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 } else if (rc == -ENODATA) {
459 /* haven't initialize LMA xattr */
467 * Helper function to retrieve DMU object id from fid for accounting object
469 static dnode_t *osd_quota_fid2dmu(const struct osd_device *osd,
470 const struct lu_fid *fid)
474 LASSERT(fid_is_acct(fid));
476 switch (fid_oid(fid)) {
478 dn = osd->od_userused_dn;
481 dn = osd->od_groupused_dn;
483 #ifdef ZFS_PROJINHERIT
484 case ACCT_PROJECT_OID:
485 dn = osd->od_projectused_dn;
496 * Concurrency: no concurrent access is possible that early in object
499 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
500 const struct lu_object_conf *conf)
502 struct osd_object *obj = osd_obj(l);
503 struct osd_device *osd = osd_obj2dev(obj);
504 const struct lu_fid *fid = lu_object_fid(l);
509 LASSERT(osd_invariant(obj));
511 if (fid_is_otable_it(&l->lo_header->loh_fid)) {
512 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
513 l->lo_header->loh_attr |= LOHA_EXISTS;
517 if (conf != NULL && conf->loc_flags & LOC_F_NEW)
520 if (unlikely(fid_is_acct(fid))) {
521 obj->oo_dn = osd_quota_fid2dmu(osd, fid);
523 obj->oo_dt.do_index_ops = &osd_acct_index_ops;
524 l->lo_header->loh_attr |= LOHA_EXISTS;
530 rc = osd_fid_lookup(env, osd, fid, &oid);
532 LASSERT(obj->oo_dn == NULL);
533 rc = __osd_obj2dnode(osd->od_os, oid, &obj->oo_dn);
534 /* EEXIST will be returned if object is being deleted in ZFS */
540 CERROR("%s: lookup "DFID"/%#llx failed: rc = %d\n",
541 osd->od_svname, PFID(lu_object_fid(l)), oid, rc);
544 rc = osd_object_init0(env, obj);
548 rc = osd_check_lma(env, obj);
551 } else if (rc == -ENOENT) {
554 LASSERT(osd_invariant(obj));
560 * Concurrency: no concurrent access is possible that late in object
563 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
565 struct osd_object *obj = osd_obj(l);
567 LASSERT(osd_invariant(obj));
569 dt_object_fini(&obj->oo_dt);
570 OBD_SLAB_FREE_PTR(obj, osd_object_kmem);
574 osd_object_unlinked_add(struct osd_object *obj, struct osd_thandle *oh)
578 LASSERT(obj->oo_destroy == OSD_DESTROY_ASYNC);
580 /* the object is supposed to be exclusively locked by
581 * the caller (osd_destroy()), while the transaction
582 * (oh) is per-thread and not shared */
583 if (likely(list_empty(&obj->oo_unlinked_linkage))) {
584 list_add(&obj->oo_unlinked_linkage, &oh->ot_unlinked_list);
591 /* Default to max data size covered by a level-1 indirect block */
592 static unsigned long osd_sync_destroy_max_size =
593 1UL << (DN_MAX_INDBLKSHIFT - SPA_BLKPTRSHIFT + SPA_MAXBLOCKSHIFT);
594 module_param(osd_sync_destroy_max_size, ulong, 0444);
595 MODULE_PARM_DESC(osd_sync_destroy_max_size, "Maximum object size to use synchronous destroy.");
598 osd_object_set_destroy_type(struct osd_object *obj)
601 * Lock-less OST_WRITE can race with OST_DESTROY, so set destroy type
602 * only once and use it consistently thereafter.
604 down_write(&obj->oo_guard);
605 if (obj->oo_destroy == OSD_DESTROY_NONE) {
606 if (obj->oo_attr.la_size <= osd_sync_destroy_max_size)
607 obj->oo_destroy = OSD_DESTROY_SYNC;
608 else /* Larger objects are destroyed asynchronously */
609 obj->oo_destroy = OSD_DESTROY_ASYNC;
611 up_write(&obj->oo_guard);
614 static int osd_declare_destroy(const struct lu_env *env, struct dt_object *dt,
617 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
618 struct osd_object *obj = osd_dt_obj(dt);
619 struct osd_device *osd = osd_obj2dev(obj);
620 struct osd_thandle *oh;
627 LASSERT(dt_object_exists(dt));
629 oh = container_of0(th, struct osd_thandle, ot_super);
630 LASSERT(oh->ot_tx != NULL);
632 /* declare that we'll remove object from fid-dnode mapping */
633 zapid = osd_get_name_n_idx(env, osd, fid, NULL, 0, &dn);
634 osd_tx_hold_zap(oh->ot_tx, zapid, dn, FALSE, NULL);
636 osd_declare_xattrs_destroy(env, obj, oh);
639 rc = osd_declare_quota(env, osd, obj->oo_attr.la_uid,
640 obj->oo_attr.la_gid, obj->oo_attr.la_projid,
641 -1, oh, NULL, OSD_QID_INODE);
645 /* data to be truncated */
646 rc = osd_declare_quota(env, osd, obj->oo_attr.la_uid,
647 obj->oo_attr.la_gid, obj->oo_attr.la_projid,
648 0, oh, NULL, OSD_QID_BLK);
652 osd_object_set_destroy_type(obj);
653 if (obj->oo_destroy == OSD_DESTROY_SYNC)
654 dmu_tx_hold_free(oh->ot_tx, obj->oo_dn->dn_object,
657 osd_tx_hold_zap(oh->ot_tx, osd->od_unlinked->dn_object,
658 osd->od_unlinked, TRUE, NULL);
660 /* will help to find FID->ino when this object is being
661 * added to PENDING/ */
662 osd_idc_find_and_init(env, osd, obj);
667 static int osd_destroy(const struct lu_env *env, struct dt_object *dt,
