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.
31 * Copyright (c) 2012, 2013, Intel Corporation.
32 * Use is subject to license terms.
35 * This file is part of Lustre, http://www.lustre.org/
36 * Lustre is a trademark of Sun Microsystems, Inc.
38 * lustre/osd-zfs/osd_object.c
40 * Author: Alex Zhuravlev <bzzz@whamcloud.com>
41 * Author: Mike Pershin <tappro@whamcloud.com>
42 * Author: Johann Lombardi <johann@whamcloud.com>
46 # define EXPORT_SYMTAB
48 #define DEBUG_SUBSYSTEM S_OSD
50 #include <lustre_ver.h>
51 #include <libcfs/libcfs.h>
52 #include <obd_support.h>
53 #include <lustre_net.h>
55 #include <obd_class.h>
56 #include <lustre_disk.h>
57 #include <lustre_fid.h>
59 #include "osd_internal.h"
61 #include <sys/dnode.h>
66 #include <sys/spa_impl.h>
67 #include <sys/zfs_znode.h>
68 #include <sys/dmu_tx.h>
69 #include <sys/dmu_objset.h>
70 #include <sys/dsl_prop.h>
71 #include <sys/sa_impl.h>
74 static char *osd_obj_tag = "osd_object";
76 static struct dt_object_operations osd_obj_ops;
77 static struct lu_object_operations osd_lu_obj_ops;
78 extern struct dt_body_operations osd_body_ops;
79 static struct dt_object_operations osd_obj_otable_it_ops;
81 extern struct kmem_cache *osd_object_kmem;
84 osd_object_sa_fini(struct osd_object *obj)
87 sa_handle_destroy(obj->oo_sa_hdl);
88 obj->oo_sa_hdl = NULL;
93 osd_object_sa_init(struct osd_object *obj, udmu_objset_t *uos)
97 LASSERT(obj->oo_sa_hdl == NULL);
98 LASSERT(obj->oo_db != NULL);
100 rc = -sa_handle_get(uos->os, obj->oo_db->db_object, obj,
101 SA_HDL_PRIVATE, &obj->oo_sa_hdl);
105 /* Cache the xattr object id, valid for the life of the object */
106 rc = -sa_lookup(obj->oo_sa_hdl, SA_ZPL_XATTR(uos), &obj->oo_xattr, 8);
108 obj->oo_xattr = ZFS_NO_OBJECT;
111 osd_object_sa_fini(obj);
118 * Add object to list of dirty objects in tx handle.
121 osd_object_sa_dirty_add(struct osd_object *obj, struct osd_thandle *oh)
123 if (!cfs_list_empty(&obj->oo_sa_linkage))
126 down(&oh->ot_sa_lock);
127 write_lock(&obj->oo_attr_lock);
128 if (likely(cfs_list_empty(&obj->oo_sa_linkage)))
129 cfs_list_add(&obj->oo_sa_linkage, &oh->ot_sa_list);
130 write_unlock(&obj->oo_attr_lock);
135 * Release spill block dbuf hold for all dirty SAs.
137 void osd_object_sa_dirty_rele(struct osd_thandle *oh)
139 struct osd_object *obj;
141 down(&oh->ot_sa_lock);
142 while (!cfs_list_empty(&oh->ot_sa_list)) {
143 obj = cfs_list_entry(oh->ot_sa_list.next,
144 struct osd_object, oo_sa_linkage);
145 sa_spill_rele(obj->oo_sa_hdl);
146 write_lock(&obj->oo_attr_lock);
147 cfs_list_del_init(&obj->oo_sa_linkage);
148 write_unlock(&obj->oo_attr_lock);
154 * Update the SA and add the object to the dirty list.
156 int osd_object_sa_update(struct osd_object *obj, sa_attr_type_t type,
157 void *buf, uint32_t buflen, struct osd_thandle *oh)
161 LASSERT(obj->oo_sa_hdl != NULL);
162 LASSERT(oh->ot_tx != NULL);
164 rc = -sa_update(obj->oo_sa_hdl, type, buf, buflen, oh->ot_tx);
165 osd_object_sa_dirty_add(obj, oh);
171 * Bulk update the SA and add the object to the dirty list.
174 osd_object_sa_bulk_update(struct osd_object *obj, sa_bulk_attr_t *attrs,
175 int count, struct osd_thandle *oh)
179 LASSERT(obj->oo_sa_hdl != NULL);
180 LASSERT(oh->ot_tx != NULL);
182 rc = -sa_bulk_update(obj->oo_sa_hdl, attrs, count, oh->ot_tx);
183 osd_object_sa_dirty_add(obj, oh);
189 * Retrieve the attributes of a DMU object
191 int __osd_object_attr_get(const struct lu_env *env, udmu_objset_t *uos,
192 struct osd_object *obj, struct lu_attr *la)
194 struct osa_attr *osa = &osd_oti_get(env)->oti_osa;
196 sa_bulk_attr_t *bulk;
201 LASSERT(obj->oo_db != NULL);
203 rc = -sa_handle_get(uos->os, obj->oo_db->db_object, NULL,
204 SA_HDL_PRIVATE, &sa_hdl);
208 OBD_ALLOC(bulk, sizeof(sa_bulk_attr_t) * 9);
210 GOTO(out_sa, rc = -ENOMEM);
212 la->la_valid |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE | LA_TYPE |
213 LA_SIZE | LA_UID | LA_GID | LA_FLAGS | LA_NLINK;
215 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(uos), NULL, osa->atime, 16);
216 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(uos), NULL, osa->mtime, 16);
217 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(uos), NULL, osa->ctime, 16);
218 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(uos), NULL, &osa->mode, 8);
219 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(uos), NULL, &osa->size, 8);
220 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(uos), NULL, &osa->nlink, 8);
221 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(uos), NULL, &osa->uid, 8);
222 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(uos), NULL, &osa->gid, 8);
223 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(uos), NULL, &osa->flags, 8);
225 rc = -sa_bulk_lookup(sa_hdl, bulk, cnt);
229 la->la_atime = osa->atime[0];
230 la->la_mtime = osa->mtime[0];
231 la->la_ctime = osa->ctime[0];
232 la->la_mode = osa->mode;
233 la->la_uid = osa->uid;
234 la->la_gid = osa->gid;
235 la->la_nlink = osa->nlink;
236 la->la_flags = attrs_zfs2fs(osa->flags);
237 la->la_size = osa->size;
239 if (S_ISCHR(la->la_mode) || S_ISBLK(la->la_mode)) {
240 rc = -sa_lookup(sa_hdl, SA_ZPL_RDEV(uos), &osa->rdev, 8);
243 la->la_rdev = osa->rdev;
244 la->la_valid |= LA_RDEV;
247 OBD_FREE(bulk, sizeof(sa_bulk_attr_t) * 9);
249 sa_handle_destroy(sa_hdl);
254 int __osd_obj2dbuf(const struct lu_env *env, objset_t *os,
255 uint64_t oid, dmu_buf_t **dbp, void *tag)
257 dmu_object_info_t *doi = &osd_oti_get(env)->oti_doi;
262 rc = -sa_buf_hold(os, oid, tag, dbp);
266 dmu_object_info_from_db(*dbp, doi);
267 if (unlikely (oid != DMU_USERUSED_OBJECT &&
268 oid != DMU_GROUPUSED_OBJECT && doi->doi_bonus_type != DMU_OT_SA)) {
269 sa_buf_rele(*dbp, tag);
275 LASSERT((*dbp)->db_object == oid);
276 LASSERT((*dbp)->db_offset == -1);
277 LASSERT((*dbp)->db_data != NULL);
283 * Concurrency: no concurrent access is possible that early in object
286 struct lu_object *osd_object_alloc(const struct lu_env *env,
287 const struct lu_object_header *hdr,
290 struct osd_object *mo;
292 OBD_SLAB_ALLOC_PTR_GFP(mo, osd_object_kmem, __GFP_IO);
296 l = &mo->oo_dt.do_lu;
297 dt_object_init(&mo->oo_dt, NULL, d);
298 mo->oo_dt.do_ops = &osd_obj_ops;
299 l->lo_ops = &osd_lu_obj_ops;
300 CFS_INIT_LIST_HEAD(&mo->oo_sa_linkage);
301 init_rwsem(&mo->oo_sem);
302 sema_init(&mo->oo_guard, 1);
303 rwlock_init(&mo->oo_attr_lock);
311 * Concurrency: shouldn't matter.
