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.
116 osd_object_sa_dirty_add(struct osd_object *obj, struct osd_thandle *oh)
118 if (!list_empty(&obj->oo_sa_linkage))
121 down(&oh->ot_sa_lock);
122 write_lock(&obj->oo_attr_lock);
123 if (likely(list_empty(&obj->oo_sa_linkage)))
124 list_add(&obj->oo_sa_linkage, &oh->ot_sa_list);
125 write_unlock(&obj->oo_attr_lock);
130 * Release spill block dbuf hold for all dirty SAs.
132 void osd_object_sa_dirty_rele(struct osd_thandle *oh)
134 struct osd_object *obj;
136 down(&oh->ot_sa_lock);
137 while (!list_empty(&oh->ot_sa_list)) {
138 obj = list_entry(oh->ot_sa_list.next,
139 struct osd_object, oo_sa_linkage);
140 sa_spill_rele(obj->oo_sa_hdl);
141 write_lock(&obj->oo_attr_lock);
142 list_del_init(&obj->oo_sa_linkage);
143 write_unlock(&obj->oo_attr_lock);
149 * Update the SA and add the object to the dirty list.
151 int osd_object_sa_update(struct osd_object *obj, sa_attr_type_t type,
152 void *buf, uint32_t buflen, struct osd_thandle *oh)
156 LASSERT(obj->oo_sa_hdl != NULL);
157 LASSERT(oh->ot_tx != NULL);
159 rc = -sa_update(obj->oo_sa_hdl, type, buf, buflen, oh->ot_tx);
160 osd_object_sa_dirty_add(obj, oh);
166 * Bulk update the SA and add the object to the dirty list.
169 osd_object_sa_bulk_update(struct osd_object *obj, sa_bulk_attr_t *attrs,
170 int count, struct osd_thandle *oh)
174 LASSERT(obj->oo_sa_hdl != NULL);
175 LASSERT(oh->ot_tx != NULL);
177 rc = -sa_bulk_update(obj->oo_sa_hdl, attrs, count, oh->ot_tx);
178 osd_object_sa_dirty_add(obj, oh);
184 * Retrieve the attributes of a DMU object
186 int __osd_object_attr_get(const struct lu_env *env, struct osd_device *o,
187 struct osd_object *obj, struct lu_attr *la)
189 struct osa_attr *osa = &osd_oti_get(env)->oti_osa;
191 sa_bulk_attr_t *bulk;
196 LASSERT(obj->oo_db != NULL);
198 rc = -sa_handle_get(o->od_os, obj->oo_db->db_object, NULL,
199 SA_HDL_PRIVATE, &sa_hdl);
203 OBD_ALLOC(bulk, sizeof(sa_bulk_attr_t) * 9);
205 GOTO(out_sa, rc = -ENOMEM);
207 la->la_valid |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE | LA_TYPE |
208 LA_SIZE | LA_UID | LA_GID | LA_FLAGS | LA_NLINK;
210 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(o), NULL, osa->atime, 16);
211 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(o), NULL, osa->mtime, 16);
212 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(o), NULL, osa->ctime, 16);
213 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(o), NULL, &osa->mode, 8);
214 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(o), NULL, &osa->size, 8);
215 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(o), NULL, &osa->nlink, 8);
216 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(o), NULL, &osa->uid, 8);
217 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(o), NULL, &osa->gid, 8);
218 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(o), NULL, &osa->flags, 8);
220 rc = -sa_bulk_lookup(sa_hdl, bulk, cnt);
224 la->la_atime = osa->atime[0];
225 la->la_mtime = osa->mtime[0];
226 la->la_ctime = osa->ctime[0];
227 la->la_mode = osa->mode;
228 la->la_uid = osa->uid;
229 la->la_gid = osa->gid;
230 la->la_nlink = osa->nlink;
231 la->la_flags = attrs_zfs2fs(osa->flags);
232 la->la_size = osa->size;
234 /* Try to get extra flag from LMA. Right now, only LMAI_ORPHAN
235 * flags is stored in LMA, and it is only for orphan directory */
236 if (S_ISDIR(la->la_mode) && dt_object_exists(&obj->oo_dt)) {
237 struct osd_thread_info *info = osd_oti_get(env);
238 struct lustre_mdt_attrs *lma;
241 lma = (struct lustre_mdt_attrs *)info->oti_buf;
243 buf.lb_len = sizeof(info->oti_buf);
244 rc = osd_xattr_get(env, &obj->oo_dt, &buf, XATTR_NAME_LMA);
247 lma->lma_incompat = le32_to_cpu(lma->lma_incompat);
249 lma_to_lustre_flags(lma->lma_incompat);
251 } else if (rc == -ENODATA) {
256 if (S_ISCHR(la->la_mode) || S_ISBLK(la->la_mode)) {
257 rc = -sa_lookup(sa_hdl, SA_ZPL_RDEV(o), &osa->rdev, 8);
260 la->la_rdev = osa->rdev;
261 la->la_valid |= LA_RDEV;
264 OBD_FREE(bulk, sizeof(sa_bulk_attr_t) * 9);
266 sa_handle_destroy(sa_hdl);
271 int __osd_obj2dbuf(const struct lu_env *env, objset_t *os,
272 uint64_t oid, dmu_buf_t **dbp)
274 dmu_object_info_t *doi = &osd_oti_get(env)->oti_doi;
277 rc = -sa_buf_hold(os, oid, osd_obj_tag, dbp);
281 dmu_object_info_from_db(*dbp, doi);
282 if (unlikely (oid != DMU_USERUSED_OBJECT &&
283 oid != DMU_GROUPUSED_OBJECT && doi->doi_bonus_type != DMU_OT_SA)) {
284 sa_buf_rele(*dbp, osd_obj_tag);
290 LASSERT((*dbp)->db_object == oid);
291 LASSERT((*dbp)->db_offset == -1);
292 LASSERT((*dbp)->db_data != NULL);
298 * Concurrency: no concurrent access is possible that early in object
301 struct lu_object *osd_object_alloc(const struct lu_env *env,
302 const struct lu_object_header *hdr,
305 struct osd_object *mo;
307 OBD_SLAB_ALLOC_PTR_GFP(mo, osd_object_kmem, GFP_NOFS);
311 l = &mo->oo_dt.do_lu;
312 dt_object_init(&mo->oo_dt, NULL, d);
313 mo->oo_dt.do_ops = &osd_obj_ops;
314 l->lo_ops = &osd_lu_obj_ops;
315 INIT_LIST_HEAD(&mo->oo_sa_linkage);
316 INIT_LIST_HEAD(&mo->oo_unlinked_linkage);
317 init_rwsem(&mo->oo_sem);
318 init_rwsem(&mo->oo_guard);
319 rwlock_init(&mo->oo_attr_lock);
320 mo->oo_destroy = OSD_DESTROY_NONE;
328 * Concurrency: shouldn't matter.
