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.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2009, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2012, 2015, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/osd-zfs/osd_object.c
38 * Author: Alex Zhuravlev <bzzz@whamcloud.com>
39 * Author: Mike Pershin <tappro@whamcloud.com>
40 * Author: Johann Lombardi <johann@whamcloud.com>
43 #define DEBUG_SUBSYSTEM S_OSD
45 #include <lustre_ver.h>
46 #include <libcfs/libcfs.h>
47 #include <obd_support.h>
48 #include <lustre_net.h>
50 #include <obd_class.h>
51 #include <lustre_disk.h>
52 #include <lustre_fid.h>
54 #include "osd_internal.h"
56 #include <sys/dnode.h>
61 #include <sys/spa_impl.h>
62 #include <sys/zfs_znode.h>
63 #include <sys/dmu_tx.h>
64 #include <sys/dmu_objset.h>
65 #include <sys/dsl_prop.h>
66 #include <sys/sa_impl.h>
69 char *osd_obj_tag = "osd_object";
71 static struct dt_object_operations osd_obj_ops;
72 static struct lu_object_operations osd_lu_obj_ops;
73 extern struct dt_body_operations osd_body_ops;
74 static struct dt_object_operations osd_obj_otable_it_ops;
76 extern struct kmem_cache *osd_object_kmem;
79 osd_object_sa_fini(struct osd_object *obj)
82 sa_handle_destroy(obj->oo_sa_hdl);
83 obj->oo_sa_hdl = NULL;
88 osd_object_sa_init(struct osd_object *obj, struct osd_device *o)
92 LASSERT(obj->oo_sa_hdl == NULL);
93 LASSERT(obj->oo_db != NULL);
95 rc = -sa_handle_get(o->od_os, obj->oo_db->db_object, obj,
96 SA_HDL_PRIVATE, &obj->oo_sa_hdl);
100 /* Cache the xattr object id, valid for the life of the object */
101 rc = -sa_lookup(obj->oo_sa_hdl, SA_ZPL_XATTR(o), &obj->oo_xattr, 8);
103 obj->oo_xattr = ZFS_NO_OBJECT;
106 osd_object_sa_fini(obj);
113 * Add object to list of dirty objects in tx handle.
115 void osd_object_sa_dirty_add(struct osd_object *obj, struct osd_thandle *oh)
117 if (!list_empty(&obj->oo_sa_linkage))
120 write_lock(&obj->oo_attr_lock);
121 if (likely(list_empty(&obj->oo_sa_linkage)))
122 list_add(&obj->oo_sa_linkage, &oh->ot_sa_list);
123 write_unlock(&obj->oo_attr_lock);
127 * Release spill block dbuf hold for all dirty SAs.
129 void osd_object_sa_dirty_rele(const struct lu_env *env, struct osd_thandle *oh)
131 struct osd_object *obj;
133 while (!list_empty(&oh->ot_sa_list)) {
134 obj = list_entry(oh->ot_sa_list.next,
135 struct osd_object, oo_sa_linkage);
136 write_lock(&obj->oo_attr_lock);
137 list_del_init(&obj->oo_sa_linkage);
138 if (obj->oo_late_xattr) {
139 LASSERT(oh->ot_assigned != 0);
140 /* we need oo_guard to protect oo_sa_xattr
141 * consistency, but if we just take it, then
142 * we break lock ordering. in majority of
143 * cases the callers don't try to set EAs
144 * very often in concurrent manner, so hopefully
145 * we don't need to retake locks too often */
146 if (down_read_trylock(&obj->oo_guard)) {
147 __osd_sa_xattr_update(env, obj, oh);
148 up_read(&obj->oo_guard);
150 down_read(&obj->oo_guard);
151 __osd_sa_xattr_update(env, obj, oh);
152 up_read(&obj->oo_guard);
155 sa_spill_rele(obj->oo_sa_hdl);
156 write_unlock(&obj->oo_attr_lock);
161 * Update the SA and add the object to the dirty list.
163 int osd_object_sa_update(struct osd_object *obj, sa_attr_type_t type,
164 void *buf, uint32_t buflen, struct osd_thandle *oh)
168 LASSERT(obj->oo_sa_hdl != NULL);
169 LASSERT(oh->ot_tx != NULL);
171 rc = -sa_update(obj->oo_sa_hdl, type, buf, buflen, oh->ot_tx);
172 osd_object_sa_dirty_add(obj, oh);
178 * Bulk update the SA and add the object to the dirty list.
181 osd_object_sa_bulk_update(struct osd_object *obj, sa_bulk_attr_t *attrs,
182 int count, struct osd_thandle *oh)
186 LASSERT(obj->oo_sa_hdl != NULL);
187 LASSERT(oh->ot_tx != NULL);
189 rc = -sa_bulk_update(obj->oo_sa_hdl, attrs, count, oh->ot_tx);
190 osd_object_sa_dirty_add(obj, oh);
196 * Retrieve the attributes of a DMU object
198 int __osd_object_attr_get(const struct lu_env *env, struct osd_device *o,
199 struct osd_object *obj, struct lu_attr *la)
201 struct osa_attr *osa = &osd_oti_get(env)->oti_osa;
202 sa_bulk_attr_t *bulk = osd_oti_get(env)->oti_attr_bulk;
208 LASSERT(obj->oo_db != NULL);
210 rc = -sa_handle_get(o->od_os, obj->oo_db->db_object, NULL,
211 SA_HDL_PRIVATE, &sa_hdl);
215 la->la_valid |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE | LA_TYPE |
216 LA_SIZE | LA_UID | LA_GID | LA_FLAGS | LA_NLINK;
218 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(o), NULL, osa->atime, 16);
219 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(o), NULL, osa->mtime, 16);
220 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(o), NULL, osa->ctime, 16);
221 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(o), NULL, &osa->mode, 8);
222 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(o), NULL, &osa->size, 8);
223 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(o), NULL, &osa->nlink, 8);
224 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(o), NULL, &osa->uid, 8);
225 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(o), NULL, &osa->gid, 8);
226 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(o), NULL, &osa->flags, 8);
227 LASSERT(cnt <= ARRAY_SIZE(osd_oti_get(env)->oti_attr_bulk));
229 rc = -sa_bulk_lookup(sa_hdl, bulk, cnt);
233 la->la_atime = osa->atime[0];
234 la->la_mtime = osa->mtime[0];
235 la->la_ctime = osa->ctime[0];
236 la->la_mode = osa->mode;
237 la->la_uid = osa->uid;
238 la->la_gid = osa->gid;
239 la->la_nlink = osa->nlink;
240 la->la_flags = attrs_zfs2fs(osa->flags);
241 la->la_size = osa->size;
243 /* Try to get extra flag from LMA. Right now, only LMAI_ORPHAN