313 int osd_object_init0(const struct lu_env *env, struct osd_object *obj)
315 struct osd_device *osd = osd_obj2dev(obj);
316 udmu_objset_t *uos = &osd->od_objset;
317 const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
321 if (obj->oo_db == NULL)
326 rc = osd_object_sa_init(obj, uos);
330 /* cache attrs in object */
331 rc = __osd_object_attr_get(env, &osd->od_objset,
336 if (likely(!fid_is_acct(fid)))
337 /* no body operations for accounting objects */
338 obj->oo_dt.do_body_ops = &osd_body_ops;
341 * initialize object before marking it existing
343 obj->oo_dt.do_lu.lo_header->loh_attr |= obj->oo_attr.la_mode & S_IFMT;
346 obj->oo_dt.do_lu.lo_header->loh_attr |= LOHA_EXISTS;
351 static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
353 struct osd_thread_info *info = osd_oti_get(env);
356 struct lustre_mdt_attrs *lma;
359 CLASSERT(sizeof(info->oti_buf) >= sizeof(*lma));
360 lma = (struct lustre_mdt_attrs *)info->oti_buf;
362 buf.lb_len = sizeof(info->oti_buf);
364 rc = osd_xattr_get(env, &obj->oo_dt, &buf, XATTR_NAME_LMA, BYPASS_CAPA);
367 lustre_lma_swab(lma);
368 if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
369 CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT))) {
370 CWARN("%s: unsupported incompat LMA feature(s) %#x for "
371 "fid = "DFID"\n", osd_obj2dev(obj)->od_svname,
372 lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
373 PFID(lu_object_fid(&obj->oo_dt.do_lu)));
376 } else if (rc == -ENODATA) {
377 /* haven't initialize LMA xattr */
385 * Concurrency: no concurrent access is possible that early in object
388 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
389 const struct lu_object_conf *conf)
391 struct osd_object *obj = osd_obj(l);
392 struct osd_device *osd = osd_obj2dev(obj);
397 LASSERT(osd_invariant(obj));
399 if (fid_is_otable_it(&l->lo_header->loh_fid)) {
400 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
401 l->lo_header->loh_attr |= LOHA_EXISTS;
405 rc = osd_fid_lookup(env, osd, lu_object_fid(l), &oid);
407 LASSERT(obj->oo_db == NULL);
408 rc = __osd_obj2dbuf(env, osd->od_objset.os, oid,
409 &obj->oo_db, osd_obj_tag);
411 CERROR("%s: lookup "DFID"/"LPX64" failed: rc = %d\n",
412 osd->od_svname, PFID(lu_object_fid(l)), oid, rc);
416 rc = osd_object_init0(env, obj);
420 rc = osd_check_lma(env, obj);
423 } else if (rc == -ENOENT) {
426 LASSERT(osd_invariant(obj));
432 * Concurrency: no concurrent access is possible that late in object
435 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
437 struct osd_object *obj = osd_obj(l);
439 LASSERT(osd_invariant(obj));
441 dt_object_fini(&obj->oo_dt);
442 OBD_SLAB_FREE_PTR(obj, osd_object_kmem);
445 static void __osd_declare_object_destroy(const struct lu_env *env,
446 struct osd_object *obj,
447 struct osd_thandle *oh)
449 struct osd_device *osd = osd_obj2dev(obj);
450 udmu_objset_t *uos = &osd->od_objset;
451 dmu_buf_t *db = obj->oo_db;
452 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
453 uint64_t oid = db->db_object, xid;
454 dmu_tx_t *tx = oh->ot_tx;
458 dmu_tx_hold_free(tx, oid, 0, DMU_OBJECT_END);
460 /* zap holding xattrs */
461 if (obj->oo_xattr != ZFS_NO_OBJECT) {
464 dmu_tx_hold_free(tx, oid, 0, DMU_OBJECT_END);
466 rc = -udmu_zap_cursor_init(&zc, uos, oid, 0);
470 while ((rc = -zap_cursor_retrieve(zc, za)) == 0) {
471 BUG_ON(za->za_integer_length != sizeof(uint64_t));
472 BUG_ON(za->za_num_integers != 1);
474 rc = -zap_lookup(uos->os, obj->oo_xattr, za->za_name,
475 sizeof(uint64_t), 1, &xid);
477 CERROR("%s: xattr lookup failed: rc = %d\n",
481 dmu_tx_hold_free(tx, xid, 0, DMU_OBJECT_END);
483 zap_cursor_advance(zc);
488 udmu_zap_cursor_fini(zc);
491 if (rc && tx->tx_err == 0)
495 static int osd_declare_object_destroy(const struct lu_env *env,
496 struct dt_object *dt,
499 char *buf = osd_oti_get(env)->oti_str;
500 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
501 struct osd_object *obj = osd_dt_obj(dt);
502 struct osd_device *osd = osd_obj2dev(obj);
503 struct osd_thandle *oh;
509 LASSERT(dt_object_exists(dt));
511 oh = container_of0(th, struct osd_thandle, ot_super);
512 LASSERT(oh->ot_tx != NULL);
514 /* declare that we'll destroy the object */
515 __osd_declare_object_destroy(env, obj, oh);
517 /* declare that we'll remove object from fid-dnode mapping */
518 zapid = osd_get_name_n_idx(env, osd, fid, buf);
519 dmu_tx_hold_bonus(oh->ot_tx, zapid);
520 dmu_tx_hold_zap(oh->ot_tx, zapid, 0, buf);
522 /* declare that we'll remove object from inode accounting ZAPs */
523 dmu_tx_hold_bonus(oh->ot_tx, osd->od_iusr_oid);
524 dmu_tx_hold_zap(oh->ot_tx, osd->od_iusr_oid, 0, buf);
525 dmu_tx_hold_bonus(oh->ot_tx, osd->od_igrp_oid);
526 dmu_tx_hold_zap(oh->ot_tx, osd->od_igrp_oid, 0, buf);
529 rc = osd_declare_quota(env, osd, obj->oo_attr.la_uid,
530 obj->oo_attr.la_gid, -1, oh, false, NULL, false);
534 /* data to be truncated */
535 rc = osd_declare_quota(env, osd, obj->oo_attr.la_uid,
536 obj->oo_attr.la_gid, 0, oh, true, NULL, false);
540 int __osd_object_free(udmu_objset_t *uos, uint64_t oid, dmu_tx_t *tx)
542 LASSERT(uos->objects != 0);
543 spin_lock(&uos->lock);
545 spin_unlock(&uos->lock);
547 return -dmu_object_free(uos->os, oid, tx);
551 * Delete a DMU object
553 * The transaction passed to this routine must have
554 * dmu_tx_hold_free(tx, oid, 0, DMU_OBJECT_END) called
555 * and then assigned to a transaction group.