670 struct osd_thread_info *info = osd_oti_get(env);
671 char *buf = info->oti_str;
672 struct osd_object *obj = osd_dt_obj(dt);
673 struct osd_device *osd = osd_obj2dev(obj);
674 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
675 struct osd_thandle *oh;
681 down_write(&obj->oo_guard);
683 if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
684 GOTO(out, rc = -ENOENT);
686 LASSERT(obj->oo_dn != NULL);
688 oh = container_of0(th, struct osd_thandle, ot_super);
690 LASSERT(oh->ot_tx != NULL);
692 /* remove obj ref from index dir (it depends) */
693 zapid = osd_get_name_n_idx(env, osd, fid, buf,
694 sizeof(info->oti_str), &zdn);
695 rc = osd_zap_remove(osd, zapid, zdn, buf, oh->ot_tx);
697 CERROR("%s: zap_remove(%s) failed: rc = %d\n",
698 osd->od_svname, buf, rc);
702 rc = osd_xattrs_destroy(env, obj, oh);
704 CERROR("%s: cannot destroy xattrs for %s: rc = %d\n",
705 osd->od_svname, buf, rc);
709 oid = obj->oo_dn->dn_object;
710 if (unlikely(obj->oo_destroy == OSD_DESTROY_NONE)) {
711 /* this may happen if the destroy wasn't declared
712 * e.g. when the object is created and then destroyed
713 * in the same transaction - we don't need additional
714 * space for destroy specifically */
715 LASSERT(obj->oo_attr.la_size <= osd_sync_destroy_max_size);
716 rc = -dmu_object_free(osd->od_os, oid, oh->ot_tx);
718 CERROR("%s: failed to free %s %llu: rc = %d\n",
719 osd->od_svname, buf, oid, rc);
720 } else if (obj->oo_destroy == OSD_DESTROY_SYNC) {
721 rc = -dmu_object_free(osd->od_os, oid, oh->ot_tx);
723 CERROR("%s: failed to free %s %llu: rc = %d\n",
724 osd->od_svname, buf, oid, rc);
725 } else { /* asynchronous destroy */
726 char *key = info->oti_key;
728 rc = osd_object_unlinked_add(obj, oh);
732 snprintf(key, sizeof(info->oti_key), "%llx", oid);
733 rc = osd_zap_add(osd, osd->od_unlinked->dn_object,
734 osd->od_unlinked, key, 8, 1, &oid, oh->ot_tx);
736 CERROR("%s: zap_add_int() failed %s %llu: rc = %d\n",
737 osd->od_svname, buf, oid, rc);
741 /* not needed in the cache anymore */
742 set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
744 obj->oo_destroyed = 1;
745 up_write(&obj->oo_guard);
749 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
751 struct osd_object *obj = osd_obj(l);
752 const struct lu_fid *fid = lu_object_fid(l);
755 if (likely(!fid_is_acct(fid))) {
756 osd_object_sa_fini(obj);
757 if (obj->oo_sa_xattr) {
758 nvlist_free(obj->oo_sa_xattr);
759 obj->oo_sa_xattr = NULL;
761 osd_dnode_rele(obj->oo_dn);
762 list_del(&obj->oo_sa_linkage);
769 * Concurrency: ->loo_object_release() is called under site spin-lock.
771 static void osd_object_release(const struct lu_env *env,
777 * Concurrency: shouldn't matter.
779 static int osd_object_print(const struct lu_env *env, void *cookie,
780 lu_printer_t p, const struct lu_object *l)
782 struct osd_object *o = osd_obj(l);
784 return (*p)(env, cookie, LUSTRE_OSD_ZFS_NAME"-object@%p", o);
787 static void osd_read_lock(const struct lu_env *env, struct dt_object *dt,
790 struct osd_object *obj = osd_dt_obj(dt);
792 LASSERT(osd_invariant(obj));
794 down_read_nested(&obj->oo_sem, role);
797 static void osd_write_lock(const struct lu_env *env, struct dt_object *dt,
800 struct osd_object *obj = osd_dt_obj(dt);
802 LASSERT(osd_invariant(obj));
804 down_write_nested(&obj->oo_sem, role);
807 static void osd_read_unlock(const struct lu_env *env, struct dt_object *dt)
809 struct osd_object *obj = osd_dt_obj(dt);
811 LASSERT(osd_invariant(obj));
812 up_read(&obj->oo_sem);
815 static void osd_write_unlock(const struct lu_env *env, struct dt_object *dt)
817 struct osd_object *obj = osd_dt_obj(dt);
819 LASSERT(osd_invariant(obj));
820 up_write(&obj->oo_sem);
823 static int osd_write_locked(const struct lu_env *env, struct dt_object *dt)
825 struct osd_object *obj = osd_dt_obj(dt);
828 LASSERT(osd_invariant(obj));
830 if (down_write_trylock(&obj->oo_sem)) {
832 up_write(&obj->oo_sem);
837 static int osd_attr_get(const struct lu_env *env,
838 struct dt_object *dt,
839 struct lu_attr *attr)
841 struct osd_object *obj = osd_dt_obj(dt);
846 down_read(&obj->oo_guard);
848 if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
849 GOTO(out, rc = -ENOENT);
851 if (unlikely(fid_is_acct(lu_object_fid(&dt->do_lu))))
854 LASSERT(osd_invariant(obj));
857 read_lock(&obj->oo_attr_lock);
858 *attr = obj->oo_attr;
859 if (obj->oo_lma_flags & LUSTRE_ORPHAN_FL)
860 attr->la_flags |= LUSTRE_ORPHAN_FL;
861 read_unlock(&obj->oo_attr_lock);
863 /* with ZFS_DEBUG zrl_add_debug() called by DB_DNODE_ENTER()
864 * from within sa_object_size() can block on a mutex, so