244 * flags is stored in LMA, and it is only for orphan directory */
245 if (S_ISDIR(la->la_mode) && dt_object_exists(&obj->oo_dt)) {
246 struct osd_thread_info *info = osd_oti_get(env);
247 struct lustre_mdt_attrs *lma;
250 lma = (struct lustre_mdt_attrs *)info->oti_buf;
252 buf.lb_len = sizeof(info->oti_buf);
253 rc = osd_xattr_get(env, &obj->oo_dt, &buf, XATTR_NAME_LMA);
256 lma->lma_incompat = le32_to_cpu(lma->lma_incompat);
258 lma_to_lustre_flags(lma->lma_incompat);
260 } else if (rc == -ENODATA) {
265 if (S_ISCHR(la->la_mode) || S_ISBLK(la->la_mode)) {
266 rc = -sa_lookup(sa_hdl, SA_ZPL_RDEV(o), &osa->rdev, 8);
269 la->la_rdev = osa->rdev;
270 la->la_valid |= LA_RDEV;
273 sa_handle_destroy(sa_hdl);
278 int __osd_obj2dbuf(const struct lu_env *env, objset_t *os,
279 uint64_t oid, dmu_buf_t **dbp)
281 dmu_object_info_t *doi = &osd_oti_get(env)->oti_doi;
284 rc = -sa_buf_hold(os, oid, osd_obj_tag, dbp);
288 dmu_object_info_from_db(*dbp, doi);
289 if (unlikely (oid != DMU_USERUSED_OBJECT &&
290 oid != DMU_GROUPUSED_OBJECT && doi->doi_bonus_type != DMU_OT_SA)) {
291 sa_buf_rele(*dbp, osd_obj_tag);
297 LASSERT((*dbp)->db_object == oid);
298 LASSERT((*dbp)->db_offset == -1);
299 LASSERT((*dbp)->db_data != NULL);
305 * Concurrency: no concurrent access is possible that early in object
308 struct lu_object *osd_object_alloc(const struct lu_env *env,
309 const struct lu_object_header *hdr,
312 struct osd_object *mo;
314 OBD_SLAB_ALLOC_PTR_GFP(mo, osd_object_kmem, GFP_NOFS);
318 l = &mo->oo_dt.do_lu;
319 dt_object_init(&mo->oo_dt, NULL, d);
320 mo->oo_dt.do_ops = &osd_obj_ops;
321 l->lo_ops = &osd_lu_obj_ops;
322 INIT_LIST_HEAD(&mo->oo_sa_linkage);
323 INIT_LIST_HEAD(&mo->oo_unlinked_linkage);
324 init_rwsem(&mo->oo_sem);
325 init_rwsem(&mo->oo_guard);
326 rwlock_init(&mo->oo_attr_lock);
327 mo->oo_destroy = OSD_DESTROY_NONE;
335 * Concurrency: shouldn't matter.
337 int osd_object_init0(const struct lu_env *env, struct osd_object *obj)
339 struct osd_device *osd = osd_obj2dev(obj);
340 const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
344 if (obj->oo_db == NULL)
349 rc = osd_object_sa_init(obj, osd);
353 /* cache attrs in object */
354 rc = __osd_object_attr_get(env, osd, obj, &obj->oo_attr);
358 if (likely(!fid_is_acct(fid)))
359 /* no body operations for accounting objects */
360 obj->oo_dt.do_body_ops = &osd_body_ops;
363 * initialize object before marking it existing
365 obj->oo_dt.do_lu.lo_header->loh_attr |= obj->oo_attr.la_mode & S_IFMT;
368 obj->oo_dt.do_lu.lo_header->loh_attr |= LOHA_EXISTS;
373 static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
375 struct osd_thread_info *info = osd_oti_get(env);
378 struct lustre_mdt_attrs *lma;
381 CLASSERT(sizeof(info->oti_buf) >= sizeof(*lma));
382 lma = (struct lustre_mdt_attrs *)info->oti_buf;
384 buf.lb_len = sizeof(info->oti_buf);
386 rc = osd_xattr_get(env, &obj->oo_dt, &buf, XATTR_NAME_LMA);
389 lustre_lma_swab(lma);
390 if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
391 CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT))) {
392 CWARN("%s: unsupported incompat LMA feature(s) %#x for "
393 "fid = "DFID"\n", osd_obj2dev(obj)->od_svname,
394 lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
395 PFID(lu_object_fid(&obj->oo_dt.do_lu)));
398 } else if (rc == -ENODATA) {
399 /* haven't initialize LMA xattr */
407 * Concurrency: no concurrent access is possible that early in object
410 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
411 const struct lu_object_conf *conf)
413 struct osd_object *obj = osd_obj(l);
414 struct osd_device *osd = osd_obj2dev(obj);
419 LASSERT(osd_invariant(obj));
421 if (fid_is_otable_it(&l->lo_header->loh_fid)) {
422 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
423 l->lo_header->loh_attr |= LOHA_EXISTS;
427 if (conf != NULL && conf->loc_flags & LOC_F_NEW)
430 rc = osd_fid_lookup(env, osd, lu_object_fid(l), &oid);
432 LASSERT(obj->oo_db == NULL);
433 rc = __osd_obj2dbuf(env, osd->od_os, oid, &obj->oo_db);
434 /* EEXIST will be returned if object is being deleted in ZFS */
440 CERROR("%s: lookup "DFID"/"LPX64" failed: rc = %d\n",
441 osd->od_svname, PFID(lu_object_fid(l)), oid, rc);
445 rc = osd_object_init0(env, obj);
449 rc = osd_check_lma(env, obj);
452 } else if (rc == -ENOENT) {
455 LASSERT(osd_invariant(obj));
461 * Concurrency: no concurrent access is possible that late in object
464 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
466 struct osd_object *obj = osd_obj(l);
468 LASSERT(osd_invariant(obj));
470 dt_object_fini(&obj->oo_dt);
471 OBD_SLAB_FREE_PTR(obj, osd_object_kmem);
475 osd_object_unlinked_add(struct osd_object *obj, struct osd_thandle *oh)
479 LASSERT(obj->oo_destroy == OSD_DESTROY_ASYNC);
481 /* the object is supposed to be exclusively locked by
482 * the caller (osd_object_destroy()), while the transaction
483 * (oh) is per-thread and not shared */
484 if (likely(list_empty(&obj->oo_unlinked_linkage))) {
485 list_add(&obj->oo_unlinked_linkage, &oh->ot_unlinked_list);
492 /* Default to max data size covered by a level-1 indirect block */
493 static unsigned long osd_sync_destroy_max_size =
494 1UL << (DN_MAX_INDBLKSHIFT - SPA_BLKPTRSHIFT + SPA_MAXBLOCKSHIFT);
495 module_param(osd_sync_destroy_max_size, ulong, 0444);
496 MODULE_PARM_DESC(osd_sync_destroy_max_size, "Maximum object size to use synchronous destroy.");
499 osd_object_set_destroy_type(struct osd_object *obj)
502 * Lock-less OST_WRITE can race with OST_DESTROY, so set destroy type
503 * only once and use it consistently thereafter.