330 int osd_object_init0(const struct lu_env *env, struct osd_object *obj)
332 struct osd_device *osd = osd_obj2dev(obj);
333 const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
337 if (obj->oo_db == NULL)
342 rc = osd_object_sa_init(obj, osd);
346 /* cache attrs in object */
347 rc = __osd_object_attr_get(env, osd, obj, &obj->oo_attr);
351 if (likely(!fid_is_acct(fid)))
352 /* no body operations for accounting objects */
353 obj->oo_dt.do_body_ops = &osd_body_ops;
356 * initialize object before marking it existing
358 obj->oo_dt.do_lu.lo_header->loh_attr |= obj->oo_attr.la_mode & S_IFMT;
361 obj->oo_dt.do_lu.lo_header->loh_attr |= LOHA_EXISTS;
366 static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
368 struct osd_thread_info *info = osd_oti_get(env);
371 struct lustre_mdt_attrs *lma;
374 CLASSERT(sizeof(info->oti_buf) >= sizeof(*lma));
375 lma = (struct lustre_mdt_attrs *)info->oti_buf;
377 buf.lb_len = sizeof(info->oti_buf);
379 rc = osd_xattr_get(env, &obj->oo_dt, &buf, XATTR_NAME_LMA);
382 lustre_lma_swab(lma);
383 if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
384 CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT))) {
385 CWARN("%s: unsupported incompat LMA feature(s) %#x for "
386 "fid = "DFID"\n", osd_obj2dev(obj)->od_svname,
387 lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
388 PFID(lu_object_fid(&obj->oo_dt.do_lu)));
391 } else if (rc == -ENODATA) {
392 /* haven't initialize LMA xattr */
400 * Concurrency: no concurrent access is possible that early in object
403 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
404 const struct lu_object_conf *conf)
406 struct osd_object *obj = osd_obj(l);
407 struct osd_device *osd = osd_obj2dev(obj);
412 LASSERT(osd_invariant(obj));
414 if (fid_is_otable_it(&l->lo_header->loh_fid)) {
415 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
416 l->lo_header->loh_attr |= LOHA_EXISTS;
420 rc = osd_fid_lookup(env, osd, lu_object_fid(l), &oid);
422 LASSERT(obj->oo_db == NULL);
423 rc = __osd_obj2dbuf(env, osd->od_os, oid, &obj->oo_db);
424 /* EEXIST will be returned if object is being deleted in ZFS */
430 CERROR("%s: lookup "DFID"/"LPX64" failed: rc = %d\n",
431 osd->od_svname, PFID(lu_object_fid(l)), oid, rc);
435 rc = osd_object_init0(env, obj);
439 rc = osd_check_lma(env, obj);
442 } else if (rc == -ENOENT) {
445 LASSERT(osd_invariant(obj));
451 * Concurrency: no concurrent access is possible that late in object
454 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
456 struct osd_object *obj = osd_obj(l);
458 LASSERT(osd_invariant(obj));
460 dt_object_fini(&obj->oo_dt);
461 OBD_SLAB_FREE_PTR(obj, osd_object_kmem);
465 osd_object_unlinked_add(struct osd_object *obj, struct osd_thandle *oh)
469 LASSERT(obj->oo_destroy == OSD_DESTROY_ASYNC);
471 /* the object is supposed to be exclusively locked by
472 * the caller (osd_object_destroy()), while the transaction
473 * (oh) is per-thread and not shared */
474 if (likely(list_empty(&obj->oo_unlinked_linkage))) {
475 list_add(&obj->oo_unlinked_linkage, &oh->ot_unlinked_list);
482 /* Default to max data size covered by a level-1 indirect block */
483 static unsigned long osd_sync_destroy_max_size =
484 1UL << (DN_MAX_INDBLKSHIFT - SPA_BLKPTRSHIFT + SPA_MAXBLOCKSHIFT);
485 CFS_MODULE_PARM(osd_sync_destroy_max_size, "ul", ulong, 0444,
486 "Maximum object size to use synchronous destroy.");
489 osd_object_set_destroy_type(struct osd_object *obj)
492 * Lock-less OST_WRITE can race with OST_DESTROY, so set destroy type
493 * only once and use it consistently thereafter.
