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, 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 <lustre_fsfilt.h>
53 #include <obd_support.h>
54 #include <lustre_net.h>
56 #include <obd_class.h>
57 #include <lustre_disk.h>
58 #include <lustre_fid.h>
60 #include "osd_internal.h"
62 #include <sys/dnode.h>
67 #include <sys/spa_impl.h>
68 #include <sys/zfs_znode.h>
69 #include <sys/dmu_tx.h>
70 #include <sys/dmu_objset.h>
71 #include <sys/dsl_prop.h>
72 #include <sys/sa_impl.h>
75 static char *osd_obj_tag = "osd_object";
77 static struct dt_object_operations osd_obj_ops;
78 static struct lu_object_operations osd_lu_obj_ops;
79 extern struct dt_body_operations osd_body_ops;
80 static struct dt_object_operations osd_obj_otable_it_ops;
82 extern cfs_mem_cache_t *osd_object_kmem;
85 osd_object_sa_fini(struct osd_object *obj)
88 sa_handle_destroy(obj->oo_sa_hdl);
89 obj->oo_sa_hdl = NULL;
94 osd_object_sa_init(struct osd_object *obj, udmu_objset_t *uos)
98 LASSERT(obj->oo_sa_hdl == NULL);
99 LASSERT(obj->oo_db != NULL);
101 rc = -sa_handle_get(uos->os, obj->oo_db->db_object, obj,
102 SA_HDL_PRIVATE, &obj->oo_sa_hdl);
106 /* Cache the xattr object id, valid for the life of the object */
107 rc = -sa_lookup(obj->oo_sa_hdl, SA_ZPL_XATTR(uos), &obj->oo_xattr, 8);
109 obj->oo_xattr = ZFS_NO_OBJECT;
112 osd_object_sa_fini(obj);
119 * Add object to list of dirty objects in tx handle.
122 osd_object_sa_dirty_add(struct osd_object *obj, struct osd_thandle *oh)
124 if (!cfs_list_empty(&obj->oo_sa_linkage))
127 down(&oh->ot_sa_lock);
128 write_lock(&obj->oo_attr_lock);
129 if (likely(cfs_list_empty(&obj->oo_sa_linkage)))
130 cfs_list_add(&obj->oo_sa_linkage, &oh->ot_sa_list);
131 write_unlock(&obj->oo_attr_lock);
136 * Release spill block dbuf hold for all dirty SAs.
138 void osd_object_sa_dirty_rele(struct osd_thandle *oh)
140 struct osd_object *obj;
142 down(&oh->ot_sa_lock);
143 while (!cfs_list_empty(&oh->ot_sa_list)) {
144 obj = cfs_list_entry(oh->ot_sa_list.next,
145 struct osd_object, oo_sa_linkage);
146 sa_spill_rele(obj->oo_sa_hdl);
147 write_lock(&obj->oo_attr_lock);
148 cfs_list_del_init(&obj->oo_sa_linkage);
149 write_unlock(&obj->oo_attr_lock);
155 * Update the SA and add the object to the dirty list.
157 int osd_object_sa_update(struct osd_object *obj, sa_attr_type_t type,
158 void *buf, uint32_t buflen, struct osd_thandle *oh)
162 LASSERT(obj->oo_sa_hdl != NULL);
163 LASSERT(oh->ot_tx != NULL);
165 rc = -sa_update(obj->oo_sa_hdl, type, buf, buflen, oh->ot_tx);
166 osd_object_sa_dirty_add(obj, oh);
172 * Bulk update the SA and add the object to the dirty list.