557 * This will release db and set it to NULL to prevent further dbuf releases.
559 static int __osd_object_destroy(const struct lu_env *env,
560 struct osd_object *obj,
561 dmu_tx_t *tx, void *tag)
563 struct osd_device *osd = osd_obj2dev(obj);
564 udmu_objset_t *uos = &osd->od_objset;
566 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
570 /* Assert that the transaction has been assigned to a
571 transaction group. */
572 LASSERT(tx->tx_txg != 0);
574 /* zap holding xattrs */
575 if (obj->oo_xattr != ZFS_NO_OBJECT) {
576 rc = -udmu_zap_cursor_init(&zc, uos, obj->oo_xattr, 0);
579 while ((rc = -zap_cursor_retrieve(zc, za)) == 0) {
580 BUG_ON(za->za_integer_length != sizeof(uint64_t));
581 BUG_ON(za->za_num_integers != 1);
583 rc = -zap_lookup(uos->os, obj->oo_xattr, za->za_name,
584 sizeof(uint64_t), 1, &xid);
586 CERROR("%s: lookup xattr %s failed: rc = %d\n",
587 osd->od_svname, za->za_name, rc);
590 rc = __osd_object_free(uos, xid, tx);
592 CERROR("%s: fetch xattr %s failed: rc = %d\n",
593 osd->od_svname, za->za_name, rc);
594 zap_cursor_advance(zc);
596 udmu_zap_cursor_fini(zc);
598 rc = __osd_object_free(uos, obj->oo_xattr, tx);
600 CERROR("%s: freeing xattr failed: rc = %d\n",
604 return __osd_object_free(uos, obj->oo_db->db_object, tx);
607 static int osd_object_destroy(const struct lu_env *env,
608 struct dt_object *dt,
611 char *buf = osd_oti_get(env)->oti_str;
612 struct osd_object *obj = osd_dt_obj(dt);
613 struct osd_device *osd = osd_obj2dev(obj);
614 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
615 struct osd_thandle *oh;
620 LASSERT(obj->oo_db != NULL);
621 LASSERT(dt_object_exists(dt));
622 LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
624 oh = container_of0(th, struct osd_thandle, ot_super);
626 LASSERT(oh->ot_tx != NULL);
628 zapid = osd_get_name_n_idx(env, osd, fid, buf);
630 /* remove obj ref from index dir (it depends) */
631 rc = -zap_remove(osd->od_objset.os, zapid, buf, oh->ot_tx);
633 CERROR("%s: zap_remove() failed: rc = %d\n",
638 /* Remove object from inode accounting. It is not fatal for the destroy
639 * operation if something goes wrong while updating accounting, but we
640 * still log an error message to notify the administrator */
641 rc = -zap_increment_int(osd->od_objset.os, osd->od_iusr_oid,
642 obj->oo_attr.la_uid, -1, oh->ot_tx);
644 CERROR("%s: failed to remove "DFID" from accounting ZAP for usr"
645 " %d: rc = %d\n", osd->od_svname, PFID(fid),
646 obj->oo_attr.la_uid, rc);
647 rc = -zap_increment_int(osd->od_objset.os, osd->od_igrp_oid,
648 obj->oo_attr.la_gid, -1, oh->ot_tx);
650 CERROR("%s: failed to remove "DFID" from accounting ZAP for grp"
651 " %d: rc = %d\n", osd->od_svname, PFID(fid),
652 obj->oo_attr.la_gid, rc);
655 rc = __osd_object_destroy(env, obj, oh->ot_tx, osd_obj_tag);
657 CERROR("%s: __osd_object_destroy() failed: rc = %d\n",
663 /* not needed in the cache anymore */
664 set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
669 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
671 struct osd_object *obj = osd_obj(l);
673 if (obj->oo_db != NULL) {
674 osd_object_sa_fini(obj);
675 if (obj->oo_sa_xattr) {
676 nvlist_free(obj->oo_sa_xattr);
677 obj->oo_sa_xattr = NULL;
679 sa_buf_rele(obj->oo_db, osd_obj_tag);
680 cfs_list_del(&obj->oo_sa_linkage);
686 * Concurrency: ->loo_object_release() is called under site spin-lock.
688 static void osd_object_release(const struct lu_env *env,
694 * Concurrency: shouldn't matter.