495 down_write(&obj->oo_guard);
496 if (obj->oo_destroy == OSD_DESTROY_NONE) {
497 if (obj->oo_attr.la_size <= osd_sync_destroy_max_size)
498 obj->oo_destroy = OSD_DESTROY_SYNC;
499 else /* Larger objects are destroyed asynchronously */
500 obj->oo_destroy = OSD_DESTROY_ASYNC;
502 up_write(&obj->oo_guard);
505 static int osd_declare_object_destroy(const struct lu_env *env,
506 struct dt_object *dt,
509 char *buf = osd_oti_get(env)->oti_str;
510 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
511 struct osd_object *obj = osd_dt_obj(dt);
512 struct osd_device *osd = osd_obj2dev(obj);
513 struct osd_thandle *oh;
519 LASSERT(dt_object_exists(dt));
521 oh = container_of0(th, struct osd_thandle, ot_super);
522 LASSERT(oh->ot_tx != NULL);
524 /* declare that we'll remove object from fid-dnode mapping */
525 zapid = osd_get_name_n_idx(env, osd, fid, buf);
526 dmu_tx_hold_bonus(oh->ot_tx, zapid);
527 dmu_tx_hold_zap(oh->ot_tx, zapid, FALSE, buf);
529 osd_declare_xattrs_destroy(env, obj, oh);
531 /* declare that we'll remove object from inode accounting ZAPs */
532 dmu_tx_hold_bonus(oh->ot_tx, osd->od_iusr_oid);
533 dmu_tx_hold_zap(oh->ot_tx, osd->od_iusr_oid, FALSE, buf);
534 dmu_tx_hold_bonus(oh->ot_tx, osd->od_igrp_oid);
535 dmu_tx_hold_zap(oh->ot_tx, osd->od_igrp_oid, FALSE, buf);
538 rc = osd_declare_quota(env, osd, obj->oo_attr.la_uid,
539 obj->oo_attr.la_gid, -1, oh, false, NULL, false);
543 /* data to be truncated */
544 rc = osd_declare_quota(env, osd, obj->oo_attr.la_uid,
545 obj->oo_attr.la_gid, 0, oh, true, NULL, false);
549 osd_object_set_destroy_type(obj);
550 if (obj->oo_destroy == OSD_DESTROY_SYNC)
551 dmu_tx_hold_free(oh->ot_tx, obj->oo_db->db_object,
554 dmu_tx_hold_zap(oh->ot_tx, osd->od_unlinkedid, TRUE, NULL);
559 static int osd_object_destroy(const struct lu_env *env,
560 struct dt_object *dt, struct thandle *th)
562 char *buf = osd_oti_get(env)->oti_str;
563 struct osd_object *obj = osd_dt_obj(dt);
564 struct osd_device *osd = osd_obj2dev(obj);
565 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
566 struct osd_thandle *oh;
571 down_write(&obj->oo_guard);
573 if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
574 GOTO(out, rc = -ENOENT);
576 LASSERT(obj->oo_db != NULL);
578 oh = container_of0(th, struct osd_thandle, ot_super);
580 LASSERT(oh->ot_tx != NULL);
582 /* remove obj ref from index dir (it depends) */
583 zapid = osd_get_name_n_idx(env, osd, fid, buf);
584 rc = -zap_remove(osd->od_os, zapid, buf, oh->ot_tx);
586 CERROR("%s: zap_remove(%s) failed: rc = %d\n",
587 osd->od_svname, buf, rc);
591 rc = osd_xattrs_destroy(env, obj, oh);
593 CERROR("%s: cannot destroy xattrs for %s: rc = %d\n",
594 osd->od_svname, buf, rc);
598 /* Remove object from inode accounting. It is not fatal for the destroy
599 * operation if something goes wrong while updating accounting, but we
600 * still log an error message to notify the administrator */
601 rc = -zap_increment_int(osd->od_os, osd->od_iusr_oid,
602 obj->oo_attr.la_uid, -1, oh->ot_tx);
604 CERROR("%s: failed to remove "DFID" from accounting ZAP for usr"
605 " %d: rc = %d\n", osd->od_svname, PFID(fid),
606 obj->oo_attr.la_uid, rc);
607 rc = -zap_increment_int(osd->od_os, osd->od_igrp_oid,
608 obj->oo_attr.la_gid, -1, oh->ot_tx);
610 CERROR("%s: failed to remove "DFID" from accounting ZAP for grp"
611 " %d: rc = %d\n", osd->od_svname, PFID(fid),
612 obj->oo_attr.la_gid, rc);
614 oid = obj->oo_db->db_object;
615 if (unlikely(obj->oo_destroy == OSD_DESTROY_NONE)) {
616 /* this may happen if the destroy wasn't declared
617 * e.g. when the object is created and then destroyed
618 * in the same transaction - we don't need additional
619 * space for destroy specifically */
620 LASSERT(obj->oo_attr.la_size <= osd_sync_destroy_max_size);
621 rc = -dmu_object_free(osd->od_os, oid, oh->ot_tx);
623 CERROR("%s: failed to free %s "LPU64": rc = %d\n",
624 osd->od_svname, buf, oid, rc);
625 } else if (obj->oo_destroy == OSD_DESTROY_SYNC) {
626 rc = -dmu_object_free(osd->od_os, oid, oh->ot_tx);
628 CERROR("%s: failed to free %s "LPU64": rc = %d\n",
629 osd->od_svname, buf, oid, rc);
630 } else { /* asynchronous destroy */
631 rc = osd_object_unlinked_add(obj, oh);
635 rc = -zap_add_int(osd->od_os, osd->od_unlinkedid,
638 CERROR("%s: zap_add_int() failed %s "LPU64": rc = %d\n",
639 osd->od_svname, buf, oid, rc);
643 /* not needed in the cache anymore */
644 set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
646 obj->oo_destroyed = 1;
647 up_write(&obj->oo_guard);
651 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
653 struct osd_object *obj = osd_obj(l);
655 if (obj->oo_db != NULL) {
656 osd_object_sa_fini(obj);
657 if (obj->oo_sa_xattr) {
658 nvlist_free(obj->oo_sa_xattr);
659 obj->oo_sa_xattr = NULL;
661 sa_buf_rele(obj->oo_db, osd_obj_tag);
662 list_del(&obj->oo_sa_linkage);
668 * Concurrency: ->loo_object_release() is called under site spin-lock.
670 static void osd_object_release(const struct lu_env *env,
676 * Concurrency: shouldn't matter.