505 down_write(&obj->oo_guard);
506 if (obj->oo_destroy == OSD_DESTROY_NONE) {
507 if (obj->oo_attr.la_size <= osd_sync_destroy_max_size)
508 obj->oo_destroy = OSD_DESTROY_SYNC;
509 else /* Larger objects are destroyed asynchronously */
510 obj->oo_destroy = OSD_DESTROY_ASYNC;
512 up_write(&obj->oo_guard);
515 static int osd_declare_object_destroy(const struct lu_env *env,
516 struct dt_object *dt,
519 char *buf = osd_oti_get(env)->oti_str;
520 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
521 struct osd_object *obj = osd_dt_obj(dt);
522 struct osd_device *osd = osd_obj2dev(obj);
523 struct osd_thandle *oh;
529 LASSERT(dt_object_exists(dt));
531 oh = container_of0(th, struct osd_thandle, ot_super);
532 LASSERT(oh->ot_tx != NULL);
534 /* declare that we'll remove object from fid-dnode mapping */
535 zapid = osd_get_name_n_idx(env, osd, fid, buf);
536 dmu_tx_hold_bonus(oh->ot_tx, zapid);
537 dmu_tx_hold_zap(oh->ot_tx, zapid, FALSE, buf);
539 osd_declare_xattrs_destroy(env, obj, oh);
541 /* declare that we'll remove object from inode accounting ZAPs */
542 dmu_tx_hold_bonus(oh->ot_tx, osd->od_iusr_oid);
543 dmu_tx_hold_zap(oh->ot_tx, osd->od_iusr_oid, FALSE, buf);
544 dmu_tx_hold_bonus(oh->ot_tx, osd->od_igrp_oid);
545 dmu_tx_hold_zap(oh->ot_tx, osd->od_igrp_oid, FALSE, buf);
548 rc = osd_declare_quota(env, osd, obj->oo_attr.la_uid,
549 obj->oo_attr.la_gid, -1, oh, false, NULL, false);
553 /* data to be truncated */
554 rc = osd_declare_quota(env, osd, obj->oo_attr.la_uid,
555 obj->oo_attr.la_gid, 0, oh, true, NULL, false);
559 osd_object_set_destroy_type(obj);
560 if (obj->oo_destroy == OSD_DESTROY_SYNC)
561 dmu_tx_hold_free(oh->ot_tx, obj->oo_db->db_object,
564 dmu_tx_hold_zap(oh->ot_tx, osd->od_unlinkedid, TRUE, NULL);
569 static int osd_object_destroy(const struct lu_env *env,
570 struct dt_object *dt, struct thandle *th)
572 char *buf = osd_oti_get(env)->oti_str;
573 struct osd_object *obj = osd_dt_obj(dt);
574 struct osd_device *osd = osd_obj2dev(obj);
575 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
576 struct osd_thandle *oh;
581 down_write(&obj->oo_guard);
583 if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
584 GOTO(out, rc = -ENOENT);
586 LASSERT(obj->oo_db != NULL);
588 oh = container_of0(th, struct osd_thandle, ot_super);
590 LASSERT(oh->ot_tx != NULL);
592 /* remove obj ref from index dir (it depends) */
593 zapid = osd_get_name_n_idx(env, osd, fid, buf);
594 rc = -zap_remove(osd->od_os, zapid, buf, oh->ot_tx);
596 CERROR("%s: zap_remove(%s) failed: rc = %d\n",
597 osd->od_svname, buf, rc);
601 rc = osd_xattrs_destroy(env, obj, oh);
603 CERROR("%s: cannot destroy xattrs for %s: rc = %d\n",
604 osd->od_svname, buf, rc);
608 /* Remove object from inode accounting. It is not fatal for the destroy
609 * operation if something goes wrong while updating accounting, but we
610 * still log an error message to notify the administrator */
611 rc = -zap_increment_int(osd->od_os, osd->od_iusr_oid,
612 obj->oo_attr.la_uid, -1, oh->ot_tx);
614 CERROR("%s: failed to remove "DFID" from accounting ZAP for usr"
615 " %d: rc = %d\n", osd->od_svname, PFID(fid),
616 obj->oo_attr.la_uid, rc);
617 rc = -zap_increment_int(osd->od_os, osd->od_igrp_oid,
618 obj->oo_attr.la_gid, -1, oh->ot_tx);
620 CERROR("%s: failed to remove "DFID" from accounting ZAP for grp"
621 " %d: rc = %d\n", osd->od_svname, PFID(fid),
622 obj->oo_attr.la_gid, rc);
624 oid = obj->oo_db->db_object;
625 if (unlikely(obj->oo_destroy == OSD_DESTROY_NONE)) {
626 /* this may happen if the destroy wasn't declared
627 * e.g. when the object is created and then destroyed
628 * in the same transaction - we don't need additional
629 * space for destroy specifically */
630 LASSERT(obj->oo_attr.la_size <= osd_sync_destroy_max_size);
631 rc = -dmu_object_free(osd->od_os, oid, oh->ot_tx);
633 CERROR("%s: failed to free %s "LPU64": rc = %d\n",
634 osd->od_svname, buf, oid, rc);
635 } else if (obj->oo_destroy == OSD_DESTROY_SYNC) {
636 rc = -dmu_object_free(osd->od_os, oid, oh->ot_tx);
638 CERROR("%s: failed to free %s "LPU64": rc = %d\n",
639 osd->od_svname, buf, oid, rc);
640 } else { /* asynchronous destroy */
641 rc = osd_object_unlinked_add(obj, oh);
645 rc = -zap_add_int(osd->od_os, osd->od_unlinkedid,
648 CERROR("%s: zap_add_int() failed %s "LPU64": rc = %d\n",
649 osd->od_svname, buf, oid, rc);
653 /* not needed in the cache anymore */
654 set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
656 obj->oo_destroyed = 1;
657 up_write(&obj->oo_guard);
661 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
663 struct osd_object *obj = osd_obj(l);
665 if (obj->oo_db != NULL) {
666 osd_object_sa_fini(obj);
667 if (obj->oo_sa_xattr) {
668 nvlist_free(obj->oo_sa_xattr);
669 obj->oo_sa_xattr = NULL;
671 sa_buf_rele(obj->oo_db, osd_obj_tag);
672 list_del(&obj->oo_sa_linkage);
678 * Concurrency: ->loo_object_release() is called under site spin-lock.
680 static void osd_object_release(const struct lu_env *env,
686 * Concurrency: shouldn't matter.