175 osd_object_sa_bulk_update(struct osd_object *obj, sa_bulk_attr_t *attrs,
176 int count, struct osd_thandle *oh)
180 LASSERT(obj->oo_sa_hdl != NULL);
181 LASSERT(oh->ot_tx != NULL);
183 rc = -sa_bulk_update(obj->oo_sa_hdl, attrs, count, oh->ot_tx);
184 osd_object_sa_dirty_add(obj, oh);
190 * Retrieve the attributes of a DMU object
192 int __osd_object_attr_get(const struct lu_env *env, udmu_objset_t *uos,
193 struct osd_object *obj, struct lu_attr *la)
195 struct osa_attr *osa = &osd_oti_get(env)->oti_osa;
197 sa_bulk_attr_t *bulk;
202 LASSERT(obj->oo_db != NULL);
204 rc = -sa_handle_get(uos->os, obj->oo_db->db_object, NULL,
205 SA_HDL_PRIVATE, &sa_hdl);
209 OBD_ALLOC(bulk, sizeof(sa_bulk_attr_t) * 9);
211 GOTO(out_sa, rc = -ENOMEM);
213 la->la_valid |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE | LA_TYPE |
214 LA_SIZE | LA_UID | LA_GID | LA_FLAGS | LA_NLINK;
216 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(uos), NULL, osa->atime, 16);
217 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(uos), NULL, osa->mtime, 16);
218 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(uos), NULL, osa->ctime, 16);
219 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(uos), NULL, &osa->mode, 8);
220 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(uos), NULL, &osa->size, 8);
221 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(uos), NULL, &osa->nlink, 8);
222 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(uos), NULL, &osa->uid, 8);
223 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(uos), NULL, &osa->gid, 8);
224 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(uos), NULL, &osa->flags, 8);
226 rc = -sa_bulk_lookup(sa_hdl, bulk, cnt);
230 la->la_atime = osa->atime[0];
231 la->la_mtime = osa->mtime[0];
232 la->la_ctime = osa->ctime[0];
233 la->la_mode = osa->mode;
234 la->la_uid = osa->uid;
235 la->la_gid = osa->gid;
236 la->la_nlink = osa->nlink;
237 la->la_flags = osa->flags;
238 la->la_size = osa->size;
240 if (S_ISCHR(la->la_mode) || S_ISBLK(la->la_mode)) {
241 rc = -sa_lookup(sa_hdl, SA_ZPL_RDEV(uos), &osa->rdev, 8);
244 la->la_rdev = osa->rdev;
245 la->la_valid |= LA_RDEV;
248 OBD_FREE(bulk, sizeof(sa_bulk_attr_t) * 9);
250 sa_handle_destroy(sa_hdl);
255 int __osd_obj2dbuf(const struct lu_env *env, objset_t *os,
256 uint64_t oid, dmu_buf_t **dbp, void *tag)
258 dmu_object_info_t *doi = &osd_oti_get(env)->oti_doi;
263 rc = -sa_buf_hold(os, oid, tag, dbp);
267 dmu_object_info_from_db(*dbp, doi);
268 if (unlikely (oid != DMU_USERUSED_OBJECT &&
269 oid != DMU_GROUPUSED_OBJECT && doi->doi_bonus_type != DMU_OT_SA)) {
270 sa_buf_rele(*dbp, tag);
276 LASSERT((*dbp)->db_object == oid);
277 LASSERT((*dbp)->db_offset == -1);
278 LASSERT((*dbp)->db_data != NULL);
284 * Concurrency: no concurrent access is possible that early in object
287 struct lu_object *osd_object_alloc(const struct lu_env *env,
288 const struct lu_object_header *hdr,
291 struct osd_object *mo;
293 OBD_SLAB_ALLOC_PTR_GFP(mo, osd_object_kmem, CFS_ALLOC_IO);
297 l = &mo->oo_dt.do_lu;
298 dt_object_init(&mo->oo_dt, NULL, d);
299 mo->oo_dt.do_ops = &osd_obj_ops;
300 l->lo_ops = &osd_lu_obj_ops;
301 CFS_INIT_LIST_HEAD(&mo->oo_sa_linkage);
302 init_rwsem(&mo->oo_sem);
303 sema_init(&mo->oo_guard, 1);
304 rwlock_init(&mo->oo_attr_lock);
312 * Concurrency: shouldn't matter.