678 static int osd_object_print(const struct lu_env *env, void *cookie,
679 lu_printer_t p, const struct lu_object *l)
681 struct osd_object *o = osd_obj(l);
683 return (*p)(env, cookie, LUSTRE_OSD_ZFS_NAME"-object@%p", o);
686 static void osd_object_read_lock(const struct lu_env *env,
687 struct dt_object *dt, unsigned role)
689 struct osd_object *obj = osd_dt_obj(dt);
691 LASSERT(osd_invariant(obj));
693 down_read_nested(&obj->oo_sem, role);
696 static void osd_object_write_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_write_nested(&obj->oo_sem, role);
706 static void osd_object_read_unlock(const struct lu_env *env,
707 struct dt_object *dt)
709 struct osd_object *obj = osd_dt_obj(dt);
711 LASSERT(osd_invariant(obj));
712 up_read(&obj->oo_sem);
715 static void osd_object_write_unlock(const struct lu_env *env,
716 struct dt_object *dt)
718 struct osd_object *obj = osd_dt_obj(dt);
720 LASSERT(osd_invariant(obj));
721 up_write(&obj->oo_sem);
724 static int osd_object_write_locked(const struct lu_env *env,
725 struct dt_object *dt)
727 struct osd_object *obj = osd_dt_obj(dt);
730 LASSERT(osd_invariant(obj));
732 if (down_write_trylock(&obj->oo_sem)) {
734 up_write(&obj->oo_sem);
739 static int osd_attr_get(const struct lu_env *env,
740 struct dt_object *dt,
741 struct lu_attr *attr)
743 struct osd_object *obj = osd_dt_obj(dt);
748 down_read(&obj->oo_guard);
750 if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
751 GOTO(out, rc = -ENOENT);
753 LASSERT(osd_invariant(obj));
756 read_lock(&obj->oo_attr_lock);
757 *attr = obj->oo_attr;
758 if (obj->oo_lma_flags & LUSTRE_ORPHAN_FL)
759 attr->la_flags |= LUSTRE_ORPHAN_FL;
760 read_unlock(&obj->oo_attr_lock);
762 /* with ZFS_DEBUG zrl_add_debug() called by DB_DNODE_ENTER()
763 * from within sa_object_size() can block on a mutex, so
764 * we can't call sa_object_size() holding rwlock */
765 sa_object_size(obj->oo_sa_hdl, &blksize, &blocks);
766 /* we do not control size of indices, so always calculate
767 * it from number of blocks reported by DMU */
768 if (S_ISDIR(attr->la_mode))
769 attr->la_size = 512 * blocks;
770 /* Block size may be not set; suggest maximal I/O transfers. */
772 blksize = osd_spa_maxblocksize(
773 dmu_objset_spa(osd_obj2dev(obj)->od_os));
775 attr->la_blksize = blksize;
776 attr->la_blocks = blocks;
777 attr->la_valid |= LA_BLOCKS | LA_BLKSIZE;
780 up_read(&obj->oo_guard);
784 /* Simple wrapper on top of qsd API which implement quota transfer for osd
785 * setattr needs. As a reminder, only the root user can change ownership of
786 * a file, that's why EDQUOT & EINPROGRESS errors are discarded */
787 static inline int qsd_transfer(const struct lu_env *env,
788 struct qsd_instance *qsd,
789 struct lquota_trans *trans, int qtype,
790 __u64 orig_id, __u64 new_id, __u64 bspace,
791 struct lquota_id_info *qi)
795 if (unlikely(qsd == NULL))
798 LASSERT(qtype >= 0 && qtype < MAXQUOTAS);
799 qi->lqi_type = qtype;
801 /* inode accounting */
802 qi->lqi_is_blk = false;
804 /* one more inode for the new owner ... */
805 qi->lqi_id.qid_uid = new_id;
807 rc = qsd_op_begin(env, qsd, trans, qi, NULL);
808 if (rc == -EDQUOT || rc == -EINPROGRESS)
813 /* and one less inode for the current id */
814 qi->lqi_id.qid_uid = orig_id;;
816 /* can't get EDQUOT when reducing usage */
817 rc = qsd_op_begin(env, qsd, trans, qi, NULL);
818 if (rc == -EINPROGRESS)
823 /* block accounting */
824 qi->lqi_is_blk = true;
826 /* more blocks for the new owner ... */
827 qi->lqi_id.qid_uid = new_id;
828 qi->lqi_space = bspace;
829 rc = qsd_op_begin(env, qsd, trans, qi, NULL);
830 if (rc == -EDQUOT || rc == -EINPROGRESS)
835 /* and finally less blocks for the current owner */
836 qi->lqi_id.qid_uid = orig_id;
837 qi->lqi_space = -bspace;
838 rc = qsd_op_begin(env, qsd, trans, qi, NULL);
839 /* can't get EDQUOT when reducing usage */
840 if (rc == -EINPROGRESS)
845 static int osd_declare_attr_set(const struct lu_env *env,
846 struct dt_object *dt,
847 const struct lu_attr *attr,
848 struct thandle *handle)
850 struct osd_thread_info *info = osd_oti_get(env);
851 char *buf = osd_oti_get(env)->oti_str;
852 struct osd_object *obj = osd_dt_obj(dt);
853 struct osd_device *osd = osd_obj2dev(obj);
854 struct osd_thandle *oh;
861 LASSERT(handle != NULL);
862 LASSERT(osd_invariant(obj));
864 oh = container_of0(handle, struct osd_thandle, ot_super);
866 down_read(&obj->oo_guard);
867 if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
870 LASSERT(obj->oo_sa_hdl != NULL);
871 LASSERT(oh->ot_tx != NULL);
872 dmu_tx_hold_sa(oh->ot_tx, obj->oo_sa_hdl, 0);
873 if (oh->ot_tx->tx_err != 0)
874 GOTO(out, rc = -oh->ot_tx->tx_err);
876 sa_object_size(obj->oo_sa_hdl, &blksize, &bspace);
877 bspace = toqb(bspace * blksize);
879 __osd_xattr_declare_set(env, obj, sizeof(struct lustre_mdt_attrs),
882 if (attr && attr->la_valid & LA_UID) {
883 /* account for user inode tracking ZAP update */
884 dmu_tx_hold_bonus(oh->ot_tx, osd->od_iusr_oid);
885 dmu_tx_hold_zap(oh->ot_tx, osd->od_iusr_oid, TRUE, buf);
887 /* quota enforcement for user */
888 if (attr->la_uid != obj->oo_attr.la_uid) {
889 rc = qsd_transfer(env, osd->od_quota_slave,
890 &oh->ot_quota_trans, USRQUOTA,
891 obj->oo_attr.la_uid, attr->la_uid,
892 bspace, &info->oti_qi);
897 if (attr && attr->la_valid & LA_GID) {
898 /* account for user inode tracking ZAP update */
899 dmu_tx_hold_bonus(oh->ot_tx, osd->od_igrp_oid);
900 dmu_tx_hold_zap(oh->ot_tx, osd->od_igrp_oid, TRUE, buf);
902 /* quota enforcement for group */
903 if (attr->la_gid != obj->oo_attr.la_gid) {
904 rc = qsd_transfer(env, osd->od_quota_slave,
905 &oh->ot_quota_trans, GRPQUOTA,
906 obj->oo_attr.la_gid, attr->la_gid,
907 bspace, &info->oti_qi);
914 up_read(&obj->oo_guard);
919 * Set the attributes of an object
921 * The transaction passed to this routine must have
922 * dmu_tx_hold_bonus(tx, oid) called and then assigned
923 * to a transaction group.