696 static int osd_object_print(const struct lu_env *env, void *cookie,
697 lu_printer_t p, const struct lu_object *l)
699 struct osd_object *o = osd_obj(l);
701 return (*p)(env, cookie, LUSTRE_OSD_ZFS_NAME"-object@%p", o);
704 static void osd_object_read_lock(const struct lu_env *env,
705 struct dt_object *dt, unsigned role)
707 struct osd_object *obj = osd_dt_obj(dt);
709 LASSERT(osd_invariant(obj));
711 down_read(&obj->oo_sem);
714 static void osd_object_write_lock(const struct lu_env *env,
715 struct dt_object *dt, unsigned role)
717 struct osd_object *obj = osd_dt_obj(dt);
719 LASSERT(osd_invariant(obj));
721 down_write(&obj->oo_sem);
724 static void osd_object_read_unlock(const struct lu_env *env,
725 struct dt_object *dt)
727 struct osd_object *obj = osd_dt_obj(dt);
729 LASSERT(osd_invariant(obj));
730 up_read(&obj->oo_sem);
733 static void osd_object_write_unlock(const struct lu_env *env,
734 struct dt_object *dt)
736 struct osd_object *obj = osd_dt_obj(dt);
738 LASSERT(osd_invariant(obj));
739 up_write(&obj->oo_sem);
742 static int osd_object_write_locked(const struct lu_env *env,
743 struct dt_object *dt)
745 struct osd_object *obj = osd_dt_obj(dt);
748 LASSERT(osd_invariant(obj));
750 if (down_write_trylock(&obj->oo_sem)) {
752 up_write(&obj->oo_sem);
757 static int osd_attr_get(const struct lu_env *env,
758 struct dt_object *dt,
759 struct lu_attr *attr,
760 struct lustre_capa *capa)
762 struct osd_object *obj = osd_dt_obj(dt);
766 LASSERT(dt_object_exists(dt));
767 LASSERT(osd_invariant(obj));
770 read_lock(&obj->oo_attr_lock);
771 *attr = obj->oo_attr;
772 read_unlock(&obj->oo_attr_lock);
774 /* with ZFS_DEBUG zrl_add_debug() called by DB_DNODE_ENTER()
775 * from within sa_object_size() can block on a mutex, so
776 * we can't call sa_object_size() holding rwlock */
777 sa_object_size(obj->oo_sa_hdl, &blksize, &blocks);
778 /* we do not control size of indices, so always calculate
779 * it from number of blocks reported by DMU */
780 if (S_ISDIR(attr->la_mode))
781 attr->la_size = 512 * blocks;
782 /* Block size may be not set; suggest maximal I/O transfers. */
784 blksize = 1ULL << SPA_MAXBLOCKSHIFT;
786 attr->la_blksize = blksize;
787 attr->la_blocks = blocks;
788 attr->la_valid |= LA_BLOCKS | LA_BLKSIZE;
793 /* Simple wrapper on top of qsd API which implement quota transfer for osd
794 * setattr needs. As a reminder, only the root user can change ownership of
795 * a file, that's why EDQUOT & EINPROGRESS errors are discarded */
796 static inline int qsd_transfer(const struct lu_env *env,
797 struct qsd_instance *qsd,
798 struct lquota_trans *trans, int qtype,
799 __u64 orig_id, __u64 new_id, __u64 bspace,
800 struct lquota_id_info *qi)
804 if (unlikely(qsd == NULL))
807 LASSERT(qtype >= 0 && qtype < MAXQUOTAS);
808 qi->lqi_type = qtype;
810 /* inode accounting */
811 qi->lqi_is_blk = false;
813 /* one more inode for the new owner ... */
814 qi->lqi_id.qid_uid = new_id;
816 rc = qsd_op_begin(env, qsd, trans, qi, NULL);
817 if (rc == -EDQUOT || rc == -EINPROGRESS)
822 /* and one less inode for the current id */
823 qi->lqi_id.qid_uid = orig_id;;
825 /* can't get EDQUOT when reducing usage */
826 rc = qsd_op_begin(env, qsd, trans, qi, NULL);
827 if (rc == -EINPROGRESS)
832 /* block accounting */
833 qi->lqi_is_blk = true;
835 /* more blocks for the new owner ... */
836 qi->lqi_id.qid_uid = new_id;
837 qi->lqi_space = bspace;
838 rc = qsd_op_begin(env, qsd, trans, qi, NULL);
839 if (rc == -EDQUOT || rc == -EINPROGRESS)
844 /* and finally less blocks for the current owner */
845 qi->lqi_id.qid_uid = orig_id;
846 qi->lqi_space = -bspace;
847 rc = qsd_op_begin(env, qsd, trans, qi, NULL);
848 /* can't get EDQUOT when reducing usage */
849 if (rc == -EINPROGRESS)
854 static int osd_declare_attr_set(const struct lu_env *env,
855 struct dt_object *dt,
856 const struct lu_attr *attr,
857 struct thandle *handle)
859 struct osd_thread_info *info = osd_oti_get(env);
860 char *buf = osd_oti_get(env)->oti_str;
861 struct osd_object *obj = osd_dt_obj(dt);
862 struct osd_device *osd = osd_obj2dev(obj);
863 struct osd_thandle *oh;
869 if (!dt_object_exists(dt)) {
870 /* XXX: sanity check that object creation is declared */
874 LASSERT(handle != NULL);
875 LASSERT(osd_invariant(obj));
877 oh = container_of0(handle, struct osd_thandle, ot_super);
879 LASSERT(obj->oo_sa_hdl != NULL);
880 dmu_tx_hold_sa(oh->ot_tx, obj->oo_sa_hdl, 0);
882 sa_object_size(obj->oo_sa_hdl, &blksize, &bspace);
883 bspace = toqb(bspace * blksize);
885 if (attr && attr->la_valid & LA_UID) {
886 /* account for user inode tracking ZAP update */
887 dmu_tx_hold_bonus(oh->ot_tx, osd->od_iusr_oid);
888 dmu_tx_hold_zap(oh->ot_tx, osd->od_iusr_oid, TRUE, buf);
890 /* quota enforcement for user */
891 if (attr->la_uid != obj->oo_attr.la_uid) {
892 rc = qsd_transfer(env, osd->od_quota_slave,
893 &oh->ot_quota_trans, USRQUOTA,
894 obj->oo_attr.la_uid, attr->la_uid,
895 bspace, &info->oti_qi);
900 if (attr && attr->la_valid & LA_GID) {
901 /* account for user inode tracking ZAP update */
902 dmu_tx_hold_bonus(oh->ot_tx, osd->od_igrp_oid);
903 dmu_tx_hold_zap(oh->ot_tx, osd->od_igrp_oid, TRUE, buf);
905 /* quota enforcement for group */
906 if (attr->la_gid != obj->oo_attr.la_gid) {
907 rc = qsd_transfer(env, osd->od_quota_slave,
908 &oh->ot_quota_trans, GRPQUOTA,
909 obj->oo_attr.la_gid, attr->la_gid,
910 bspace, &info->oti_qi);
920 * Set the attributes of an object
922 * The transaction passed to this routine must have
923 * dmu_tx_hold_bonus(tx, oid) called and then assigned
924 * to a transaction group.