314 int osd_object_init0(const struct lu_env *env, struct osd_object *obj)
316 struct osd_device *osd = osd_obj2dev(obj);
317 udmu_objset_t *uos = &osd->od_objset;
318 const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
322 if (obj->oo_db == NULL)
327 rc = osd_object_sa_init(obj, uos);
331 /* cache attrs in object */
332 rc = __osd_object_attr_get(env, &osd->od_objset,
337 if (likely(!fid_is_acct(fid)))
338 /* no body operations for accounting objects */
339 obj->oo_dt.do_body_ops = &osd_body_ops;
342 * initialize object before marking it existing
344 obj->oo_dt.do_lu.lo_header->loh_attr |= obj->oo_attr.la_mode & S_IFMT;
347 obj->oo_dt.do_lu.lo_header->loh_attr |= LOHA_EXISTS;
353 * Concurrency: no concurrent access is possible that early in object
356 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
357 const struct lu_object_conf *conf)
359 struct osd_object *obj = osd_obj(l);
360 struct osd_device *osd = osd_obj2dev(obj);
365 LASSERT(osd_invariant(obj));
367 if (fid_is_otable_it(&l->lo_header->loh_fid)) {
368 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
369 l->lo_header->loh_attr |= LOHA_EXISTS;
373 rc = osd_fid_lookup(env, osd, lu_object_fid(l), &oid);
375 LASSERT(obj->oo_db == NULL);
376 rc = __osd_obj2dbuf(env, osd->od_objset.os, oid,
377 &obj->oo_db, osd_obj_tag);
380 rc = osd_object_init0(env, obj);
382 CERROR("%s: lookup "DFID"/"LPX64" failed: rc = %d\n",
383 osd->od_svname, PFID(lu_object_fid(l)), oid, rc);
385 } else if (rc == -ENOENT) {
388 LASSERT(osd_invariant(obj));
393 * Concurrency: no concurrent access is possible that late in object
396 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
398 struct osd_object *obj = osd_obj(l);
400 LASSERT(osd_invariant(obj));
402 dt_object_fini(&obj->oo_dt);
403 OBD_SLAB_FREE_PTR(obj, osd_object_kmem);
406 static void __osd_declare_object_destroy(const struct lu_env *env,
407 struct osd_object *obj,
408 struct osd_thandle *oh)
410 struct osd_device *osd = osd_obj2dev(obj);
411 udmu_objset_t *uos = &osd->od_objset;
412 dmu_buf_t *db = obj->oo_db;
413 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
414 uint64_t oid = db->db_object, xid;
415 dmu_tx_t *tx = oh->ot_tx;
419 dmu_tx_hold_free(tx, oid, 0, DMU_OBJECT_END);
421 /* zap holding xattrs */
422 if (obj->oo_xattr != ZFS_NO_OBJECT) {
425 dmu_tx_hold_free(tx, oid, 0, DMU_OBJECT_END);
427 rc = -udmu_zap_cursor_init(&zc, uos, oid, 0);
431 while ((rc = -zap_cursor_retrieve(zc, za)) == 0) {
432 BUG_ON(za->za_integer_length != sizeof(uint64_t));
433 BUG_ON(za->za_num_integers != 1);
435 rc = -zap_lookup(uos->os, obj->oo_xattr, za->za_name,
436 sizeof(uint64_t), 1, &xid);
438 CERROR("%s: xattr lookup failed: rc = %d\n",
442 dmu_tx_hold_free(tx, xid, 0, DMU_OBJECT_END);
444 zap_cursor_advance(zc);
449 udmu_zap_cursor_fini(zc);
452 if (rc && tx->tx_err == 0)
456 static int osd_declare_object_destroy(const struct lu_env *env,
457 struct dt_object *dt,
460 char *buf = osd_oti_get(env)->oti_str;
461 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
462 struct osd_object *obj = osd_dt_obj(dt);
463 struct osd_device *osd = osd_obj2dev(obj);
464 struct osd_thandle *oh;
470 LASSERT(dt_object_exists(dt));
472 oh = container_of0(th, struct osd_thandle, ot_super);
473 LASSERT(oh->ot_tx != NULL);
475 /* declare that we'll destroy the object */
476 __osd_declare_object_destroy(env, obj, oh);
478 /* declare that we'll remove object from fid-dnode mapping */
479 zapid = osd_get_name_n_idx(env, osd, fid, buf);
480 dmu_tx_hold_bonus(oh->ot_tx, zapid);
481 dmu_tx_hold_zap(oh->ot_tx, zapid, 0, buf);
483 /* declare that we'll remove object from inode accounting ZAPs */
484 dmu_tx_hold_bonus(oh->ot_tx, osd->od_iusr_oid);
485 dmu_tx_hold_zap(oh->ot_tx, osd->od_iusr_oid, 0, buf);
486 dmu_tx_hold_bonus(oh->ot_tx, osd->od_igrp_oid);
487 dmu_tx_hold_zap(oh->ot_tx, osd->od_igrp_oid, 0, buf);
490 rc = osd_declare_quota(env, osd, obj->oo_attr.la_uid,
491 obj->oo_attr.la_gid, -1, oh, false, NULL, false);
495 /* data to be truncated */
496 rc = osd_declare_quota(env, osd, obj->oo_attr.la_uid,
497 obj->oo_attr.la_gid, 0, oh, true, NULL, false);
501 int __osd_object_free(udmu_objset_t *uos, uint64_t oid, dmu_tx_t *tx)
503 LASSERT(uos->objects != 0);
504 spin_lock(&uos->lock);
506 spin_unlock(&uos->lock);
508 return -dmu_object_free(uos->os, oid, tx);
512 * Delete a DMU object
514 * The transaction passed to this routine must have
515 * dmu_tx_hold_free(tx, oid, 0, DMU_OBJECT_END) called
516 * and then assigned to a transaction group.