925 static int osd_attr_set(const struct lu_env *env, struct dt_object *dt,
926 const struct lu_attr *la, struct thandle *handle)
928 struct osd_thread_info *info = osd_oti_get(env);
929 struct osd_object *obj = osd_dt_obj(dt);
930 struct osd_device *osd = osd_obj2dev(obj);
931 struct osd_thandle *oh;
932 struct osa_attr *osa = &info->oti_osa;
933 sa_bulk_attr_t *bulk;
934 __u64 valid = la->la_valid;
940 down_read(&obj->oo_guard);
941 if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
942 GOTO(out, rc = -ENOENT);
944 LASSERT(handle != NULL);
945 LASSERT(osd_invariant(obj));
946 LASSERT(obj->oo_sa_hdl);
948 oh = container_of0(handle, struct osd_thandle, ot_super);
949 /* Assert that the transaction has been assigned to a
950 transaction group. */
951 LASSERT(oh->ot_tx->tx_txg != 0);
953 /* Only allow set size for regular file */
954 if (!S_ISREG(dt->do_lu.lo_header->loh_attr))
955 valid &= ~(LA_SIZE | LA_BLOCKS);
960 if (valid & LA_FLAGS) {
961 struct lustre_mdt_attrs *lma;
964 if (la->la_flags & LUSTRE_LMA_FL_MASKS) {
965 CLASSERT(sizeof(info->oti_buf) >= sizeof(*lma));
966 lma = (struct lustre_mdt_attrs *)&info->oti_buf;
968 buf.lb_len = sizeof(info->oti_buf);
969 rc = osd_xattr_get(env, &obj->oo_dt, &buf,
973 le32_to_cpu(lma->lma_incompat);
975 lustre_to_lma_flags(la->la_flags);
977 cpu_to_le32(lma->lma_incompat);
979 buf.lb_len = sizeof(*lma);
980 rc = osd_xattr_set_internal(env, obj, &buf,
986 CWARN("%s: failed to set LMA flags: rc = %d\n",
993 OBD_ALLOC(bulk, sizeof(sa_bulk_attr_t) * 10);
995 GOTO(out, rc = -ENOMEM);
997 /* do both accounting updates outside oo_attr_lock below */
998 if ((valid & LA_UID) && (la->la_uid != obj->oo_attr.la_uid)) {
999 /* Update user accounting. Failure isn't fatal, but we still
1000 * log an error message */
1001 rc = -zap_increment_int(osd->od_os, osd->od_iusr_oid,
1002 la->la_uid, 1, oh->ot_tx);
1004 CERROR("%s: failed to update accounting ZAP for user "
1005 "%d (%d)\n", osd->od_svname, la->la_uid, rc);
1006 rc = -zap_increment_int(osd->od_os, osd->od_iusr_oid,
1007 obj->oo_attr.la_uid, -1, oh->ot_tx);
1009 CERROR("%s: failed to update accounting ZAP for user "
1010 "%d (%d)\n", osd->od_svname,
1011 obj->oo_attr.la_uid, rc);
1013 if ((valid & LA_GID) && (la->la_gid != obj->oo_attr.la_gid)) {
1014 /* Update group accounting. Failure isn't fatal, but we still
1015 * log an error message */
1016 rc = -zap_increment_int(osd->od_os, osd->od_igrp_oid,
1017 la->la_gid, 1, oh->ot_tx);
1019 CERROR("%s: failed to update accounting ZAP for user "
1020 "%d (%d)\n", osd->od_svname, la->la_gid, rc);
1021 rc = -zap_increment_int(osd->od_os, osd->od_igrp_oid,
1022 obj->oo_attr.la_gid, -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_gid, rc);
1029 write_lock(&obj->oo_attr_lock);
1031 if (valid & LA_ATIME) {
1032 osa->atime[0] = obj->oo_attr.la_atime = la->la_atime;
1033 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(osd), NULL,
1036 if (valid & LA_MTIME) {
1037 osa->mtime[0] = obj->oo_attr.la_mtime = la->la_mtime;
1038 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(osd), NULL,
1041 if (valid & LA_CTIME) {
1042 osa->ctime[0] = obj->oo_attr.la_ctime = la->la_ctime;
1043 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(osd), NULL,
1046 if (valid & LA_MODE) {
1047 /* mode is stored along with type, so read it first */
1048 obj->oo_attr.la_mode = (obj->oo_attr.la_mode & S_IFMT) |
1049 (la->la_mode & ~S_IFMT);
1050 osa->mode = obj->oo_attr.la_mode;
1051 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(osd), NULL,
1054 if (valid & LA_SIZE) {
1055 osa->size = obj->oo_attr.la_size = la->la_size;
1056 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(osd), NULL,
1059 if (valid & LA_NLINK) {
1060 osa->nlink = obj->oo_attr.la_nlink = la->la_nlink;
1061 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(osd), NULL,
1064 if (valid & LA_RDEV) {
1065 osa->rdev = obj->oo_attr.la_rdev = la->la_rdev;
1066 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_RDEV(osd), NULL,
1069 if (valid & LA_FLAGS) {
1070 osa->flags = attrs_fs2zfs(la->la_flags);
1071 /* many flags are not supported by zfs, so ensure a good cached
1073 obj->oo_attr.la_flags = attrs_zfs2fs(osa->flags);
1074 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(osd), NULL,
1077 if (valid & LA_UID) {
1078 osa->uid = obj->oo_attr.la_uid = la->la_uid;
1079 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(osd), NULL,
1082 if (valid & LA_GID) {
1083 osa->gid = obj->oo_attr.la_gid = la->la_gid;
1084 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(osd), NULL,
1087 obj->oo_attr.la_valid |= valid;
1088 write_unlock(&obj->oo_attr_lock);
1090 rc = osd_object_sa_bulk_update(obj, bulk, cnt, oh);
1092 OBD_FREE(bulk, sizeof(sa_bulk_attr_t) * 10);
1094 up_read(&obj->oo_guard);
1101 * XXX temporary solution.