688 static int osd_object_print(const struct lu_env *env, void *cookie,
689 lu_printer_t p, const struct lu_object *l)
691 struct osd_object *o = osd_obj(l);
693 return (*p)(env, cookie, LUSTRE_OSD_ZFS_NAME"-object@%p", o);
696 static void osd_object_read_lock(const struct lu_env *env,
697 struct dt_object *dt, unsigned role)
699 struct osd_object *obj = osd_dt_obj(dt);
701 LASSERT(osd_invariant(obj));
703 down_read_nested(&obj->oo_sem, role);
706 static void osd_object_write_lock(const struct lu_env *env,
707 struct dt_object *dt, unsigned role)
709 struct osd_object *obj = osd_dt_obj(dt);
711 LASSERT(osd_invariant(obj));
713 down_write_nested(&obj->oo_sem, role);
716 static void osd_object_read_unlock(const struct lu_env *env,
717 struct dt_object *dt)
719 struct osd_object *obj = osd_dt_obj(dt);
721 LASSERT(osd_invariant(obj));
722 up_read(&obj->oo_sem);
725 static void osd_object_write_unlock(const struct lu_env *env,
726 struct dt_object *dt)
728 struct osd_object *obj = osd_dt_obj(dt);
730 LASSERT(osd_invariant(obj));
731 up_write(&obj->oo_sem);
734 static int osd_object_write_locked(const struct lu_env *env,
735 struct dt_object *dt)
737 struct osd_object *obj = osd_dt_obj(dt);
740 LASSERT(osd_invariant(obj));
742 if (down_write_trylock(&obj->oo_sem)) {
744 up_write(&obj->oo_sem);
749 static int osd_attr_get(const struct lu_env *env,
750 struct dt_object *dt,
751 struct lu_attr *attr)
753 struct osd_object *obj = osd_dt_obj(dt);
758 down_read(&obj->oo_guard);
760 if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
761 GOTO(out, rc = -ENOENT);
763 LASSERT(osd_invariant(obj));
766 read_lock(&obj->oo_attr_lock);
767 *attr = obj->oo_attr;
768 if (obj->oo_lma_flags & LUSTRE_ORPHAN_FL)
769 attr->la_flags |= LUSTRE_ORPHAN_FL;
770 read_unlock(&obj->oo_attr_lock);
772 /* with ZFS_DEBUG zrl_add_debug() called by DB_DNODE_ENTER()
773 * from within sa_object_size() can block on a mutex, so
774 * we can't call sa_object_size() holding rwlock */
775 sa_object_size(obj->oo_sa_hdl, &blksize, &blocks);
776 /* we do not control size of indices, so always calculate
777 * it from number of blocks reported by DMU */
778 if (S_ISDIR(attr->la_mode))
779 attr->la_size = 512 * blocks;
780 /* Block size may be not set; suggest maximal I/O transfers. */
782 blksize = osd_spa_maxblocksize(
783 dmu_objset_spa(osd_obj2dev(obj)->od_os));
785 attr->la_blksize = blksize;
786 attr->la_blocks = blocks;
787 attr->la_valid |= LA_BLOCKS | LA_BLKSIZE;
790 up_read(&obj->oo_guard);
794 /* Simple wrapper on top of qsd API which implement quota transfer for osd
795 * setattr needs. As a reminder, only the root user can change ownership of
796 * a file, that's why EDQUOT & EINPROGRESS errors are discarded */
797 static inline int qsd_transfer(const struct lu_env *env,
798 struct qsd_instance *qsd,
799 struct lquota_trans *trans, int qtype,
800 __u64 orig_id, __u64 new_id, __u64 bspace,
801 struct lquota_id_info *qi)
805 if (unlikely(qsd == NULL))
808 LASSERT(qtype >= 0 && qtype < LL_MAXQUOTAS);
809 qi->lqi_type = qtype;
811 /* inode accounting */
812 qi->lqi_is_blk = false;
814 /* one more inode for the new owner ... */
815 qi->lqi_id.qid_uid = new_id;
817 rc = qsd_op_begin(env, qsd, trans, qi, NULL);
818 if (rc == -EDQUOT || rc == -EINPROGRESS)
823 /* and one less inode for the current id */
824 qi->lqi_id.qid_uid = orig_id;;
826 /* can't get EDQUOT when reducing usage */
827 rc = qsd_op_begin(env, qsd, trans, qi, NULL);
828 if (rc == -EINPROGRESS)
833 /* block accounting */
834 qi->lqi_is_blk = true;
836 /* more blocks for the new owner ... */
837 qi->lqi_id.qid_uid = new_id;
838 qi->lqi_space = bspace;
839 rc = qsd_op_begin(env, qsd, trans, qi, NULL);
840 if (rc == -EDQUOT || rc == -EINPROGRESS)
845 /* and finally less blocks for the current owner */
846 qi->lqi_id.qid_uid = orig_id;
847 qi->lqi_space = -bspace;
848 rc = qsd_op_begin(env, qsd, trans, qi, NULL);
849 /* can't get EDQUOT when reducing usage */
850 if (rc == -EINPROGRESS)
855 static int osd_declare_attr_set(const struct lu_env *env,
856 struct dt_object *dt,
857 const struct lu_attr *attr,
858 struct thandle *handle)
860 struct osd_thread_info *info = osd_oti_get(env);
861 char *buf = osd_oti_get(env)->oti_str;
862 struct osd_object *obj = osd_dt_obj(dt);
863 struct osd_device *osd = osd_obj2dev(obj);
864 struct osd_thandle *oh;
871 LASSERT(handle != NULL);
872 LASSERT(osd_invariant(obj));
874 oh = container_of0(handle, struct osd_thandle, ot_super);
876 down_read(&obj->oo_guard);
877 if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
880 LASSERT(obj->oo_sa_hdl != NULL);
881 LASSERT(oh->ot_tx != NULL);
882 dmu_tx_hold_sa(oh->ot_tx, obj->oo_sa_hdl, 0);
883 if (oh->ot_tx->tx_err != 0)
884 GOTO(out, rc = -oh->ot_tx->tx_err);
886 sa_object_size(obj->oo_sa_hdl, &blksize, &bspace);
887 bspace = toqb(bspace * blksize);
889 __osd_xattr_declare_set(env, obj, sizeof(struct lustre_mdt_attrs),
892 if (attr && attr->la_valid & LA_UID) {
893 /* account for user inode tracking ZAP update */
894 dmu_tx_hold_bonus(oh->ot_tx, osd->od_iusr_oid);
895 dmu_tx_hold_zap(oh->ot_tx, osd->od_iusr_oid, TRUE, buf);
897 /* quota enforcement for user */
898 if (attr->la_uid != obj->oo_attr.la_uid) {
899 rc = qsd_transfer(env, osd->od_quota_slave,
900 &oh->ot_quota_trans, USRQUOTA,
901 obj->oo_attr.la_uid, attr->la_uid,
902 bspace, &info->oti_qi);
907 if (attr && attr->la_valid & LA_GID) {
908 /* account for user inode tracking ZAP update */
909 dmu_tx_hold_bonus(oh->ot_tx, osd->od_igrp_oid);
910 dmu_tx_hold_zap(oh->ot_tx, osd->od_igrp_oid, TRUE, buf);
912 /* quota enforcement for group */
913 if (attr->la_gid != obj->oo_attr.la_gid) {
914 rc = qsd_transfer(env, osd->od_quota_slave,
915 &oh->ot_quota_trans, GRPQUOTA,
916 obj->oo_attr.la_gid, attr->la_gid,
917 bspace, &info->oti_qi);
924 up_read(&obj->oo_guard);
929 * Set the attributes of an object
931 * The transaction passed to this routine must have
932 * dmu_tx_hold_bonus(tx, oid) called and then assigned
933 * to a transaction group.