518 * This will release db and set it to NULL to prevent further dbuf releases.
520 static int __osd_object_destroy(const struct lu_env *env,
521 struct osd_object *obj,
522 dmu_tx_t *tx, void *tag)
524 struct osd_device *osd = osd_obj2dev(obj);
525 udmu_objset_t *uos = &osd->od_objset;
527 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
531 /* Assert that the transaction has been assigned to a
532 transaction group. */
533 LASSERT(tx->tx_txg != 0);
535 /* zap holding xattrs */
536 if (obj->oo_xattr != ZFS_NO_OBJECT) {
537 rc = -udmu_zap_cursor_init(&zc, uos, obj->oo_xattr, 0);
540 while ((rc = -zap_cursor_retrieve(zc, za)) == 0) {
541 BUG_ON(za->za_integer_length != sizeof(uint64_t));
542 BUG_ON(za->za_num_integers != 1);
544 rc = -zap_lookup(uos->os, obj->oo_xattr, za->za_name,
545 sizeof(uint64_t), 1, &xid);
547 CERROR("%s: lookup xattr %s failed: rc = %d\n",
548 osd->od_svname, za->za_name, rc);
551 rc = __osd_object_free(uos, xid, tx);
553 CERROR("%s: fetch xattr %s failed: rc = %d\n",
554 osd->od_svname, za->za_name, rc);
555 zap_cursor_advance(zc);
557 udmu_zap_cursor_fini(zc);
559 rc = __osd_object_free(uos, obj->oo_xattr, tx);
561 CERROR("%s: freeing xattr failed: rc = %d\n",
565 return __osd_object_free(uos, obj->oo_db->db_object, tx);
568 static int osd_object_destroy(const struct lu_env *env,
569 struct dt_object *dt,
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 LASSERT(obj->oo_db != NULL);
582 LASSERT(dt_object_exists(dt));
583 LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
585 oh = container_of0(th, struct osd_thandle, ot_super);
587 LASSERT(oh->ot_tx != NULL);
589 zapid = osd_get_name_n_idx(env, osd, fid, buf);
591 /* remove obj ref from index dir (it depends) */
592 rc = -zap_remove(osd->od_objset.os, zapid, buf, oh->ot_tx);
594 CERROR("%s: zap_remove() failed: rc = %d\n",
599 /* Remove object from inode accounting. It is not fatal for the destroy
600 * operation if something goes wrong while updating accounting, but we
601 * still log an error message to notify the administrator */
602 rc = -zap_increment_int(osd->od_objset.os, osd->od_iusr_oid,
603 obj->oo_attr.la_uid, -1, oh->ot_tx);
605 CERROR("%s: failed to remove "DFID" from accounting ZAP for usr"
606 " %d: rc = %d\n", osd->od_svname, PFID(fid),
607 obj->oo_attr.la_uid, rc);
608 rc = -zap_increment_int(osd->od_objset.os, osd->od_igrp_oid,
609 obj->oo_attr.la_gid, -1, oh->ot_tx);
611 CERROR("%s: failed to remove "DFID" from accounting ZAP for grp"
612 " %d: rc = %d\n", osd->od_svname, PFID(fid),
613 obj->oo_attr.la_gid, rc);
616 rc = __osd_object_destroy(env, obj, oh->ot_tx, osd_obj_tag);
618 CERROR("%s: __osd_object_destroy() failed: rc = %d\n",
624 /* not needed in the cache anymore */
625 set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
630 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
632 struct osd_object *obj = osd_obj(l);
634 if (obj->oo_db != NULL) {
635 osd_object_sa_fini(obj);
636 if (obj->oo_sa_xattr) {
637 nvlist_free(obj->oo_sa_xattr);
638 obj->oo_sa_xattr = NULL;
640 sa_buf_rele(obj->oo_db, osd_obj_tag);
641 cfs_list_del(&obj->oo_sa_linkage);
647 * Concurrency: ->loo_object_release() is called under site spin-lock.