1104 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
1105 struct dt_object *parent, struct dt_object *child,
1110 ah->dah_parent = parent;
1111 ah->dah_mode = child_mode;
1114 static int osd_declare_object_create(const struct lu_env *env,
1115 struct dt_object *dt,
1116 struct lu_attr *attr,
1117 struct dt_allocation_hint *hint,
1118 struct dt_object_format *dof,
1119 struct thandle *handle)
1121 char *buf = osd_oti_get(env)->oti_str;
1122 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1123 struct osd_object *obj = osd_dt_obj(dt);
1124 struct osd_device *osd = osd_obj2dev(obj);
1125 struct osd_thandle *oh;
1132 switch (dof->dof_type) {
1136 if (obj->oo_dt.do_body_ops == NULL)
1137 obj->oo_dt.do_body_ops = &osd_body_ops;
1143 LASSERT(handle != NULL);
1144 oh = container_of0(handle, struct osd_thandle, ot_super);
1145 LASSERT(oh->ot_tx != NULL);
1147 switch (dof->dof_type) {
1149 dt->do_index_ops = &osd_dir_ops;
1151 /* for zap create */
1152 dmu_tx_hold_zap(oh->ot_tx, DMU_NEW_OBJECT, 1, NULL);
1157 /* first, we'll create new object */
1158 dmu_tx_hold_bonus(oh->ot_tx, DMU_NEW_OBJECT);
1166 /* and we'll add it to some mapping */
1167 zapid = osd_get_name_n_idx(env, osd, fid, buf);
1168 dmu_tx_hold_bonus(oh->ot_tx, zapid);
1169 dmu_tx_hold_zap(oh->ot_tx, zapid, TRUE, buf);
1171 /* we will also update inode accounting ZAPs */
1172 dmu_tx_hold_bonus(oh->ot_tx, osd->od_iusr_oid);
1173 dmu_tx_hold_zap(oh->ot_tx, osd->od_iusr_oid, TRUE, buf);
1174 dmu_tx_hold_bonus(oh->ot_tx, osd->od_igrp_oid);
1175 dmu_tx_hold_zap(oh->ot_tx, osd->od_igrp_oid, TRUE, buf);
1177 dmu_tx_hold_sa_create(oh->ot_tx, ZFS_SA_BASE_ATTR_SIZE);
1179 __osd_xattr_declare_set(env, obj, sizeof(struct lustre_mdt_attrs),
1180 XATTR_NAME_LMA, oh);
1182 rc = osd_declare_quota(env, osd, attr->la_uid, attr->la_gid, 1, oh,
1183 false, NULL, false);
1187 int __osd_attr_init(const struct lu_env *env, struct osd_device *osd,
1188 uint64_t oid, dmu_tx_t *tx, struct lu_attr *la,
1191 sa_bulk_attr_t *bulk;
1192 sa_handle_t *sa_hdl;
1193 struct osa_attr *osa = &osd_oti_get(env)->oti_osa;
1201 gen = dmu_tx_get_txg(tx);
1203 ZFS_TIME_ENCODE(&now, crtime);
1205 osa->atime[0] = la->la_atime;
1206 osa->ctime[0] = la->la_ctime;
1207 osa->mtime[0] = la->la_mtime;
1208 osa->mode = la->la_mode;
1209 osa->uid = la->la_uid;
1210 osa->gid = la->la_gid;
1211 osa->rdev = la->la_rdev;
1212 osa->nlink = la->la_nlink;
1213 osa->flags = attrs_fs2zfs(la->la_flags);
1214 osa->size = la->la_size;
1216 /* Now add in all of the "SA" attributes */
1217 rc = -sa_handle_get(osd->od_os, oid, NULL, SA_HDL_PRIVATE, &sa_hdl);
1221 OBD_ALLOC(bulk, sizeof(sa_bulk_attr_t) * 13);
1227 * we need to create all SA below upon object create.
1229 * XXX The attribute order matters since the accounting callback relies
1230 * on static offsets (i.e. SA_*_OFFSET, see zfs_space_delta_cb()) to
1231 * look up the UID/GID attributes. Moreover, the callback does not seem
1232 * to support the spill block.
1233 * We define attributes in the same order as SA_*_OFFSET in order to
1234 * work around the problem. See ORI-610.
1237 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(osd), NULL, &osa->mode, 8);
1238 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(osd), NULL, &osa->size, 8);
1239 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GEN(osd), NULL, &gen, 8);
1240 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(osd), NULL, &osa->uid, 8);
1241 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(osd), NULL, &osa->gid, 8);
1242 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_PARENT(osd), NULL, &parent, 8);
1243 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(osd), NULL, &osa->flags, 8);
1244 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(osd), NULL, osa->atime, 16);
1245 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(osd), NULL, osa->mtime, 16);
1246 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(osd), NULL, osa->ctime, 16);
1247 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CRTIME(osd), NULL, crtime, 16);
1248 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(osd), NULL, &osa->nlink, 8);
1249 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_RDEV(osd), NULL, &osa->rdev, 8);
1251 rc = -sa_replace_all_by_template(sa_hdl, bulk, cnt, tx);
1253 OBD_FREE(bulk, sizeof(sa_bulk_attr_t) * 13);
1255 sa_handle_destroy(sa_hdl);
1260 * The transaction passed to this routine must have
1261 * dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT) called and then assigned
1262 * to a transaction group.
1264 int __osd_object_create(const struct lu_env *env, struct osd_object *obj,
1265 dmu_buf_t **dbp, dmu_tx_t *tx, struct lu_attr *la,
1270 struct osd_device *osd = osd_obj2dev(obj);
1271 const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
1272 dmu_object_type_t type = DMU_OT_PLAIN_FILE_CONTENTS;
1274 /* Assert that the transaction has been assigned to a
1275 transaction group. */
1276 LASSERT(tx->tx_txg != 0);
1278 /* Use DMU_OTN_UINT8_METADATA for local objects so their data blocks
1279 * would get an additional ditto copy */
1280 if (unlikely(S_ISREG(la->la_mode) &&
1281 fid_seq_is_local_file(fid_seq(fid))))
1282 type = DMU_OTN_UINT8_METADATA;
1284 /* Create a new DMU object. */
1285 oid = dmu_object_alloc(osd->od_os, type, 0,
1286 DMU_OT_SA, DN_MAX_BONUSLEN, tx);
1287 rc = -sa_buf_hold(osd->od_os, oid, osd_obj_tag, dbp);
1288 LASSERTF(rc == 0, "sa_buf_hold "LPU64" failed: %d\n", oid, rc);
1290 LASSERT(la->la_valid & LA_MODE);
1294 rc = __osd_attr_init(env, osd, oid, tx, la, parent);
1296 sa_buf_rele(*dbp, osd_obj_tag);
1298 dmu_object_free(osd->od_os, oid, tx);
1306 * The transaction passed to this routine must have
1307 * dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, ...) called and then assigned
1308 * to a transaction group.
1310 * Using ZAP_FLAG_HASH64 will force the ZAP to always be a FAT ZAP.