935 static int osd_attr_set(const struct lu_env *env, struct dt_object *dt,
936 const struct lu_attr *la, struct thandle *handle)
938 struct osd_thread_info *info = osd_oti_get(env);
939 sa_bulk_attr_t *bulk = osd_oti_get(env)->oti_attr_bulk;
940 struct osd_object *obj = osd_dt_obj(dt);
941 struct osd_device *osd = osd_obj2dev(obj);
942 struct osd_thandle *oh;
943 struct osa_attr *osa = &info->oti_osa;
944 __u64 valid = la->la_valid;
950 down_read(&obj->oo_guard);
951 if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
952 GOTO(out, rc = -ENOENT);
954 LASSERT(handle != NULL);
955 LASSERT(osd_invariant(obj));
956 LASSERT(obj->oo_sa_hdl);
958 oh = container_of0(handle, struct osd_thandle, ot_super);
959 /* Assert that the transaction has been assigned to a
960 transaction group. */
961 LASSERT(oh->ot_tx->tx_txg != 0);
963 /* Only allow set size for regular file */
964 if (!S_ISREG(dt->do_lu.lo_header->loh_attr))
965 valid &= ~(LA_SIZE | LA_BLOCKS);
967 if (valid & LA_CTIME && la->la_ctime == obj->oo_attr.la_ctime)
970 if (valid & LA_MTIME && la->la_mtime == obj->oo_attr.la_mtime)
973 if (valid & LA_ATIME && la->la_atime == obj->oo_attr.la_atime)
979 if (valid & LA_FLAGS) {
980 struct lustre_mdt_attrs *lma;
983 if (la->la_flags & LUSTRE_LMA_FL_MASKS) {
984 CLASSERT(sizeof(info->oti_buf) >= sizeof(*lma));
985 lma = (struct lustre_mdt_attrs *)&info->oti_buf;
987 buf.lb_len = sizeof(info->oti_buf);
988 rc = osd_xattr_get(env, &obj->oo_dt, &buf,
992 le32_to_cpu(lma->lma_incompat);
994 lustre_to_lma_flags(la->la_flags);
996 cpu_to_le32(lma->lma_incompat);
998 buf.lb_len = sizeof(*lma);
999 rc = osd_xattr_set_internal(env, obj, &buf,
1005 CWARN("%s: failed to set LMA flags: rc = %d\n",
1006 osd->od_svname, rc);
1012 /* do both accounting updates outside oo_attr_lock below */
1013 if ((valid & LA_UID) && (la->la_uid != obj->oo_attr.la_uid)) {
1014 /* Update user accounting. Failure isn't fatal, but we still
1015 * log an error message */
1016 rc = -zap_increment_int(osd->od_os, osd->od_iusr_oid,
1017 la->la_uid, 1, oh->ot_tx);
1019 CERROR("%s: failed to update accounting ZAP for user "
1020 "%d (%d)\n", osd->od_svname, la->la_uid, rc);
1021 rc = -zap_increment_int(osd->od_os, osd->od_iusr_oid,
1022 obj->oo_attr.la_uid, -1, oh->ot_tx);
1024 CERROR("%s: failed to update accounting ZAP for user "
1025 "%d (%d)\n", osd->od_svname,
1026 obj->oo_attr.la_uid, rc);
1028 if ((valid & LA_GID) && (la->la_gid != obj->oo_attr.la_gid)) {
1029 /* Update group accounting. Failure isn't fatal, but we still
1030 * log an error message */
1031 rc = -zap_increment_int(osd->od_os, osd->od_igrp_oid,
1032 la->la_gid, 1, oh->ot_tx);
1034 CERROR("%s: failed to update accounting ZAP for user "
1035 "%d (%d)\n", osd->od_svname, la->la_gid, rc);
1036 rc = -zap_increment_int(osd->od_os, osd->od_igrp_oid,
1037 obj->oo_attr.la_gid, -1, oh->ot_tx);
1039 CERROR("%s: failed to update accounting ZAP for user "
1040 "%d (%d)\n", osd->od_svname,
1041 obj->oo_attr.la_gid, rc);
1044 write_lock(&obj->oo_attr_lock);
1046 if (valid & LA_ATIME) {
1047 osa->atime[0] = obj->oo_attr.la_atime = la->la_atime;
1048 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(osd), NULL,
1051 if (valid & LA_MTIME) {
1052 osa->mtime[0] = obj->oo_attr.la_mtime = la->la_mtime;
1053 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(osd), NULL,
1056 if (valid & LA_CTIME) {
1057 osa->ctime[0] = obj->oo_attr.la_ctime = la->la_ctime;
1058 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(osd), NULL,
1061 if (valid & LA_MODE) {
1062 /* mode is stored along with type, so read it first */
1063 obj->oo_attr.la_mode = (obj->oo_attr.la_mode & S_IFMT) |
1064 (la->la_mode & ~S_IFMT);
1065 osa->mode = obj->oo_attr.la_mode;
1066 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(osd), NULL,
1069 if (valid & LA_SIZE) {
1070 osa->size = obj->oo_attr.la_size = la->la_size;
1071 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(osd), NULL,
1074 if (valid & LA_NLINK) {
1075 osa->nlink = obj->oo_attr.la_nlink = la->la_nlink;
1076 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(osd), NULL,
1079 if (valid & LA_RDEV) {
1080 osa->rdev = obj->oo_attr.la_rdev = la->la_rdev;
1081 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_RDEV(osd), NULL,
1084 if (valid & LA_FLAGS) {
1085 osa->flags = attrs_fs2zfs(la->la_flags);
1086 /* many flags are not supported by zfs, so ensure a good cached
1088 obj->oo_attr.la_flags = attrs_zfs2fs(osa->flags);
1089 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(osd), NULL,
1092 if (valid & LA_UID) {
1093 osa->uid = obj->oo_attr.la_uid = la->la_uid;
1094 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(osd), NULL,
1097 if (valid & LA_GID) {
1098 osa->gid = obj->oo_attr.la_gid = la->la_gid;
1099 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(osd), NULL,
1102 obj->oo_attr.la_valid |= valid;
1103 write_unlock(&obj->oo_attr_lock);
1105 LASSERT(cnt <= ARRAY_SIZE(osd_oti_get(env)->oti_attr_bulk));
1106 rc = osd_object_sa_bulk_update(obj, bulk, cnt, oh);
1109 up_read(&obj->oo_guard);
1116 * XXX temporary solution.