649 static void osd_object_release(const struct lu_env *env,
655 * Concurrency: shouldn't matter.
657 static int osd_object_print(const struct lu_env *env, void *cookie,
658 lu_printer_t p, const struct lu_object *l)
660 struct osd_object *o = osd_obj(l);
662 return (*p)(env, cookie, LUSTRE_OSD_ZFS_NAME"-object@%p", o);
665 static void osd_object_read_lock(const struct lu_env *env,
666 struct dt_object *dt, unsigned role)
668 struct osd_object *obj = osd_dt_obj(dt);
670 LASSERT(osd_invariant(obj));
672 down_read(&obj->oo_sem);
675 static void osd_object_write_lock(const struct lu_env *env,
676 struct dt_object *dt, unsigned role)
678 struct osd_object *obj = osd_dt_obj(dt);
680 LASSERT(osd_invariant(obj));
682 down_write(&obj->oo_sem);
685 static void osd_object_read_unlock(const struct lu_env *env,
686 struct dt_object *dt)
688 struct osd_object *obj = osd_dt_obj(dt);
690 LASSERT(osd_invariant(obj));
691 up_read(&obj->oo_sem);
694 static void osd_object_write_unlock(const struct lu_env *env,
695 struct dt_object *dt)
697 struct osd_object *obj = osd_dt_obj(dt);
699 LASSERT(osd_invariant(obj));
700 up_write(&obj->oo_sem);
703 static int osd_object_write_locked(const struct lu_env *env,
704 struct dt_object *dt)
706 struct osd_object *obj = osd_dt_obj(dt);
709 LASSERT(osd_invariant(obj));
711 if (down_write_trylock(&obj->oo_sem)) {
713 up_write(&obj->oo_sem);
718 static int osd_attr_get(const struct lu_env *env,
719 struct dt_object *dt,
720 struct lu_attr *attr,
721 struct lustre_capa *capa)
723 struct osd_object *obj = osd_dt_obj(dt);
727 LASSERT(dt_object_exists(dt));
728 LASSERT(osd_invariant(obj));
731 read_lock(&obj->oo_attr_lock);
732 *attr = obj->oo_attr;
733 read_unlock(&obj->oo_attr_lock);
735 /* with ZFS_DEBUG zrl_add_debug() called by DB_DNODE_ENTER()
736 * from within sa_object_size() can block on a mutex, so
737 * we can't call sa_object_size() holding rwlock */
738 sa_object_size(obj->oo_sa_hdl, &blksize, &blocks);
739 /* we do not control size of indices, so always calculate
740 * it from number of blocks reported by DMU */
741 if (S_ISDIR(attr->la_mode))
742 attr->la_size = 512 * blocks;
743 /* Block size may be not set; suggest maximal I/O transfers. */
745 blksize = 1ULL << SPA_MAXBLOCKSHIFT;
747 attr->la_blksize = blksize;
748 attr->la_blocks = blocks;
749 attr->la_valid |= LA_BLOCKS | LA_BLKSIZE;
754 /* Simple wrapper on top of qsd API which implement quota transfer for osd
755 * setattr needs. As a reminder, only the root user can change ownership of
756 * a file, that's why EDQUOT & EINPROGRESS errors are discarded */
757 static inline int qsd_transfer(const struct lu_env *env,
758 struct qsd_instance *qsd,
759 struct lquota_trans *trans, int qtype,
760 __u64 orig_id, __u64 new_id, __u64 bspace,
761 struct lquota_id_info *qi)
765 if (unlikely(qsd == NULL))