865 * we can't call sa_object_size() holding rwlock */
866 sa_object_size(obj->oo_sa_hdl, &blksize, &blocks);
867 /* we do not control size of indices, so always calculate
868 * it from number of blocks reported by DMU */
869 if (S_ISDIR(attr->la_mode))
870 attr->la_size = 512 * blocks;
871 /* Block size may be not set; suggest maximal I/O transfers. */
873 blksize = osd_spa_maxblocksize(
874 dmu_objset_spa(osd_obj2dev(obj)->od_os));
876 attr->la_blksize = blksize;
877 attr->la_blocks = blocks;
878 attr->la_valid |= LA_BLOCKS | LA_BLKSIZE;
881 up_read(&obj->oo_guard);
885 /* Simple wrapper on top of qsd API which implement quota transfer for osd
886 * setattr needs. As a reminder, only the root user can change ownership of
887 * a file, that's why EDQUOT & EINPROGRESS errors are discarded */
888 static inline int qsd_transfer(const struct lu_env *env,
889 struct qsd_instance *qsd,
890 struct lquota_trans *trans, int qtype,
891 __u64 orig_id, __u64 new_id, __u64 bspace,
892 struct lquota_id_info *qi)
896 if (unlikely(qsd == NULL))
899 LASSERT(qtype >= 0 && qtype < LL_MAXQUOTAS);
900 qi->lqi_type = qtype;
902 /* inode accounting */
903 qi->lqi_is_blk = false;
905 /* one more inode for the new owner ... */
906 qi->lqi_id.qid_uid = new_id;
908 rc = qsd_op_begin(env, qsd, trans, qi, NULL);
909 if (rc == -EDQUOT || rc == -EINPROGRESS)
914 /* and one less inode for the current id */
915 qi->lqi_id.qid_uid = orig_id;;
917 /* can't get EDQUOT when reducing usage */
918 rc = qsd_op_begin(env, qsd, trans, qi, NULL);
919 if (rc == -EINPROGRESS)
924 /* block accounting */
925 qi->lqi_is_blk = true;
927 /* more blocks for the new owner ... */
928 qi->lqi_id.qid_uid = new_id;
929 qi->lqi_space = bspace;
930 rc = qsd_op_begin(env, qsd, trans, qi, NULL);
931 if (rc == -EDQUOT || rc == -EINPROGRESS)
936 /* and finally less blocks for the current owner */
937 qi->lqi_id.qid_uid = orig_id;
938 qi->lqi_space = -bspace;
939 rc = qsd_op_begin(env, qsd, trans, qi, NULL);
940 /* can't get EDQUOT when reducing usage */
941 if (rc == -EINPROGRESS)
946 static int osd_declare_attr_set(const struct lu_env *env,
947 struct dt_object *dt,
948 const struct lu_attr *attr,
949 struct thandle *handle)
951 struct osd_thread_info *info = osd_oti_get(env);
952 struct osd_object *obj = osd_dt_obj(dt);
953 struct osd_device *osd = osd_obj2dev(obj);
955 struct osd_thandle *oh;
963 LASSERT(handle != NULL);
964 LASSERT(osd_invariant(obj));
966 oh = container_of0(handle, struct osd_thandle, ot_super);
968 down_read(&obj->oo_guard);
969 if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
972 LASSERT(obj->oo_sa_hdl != NULL);
973 LASSERT(oh->ot_tx != NULL);
974 /* regular attributes are part of the bonus buffer */
975 /* let's check whether this object is already part of
978 for (txh = list_head(&oh->ot_tx->tx_holds); txh;
979 txh = list_next(&oh->ot_tx->tx_holds, txh)) {
980 if (txh->txh_dnode == NULL)
982 if (txh->txh_dnode->dn_object != obj->oo_dn->dn_object)
984 /* this object is part of the transaction already
985 * we don't need to declare bonus again */
990 dmu_tx_hold_bonus(oh->ot_tx, obj->oo_dn->dn_object);
991 if (oh->ot_tx->tx_err != 0)
992 GOTO(out, rc = -oh->ot_tx->tx_err);
994 if (attr && attr->la_valid & LA_FLAGS) {
995 /* LMA is usually a part of bonus, no need to declare
999 if (attr && (attr->la_valid & (LA_UID | LA_GID | LA_PROJID))) {
1000 sa_object_size(obj->oo_sa_hdl, &blksize, &bspace);
1001 bspace = toqb(bspace * blksize);
1004 if (attr && attr->la_valid & LA_UID) {
1005 /* quota enforcement for user */
1006 if (attr->la_uid != obj->oo_attr.la_uid) {
1007 rc = qsd_transfer(env, osd->od_quota_slave,
1008 &oh->ot_quota_trans, USRQUOTA,
1009 obj->oo_attr.la_uid, attr->la_uid,
1010 bspace, &info->oti_qi);
1015 if (attr && attr->la_valid & LA_GID) {
1016 /* quota enforcement for group */
1017 if (attr->la_gid != obj->oo_attr.la_gid) {
1018 rc = qsd_transfer(env, osd->od_quota_slave,
1019 &oh->ot_quota_trans, GRPQUOTA,
1020 obj->oo_attr.la_gid, attr->la_gid,
1021 bspace, &info->oti_qi);
1026 #ifdef ZFS_PROJINHERIT
1027 if (attr && attr->la_valid & LA_PROJID) {
1028 if (!osd->od_projectused_dn)
1029 GOTO(out, rc = -EOPNOTSUPP);
1031 /* Usually, if project quota is upgradable for the device,
1032 * then the upgrade will be done before or when mount the
1033 * device. So when we come here, this project should have
1034 * project ID attribute already (that is zero by default).
1035 * Otherwise, there was something wrong during the former
1036 * upgrade, let's return failure to report that.