1119 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
1120 struct dt_object *parent, struct dt_object *child,
1125 ah->dah_parent = parent;
1126 ah->dah_mode = child_mode;
1129 static int osd_declare_object_create(const struct lu_env *env,
1130 struct dt_object *dt,
1131 struct lu_attr *attr,
1132 struct dt_allocation_hint *hint,
1133 struct dt_object_format *dof,
1134 struct thandle *handle)
1136 char *buf = osd_oti_get(env)->oti_str;
1137 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1138 struct osd_object *obj = osd_dt_obj(dt);
1139 struct osd_device *osd = osd_obj2dev(obj);
1140 struct osd_thandle *oh;
1147 switch (dof->dof_type) {
1151 if (obj->oo_dt.do_body_ops == NULL)
1152 obj->oo_dt.do_body_ops = &osd_body_ops;
1158 LASSERT(handle != NULL);
1159 oh = container_of0(handle, struct osd_thandle, ot_super);
1160 LASSERT(oh->ot_tx != NULL);
1162 switch (dof->dof_type) {
1164 dt->do_index_ops = &osd_dir_ops;
1166 /* for zap create */
1167 dmu_tx_hold_zap(oh->ot_tx, DMU_NEW_OBJECT, 1, NULL);
1172 /* first, we'll create new object */
1173 dmu_tx_hold_bonus(oh->ot_tx, DMU_NEW_OBJECT);
1181 /* and we'll add it to some mapping */
1182 zapid = osd_get_name_n_idx(env, osd, fid, buf);
1183 dmu_tx_hold_bonus(oh->ot_tx, zapid);
1184 dmu_tx_hold_zap(oh->ot_tx, zapid, TRUE, buf);
1186 /* we will also update inode accounting ZAPs */
1187 dmu_tx_hold_bonus(oh->ot_tx, osd->od_iusr_oid);
1188 dmu_tx_hold_zap(oh->ot_tx, osd->od_iusr_oid, TRUE, buf);
1189 dmu_tx_hold_bonus(oh->ot_tx, osd->od_igrp_oid);
1190 dmu_tx_hold_zap(oh->ot_tx, osd->od_igrp_oid, TRUE, buf);
1192 dmu_tx_hold_sa_create(oh->ot_tx, ZFS_SA_BASE_ATTR_SIZE);
1194 __osd_xattr_declare_set(env, obj, sizeof(struct lustre_mdt_attrs),
1195 XATTR_NAME_LMA, oh);
1197 rc = osd_declare_quota(env, osd, attr->la_uid, attr->la_gid, 1, oh,
1198 false, NULL, false);
1202 int __osd_attr_init(const struct lu_env *env, struct osd_device *osd,
1203 uint64_t oid, dmu_tx_t *tx, struct lu_attr *la,
1206 sa_handle_t *sa_hdl;
1207 sa_bulk_attr_t *bulk = osd_oti_get(env)->oti_attr_bulk;
1208 struct osa_attr *osa = &osd_oti_get(env)->oti_osa;
1216 gen = dmu_tx_get_txg(tx);
1218 ZFS_TIME_ENCODE(&now, crtime);
1220 osa->atime[0] = la->la_atime;
1221 osa->ctime[0] = la->la_ctime;
1222 osa->mtime[0] = la->la_mtime;
1223 osa->mode = la->la_mode;
1224 osa->uid = la->la_uid;
1225 osa->gid = la->la_gid;
1226 osa->rdev = la->la_rdev;
1227 osa->nlink = la->la_nlink;
1228 osa->flags = attrs_fs2zfs(la->la_flags);
1229 osa->size = la->la_size;
1231 /* Now add in all of the "SA" attributes */
1232 rc = -sa_handle_get(osd->od_os, oid, NULL, SA_HDL_PRIVATE, &sa_hdl);
1237 * we need to create all SA below upon object create.
1239 * XXX The attribute order matters since the accounting callback relies
1240 * on static offsets (i.e. SA_*_OFFSET, see zfs_space_delta_cb()) to
1241 * look up the UID/GID attributes. Moreover, the callback does not seem
1242 * to support the spill block.
1243 * We define attributes in the same order as SA_*_OFFSET in order to
1244 * work around the problem. See ORI-610.
1247 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(osd), NULL, &osa->mode, 8);
1248 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(osd), NULL, &osa->size, 8);
1249 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GEN(osd), NULL, &gen, 8);
1250 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(osd), NULL, &osa->uid, 8);
1251 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(osd), NULL, &osa->gid, 8);
1252 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_PARENT(osd), NULL, &parent, 8);
1253 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(osd), NULL, &osa->flags, 8);
1254 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(osd), NULL, osa->atime, 16);
1255 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(osd), NULL, osa->mtime, 16);
1256 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(osd), NULL, osa->ctime, 16);
1257 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CRTIME(osd), NULL, crtime, 16);
1258 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(osd), NULL, &osa->nlink, 8);
1259 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_RDEV(osd), NULL, &osa->rdev, 8);
1260 LASSERT(cnt <= ARRAY_SIZE(osd_oti_get(env)->oti_attr_bulk));
1262 rc = -sa_replace_all_by_template(sa_hdl, bulk, cnt, tx);
1264 sa_handle_destroy(sa_hdl);
1269 * The transaction passed to this routine must have
1270 * dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT) called and then assigned
1271 * to a transaction group.
1273 int __osd_object_create(const struct lu_env *env, struct osd_object *obj,
1274 dmu_buf_t **dbp, dmu_tx_t *tx, struct lu_attr *la,
1279 struct osd_device *osd = osd_obj2dev(obj);
1280 const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
1281 dmu_object_type_t type = DMU_OT_PLAIN_FILE_CONTENTS;
1283 /* Assert that the transaction has been assigned to a
1284 transaction group. */
1285 LASSERT(tx->tx_txg != 0);
1287 /* Use DMU_OTN_UINT8_METADATA for local objects so their data blocks
1288 * would get an additional ditto copy */
1289 if (unlikely(S_ISREG(la->la_mode) &&
1290 fid_seq_is_local_file(fid_seq(fid))))
1291 type = DMU_OTN_UINT8_METADATA;
1293 /* Create a new DMU object using the default dnode size. */
1294 oid = osd_dmu_object_alloc(osd->od_os, type, 0, 0, tx);
1295 rc = -sa_buf_hold(osd->od_os, oid, osd_obj_tag, dbp);
1296 LASSERTF(rc == 0, "sa_buf_hold "LPU64" failed: %d\n", oid, rc);
1298 LASSERT(la->la_valid & LA_MODE);
1302 rc = __osd_attr_init(env, osd, oid, tx, la, parent);
1304 sa_buf_rele(*dbp, osd_obj_tag);
1306 dmu_object_free(osd->od_os, oid, tx);
1314 * The transaction passed to this routine must have
1315 * dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, ...) called and then assigned
1316 * to a transaction group.