926 static int osd_attr_set(const struct lu_env *env, struct dt_object *dt,
927 const struct lu_attr *la, struct thandle *handle,
928 struct lustre_capa *capa)
930 struct osd_object *obj = osd_dt_obj(dt);
931 struct osd_device *osd = osd_obj2dev(obj);
932 udmu_objset_t *uos = &osd->od_objset;
933 struct osd_thandle *oh;
934 struct osa_attr *osa = &osd_oti_get(env)->oti_osa;
935 sa_bulk_attr_t *bulk;
940 LASSERT(handle != NULL);
941 LASSERT(dt_object_exists(dt));
942 LASSERT(osd_invariant(obj));
943 LASSERT(obj->oo_sa_hdl);
945 oh = container_of0(handle, struct osd_thandle, ot_super);
946 /* Assert that the transaction has been assigned to a
947 transaction group. */
948 LASSERT(oh->ot_tx->tx_txg != 0);
950 if (la->la_valid == 0)
953 OBD_ALLOC(bulk, sizeof(sa_bulk_attr_t) * 10);
957 /* do both accounting updates outside oo_attr_lock below */
958 if ((la->la_valid & LA_UID) && (la->la_uid != obj->oo_attr.la_uid)) {
959 /* Update user accounting. Failure isn't fatal, but we still
960 * log an error message */
961 rc = -zap_increment_int(osd->od_objset.os, osd->od_iusr_oid,
962 la->la_uid, 1, oh->ot_tx);
964 CERROR("%s: failed to update accounting ZAP for user "
965 "%d (%d)\n", osd->od_svname, la->la_uid, rc);
966 rc = -zap_increment_int(osd->od_objset.os, osd->od_iusr_oid,
967 obj->oo_attr.la_uid, -1, oh->ot_tx);
969 CERROR("%s: failed to update accounting ZAP for user "
970 "%d (%d)\n", osd->od_svname,
971 obj->oo_attr.la_uid, rc);
973 if ((la->la_valid & LA_GID) && (la->la_gid != obj->oo_attr.la_gid)) {
974 /* Update group accounting. Failure isn't fatal, but we still
975 * log an error message */
976 rc = -zap_increment_int(osd->od_objset.os, osd->od_igrp_oid,
977 la->la_gid, 1, oh->ot_tx);
979 CERROR("%s: failed to update accounting ZAP for user "
980 "%d (%d)\n", osd->od_svname, la->la_gid, rc);
981 rc = -zap_increment_int(osd->od_objset.os, osd->od_igrp_oid,
982 obj->oo_attr.la_gid, -1, oh->ot_tx);
984 CERROR("%s: failed to update accounting ZAP for user "
985 "%d (%d)\n", osd->od_svname,
986 obj->oo_attr.la_gid, rc);
989 write_lock(&obj->oo_attr_lock);
991 if (la->la_valid & LA_ATIME) {
992 osa->atime[0] = obj->oo_attr.la_atime = la->la_atime;
993 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(uos), NULL,
996 if (la->la_valid & LA_MTIME) {
997 osa->mtime[0] = obj->oo_attr.la_mtime = la->la_mtime;
998 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(uos), NULL,
1001 if (la->la_valid & LA_CTIME) {
1002 osa->ctime[0] = obj->oo_attr.la_ctime = la->la_ctime;
1003 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(uos), NULL,
1006 if (la->la_valid & LA_MODE) {
1007 /* mode is stored along with type, so read it first */
1008 obj->oo_attr.la_mode = (obj->oo_attr.la_mode & S_IFMT) |
1009 (la->la_mode & ~S_IFMT);
1010 osa->mode = obj->oo_attr.la_mode;
1011 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(uos), NULL,
1014 if (la->la_valid & LA_SIZE) {
1015 osa->size = obj->oo_attr.la_size = la->la_size;
1016 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(uos), NULL,
1019 if (la->la_valid & LA_NLINK) {
1020 osa->nlink = obj->oo_attr.la_nlink = la->la_nlink;
1021 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(uos), NULL,
1024 if (la->la_valid & LA_RDEV) {
1025 osa->rdev = obj->oo_attr.la_rdev = la->la_rdev;
1026 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_RDEV(uos), NULL,
1029 if (la->la_valid & LA_FLAGS) {
1030 osa->flags = attrs_fs2zfs(la->la_flags);
1031 /* many flags are not supported by zfs, so ensure a good cached
1033 obj->oo_attr.la_flags = attrs_zfs2fs(osa->flags);
1034 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(uos), NULL,
1037 if (la->la_valid & LA_UID) {
1038 osa->uid = obj->oo_attr.la_uid = la->la_uid;
1039 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(uos), NULL,
1042 if (la->la_valid & LA_GID) {
1043 osa->gid = obj->oo_attr.la_gid = la->la_gid;
1044 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(uos), NULL,
1047 obj->oo_attr.la_valid |= la->la_valid;
1048 write_unlock(&obj->oo_attr_lock);
1050 rc = osd_object_sa_bulk_update(obj, bulk, cnt, oh);
1052 OBD_FREE(bulk, sizeof(sa_bulk_attr_t) * 10);
1059 * XXX temporary solution.
1062 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
1063 struct dt_object *parent, struct dt_object *child,
1064 cfs_umode_t child_mode)
1068 memset(ah, 0, sizeof(*ah));
1069 ah->dah_parent = parent;
1070 ah->dah_mode = child_mode;
1073 static int osd_declare_object_create(const struct lu_env *env,
1074 struct dt_object *dt,
1075 struct lu_attr *attr,
1076 struct dt_allocation_hint *hint,
1077 struct dt_object_format *dof,
1078 struct thandle *handle)
1080 char *buf = osd_oti_get(env)->oti_str;
1081 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1082 struct osd_object *obj = osd_dt_obj(dt);
1083 struct osd_device *osd = osd_obj2dev(obj);
1084 struct osd_thandle *oh;
1091 switch (dof->dof_type) {
1095 if (obj->oo_dt.do_body_ops == NULL)
1096 obj->oo_dt.do_body_ops = &osd_body_ops;
1102 LASSERT(handle != NULL);
1103 oh = container_of0(handle, struct osd_thandle, ot_super);
1104 LASSERT(oh->ot_tx != NULL);
1106 switch (dof->dof_type) {
1109 /* for zap create */
1110 dmu_tx_hold_zap(oh->ot_tx, DMU_NEW_OBJECT, 1, NULL);
1115 /* first, we'll create new object */
1116 dmu_tx_hold_bonus(oh->ot_tx, DMU_NEW_OBJECT);
1124 /* and we'll add it to some mapping */
1125 zapid = osd_get_name_n_idx(env, osd, fid, buf);
1126 dmu_tx_hold_bonus(oh->ot_tx, zapid);
1127 dmu_tx_hold_zap(oh->ot_tx, zapid, TRUE, buf);
1129 /* we will also update inode accounting ZAPs */
1130 dmu_tx_hold_bonus(oh->ot_tx, osd->od_iusr_oid);
1131 dmu_tx_hold_zap(oh->ot_tx, osd->od_iusr_oid, TRUE, buf);
1132 dmu_tx_hold_bonus(oh->ot_tx, osd->od_igrp_oid);
1133 dmu_tx_hold_zap(oh->ot_tx, osd->od_igrp_oid, TRUE, buf);
1135 dmu_tx_hold_sa_create(oh->ot_tx, ZFS_SA_BASE_ATTR_SIZE);
1137 __osd_xattr_declare_set(env, obj, sizeof(struct lustre_mdt_attrs),
1138 XATTR_NAME_LMA, oh);
1140 rc = osd_declare_quota(env, osd, attr->la_uid, attr->la_gid, 1, oh,
1141 false, NULL, false);
1145 int __osd_attr_init(const struct lu_env *env, udmu_objset_t *uos, uint64_t oid,
1146 dmu_tx_t *tx, struct lu_attr *la, uint64_t parent)
1148 sa_bulk_attr_t *bulk;
1149 sa_handle_t *sa_hdl;
1150 struct osa_attr *osa = &osd_oti_get(env)->oti_osa;
1158 gen = dmu_tx_get_txg(tx);
1160 ZFS_TIME_ENCODE(&now, crtime);
1162 osa->atime[0] = la->la_atime;
1163 osa->ctime[0] = la->la_ctime;
1164 osa->mtime[0] = la->la_mtime;
1165 osa->mode = la->la_mode;
1166 osa->uid = la->la_uid;
1167 osa->gid = la->la_gid;
1168 osa->rdev = la->la_rdev;
1169 osa->nlink = la->la_nlink;
1170 osa->flags = attrs_fs2zfs(la->la_flags);
1171 osa->size = la->la_size;
1173 /* Now add in all of the "SA" attributes */
1174 rc = -sa_handle_get(uos->os, oid, NULL, SA_HDL_PRIVATE, &sa_hdl);
1178 OBD_ALLOC(bulk, sizeof(sa_bulk_attr_t) * 13);
1184 * we need to create all SA below upon object create.