768 LASSERT(qtype >= 0 && qtype < MAXQUOTAS);
769 qi->lqi_type = qtype;
771 /* inode accounting */
772 qi->lqi_is_blk = false;
774 /* one more inode for the new owner ... */
775 qi->lqi_id.qid_uid = new_id;
777 rc = qsd_op_begin(env, qsd, trans, qi, NULL);
778 if (rc == -EDQUOT || rc == -EINPROGRESS)
783 /* and one less inode for the current id */
784 qi->lqi_id.qid_uid = orig_id;;
786 /* can't get EDQUOT when reducing usage */
787 rc = qsd_op_begin(env, qsd, trans, qi, NULL);
788 if (rc == -EINPROGRESS)
793 /* block accounting */
794 qi->lqi_is_blk = true;
796 /* more blocks for the new owner ... */
797 qi->lqi_id.qid_uid = new_id;
798 qi->lqi_space = bspace;
799 rc = qsd_op_begin(env, qsd, trans, qi, NULL);
800 if (rc == -EDQUOT || rc == -EINPROGRESS)
805 /* and finally less blocks for the current owner */
806 qi->lqi_id.qid_uid = orig_id;
807 qi->lqi_space = -bspace;
808 rc = qsd_op_begin(env, qsd, trans, qi, NULL);
809 /* can't get EDQUOT when reducing usage */
810 if (rc == -EINPROGRESS)
815 static int osd_declare_attr_set(const struct lu_env *env,
816 struct dt_object *dt,
817 const struct lu_attr *attr,
818 struct thandle *handle)
820 struct osd_thread_info *info = osd_oti_get(env);
821 char *buf = osd_oti_get(env)->oti_str;
822 struct osd_object *obj = osd_dt_obj(dt);
823 struct osd_device *osd = osd_obj2dev(obj);
824 struct osd_thandle *oh;
830 if (!dt_object_exists(dt)) {
831 /* XXX: sanity check that object creation is declared */
835 LASSERT(handle != NULL);
836 LASSERT(osd_invariant(obj));
838 oh = container_of0(handle, struct osd_thandle, ot_super);
840 LASSERT(obj->oo_sa_hdl != NULL);
841 dmu_tx_hold_sa(oh->ot_tx, obj->oo_sa_hdl, 0);
843 sa_object_size(obj->oo_sa_hdl, &blksize, &bspace);
844 bspace = toqb(bspace * blksize);
846 if (attr && attr->la_valid & LA_UID) {
847 /* account for user inode tracking ZAP update */
848 dmu_tx_hold_bonus(oh->ot_tx, osd->od_iusr_oid);
849 dmu_tx_hold_zap(oh->ot_tx, osd->od_iusr_oid, TRUE, buf);
851 /* quota enforcement for user */
852 if (attr->la_uid != obj->oo_attr.la_uid) {
853 rc = qsd_transfer(env, osd->od_quota_slave,
854 &oh->ot_quota_trans, USRQUOTA,
855 obj->oo_attr.la_uid, attr->la_uid,
856 bspace, &info->oti_qi);
861 if (attr && attr->la_valid & LA_GID) {
862 /* account for user inode tracking ZAP update */
863 dmu_tx_hold_bonus(oh->ot_tx, osd->od_igrp_oid);
864 dmu_tx_hold_zap(oh->ot_tx, osd->od_igrp_oid, TRUE, buf);
866 /* quota enforcement for group */
867 if (attr->la_gid != obj->oo_attr.la_gid) {
868 rc = qsd_transfer(env, osd->od_quota_slave,
869 &oh->ot_quota_trans, GRPQUOTA,
870 obj->oo_attr.la_gid, attr->la_gid,
871 bspace, &info->oti_qi);
881 * Set the attributes of an object
883 * The transaction passed to this routine must have
884 * dmu_tx_hold_bonus(tx, oid) called and then assigned
885 * to a transaction group.