1038 * Please note that, different from other attributes, you
1039 * can NOT simply set the project ID attribute under such
1040 * case, because adding (NOT change) project ID attribute
1041 * needs to change the object's attribute layout to match
1042 * zfs backend quota accounting requirement. */
1043 if (unlikely(!obj->oo_with_projid))
1044 GOTO(out, rc = -ENXIO);
1046 /* quota enforcement for project */
1047 if (attr->la_projid != obj->oo_attr.la_projid) {
1048 rc = qsd_transfer(env, osd->od_quota_slave,
1049 &oh->ot_quota_trans, PRJQUOTA,
1050 obj->oo_attr.la_projid,
1051 attr->la_projid, bspace,
1059 up_read(&obj->oo_guard);
1064 * Set the attributes of an object
1066 * The transaction passed to this routine must have
1067 * dmu_tx_hold_bonus(tx, oid) called and then assigned
1068 * to a transaction group.
1070 static int osd_attr_set(const struct lu_env *env, struct dt_object *dt,
1071 const struct lu_attr *la, struct thandle *handle)
1073 struct osd_thread_info *info = osd_oti_get(env);
1074 sa_bulk_attr_t *bulk = osd_oti_get(env)->oti_attr_bulk;
1075 struct osd_object *obj = osd_dt_obj(dt);
1076 struct osd_device *osd = osd_obj2dev(obj);
1077 struct osd_thandle *oh;
1078 struct osa_attr *osa = &info->oti_osa;
1079 __u64 valid = la->la_valid;
1085 down_read(&obj->oo_guard);
1086 if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
1087 GOTO(out, rc = -ENOENT);
1089 LASSERT(handle != NULL);
1090 LASSERT(osd_invariant(obj));
1091 LASSERT(obj->oo_sa_hdl);
1093 oh = container_of0(handle, struct osd_thandle, ot_super);
1094 /* Assert that the transaction has been assigned to a
1095 transaction group. */
1096 LASSERT(oh->ot_tx->tx_txg != 0);
1098 /* Only allow set size for regular file */
1099 if (!S_ISREG(dt->do_lu.lo_header->loh_attr))
1100 valid &= ~(LA_SIZE | LA_BLOCKS);
1102 if (valid & LA_CTIME && la->la_ctime == obj->oo_attr.la_ctime)
1105 if (valid & LA_MTIME && la->la_mtime == obj->oo_attr.la_mtime)
1108 if (valid & LA_ATIME && la->la_atime == obj->oo_attr.la_atime)
1114 if (valid & LA_FLAGS) {
1115 struct lustre_mdt_attrs *lma;
1118 if (la->la_flags & LUSTRE_LMA_FL_MASKS) {
1119 CLASSERT(sizeof(info->oti_buf) >= sizeof(*lma));
1120 lma = (struct lustre_mdt_attrs *)&info->oti_buf;
1122 buf.lb_len = sizeof(info->oti_buf);
1123 rc = osd_xattr_get(env, &obj->oo_dt, &buf,
1127 le32_to_cpu(lma->lma_incompat);
1128 lma->lma_incompat |=
1129 lustre_to_lma_flags(la->la_flags);
1131 cpu_to_le32(lma->lma_incompat);
1133 buf.lb_len = sizeof(*lma);
1134 rc = osd_xattr_set_internal(env, obj, &buf,
1140 CWARN("%s: failed to set LMA flags: rc = %d\n",
1141 osd->od_svname, rc);
1147 write_lock(&obj->oo_attr_lock);
1150 if (valid & LA_PROJID) {
1151 #ifdef ZFS_PROJINHERIT
1152 /* osd_declare_attr_set() must be called firstly.
1153 * If osd::od_projectused_dn is not set, then we
1154 * can not arrive at here. */
1155 LASSERT(osd->od_projectused_dn);
1156 LASSERT(obj->oo_with_projid);
1158 osa->projid = obj->oo_attr.la_projid = la->la_projid;
1159 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_PROJID(osd), NULL,
1162 valid &= ~LA_PROJID;
1166 if (valid & LA_ATIME) {
1167 osa->atime[0] = obj->oo_attr.la_atime = la->la_atime;
1168 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(osd), NULL,
1171 if (valid & LA_MTIME) {
1172 osa->mtime[0] = obj->oo_attr.la_mtime = la->la_mtime;
1173 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(osd), NULL,
1176 if (valid & LA_CTIME) {
1177 osa->ctime[0] = obj->oo_attr.la_ctime = la->la_ctime;
1178 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(osd), NULL,
1181 if (valid & LA_MODE) {
1182 /* mode is stored along with type, so read it first */
1183 obj->oo_attr.la_mode = (obj->oo_attr.la_mode & S_IFMT) |
1184 (la->la_mode & ~S_IFMT);
1185 osa->mode = obj->oo_attr.la_mode;
1186 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(osd), NULL,
1189 if (valid & LA_SIZE) {
1190 osa->size = obj->oo_attr.la_size = la->la_size;
1191 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(osd), NULL,
1194 if (valid & LA_NLINK) {
1195 osa->nlink = obj->oo_attr.la_nlink = la->la_nlink;
1196 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(osd), NULL,
1199 if (valid & LA_RDEV) {
1200 osa->rdev = obj->oo_attr.la_rdev = la->la_rdev;
1201 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_RDEV(osd), NULL,
1204 if (valid & LA_FLAGS) {
1205 osa->flags = attrs_fs2zfs(la->la_flags);
1206 /* many flags are not supported by zfs, so ensure a good cached
1208 obj->oo_attr.la_flags = attrs_zfs2fs(osa->flags);
1209 #ifdef ZFS_PROJINHERIT
1210 if (obj->oo_with_projid)
1211 osa->flags |= ZFS_PROJID;
1213 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(osd), NULL,
1216 if (valid & LA_UID) {
1217 osa->uid = obj->oo_attr.la_uid = la->la_uid;
1218 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(osd), NULL,
1221 if (valid & LA_GID) {
1222 osa->gid = obj->oo_attr.la_gid = la->la_gid;
1223 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(osd), NULL,
1226 obj->oo_attr.la_valid |= valid;
1227 write_unlock(&obj->oo_attr_lock);
1229 LASSERT(cnt <= ARRAY_SIZE(osd_oti_get(env)->oti_attr_bulk));
1230 rc = osd_object_sa_bulk_update(obj, bulk, cnt, oh);
1233 up_read(&obj->oo_guard);
1240 * XXX temporary solution.