1186 * XXX The attribute order matters since the accounting callback relies
1187 * on static offsets (i.e. SA_*_OFFSET, see zfs_space_delta_cb()) to
1188 * look up the UID/GID attributes. Moreover, the callback does not seem
1189 * to support the spill block.
1190 * We define attributes in the same order as SA_*_OFFSET in order to
1191 * work around the problem. See ORI-610.
1194 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(uos), NULL, &osa->mode, 8);
1195 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(uos), NULL, &osa->size, 8);
1196 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GEN(uos), NULL, &gen, 8);
1197 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(uos), NULL, &osa->uid, 8);
1198 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(uos), NULL, &osa->gid, 8);
1199 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_PARENT(uos), NULL, &parent, 8);
1200 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(uos), NULL, &osa->flags, 8);
1201 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(uos), NULL, osa->atime, 16);
1202 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(uos), NULL, osa->mtime, 16);
1203 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(uos), NULL, osa->ctime, 16);
1204 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CRTIME(uos), NULL, crtime, 16);
1205 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(uos), NULL, &osa->nlink, 8);
1206 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_RDEV(uos), NULL, &osa->rdev, 8);
1208 rc = -sa_replace_all_by_template(sa_hdl, bulk, cnt, tx);
1210 OBD_FREE(bulk, sizeof(sa_bulk_attr_t) * 13);
1212 sa_handle_destroy(sa_hdl);
1217 * The transaction passed to this routine must have
1218 * dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT) called and then assigned
1219 * to a transaction group.
1221 int __osd_object_create(const struct lu_env *env, udmu_objset_t *uos,
1222 dmu_buf_t **dbp, dmu_tx_t *tx, struct lu_attr *la,
1223 uint64_t parent, void *tag)
1229 spin_lock(&uos->lock);
1231 spin_unlock(&uos->lock);
1233 /* Assert that the transaction has been assigned to a
1234 transaction group. */
1235 LASSERT(tx->tx_txg != 0);
1237 /* Create a new DMU object. */
1238 oid = dmu_object_alloc(uos->os, DMU_OT_PLAIN_FILE_CONTENTS, 0,
1239 DMU_OT_SA, DN_MAX_BONUSLEN, tx);
1240 rc = -sa_buf_hold(uos->os, oid, tag, dbp);
1244 LASSERT(la->la_valid & LA_MODE);
1248 return __osd_attr_init(env, uos, oid, tx, la, parent);
1252 * The transaction passed to this routine must have
1253 * dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, ...) called and then assigned
1254 * to a transaction group.
1256 * Using ZAP_FLAG_HASH64 will force the ZAP to always be a FAT ZAP.
1257 * This is fine for directories today, because storing the FID in the dirent
1258 * will also require a FAT ZAP. If there is a new type of micro ZAP created
1259 * then we might need to re-evaluate the use of this flag and instead do
1260 * a conversion from the different internal ZAP hash formats being used. */
1261 int __osd_zap_create(const struct lu_env *env, udmu_objset_t *uos,
1262 dmu_buf_t **zap_dbp, dmu_tx_t *tx,
1263 struct lu_attr *la, uint64_t parent,
1264 void *tag, zap_flags_t flags)
1271 spin_lock(&uos->lock);
1273 spin_unlock(&uos->lock);
1275 /* Assert that the transaction has been assigned to a
1276 transaction group. */
1277 LASSERT(tx->tx_txg != 0);
1279 oid = zap_create_flags(uos->os, 0, flags | ZAP_FLAG_HASH64,
1280 DMU_OT_DIRECTORY_CONTENTS, 12, 12,
1281 DMU_OT_SA, DN_MAX_BONUSLEN, tx);
1283 rc = -sa_buf_hold(uos->os, oid, tag, zap_dbp);
1287 LASSERT(la->la_valid & LA_MODE);
1291 return __osd_attr_init(env, uos, oid, tx, la, parent);
1294 static dmu_buf_t *osd_mkidx(const struct lu_env *env, struct osd_device *osd,
1295 struct lu_attr *la, uint64_t parent,
1296 struct osd_thandle *oh)
1301 /* Index file should be created as regular file in order not to confuse
1302 * ZPL which could interpret them as directory.