1318 * Using ZAP_FLAG_HASH64 will force the ZAP to always be a FAT ZAP.
1319 * This is fine for directories today, because storing the FID in the dirent
1320 * will also require a FAT ZAP. If there is a new type of micro ZAP created
1321 * then we might need to re-evaluate the use of this flag and instead do
1322 * a conversion from the different internal ZAP hash formats being used. */
1323 int __osd_zap_create(const struct lu_env *env, struct osd_device *osd,
1324 dmu_buf_t **zap_dbp, dmu_tx_t *tx,
1325 struct lu_attr *la, uint64_t parent, zap_flags_t flags)
1330 /* Assert that the transaction has been assigned to a
1331 transaction group. */
1332 LASSERT(tx->tx_txg != 0);
1334 oid = osd_zap_create_flags(osd->od_os, 0, flags | ZAP_FLAG_HASH64,
1335 DMU_OT_DIRECTORY_CONTENTS,
1336 14, /* == ZFS fzap_default_blockshift */
1337 DN_MAX_INDBLKSHIFT, /* indirect blockshift */
1340 rc = -sa_buf_hold(osd->od_os, oid, osd_obj_tag, zap_dbp);
1344 LASSERT(la->la_valid & LA_MODE);
1348 return __osd_attr_init(env, osd, oid, tx, la, parent);
1351 static dmu_buf_t *osd_mkidx(const struct lu_env *env, struct osd_object *obj,
1352 struct lu_attr *la, uint64_t parent,
1353 struct osd_thandle *oh)
1358 /* Index file should be created as regular file in order not to confuse
1359 * ZPL which could interpret them as directory.
1360 * We set ZAP_FLAG_UINT64_KEY to let ZFS know than we are going to use
1362 LASSERT(S_ISREG(la->la_mode));
1363 rc = __osd_zap_create(env, osd_obj2dev(obj), &db, oh->ot_tx, la, parent,
1364 ZAP_FLAG_UINT64_KEY);
1370 static dmu_buf_t *osd_mkdir(const struct lu_env *env, struct osd_object *obj,
1371 struct lu_attr *la, uint64_t parent,
1372 struct osd_thandle *oh)
1377 LASSERT(S_ISDIR(la->la_mode));
1378 rc = __osd_zap_create(env, osd_obj2dev(obj), &db,
1379 oh->ot_tx, la, parent, 0);
1385 static dmu_buf_t *osd_mkreg(const struct lu_env *env, struct osd_object *obj,
1386 struct lu_attr *la, uint64_t parent,
1387 struct osd_thandle *oh)
1389 const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
1392 struct osd_device *osd = osd_obj2dev(obj);
1394 LASSERT(S_ISREG(la->la_mode));
1395 rc = __osd_object_create(env, obj, &db, oh->ot_tx, la, parent);
1400 * XXX: This heuristic is non-optimal. It would be better to
1401 * increase the blocksize up to osd->od_max_blksz during the write.
1402 * This is exactly how the ZPL behaves and it ensures that the right
1403 * blocksize is selected based on the file size rather than the
1404 * making broad assumptions based on the osd type.
1406 if ((fid_is_idif(fid) || fid_is_norm(fid)) && osd->od_is_ost) {
1407 rc = -dmu_object_set_blocksize(osd->od_os, db->db_object,
1408 osd->od_max_blksz, 0, oh->ot_tx);
1410 CERROR("%s: can't change blocksize: %d\n",
1411 osd->od_svname, rc);
1419 static dmu_buf_t *osd_mksym(const struct lu_env *env, struct osd_object *obj,
1420 struct lu_attr *la, uint64_t parent,
1421 struct osd_thandle *oh)
1426 LASSERT(S_ISLNK(la->la_mode));
1427 rc = __osd_object_create(env, obj, &db, oh->ot_tx, la, parent);
1433 static dmu_buf_t *osd_mknod(const struct lu_env *env, struct osd_object *obj,
1434 struct lu_attr *la, uint64_t parent,
1435 struct osd_thandle *oh)
1440 la->la_valid = LA_MODE;
1441 if (S_ISCHR(la->la_mode) || S_ISBLK(la->la_mode))
1442 la->la_valid |= LA_RDEV;
1444 rc = __osd_object_create(env, obj, &db, oh->ot_tx, la, parent);
1450 typedef dmu_buf_t *(*osd_obj_type_f)(const struct lu_env *env,
1451 struct osd_object *obj,
1454 struct osd_thandle *oh);
1456 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
1458 osd_obj_type_f result;
1484 * Primitives for directory (i.e. ZAP) handling
1486 static inline int osd_init_lma(const struct lu_env *env, struct osd_object *obj,
1487 const struct lu_fid *fid, struct osd_thandle *oh)
1489 struct osd_thread_info *info = osd_oti_get(env);
1490 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
1494 lustre_lma_init(lma, fid, 0, 0);
1495 lustre_lma_swab(lma);
1497 buf.lb_len = sizeof(*lma);
1499 rc = osd_xattr_set_internal(env, obj, &buf, XATTR_NAME_LMA,
1500 LU_XATTR_CREATE, oh);
1506 * Concurrency: @dt is write locked.
1508 static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
1509 struct lu_attr *attr,
1510 struct dt_allocation_hint *hint,
1511 struct dt_object_format *dof,
1514 struct zpl_direntry *zde = &osd_oti_get(env)->oti_zde.lzd_reg;
1515 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1516 struct osd_object *obj = osd_dt_obj(dt);
1517 struct osd_device *osd = osd_obj2dev(obj);
1518 char *buf = osd_oti_get(env)->oti_str;
1519 struct osd_thandle *oh;
1526 /* concurrent create declarations should not see
1527 * the object inconsistent (db, attr, etc).
1528 * in regular cases acquisition should be cheap */
1529 down_write(&obj->oo_guard);
1531 if (unlikely(dt_object_exists(dt)))
1532 GOTO(out, rc = -EEXIST);
1534 LASSERT(osd_invariant(obj));
1535 LASSERT(dof != NULL);
1537 LASSERT(th != NULL);
1538 oh = container_of0(th, struct osd_thandle, ot_super);
1541 * XXX missing: Quote handling.