1243 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
1244 struct dt_object *parent, struct dt_object *child,
1249 ah->dah_parent = parent;
1250 ah->dah_mode = child_mode;
1252 if (parent != NULL && !dt_object_remote(parent)) {
1253 /* will help to find FID->ino at dt_insert("..") */
1254 struct osd_object *pobj = osd_dt_obj(parent);
1256 osd_idc_find_and_init(env, osd_obj2dev(pobj), pobj);
1260 static int osd_declare_create(const struct lu_env *env, struct dt_object *dt,
1261 struct lu_attr *attr,
1262 struct dt_allocation_hint *hint,
1263 struct dt_object_format *dof,
1264 struct thandle *handle)
1266 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1267 struct osd_object *obj = osd_dt_obj(dt);
1268 struct osd_device *osd = osd_obj2dev(obj);
1269 struct osd_thandle *oh;
1277 switch (dof->dof_type) {
1281 if (obj->oo_dt.do_body_ops == NULL)
1282 obj->oo_dt.do_body_ops = &osd_body_ops;
1288 LASSERT(handle != NULL);
1289 oh = container_of0(handle, struct osd_thandle, ot_super);
1290 LASSERT(oh->ot_tx != NULL);
1292 /* this is the minimum set of EAs on every Lustre object */
1293 obj->oo_ea_in_bonus = ZFS_SA_BASE_ATTR_SIZE +
1294 sizeof(__u64) + /* VBR VERSION */
1295 sizeof(struct lustre_mdt_attrs); /* LMA */
1296 /* reserve 32 bytes for extra stuff like ACLs */
1297 dnode_size = size_roundup_power2(obj->oo_ea_in_bonus + 32);
1299 switch (dof->dof_type) {
1301 dt->do_index_ops = &osd_dir_ops;
1303 /* for zap create */
1304 dmu_tx_hold_zap(oh->ot_tx, DMU_NEW_OBJECT, FALSE, NULL);
1305 dmu_tx_hold_sa_create(oh->ot_tx, dnode_size);
1310 /* first, we'll create new object */
1311 dmu_tx_hold_sa_create(oh->ot_tx, dnode_size);
1319 /* and we'll add it to some mapping */
1320 zapid = osd_get_name_n_idx(env, osd, fid, NULL, 0, &dn);
1321 osd_tx_hold_zap(oh->ot_tx, zapid, dn, TRUE, NULL);
1323 /* will help to find FID->ino mapping at dt_insert() */
1324 osd_idc_find_and_init(env, osd, obj);
1326 rc = osd_declare_quota(env, osd, attr->la_uid, attr->la_gid,
1327 attr->la_projid, 1, oh, NULL, OSD_QID_INODE);
1332 int __osd_attr_init(const struct lu_env *env, struct osd_device *osd,
1333 struct osd_object *obj, sa_handle_t *sa_hdl, dmu_tx_t *tx,
1334 struct lu_attr *la, uint64_t parent,
1337 sa_bulk_attr_t *bulk = osd_oti_get(env)->oti_attr_bulk;
1338 struct osa_attr *osa = &osd_oti_get(env)->oti_osa;
1344 char *dxattr = NULL;
1350 gen = dmu_tx_get_txg(tx);
1352 ZFS_TIME_ENCODE(&now, crtime);
1354 osa->atime[0] = la->la_atime;
1355 osa->ctime[0] = la->la_ctime;
1356 osa->mtime[0] = la->la_mtime;
1357 osa->mode = la->la_mode;
1358 osa->uid = la->la_uid;
1359 osa->gid = la->la_gid;
1360 osa->rdev = la->la_rdev;
1361 osa->nlink = la->la_nlink;
1362 if (la->la_valid & LA_FLAGS)
1363 osa->flags = attrs_fs2zfs(la->la_flags);
1366 osa->size = la->la_size;
1367 #ifdef ZFS_PROJINHERIT
1368 if (osd->od_projectused_dn) {
1369 if (la->la_valid & LA_PROJID)
1370 osa->projid = la->la_projid;
1372 osa->projid = ZFS_DEFAULT_PROJID;
1373 osa->flags |= ZFS_PROJID;
1375 obj->oo_with_projid = 1;
1377 osa->flags &= ~ZFS_PROJID;
1382 * we need to create all SA below upon object create.
1384 * XXX The attribute order matters since the accounting callback relies
1385 * on static offsets (i.e. SA_*_OFFSET, see zfs_space_delta_cb()) to
1386 * look up the UID/GID/PROJID attributes. Moreover, the callback does
1387 * not seem to support the spill block.
1388 * We define attributes in the same order as SA_*_OFFSET in order to
1389 * work around the problem. See ORI-610.