1303 * We set ZAP_FLAG_UINT64_KEY to let ZFS know than we are going to use
1305 LASSERT(S_ISREG(la->la_mode));
1306 rc = __osd_zap_create(env, &osd->od_objset, &db, oh->ot_tx, la,
1307 parent, osd_obj_tag, ZAP_FLAG_UINT64_KEY);
1313 static dmu_buf_t *osd_mkdir(const struct lu_env *env, struct osd_device *osd,
1314 struct lu_attr *la, uint64_t parent,
1315 struct osd_thandle *oh)
1320 LASSERT(S_ISDIR(la->la_mode));
1321 rc = __osd_zap_create(env, &osd->od_objset, &db, oh->ot_tx, la,
1322 parent, osd_obj_tag, 0);
1328 static dmu_buf_t* osd_mkreg(const struct lu_env *env, struct osd_device *osd,
1329 struct lu_attr *la, uint64_t parent,
1330 struct osd_thandle *oh)
1335 LASSERT(S_ISREG(la->la_mode));
1336 rc = __osd_object_create(env, &osd->od_objset, &db, oh->ot_tx, la,
1337 parent, osd_obj_tag);
1342 * XXX: a hack, OST to use bigger blocksize. we need
1343 * a method in OSD API to control this from OFD/MDD
1345 if (!lu_device_is_md(osd2lu_dev(osd))) {
1346 rc = -dmu_object_set_blocksize(osd->od_objset.os,
1348 128 << 10, 0, oh->ot_tx);
1350 CERROR("%s: can't change blocksize: %d\n",
1351 osd->od_svname, rc);
1359 static dmu_buf_t *osd_mksym(const struct lu_env *env, struct osd_device *osd,
1360 struct lu_attr *la, uint64_t parent,
1361 struct osd_thandle *oh)
1366 LASSERT(S_ISLNK(la->la_mode));
1367 rc = __osd_object_create(env, &osd->od_objset, &db, oh->ot_tx, la,
1368 parent, osd_obj_tag);
1374 static dmu_buf_t *osd_mknod(const struct lu_env *env, struct osd_device *osd,
1375 struct lu_attr *la, uint64_t parent,
1376 struct osd_thandle *oh)
1381 la->la_valid = LA_MODE;
1382 if (S_ISCHR(la->la_mode) || S_ISBLK(la->la_mode))
1383 la->la_valid |= LA_RDEV;
1385 rc = __osd_object_create(env, &osd->od_objset, &db, oh->ot_tx, la,
1386 parent, osd_obj_tag);
1392 typedef dmu_buf_t *(*osd_obj_type_f)(const struct lu_env *env,
1393 struct osd_device *osd,
1396 struct osd_thandle *oh);
1398 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
1400 osd_obj_type_f result;
1426 * Primitives for directory (i.e. ZAP) handling
1428 static inline int osd_init_lma(const struct lu_env *env, struct osd_object *obj,
1429 const struct lu_fid *fid, struct osd_thandle *oh)
1431 struct osd_thread_info *info = osd_oti_get(env);
1432 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
1436 lustre_lma_init(lma, fid, 0);
1437 lustre_lma_swab(lma);
1439 buf.lb_len = sizeof(*lma);
1441 rc = osd_xattr_set_internal(env, obj, &buf, XATTR_NAME_LMA,
1442 LU_XATTR_CREATE, oh, BYPASS_CAPA);
1448 * Concurrency: @dt is write locked.
1450 static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
1451 struct lu_attr *attr,
1452 struct dt_allocation_hint *hint,
1453 struct dt_object_format *dof,
1456 struct zpl_direntry *zde = &osd_oti_get(env)->oti_zde.lzd_reg;
1457 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1458 struct osd_object *obj = osd_dt_obj(dt);
1459 struct osd_device *osd = osd_obj2dev(obj);
1460 char *buf = osd_oti_get(env)->oti_str;
1461 struct osd_thandle *oh;
1468 /* concurrent create declarations should not see
1469 * the object inconsistent (db, attr, etc).
1470 * in regular cases acquisition should be cheap */
1471 down(&obj->oo_guard);
1473 LASSERT(osd_invariant(obj));
1474 LASSERT(!dt_object_exists(dt));
1475 LASSERT(dof != NULL);
1477 LASSERT(th != NULL);
1478 oh = container_of0(th, struct osd_thandle, ot_super);
1481 * XXX missing: Quote handling.
1484 LASSERT(obj->oo_db == NULL);
1486 /* to follow ZFS on-disk format we need
1487 * to initialize parent dnode properly */
1489 if (hint && hint->dah_parent)
1490 zapid = osd_dt_obj(hint->dah_parent)->oo_db->db_object;
1492 db = osd_create_type_f(dof->dof_type)(env, osd, attr, zapid, oh);
1494 GOTO(out, rc = PTR_ERR(th));
1497 zde->zde_dnode = db->db_object;
1498 zde->zde_type = IFTODT(attr->la_mode & S_IFMT);
1500 zapid = osd_get_name_n_idx(env, osd, fid, buf);
1502 rc = -zap_add(osd->od_objset.os, zapid, buf, 8, 1, zde, oh->ot_tx);
1506 /* Add new object to inode accounting.
1507 * Errors are not considered as fatal */
1508 rc = -zap_increment_int(osd->od_objset.os, osd->od_iusr_oid,
1509 (attr->la_valid & LA_UID) ? attr->la_uid : 0, 1,
1512 CERROR("%s: failed to add "DFID" to accounting ZAP for usr %d "
1513 "(%d)\n", osd->od_svname, PFID(fid), attr->la_uid, rc);
1514 rc = -zap_increment_int(osd->od_objset.os, osd->od_igrp_oid,
1515 (attr->la_valid & LA_GID) ? attr->la_gid : 0, 1,
1518 CERROR("%s: failed to add "DFID" to accounting ZAP for grp %d "
1519 "(%d)\n", osd->od_svname, PFID(fid), attr->la_gid, rc);
1521 /* configure new osd object */
1523 rc = osd_object_init0(env, obj);
1524 LASSERT(ergo(rc == 0, dt_object_exists(dt)));
1525 LASSERT(osd_invariant(obj));
1527 rc = osd_init_lma(env, obj, fid, oh);
1529 CERROR("%s: can not set LMA on "DFID": rc = %d\n",
1530 osd->od_svname, PFID(fid), rc);
1531 /* ignore errors during LMA initialization */
1540 static int osd_declare_object_ref_add(const struct lu_env *env,
1541 struct dt_object *dt,
1544 return osd_declare_attr_set(env, dt, NULL, th);
1548 * Concurrency: @dt is write locked.
1550 static int osd_object_ref_add(const struct lu_env *env,
1551 struct dt_object *dt,
1552 struct thandle *handle)
1554 struct osd_object *obj = osd_dt_obj(dt);
1555 struct osd_thandle *oh;
1556 struct osd_device *osd = osd_obj2dev(obj);
1557 udmu_objset_t *uos = &osd->od_objset;
1563 LASSERT(osd_invariant(obj));
1564 LASSERT(dt_object_exists(dt));
1565 LASSERT(obj->oo_sa_hdl != NULL);
1567 oh = container_of0(handle, struct osd_thandle, ot_super);
1569 write_lock(&obj->oo_attr_lock);
1570 nlink = ++obj->oo_attr.la_nlink;
1571 write_unlock(&obj->oo_attr_lock);
1573 rc = osd_object_sa_update(obj, SA_ZPL_LINKS(uos), &nlink, 8, oh);
1577 static int osd_declare_object_ref_del(const struct lu_env *env,
1578 struct dt_object *dt,
1579 struct thandle *handle)
1581 return osd_declare_attr_set(env, dt, NULL, handle);
1585 * Concurrency: @dt is write locked.