1311 * This is fine for directories today, because storing the FID in the dirent
1312 * will also require a FAT ZAP. If there is a new type of micro ZAP created
1313 * then we might need to re-evaluate the use of this flag and instead do
1314 * a conversion from the different internal ZAP hash formats being used. */
1315 int __osd_zap_create(const struct lu_env *env, struct osd_device *osd,
1316 dmu_buf_t **zap_dbp, dmu_tx_t *tx,
1317 struct lu_attr *la, uint64_t parent, zap_flags_t flags)
1322 /* Assert that the transaction has been assigned to a
1323 transaction group. */
1324 LASSERT(tx->tx_txg != 0);
1326 oid = zap_create_flags(osd->od_os, 0, flags | ZAP_FLAG_HASH64,
1327 DMU_OT_DIRECTORY_CONTENTS,
1328 14, /* == ZFS fzap_default_block_shift */
1329 DN_MAX_INDBLKSHIFT, /* indirect block shift */
1330 DMU_OT_SA, DN_MAX_BONUSLEN, tx);
1332 rc = -sa_buf_hold(osd->od_os, oid, osd_obj_tag, zap_dbp);
1336 LASSERT(la->la_valid & LA_MODE);
1340 return __osd_attr_init(env, osd, oid, tx, la, parent);
1343 static dmu_buf_t *osd_mkidx(const struct lu_env *env, struct osd_object *obj,
1344 struct lu_attr *la, uint64_t parent,
1345 struct osd_thandle *oh)
1350 /* Index file should be created as regular file in order not to confuse
1351 * ZPL which could interpret them as directory.
1352 * We set ZAP_FLAG_UINT64_KEY to let ZFS know than we are going to use
1354 LASSERT(S_ISREG(la->la_mode));
1355 rc = __osd_zap_create(env, osd_obj2dev(obj), &db, oh->ot_tx, la, parent,
1356 ZAP_FLAG_UINT64_KEY);
1362 static dmu_buf_t *osd_mkdir(const struct lu_env *env, struct osd_object *obj,
1363 struct lu_attr *la, uint64_t parent,
1364 struct osd_thandle *oh)
1369 LASSERT(S_ISDIR(la->la_mode));
1370 rc = __osd_zap_create(env, osd_obj2dev(obj), &db,
1371 oh->ot_tx, la, parent, 0);
1377 static dmu_buf_t *osd_mkreg(const struct lu_env *env, struct osd_object *obj,
1378 struct lu_attr *la, uint64_t parent,
1379 struct osd_thandle *oh)
1383 struct osd_device *osd = osd_obj2dev(obj);
1385 LASSERT(S_ISREG(la->la_mode));
1386 rc = __osd_object_create(env, obj, &db, oh->ot_tx, la, parent);
1391 * XXX: This heuristic is non-optimal. It would be better to
1392 * increase the blocksize up to osd->od_max_blksz during the write.
1393 * This is exactly how the ZPL behaves and it ensures that the right
1394 * blocksize is selected based on the file size rather than the
1395 * making broad assumptions based on the osd type.
1397 if (!lu_device_is_md(osd2lu_dev(osd))) {
1398 rc = -dmu_object_set_blocksize(osd->od_os, db->db_object,
1399 osd->od_max_blksz, 0, oh->ot_tx);
1401 CERROR("%s: can't change blocksize: %d\n",
1402 osd->od_svname, rc);
1410 static dmu_buf_t *osd_mksym(const struct lu_env *env, struct osd_object *obj,
1411 struct lu_attr *la, uint64_t parent,
1412 struct osd_thandle *oh)
1417 LASSERT(S_ISLNK(la->la_mode));
1418 rc = __osd_object_create(env, obj, &db, oh->ot_tx, la, parent);
1424 static dmu_buf_t *osd_mknod(const struct lu_env *env, struct osd_object *obj,
1425 struct lu_attr *la, uint64_t parent,
1426 struct osd_thandle *oh)
1431 la->la_valid = LA_MODE;
1432 if (S_ISCHR(la->la_mode) || S_ISBLK(la->la_mode))
1433 la->la_valid |= LA_RDEV;
1435 rc = __osd_object_create(env, obj, &db, oh->ot_tx, la, parent);
1441 typedef dmu_buf_t *(*osd_obj_type_f)(const struct lu_env *env,
1442 struct osd_object *obj,
1445 struct osd_thandle *oh);
1447 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
1449 osd_obj_type_f result;
1475 * Primitives for directory (i.e. ZAP) handling
1477 static inline int osd_init_lma(const struct lu_env *env, struct osd_object *obj,
1478 const struct lu_fid *fid, struct osd_thandle *oh)
1480 struct osd_thread_info *info = osd_oti_get(env);
1481 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
1485 lustre_lma_init(lma, fid, 0, 0);
1486 lustre_lma_swab(lma);
1488 buf.lb_len = sizeof(*lma);
1490 rc = osd_xattr_set_internal(env, obj, &buf, XATTR_NAME_LMA,
1491 LU_XATTR_CREATE, oh);
1497 * Concurrency: @dt is write locked.
1499 static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
1500 struct lu_attr *attr,
1501 struct dt_allocation_hint *hint,
1502 struct dt_object_format *dof,
1505 struct zpl_direntry *zde = &osd_oti_get(env)->oti_zde.lzd_reg;
1506 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1507 struct osd_object *obj = osd_dt_obj(dt);
1508 struct osd_device *osd = osd_obj2dev(obj);
1509 char *buf = osd_oti_get(env)->oti_str;
1510 struct osd_thandle *oh;
1517 /* concurrent create declarations should not see
1518 * the object inconsistent (db, attr, etc).
1519 * in regular cases acquisition should be cheap */
1520 down_write(&obj->oo_guard);
1522 if (unlikely(dt_object_exists(dt)))
1523 GOTO(out, rc = -EEXIST);
1525 LASSERT(osd_invariant(obj));
1526 LASSERT(dof != NULL);
1528 LASSERT(th != NULL);
1529 oh = container_of0(th, struct osd_thandle, ot_super);
1532 * XXX missing: Quote handling.
1535 LASSERT(obj->oo_db == NULL);
1537 /* to follow ZFS on-disk format we need
1538 * to initialize parent dnode properly */
1540 if (hint != NULL && hint->dah_parent != NULL &&
1541 !dt_object_remote(hint->dah_parent))
1542 zapid = osd_dt_obj(hint->dah_parent)->oo_db->db_object;
1544 db = osd_create_type_f(dof->dof_type)(env, obj, attr, zapid, oh);
1546 GOTO(out, rc = PTR_ERR(db));
1549 zde->zde_dnode = db->db_object;
1550 zde->zde_type = IFTODT(attr->la_mode & S_IFMT);
1552 zapid = osd_get_name_n_idx(env, osd, fid, buf);
1554 rc = -zap_add(osd->od_os, zapid, buf, 8, 1, zde, oh->ot_tx);
1558 /* Add new object to inode accounting.