1392 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(osd), NULL, &osa->mode, 8);
1393 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(osd), NULL, &osa->size, 8);
1394 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GEN(osd), NULL, &gen, 8);
1395 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(osd), NULL, &osa->uid, 8);
1396 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(osd), NULL, &osa->gid, 8);
1397 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_PARENT(osd), NULL, &parent, 8);
1398 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(osd), NULL, &osa->flags, 8);
1399 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(osd), NULL, osa->atime, 16);
1400 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(osd), NULL, osa->mtime, 16);
1401 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(osd), NULL, osa->ctime, 16);
1402 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CRTIME(osd), NULL, crtime, 16);
1403 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(osd), NULL, &osa->nlink, 8);
1404 #ifdef ZFS_PROJINHERIT
1405 if (osd->od_projectused_dn)
1406 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_PROJID(osd), NULL,
1409 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_RDEV(osd), NULL, &osa->rdev, 8);
1410 LASSERT(cnt <= ARRAY_SIZE(osd_oti_get(env)->oti_attr_bulk));
1413 rc = -nvlist_size(xattr, &sa_size, NV_ENCODE_XDR);
1416 dxattr = osd_zio_buf_alloc(sa_size);
1419 rc = -nvlist_pack(xattr, &dxattr, &sa_size,
1420 NV_ENCODE_XDR, KM_SLEEP);
1423 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_DXATTR(osd),
1424 NULL, dxattr, sa_size);
1427 rc = -sa_replace_all_by_template(sa_hdl, bulk, cnt, tx);
1429 osd_zio_buf_free(dxattr, sa_size);
1434 static int osd_find_new_dnode(const struct lu_env *env, dmu_tx_t *tx,
1435 uint64_t oid, dnode_t **dnp)
1440 /* take dnode_t from tx to save on dnode#->dnode_t lookup */
1441 for (txh = list_tail(&tx->tx_holds); txh;
1442 txh = list_prev(&tx->tx_holds, txh)) {
1443 dnode_t *dn = txh->txh_dnode;
1448 if (dn->dn_object != oid)
1452 rw_enter(&dn->dn_struct_rwlock, RW_WRITER);
1453 if (dn->dn_bonus == NULL)
1454 dbuf_create_bonus(dn);
1455 rw_exit(&dn->dn_struct_rwlock);
1459 LASSERT(dn->dn_handle);
1461 if (refcount_add(&db->db_holds, osd_obj_tag) == 1) {
1462 refcount_add(&dn->dn_holds, tag);
1463 atomic_inc_32(&dn->dn_dbufs_count);
1466 dbuf_read(db, NULL, DB_RF_MUST_SUCCEED | DB_RF_NOPREFETCH);
1470 if (unlikely(*dnp == NULL))
1471 rc = __osd_obj2dnode(tx->tx_objset, oid, dnp);
1476 #ifdef HAVE_DMU_OBJECT_ALLOC_DNSIZE
1477 static int osd_find_dnsize(struct osd_object *obj)
1479 struct osd_device *osd = osd_obj2dev(obj);
1482 if (osd->od_dnsize == ZFS_DNSIZE_AUTO) {
1483 dnsize = DNODE_MIN_SIZE;
1485 if (DN_BONUS_SIZE(dnsize) >= obj->oo_ea_in_bonus + 32)
1488 } while (dnsize < DNODE_MAX_SIZE);
1489 if (dnsize > DNODE_MAX_SIZE)
1490 dnsize = DNODE_MAX_SIZE;
1491 } else if (osd->od_dnsize == ZFS_DNSIZE_1K) {
1493 } else if (osd->od_dnsize == ZFS_DNSIZE_2K) {
1495 } else if (osd->od_dnsize == ZFS_DNSIZE_4K) {
1497 } else if (osd->od_dnsize == ZFS_DNSIZE_8K) {
1499 } else if (osd->od_dnsize == ZFS_DNSIZE_16K) {
1502 dnsize = DNODE_MIN_SIZE;
1507 static int inline osd_find_dnsize(struct osd_object *obj)
1509 return DN_MAX_BONUSLEN;
1514 * The transaction passed to this routine must have
1515 * dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT) called and then assigned
1516 * to a transaction group.
1518 int __osd_object_create(const struct lu_env *env, struct osd_object *obj,
1519 dnode_t **dnp, dmu_tx_t *tx, struct lu_attr *la)
1521 struct osd_device *osd = osd_obj2dev(obj);
1522 const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
1523 dmu_object_type_t type = DMU_OT_PLAIN_FILE_CONTENTS;
1526 /* Use DMU_OTN_UINT8_METADATA for local objects so their data blocks
1527 * would get an additional ditto copy */
1528 if (unlikely(S_ISREG(la->la_mode) &&
1529 fid_seq_is_local_file(fid_seq(fid))))
1530 type = DMU_OTN_UINT8_METADATA;
1532 /* Create a new DMU object using the default dnode size. */
1533 oid = osd_dmu_object_alloc(osd->od_os, type, 0,
1534 osd_find_dnsize(obj), tx);
1536 LASSERT(la->la_valid & LA_MODE);
1540 return osd_find_new_dnode(env, tx, oid, dnp);
1544 * The transaction passed to this routine must have
1545 * dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, ...) called and then assigned
1546 * to a transaction group.
1548 * Using ZAP_FLAG_HASH64 will force the ZAP to always be a FAT ZAP.
1549 * This is fine for directories today, because storing the FID in the dirent
1550 * will also require a FAT ZAP. If there is a new type of micro ZAP created
1551 * then we might need to re-evaluate the use of this flag and instead do
1552 * a conversion from the different internal ZAP hash formats being used. */
1553 int __osd_zap_create(const struct lu_env *env, struct osd_device *osd,
1554 dnode_t **dnp, dmu_tx_t *tx, struct lu_attr *la,
1555 unsigned dnsize, zap_flags_t flags)
1559 /* Assert that the transaction has been assigned to a
1560 transaction group. */
1561 LASSERT(tx->tx_txg != 0);
1564 oid = osd_zap_create_flags(osd->od_os, 0, flags | ZAP_FLAG_HASH64,
1565 DMU_OT_DIRECTORY_CONTENTS,
1566 14, /* == ZFS fzap_default_blockshift */
1567 DN_MAX_INDBLKSHIFT, /* indirect blockshift */
1573 return osd_find_new_dnode(env, tx, oid, dnp);
1576 static dnode_t *osd_mkidx(const struct lu_env *env, struct osd_object *obj,
1577 struct lu_attr *la, struct osd_thandle *oh)
1582 /* Index file should be created as regular file in order not to confuse
1583 * ZPL which could interpret them as directory.