1587 static int osd_object_ref_del(const struct lu_env *env,
1588 struct dt_object *dt,
1589 struct thandle *handle)
1591 struct osd_object *obj = osd_dt_obj(dt);
1592 struct osd_thandle *oh;
1593 struct osd_device *osd = osd_obj2dev(obj);
1594 udmu_objset_t *uos = &osd->od_objset;
1600 LASSERT(osd_invariant(obj));
1601 LASSERT(dt_object_exists(dt));
1602 LASSERT(obj->oo_sa_hdl != NULL);
1604 oh = container_of0(handle, struct osd_thandle, ot_super);
1605 LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
1607 write_lock(&obj->oo_attr_lock);
1608 nlink = --obj->oo_attr.la_nlink;
1609 write_unlock(&obj->oo_attr_lock);
1611 rc = osd_object_sa_update(obj, SA_ZPL_LINKS(uos), &nlink, 8, oh);
1615 static int capa_is_sane(const struct lu_env *env, struct osd_device *dev,
1616 struct lustre_capa *capa, struct lustre_capa_key *keys)
1618 struct osd_thread_info *oti = osd_oti_get(env);
1619 struct obd_capa *oc;
1623 oc = capa_lookup(dev->od_capa_hash, capa, 0);
1625 if (capa_is_expired(oc)) {
1626 DEBUG_CAPA(D_ERROR, capa, "expired");
1633 spin_lock(&capa_lock);
1634 for (i = 0; i < 2; i++) {
1635 if (keys[i].lk_keyid == capa->lc_keyid) {
1636 oti->oti_capa_key = keys[i];
1640 spin_unlock(&capa_lock);
1643 DEBUG_CAPA(D_ERROR, capa, "no matched capa key");
1647 rc = capa_hmac(oti->oti_capa.lc_hmac, capa, oti->oti_capa_key.lk_key);
1650 if (memcmp(oti->oti_capa.lc_hmac, capa->lc_hmac, sizeof(capa->lc_hmac)))
1652 DEBUG_CAPA(D_ERROR, capa, "HMAC mismatch");
1656 oc = capa_add(dev->od_capa_hash, capa);
1662 static int osd_object_auth(const struct lu_env *env, struct dt_object *dt,
1663 struct lustre_capa *capa, __u64 opc)
1665 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1666 struct osd_device *dev = osd_dev(dt->do_lu.lo_dev);
1669 if (!dev->od_fl_capa)
1672 if (capa == BYPASS_CAPA)
1676 CERROR("no capability is provided for fid "DFID"\n", PFID(fid));
1680 if (!lu_fid_eq(fid, &capa->lc_fid)) {
1681 DEBUG_CAPA(D_ERROR, capa, "fid "DFID" mismatch with",PFID(fid));
1685 if (!capa_opc_supported(capa, opc)) {
1686 DEBUG_CAPA(D_ERROR, capa, "opc "LPX64" not supported by", opc);
1690 if ((rc = capa_is_sane(env, dev, capa, dev->od_capa_keys))) {
1691 DEBUG_CAPA(D_ERROR, capa, "insane (rc %d)", rc);
1698 static struct obd_capa *osd_capa_get(const struct lu_env *env,
1699 struct dt_object *dt,
1700 struct lustre_capa *old,
1703 struct osd_thread_info *info = osd_oti_get(env);
1704 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1705 struct osd_object *obj = osd_dt_obj(dt);
1706 struct osd_device *dev = osd_obj2dev(obj);
1707 struct lustre_capa_key *key = &info->oti_capa_key;
1708 struct lustre_capa *capa = &info->oti_capa;
1709 struct obd_capa *oc;
1713 if (!dev->od_fl_capa)
1714 RETURN(ERR_PTR(-ENOENT));
1716 LASSERT(dt_object_exists(dt));
1717 LASSERT(osd_invariant(obj));
1719 /* renewal sanity check */
1720 if (old && osd_object_auth(env, dt, old, opc))
1721 RETURN(ERR_PTR(-EACCES));
1723 capa->lc_fid = *fid;
1726 capa->lc_flags = dev->od_capa_alg << 24;
1727 capa->lc_timeout = dev->od_capa_timeout;
1728 capa->lc_expiry = 0;
1730 oc = capa_lookup(dev->od_capa_hash, capa, 1);
1732 LASSERT(!capa_is_expired(oc));
1736 spin_lock(&capa_lock);
1737 *key = dev->od_capa_keys[1];
1738 spin_unlock(&capa_lock);
1740 capa->lc_keyid = key->lk_keyid;
1741 capa->lc_expiry = cfs_time_current_sec() + dev->od_capa_timeout;
1743 rc = capa_hmac(capa->lc_hmac, capa, key->lk_key);
1745 DEBUG_CAPA(D_ERROR, capa, "HMAC failed: %d for", rc);
1747 RETURN(ERR_PTR(rc));
1750 oc = capa_add(dev->od_capa_hash, capa);
1754 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt)
1756 struct osd_device *osd = osd_obj2dev(osd_dt_obj(dt));
1759 /* XXX: no other option than syncing the whole filesystem until we
1761 txg_wait_synced(dmu_objset_pool(osd->od_objset.os), 0ULL);
1766 static struct dt_object_operations osd_obj_ops = {
1767 .do_read_lock = osd_object_read_lock,
1768 .do_write_lock = osd_object_write_lock,
1769 .do_read_unlock = osd_object_read_unlock,
1770 .do_write_unlock = osd_object_write_unlock,
1771 .do_write_locked = osd_object_write_locked,
1772 .do_attr_get = osd_attr_get,
1773 .do_declare_attr_set = osd_declare_attr_set,
1774 .do_attr_set = osd_attr_set,
1775 .do_ah_init = osd_ah_init,
1776 .do_declare_create = osd_declare_object_create,
1777 .do_create = osd_object_create,
1778 .do_declare_destroy = osd_declare_object_destroy,
1779 .do_destroy = osd_object_destroy,
1780 .do_index_try = osd_index_try,
1781 .do_declare_ref_add = osd_declare_object_ref_add,
1782 .do_ref_add = osd_object_ref_add,
1783 .do_declare_ref_del = osd_declare_object_ref_del,
1784 .do_ref_del = osd_object_ref_del,
1785 .do_xattr_get = osd_xattr_get,
1786 .do_declare_xattr_set = osd_declare_xattr_set,
1787 .do_xattr_set = osd_xattr_set,
1788 .do_declare_xattr_del = osd_declare_xattr_del,
1789 .do_xattr_del = osd_xattr_del,
1790 .do_xattr_list = osd_xattr_list,
1791 .do_capa_get = osd_capa_get,
1792 .do_object_sync = osd_object_sync,
1795 static struct lu_object_operations osd_lu_obj_ops = {
1796 .loo_object_init = osd_object_init,
1797 .loo_object_delete = osd_object_delete,
1798 .loo_object_release = osd_object_release,
1799 .loo_object_free = osd_object_free,
1800 .loo_object_print = osd_object_print,
1801 .loo_object_invariant = osd_object_invariant,
1804 static int osd_otable_it_attr_get(const struct lu_env *env,
1805 struct dt_object *dt,
1806 struct lu_attr *attr,
1807 struct lustre_capa *capa)
1813 static struct dt_object_operations osd_obj_otable_it_ops = {
1814 .do_attr_get = osd_otable_it_attr_get,
1815 .do_index_try = osd_index_try,