1559 * Errors are not considered as fatal */
1560 rc = -zap_increment_int(osd->od_os, osd->od_iusr_oid,
1561 (attr->la_valid & LA_UID) ? attr->la_uid : 0, 1,
1564 CERROR("%s: failed to add "DFID" to accounting ZAP for usr %d "
1565 "(%d)\n", osd->od_svname, PFID(fid), attr->la_uid, rc);
1566 rc = -zap_increment_int(osd->od_os, osd->od_igrp_oid,
1567 (attr->la_valid & LA_GID) ? attr->la_gid : 0, 1,
1570 CERROR("%s: failed to add "DFID" to accounting ZAP for grp %d "
1571 "(%d)\n", osd->od_svname, PFID(fid), attr->la_gid, rc);
1573 /* configure new osd object */
1575 rc = osd_object_init0(env, obj);
1576 LASSERT(ergo(rc == 0, dt_object_exists(dt)));
1577 LASSERT(osd_invariant(obj));
1579 rc = osd_init_lma(env, obj, fid, oh);
1581 CERROR("%s: can not set LMA on "DFID": rc = %d\n",
1582 osd->od_svname, PFID(fid), rc);
1583 /* ignore errors during LMA initialization */
1588 up_write(&obj->oo_guard);
1592 static int osd_declare_object_ref_add(const struct lu_env *env,
1593 struct dt_object *dt,
1596 return osd_declare_attr_set(env, dt, NULL, th);
1600 * Concurrency: @dt is write locked.
1602 static int osd_object_ref_add(const struct lu_env *env,
1603 struct dt_object *dt,
1604 struct thandle *handle)
1606 struct osd_object *obj = osd_dt_obj(dt);
1607 struct osd_thandle *oh;
1608 struct osd_device *osd = osd_obj2dev(obj);
1614 down_read(&obj->oo_guard);
1615 if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
1616 GOTO(out, rc = -ENOENT);
1618 LASSERT(osd_invariant(obj));
1619 LASSERT(obj->oo_sa_hdl != NULL);
1621 oh = container_of0(handle, struct osd_thandle, ot_super);
1623 write_lock(&obj->oo_attr_lock);
1624 nlink = ++obj->oo_attr.la_nlink;
1625 write_unlock(&obj->oo_attr_lock);
1627 rc = osd_object_sa_update(obj, SA_ZPL_LINKS(osd), &nlink, 8, oh);
1630 up_read(&obj->oo_guard);
1634 static int osd_declare_object_ref_del(const struct lu_env *env,
1635 struct dt_object *dt,
1636 struct thandle *handle)
1638 return osd_declare_attr_set(env, dt, NULL, handle);
1642 * Concurrency: @dt is write locked.
1644 static int osd_object_ref_del(const struct lu_env *env,
1645 struct dt_object *dt,
1646 struct thandle *handle)
1648 struct osd_object *obj = osd_dt_obj(dt);
1649 struct osd_thandle *oh;
1650 struct osd_device *osd = osd_obj2dev(obj);
1656 down_read(&obj->oo_guard);
1658 if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
1659 GOTO(out, rc = -ENOENT);
1661 LASSERT(osd_invariant(obj));
1662 LASSERT(obj->oo_sa_hdl != NULL);
1664 oh = container_of0(handle, struct osd_thandle, ot_super);
1665 LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
1667 write_lock(&obj->oo_attr_lock);
1668 nlink = --obj->oo_attr.la_nlink;
1669 write_unlock(&obj->oo_attr_lock);
1671 rc = osd_object_sa_update(obj, SA_ZPL_LINKS(osd), &nlink, 8, oh);
1674 up_read(&obj->oo_guard);
1678 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
1679 __u64 start, __u64 end)
1681 struct osd_device *osd = osd_obj2dev(osd_dt_obj(dt));
1684 /* XXX: no other option than syncing the whole filesystem until we
1685 * support ZIL. If the object tracked the txg that it was last
1686 * modified in, it could pass that txg here instead of "0". Maybe
1687 * the changes are already committed, so no wait is needed at all? */
1688 txg_wait_synced(dmu_objset_pool(osd->od_os), 0ULL);
1693 static struct dt_object_operations osd_obj_ops = {
1694 .do_read_lock = osd_object_read_lock,
1695 .do_write_lock = osd_object_write_lock,
1696 .do_read_unlock = osd_object_read_unlock,
1697 .do_write_unlock = osd_object_write_unlock,
1698 .do_write_locked = osd_object_write_locked,
1699 .do_attr_get = osd_attr_get,
1700 .do_declare_attr_set = osd_declare_attr_set,
1701 .do_attr_set = osd_attr_set,
1702 .do_ah_init = osd_ah_init,
1703 .do_declare_create = osd_declare_object_create,
1704 .do_create = osd_object_create,
1705 .do_declare_destroy = osd_declare_object_destroy,
1706 .do_destroy = osd_object_destroy,
1707 .do_index_try = osd_index_try,
1708 .do_declare_ref_add = osd_declare_object_ref_add,
1709 .do_ref_add = osd_object_ref_add,
1710 .do_declare_ref_del = osd_declare_object_ref_del,
1711 .do_ref_del = osd_object_ref_del,
1712 .do_xattr_get = osd_xattr_get,
1713 .do_declare_xattr_set = osd_declare_xattr_set,
1714 .do_xattr_set = osd_xattr_set,
1715 .do_declare_xattr_del = osd_declare_xattr_del,
1716 .do_xattr_del = osd_xattr_del,
1717 .do_xattr_list = osd_xattr_list,
1718 .do_object_sync = osd_object_sync,
1721 static struct lu_object_operations osd_lu_obj_ops = {
1722 .loo_object_init = osd_object_init,
1723 .loo_object_delete = osd_object_delete,
1724 .loo_object_release = osd_object_release,
1725 .loo_object_free = osd_object_free,
1726 .loo_object_print = osd_object_print,
1727 .loo_object_invariant = osd_object_invariant,
1730 static int osd_otable_it_attr_get(const struct lu_env *env,
1731 struct dt_object *dt,
1732 struct lu_attr *attr)
1738 static struct dt_object_operations osd_obj_otable_it_ops = {
1739 .do_attr_get = osd_otable_it_attr_get,
1740 .do_index_try = osd_index_try,