1584 * We set ZAP_FLAG_UINT64_KEY to let ZFS know than we are going to use
1586 LASSERT(S_ISREG(la->la_mode));
1587 rc = __osd_zap_create(env, osd_obj2dev(obj), &dn, oh->ot_tx, la,
1588 osd_find_dnsize(obj), ZAP_FLAG_UINT64_KEY);
1594 static dnode_t *osd_mkdir(const struct lu_env *env, struct osd_object *obj,
1595 struct lu_attr *la, struct osd_thandle *oh)
1600 LASSERT(S_ISDIR(la->la_mode));
1601 rc = __osd_zap_create(env, osd_obj2dev(obj), &dn, oh->ot_tx, la,
1602 osd_find_dnsize(obj), 0);
1608 static dnode_t *osd_mkreg(const struct lu_env *env, struct osd_object *obj,
1609 struct lu_attr *la, struct osd_thandle *oh)
1611 const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
1612 struct osd_device *osd = osd_obj2dev(obj);
1616 LASSERT(S_ISREG(la->la_mode));
1617 rc = __osd_object_create(env, obj, &dn, oh->ot_tx, la);
1621 if ((fid_is_idif(fid) || fid_is_norm(fid) || fid_is_echo(fid))) {
1622 /* The minimum block size must be at least page size otherwise
1623 * it will break the assumption in tgt_thread_big_cache where
1624 * the array size is PTLRPC_MAX_BRW_PAGES. It will also affect
1625 * RDMA due to subpage transfer size */
1626 rc = -dmu_object_set_blocksize(osd->od_os, dn->dn_object,
1627 PAGE_SIZE, 0, oh->ot_tx);
1629 CERROR("%s: can't change blocksize: %d\n",
1630 osd->od_svname, rc);
1638 static dnode_t *osd_mksym(const struct lu_env *env, struct osd_object *obj,
1639 struct lu_attr *la, struct osd_thandle *oh)
1644 LASSERT(S_ISLNK(la->la_mode));
1645 rc = __osd_object_create(env, obj, &dn, oh->ot_tx, la);
1651 static dnode_t *osd_mknod(const struct lu_env *env, struct osd_object *obj,
1652 struct lu_attr *la, struct osd_thandle *oh)
1657 if (S_ISCHR(la->la_mode) || S_ISBLK(la->la_mode))
1658 la->la_valid |= LA_RDEV;
1660 rc = __osd_object_create(env, obj, &dn, oh->ot_tx, la);
1666 typedef dnode_t *(*osd_obj_type_f)(const struct lu_env *env,
1667 struct osd_object *obj,
1669 struct osd_thandle *oh);
1671 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
1673 osd_obj_type_f result;
1699 * Concurrency: @dt is write locked.
1701 static int osd_create(const struct lu_env *env, struct dt_object *dt,
1702 struct lu_attr *attr, struct dt_allocation_hint *hint,
1703 struct dt_object_format *dof, struct thandle *th)
1705 struct osd_thread_info *info = osd_oti_get(env);
1706 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
1707 struct zpl_direntry *zde = &info->oti_zde.lzd_reg;
1708 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1709 struct osd_object *obj = osd_dt_obj(dt);
1710 struct osd_device *osd = osd_obj2dev(obj);
1711 char *buf = info->oti_str;
1712 struct osd_thandle *oh;
1713 dnode_t *dn = NULL, *zdn = NULL;
1714 uint64_t zapid, parent = 0;
1719 LASSERT(!fid_is_acct(fid));
1721 /* concurrent create declarations should not see
1722 * the object inconsistent (db, attr, etc).
1723 * in regular cases acquisition should be cheap */
1724 down_write(&obj->oo_guard);
1726 if (unlikely(dt_object_exists(dt)))
1727 GOTO(out, rc = -EEXIST);
1729 LASSERT(osd_invariant(obj));
1730 LASSERT(dof != NULL);
1732 LASSERT(th != NULL);
1733 oh = container_of0(th, struct osd_thandle, ot_super);
1735 LASSERT(obj->oo_dn == NULL);
1737 /* to follow ZFS on-disk format we need
1738 * to initialize parent dnode properly */
1739 if (hint != NULL && hint->dah_parent != NULL &&
1740 !dt_object_remote(hint->dah_parent))
1741 parent = osd_dt_obj(hint->dah_parent)->oo_dn->dn_object;
1743 /* we may fix some attributes, better do not change the source */
1744 obj->oo_attr = *attr;
1745 obj->oo_attr.la_valid |= LA_SIZE | LA_NLINK | LA_TYPE;
1747 #ifdef ZFS_PROJINHERIT
1748 if (osd->od_projectused_dn) {
1749 if (!(obj->oo_attr.la_valid & LA_PROJID))
1750 obj->oo_attr.la_projid = ZFS_DEFAULT_PROJID;
1751 obj->oo_with_projid = 1;
1755 dn = osd_create_type_f(dof->dof_type)(env, obj, &obj->oo_attr, oh);
1763 zde->zde_dnode = dn->dn_object;
1764 zde->zde_type = IFTODT(attr->la_mode & S_IFMT);
1766 zapid = osd_get_name_n_idx(env, osd, fid, buf,
1767 sizeof(info->oti_str), &zdn);
1768 rc = osd_zap_add(osd, zapid, zdn, buf, 8, 1, zde, oh->ot_tx);
1772 /* Now add in all of the "SA" attributes */
1773 rc = osd_sa_handle_get(obj);
1777 rc = -nvlist_alloc(&obj->oo_sa_xattr, NV_UNIQUE_NAME, KM_SLEEP);
1781 /* initialize LMA */
1782 lustre_lma_init(lma, fid, 0, 0);
1783 lustre_lma_swab(lma);
1784 rc = -nvlist_add_byte_array(obj->oo_sa_xattr, XATTR_NAME_LMA,
1785 (uchar_t *)lma, sizeof(*lma));
1789 /* configure new osd object */
1790 obj->oo_parent = parent != 0 ? parent : zapid;
1791 obj->oo_late_attr_set = 1;
1792 rc = __osd_sa_xattr_schedule_update(env, obj, oh);
1796 /* XXX: oo_lma_flags */
1797 obj->oo_dt.do_lu.lo_header->loh_attr |= obj->oo_attr.la_mode & S_IFMT;
1798 if (likely(!fid_is_acct(lu_object_fid(&obj->oo_dt.do_lu))))
1799 /* no body operations for accounting objects */
1800 obj->oo_dt.do_body_ops = &osd_body_ops;
1802 osd_idc_find_and_init(env, osd, obj);
1805 if (unlikely(rc && dn)) {
1806 dmu_object_free(osd->od_os, dn->dn_object, oh->ot_tx);
1810 obj->oo_dt.do_lu.lo_header->loh_attr |= LOHA_EXISTS;
1812 up_write(&obj->oo_guard);
1816 static int osd_declare_ref_add(const struct lu_env *env, struct dt_object *dt,
1819 return osd_declare_attr_set(env, dt, NULL, th);
1823 * Concurrency: @dt is write locked.