1544 LASSERT(obj->oo_db == NULL);
1546 /* to follow ZFS on-disk format we need
1547 * to initialize parent dnode properly */
1549 if (hint != NULL && hint->dah_parent != NULL &&
1550 !dt_object_remote(hint->dah_parent))
1551 zapid = osd_dt_obj(hint->dah_parent)->oo_db->db_object;
1553 db = osd_create_type_f(dof->dof_type)(env, obj, attr, zapid, oh);
1555 GOTO(out, rc = PTR_ERR(db));
1558 zde->zde_dnode = db->db_object;
1559 zde->zde_type = IFTODT(attr->la_mode & S_IFMT);
1561 zapid = osd_get_name_n_idx(env, osd, fid, buf);
1563 rc = -zap_add(osd->od_os, zapid, buf, 8, 1, zde, oh->ot_tx);
1567 /* Add new object to inode accounting.
1568 * Errors are not considered as fatal */
1569 rc = -zap_increment_int(osd->od_os, osd->od_iusr_oid,
1570 (attr->la_valid & LA_UID) ? attr->la_uid : 0, 1,
1573 CERROR("%s: failed to add "DFID" to accounting ZAP for usr %d "
1574 "(%d)\n", osd->od_svname, PFID(fid), attr->la_uid, rc);
1575 rc = -zap_increment_int(osd->od_os, osd->od_igrp_oid,
1576 (attr->la_valid & LA_GID) ? attr->la_gid : 0, 1,
1579 CERROR("%s: failed to add "DFID" to accounting ZAP for grp %d "
1580 "(%d)\n", osd->od_svname, PFID(fid), attr->la_gid, rc);
1582 /* configure new osd object */
1584 rc = osd_object_init0(env, obj);
1585 LASSERT(ergo(rc == 0, dt_object_exists(dt)));
1586 LASSERT(osd_invariant(obj));
1588 rc = osd_init_lma(env, obj, fid, oh);
1590 CERROR("%s: can not set LMA on "DFID": rc = %d\n",
1591 osd->od_svname, PFID(fid), rc);
1594 up_write(&obj->oo_guard);
1598 static int osd_declare_object_ref_add(const struct lu_env *env,
1599 struct dt_object *dt,
1602 return osd_declare_attr_set(env, dt, NULL, th);
1606 * Concurrency: @dt is write locked.
1608 static int osd_object_ref_add(const struct lu_env *env,
1609 struct dt_object *dt,
1610 struct thandle *handle)
1612 struct osd_object *obj = osd_dt_obj(dt);
1613 struct osd_thandle *oh;
1614 struct osd_device *osd = osd_obj2dev(obj);
1620 down_read(&obj->oo_guard);
1621 if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
1622 GOTO(out, rc = -ENOENT);
1624 LASSERT(osd_invariant(obj));
1625 LASSERT(obj->oo_sa_hdl != NULL);
1627 oh = container_of0(handle, struct osd_thandle, ot_super);
1629 write_lock(&obj->oo_attr_lock);
1630 nlink = ++obj->oo_attr.la_nlink;
1631 write_unlock(&obj->oo_attr_lock);
1633 rc = osd_object_sa_update(obj, SA_ZPL_LINKS(osd), &nlink, 8, oh);
1636 up_read(&obj->oo_guard);
1640 static int osd_declare_object_ref_del(const struct lu_env *env,
1641 struct dt_object *dt,
1642 struct thandle *handle)
1644 return osd_declare_attr_set(env, dt, NULL, handle);
1648 * Concurrency: @dt is write locked.
1650 static int osd_object_ref_del(const struct lu_env *env,
1651 struct dt_object *dt,
1652 struct thandle *handle)
1654 struct osd_object *obj = osd_dt_obj(dt);
1655 struct osd_thandle *oh;
1656 struct osd_device *osd = osd_obj2dev(obj);
1662 down_read(&obj->oo_guard);
1664 if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
1665 GOTO(out, rc = -ENOENT);
1667 LASSERT(osd_invariant(obj));
1668 LASSERT(obj->oo_sa_hdl != NULL);
1670 oh = container_of0(handle, struct osd_thandle, ot_super);
1671 LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
1673 write_lock(&obj->oo_attr_lock);
1674 nlink = --obj->oo_attr.la_nlink;
1675 write_unlock(&obj->oo_attr_lock);
1677 rc = osd_object_sa_update(obj, SA_ZPL_LINKS(osd), &nlink, 8, oh);
1680 up_read(&obj->oo_guard);
1684 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
1685 __u64 start, __u64 end)
1687 struct osd_device *osd = osd_obj2dev(osd_dt_obj(dt));
1690 /* XXX: no other option than syncing the whole filesystem until we
1691 * support ZIL. If the object tracked the txg that it was last
1692 * modified in, it could pass that txg here instead of "0". Maybe
1693 * the changes are already committed, so no wait is needed at all? */
1694 txg_wait_synced(dmu_objset_pool(osd->od_os), 0ULL);
1699 static int osd_invalidate(const struct lu_env *env, struct dt_object *dt)
1704 static struct dt_object_operations osd_obj_ops = {
1705 .do_read_lock = osd_object_read_lock,
1706 .do_write_lock = osd_object_write_lock,
1707 .do_read_unlock = osd_object_read_unlock,
1708 .do_write_unlock = osd_object_write_unlock,
1709 .do_write_locked = osd_object_write_locked,
1710 .do_attr_get = osd_attr_get,
1711 .do_declare_attr_set = osd_declare_attr_set,
1712 .do_attr_set = osd_attr_set,
1713 .do_ah_init = osd_ah_init,
1714 .do_declare_create = osd_declare_object_create,
1715 .do_create = osd_object_create,
1716 .do_declare_destroy = osd_declare_object_destroy,
1717 .do_destroy = osd_object_destroy,
1718 .do_index_try = osd_index_try,
1719 .do_declare_ref_add = osd_declare_object_ref_add,
1720 .do_ref_add = osd_object_ref_add,
1721 .do_declare_ref_del = osd_declare_object_ref_del,
1722 .do_ref_del = osd_object_ref_del,
1723 .do_xattr_get = osd_xattr_get,
1724 .do_declare_xattr_set = osd_declare_xattr_set,
1725 .do_xattr_set = osd_xattr_set,
1726 .do_declare_xattr_del = osd_declare_xattr_del,
1727 .do_xattr_del = osd_xattr_del,
1728 .do_xattr_list = osd_xattr_list,
1729 .do_object_sync = osd_object_sync,
1730 .do_invalidate = osd_invalidate,
1733 static struct lu_object_operations osd_lu_obj_ops = {
1734 .loo_object_init = osd_object_init,
1735 .loo_object_delete = osd_object_delete,
1736 .loo_object_release = osd_object_release,
1737 .loo_object_free = osd_object_free,
1738 .loo_object_print = osd_object_print,
1739 .loo_object_invariant = osd_object_invariant,
1742 static int osd_otable_it_attr_get(const struct lu_env *env,
1743 struct dt_object *dt,
1744 struct lu_attr *attr)
1750 static struct dt_object_operations osd_obj_otable_it_ops = {
1751 .do_attr_get = osd_otable_it_attr_get,
1752 .do_index_try = osd_index_try,