikely(rc && dn)) {
1829 dmu_object_free(osd->od_os, dn->dn_object, oh->ot_tx);
1833 obj->oo_dt.do_lu.lo_header->loh_attr |= LOHA_EXISTS;
1835 up_write(&obj->oo_guard);
1839 static int osd_declare_ref_add(const struct lu_env *env, struct dt_object *dt,
1842 return osd_declare_attr_set(env, dt, NULL, th);
1846 * Concurrency: @dt is write locked.
1848 static int osd_ref_add(const struct lu_env *env, struct dt_object *dt,
1849 struct thandle *handle)
1851 struct osd_object *obj = osd_dt_obj(dt);
1852 struct osd_thandle *oh;
1853 struct osd_device *osd = osd_obj2dev(obj);
1859 down_read(&obj->oo_guard);
1860 if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
1861 GOTO(out, rc = -ENOENT);
1863 LASSERT(osd_invariant(obj));
1864 LASSERT(obj->oo_sa_hdl != NULL);
1866 oh = container_of0(handle, struct osd_thandle, ot_super);
1868 write_lock(&obj->oo_attr_lock);
1869 nlink = ++obj->oo_attr.la_nlink;
1870 write_unlock(&obj->oo_attr_lock);
1872 rc = osd_object_sa_update(obj, SA_ZPL_LINKS(osd), &nlink, 8, oh);
1875 up_read(&obj->oo_guard);
1879 static int osd_declare_ref_del(const struct lu_env *env, struct dt_object *dt,
1880 struct thandle *handle)
1882 return osd_declare_attr_set(env, dt, NULL, handle);
1886 * Concurrency: @dt is write locked.
1888 static int osd_ref_del(const struct lu_env *env, struct dt_object *dt,
1889 struct thandle *handle)
1891 struct osd_object *obj = osd_dt_obj(dt);
1892 struct osd_thandle *oh;
1893 struct osd_device *osd = osd_obj2dev(obj);
1899 down_read(&obj->oo_guard);
1901 if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
1902 GOTO(out, rc = -ENOENT);
1904 LASSERT(osd_invariant(obj));
1905 LASSERT(obj->oo_sa_hdl != NULL);
1907 oh = container_of0(handle, struct osd_thandle, ot_super);
1908 LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
1910 write_lock(&obj->oo_attr_lock);
1911 nlink = --obj->oo_attr.la_nlink;
1912 write_unlock(&obj->oo_attr_lock);
1914 rc = osd_object_sa_update(obj, SA_ZPL_LINKS(osd), &nlink, 8, oh);
1917 up_read(&obj->oo_guard);
1921 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
1922 __u64 start, __u64 end)
1924 struct osd_device *osd = osd_obj2dev(osd_dt_obj(dt));
1927 /* XXX: no other option than syncing the whole filesystem until we
1928 * support ZIL. If the object tracked the txg that it was last
1929 * modified in, it could pass that txg here instead of "0". Maybe
1930 * the changes are already committed, so no wait is needed at all? */
1931 if (!osd->od_dt_dev.dd_rdonly)
1932 txg_wait_synced(dmu_objset_pool(osd->od_os), 0ULL);
1937 static int osd_invalidate(const struct lu_env *env, struct dt_object *dt)
1942 static struct dt_object_operations osd_obj_ops = {
1943 .do_read_lock = osd_read_lock,
1944 .do_write_lock = osd_write_lock,
1945 .do_read_unlock = osd_read_unlock,
1946 .do_write_unlock = osd_write_unlock,
1947 .do_write_locked = osd_write_locked,
1948 .do_attr_get = osd_attr_get,
1949 .do_declare_attr_set = osd_declare_attr_set,
1950 .do_attr_set = osd_attr_set,
1951 .do_ah_init = osd_ah_init,
1952 .do_declare_create = osd_declare_create,
1953 .do_create = osd_create,
1954 .do_declare_destroy = osd_declare_destroy,
1955 .do_destroy = osd_destroy,
1956 .do_index_try = osd_index_try,
1957 .do_declare_ref_add = osd_declare_ref_add,
1958 .do_ref_add = osd_ref_add,
1959 .do_declare_ref_del = osd_declare_ref_del,
1960 .do_ref_del = osd_ref_del,
1961 .do_xattr_get = osd_xattr_get,
1962 .do_declare_xattr_set = osd_declare_xattr_set,
1963 .do_xattr_set = osd_xattr_set,
1964 .do_declare_xattr_del = osd_declare_xattr_del,
1965 .do_xattr_del = osd_xattr_del,
1966 .do_xattr_list = osd_xattr_list,
1967 .do_object_sync = osd_object_sync,
1968 .do_invalidate = osd_invalidate,
1971 static struct lu_object_operations osd_lu_obj_ops = {
1972 .loo_object_init = osd_object_init,
1973 .loo_object_delete = osd_object_delete,
1974 .loo_object_release = osd_object_release,
1975 .loo_object_free = osd_object_free,
1976 .loo_object_print = osd_object_print,
1977 .loo_object_invariant = osd_object_invariant,
1980 static int osd_otable_it_attr_get(const struct lu_env *env,
1981 struct dt_object *dt,
1982 struct lu_attr *attr)
1988 static struct dt_object_operations osd_obj_otable_it_ops = {
1989 .do_attr_get = osd_otable_it_attr_get,
1990 .do_index_try = osd_index_try,