1825 static int osd_ref_add(const struct lu_env *env, struct dt_object *dt,
1826 struct thandle *handle)
1828 struct osd_object *obj = osd_dt_obj(dt);
1829 struct osd_thandle *oh;
1830 struct osd_device *osd = osd_obj2dev(obj);
1836 down_read(&obj->oo_guard);
1837 if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
1838 GOTO(out, rc = -ENOENT);
1840 LASSERT(osd_invariant(obj));
1841 LASSERT(obj->oo_sa_hdl != NULL);
1843 oh = container_of0(handle, struct osd_thandle, ot_super);
1845 write_lock(&obj->oo_attr_lock);
1846 nlink = ++obj->oo_attr.la_nlink;
1847 write_unlock(&obj->oo_attr_lock);
1849 rc = osd_object_sa_update(obj, SA_ZPL_LINKS(osd), &nlink, 8, oh);
1852 up_read(&obj->oo_guard);
1856 static int osd_declare_ref_del(const struct lu_env *env, struct dt_object *dt,
1857 struct thandle *handle)
1859 return osd_declare_attr_set(env, dt, NULL, handle);
1863 * Concurrency: @dt is write locked.
1865 static int osd_ref_del(const struct lu_env *env, struct dt_object *dt,
1866 struct thandle *handle)
1868 struct osd_object *obj = osd_dt_obj(dt);
1869 struct osd_thandle *oh;
1870 struct osd_device *osd = osd_obj2dev(obj);
1876 down_read(&obj->oo_guard);
1878 if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
1879 GOTO(out, rc = -ENOENT);
1881 LASSERT(osd_invariant(obj));
1882 LASSERT(obj->oo_sa_hdl != NULL);
1884 oh = container_of0(handle, struct osd_thandle, ot_super);
1885 LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
1887 write_lock(&obj->oo_attr_lock);
1888 nlink = --obj->oo_attr.la_nlink;
1889 write_unlock(&obj->oo_attr_lock);
1891 rc = osd_object_sa_update(obj, SA_ZPL_LINKS(osd), &nlink, 8, oh);
1894 up_read(&obj->oo_guard);
1898 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
1899 __u64 start, __u64 end)
1901 struct osd_device *osd = osd_obj2dev(osd_dt_obj(dt));
1904 /* XXX: no other option than syncing the whole filesystem until we
1905 * support ZIL. If the object tracked the txg that it was last
1906 * modified in, it could pass that txg here instead of "0". Maybe
1907 * the changes are already committed, so no wait is needed at all? */
1908 if (!osd->od_dt_dev.dd_rdonly)
1909 txg_wait_synced(dmu_objset_pool(osd->od_os), 0ULL);
1914 static int osd_invalidate(const struct lu_env *env, struct dt_object *dt)
1919 static struct dt_object_operations osd_obj_ops = {
1920 .do_read_lock = osd_read_lock,
1921 .do_write_lock = osd_write_lock,
1922 .do_read_unlock = osd_read_unlock,
1923 .do_write_unlock = osd_write_unlock,
1924 .do_write_locked = osd_write_locked,
1925 .do_attr_get = osd_attr_get,
1926 .do_declare_attr_set = osd_declare_attr_set,
1927 .do_attr_set = osd_attr_set,
1928 .do_ah_init = osd_ah_init,
1929 .do_declare_create = osd_declare_create,
1930 .do_create = osd_create,
1931 .do_declare_destroy = osd_declare_destroy,
1932 .do_destroy = osd_destroy,
1933 .do_index_try = osd_index_try,
1934 .do_declare_ref_add = osd_declare_ref_add,
1935 .do_ref_add = osd_ref_add,
1936 .do_declare_ref_del = osd_declare_ref_del,
1937 .do_ref_del = osd_ref_del,
1938 .do_xattr_get = osd_xattr_get,
1939 .do_declare_xattr_set = osd_declare_xattr_set,
1940 .do_xattr_set = osd_xattr_set,
1941 .do_declare_xattr_del = osd_declare_xattr_del,
1942 .do_xattr_del = osd_xattr_del,
1943 .do_xattr_list = osd_xattr_list,
1944 .do_object_sync = osd_object_sync,
1945 .do_invalidate = osd_invalidate,
1948 static struct lu_object_operations osd_lu_obj_ops = {
1949 .loo_object_init = osd_object_init,
1950 .loo_object_delete = osd_object_delete,
1951 .loo_object_release = osd_object_release,
1952 .loo_object_free = osd_object_free,
1953 .loo_object_print = osd_object_print,
1954 .loo_object_invariant = osd_object_invariant,
1957 static int osd_otable_it_attr_get(const struct lu_env *env,
1958 struct dt_object *dt,
1959 struct lu_attr *attr)
1965 static struct dt_object_operations osd_obj_otable_it_ops = {
1966 .do_attr_get = osd_otable_it_attr_get,
1967 .do_index_try = osd_index_try,