887 static int osd_attr_set(const struct lu_env *env, struct dt_object *dt,
888 const struct lu_attr *la, struct thandle *handle,
889 struct lustre_capa *capa)
891 struct osd_object *obj = osd_dt_obj(dt);
892 struct osd_device *osd = osd_obj2dev(obj);
893 udmu_objset_t *uos = &osd->od_objset;
894 struct osd_thandle *oh;
895 struct osa_attr *osa = &osd_oti_get(env)->oti_osa;
896 sa_bulk_attr_t *bulk;
901 LASSERT(handle != NULL);
902 LASSERT(dt_object_exists(dt));
903 LASSERT(osd_invariant(obj));
904 LASSERT(obj->oo_sa_hdl);
906 oh = container_of0(handle, struct osd_thandle, ot_super);
907 /* Assert that the transaction has been assigned to a
908 transaction group. */
909 LASSERT(oh->ot_tx->tx_txg != 0);
911 if (la->la_valid == 0)
914 OBD_ALLOC(bulk, sizeof(sa_bulk_attr_t) * 10);
918 /* do both accounting updates outside oo_attr_lock below */
919 if ((la->la_valid & LA_UID) && (la->la_uid != obj->oo_attr.la_uid)) {
920 /* Update user accounting. Failure isn't fatal, but we still
921 * log an error message */
922 rc = -zap_increment_int(osd->od_objset.os, osd->od_iusr_oid,
923 la->la_uid, 1, oh->ot_tx);
925 CERROR("%s: failed to update accounting ZAP for user "
926 "%d (%d)\n", osd->od_svname, la->la_uid, rc);
927 rc = -zap_increment_int(osd->od_objset.os, osd->od_iusr_oid,
928 obj->oo_attr.la_uid, -1, oh->ot_tx);
930 CERROR("%s: failed to update accounting ZAP for user "
931 "%d (%d)\n", osd->od_svname,
932 obj->oo_attr.la_uid, rc);
934 if ((la->la_valid & LA_GID) && (la->la_gid != obj->oo_attr.la_gid)) {
935 /* Update group accounting. Failure isn't fatal, but we still
936 * log an error message */
937 rc = -zap_increment_int(osd->od_objset.os, osd->od_igrp_oid,
938 la->la_gid, 1, oh->ot_tx);
940 CERROR("%s: failed to update accounting ZAP for user "
941 "%d (%d)\n", osd->od_svname, la->la_gid, rc);
942 rc = -zap_increment_int(osd->od_objset.os, osd->od_igrp_oid,
943 obj->oo_attr.la_gid, -1, oh->ot_tx);
945 CERROR("%s: failed to update accounting ZAP for user "
946 "%d (%d)\n", osd->od_svname,
947 obj->oo_attr.la_gid, rc);
950 write_lock(&obj->oo_attr_lock);
952 if (la->la_valid & LA_ATIME) {
953 osa->atime[0] = obj->oo_attr.la_atime = la->la_atime;
954 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(uos), NULL,
957 if (la->la_valid & LA_MTIME) {
958 osa->mtime[0] = obj->oo_attr.la_mtime = la->la_mtime;
959 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(uos), NULL,
962 if (la->la_valid & LA_CTIME) {
963 osa->ctime[0] = obj->oo_attr.la_ctime = la->la_ctime;
964 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(uos), NULL,
967 if (la->la_valid & LA_MODE) {
968 /* mode is stored along with type, so read it first */
969 obj->oo_attr.la_mode = (obj->oo_attr.la_mode & S_IFMT) |
970 (la->la_mode & ~S_IFMT);
971 osa->mode = obj->oo_attr.la_mode;
972 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(uos), NULL,
975 if (la->la_valid & LA_SIZE) {
976 osa->size = obj->oo_attr.la_size = la->la_size;
977 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(uos), NULL,
980 if (la->la_valid & LA_NLINK) {
981 osa->nlink = obj->oo_attr.la_nlink = la->la_nlink;
982 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(uos), NULL,
985 if (la->la_valid & LA_RDEV) {
986 osa->rdev = obj->oo_attr.la_rdev = la->la_rdev;
987 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_RDEV(uos), NULL,
990 if (la->la_valid & LA_FLAGS) {
991 osa->flags = obj->oo_attr.la_flags = la->la_flags;
992 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(uos), NULL,
995 if (la->la_valid & LA_UID) {
996 osa->uid = obj->oo_attr.la_uid = la->la_uid;
997 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(uos), NULL,
1000 if (la->la_valid & LA_GID) {
1001 osa->gid = obj->oo_attr.la_gid = la->la_gid;
1002 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(uos), NULL,
1005 obj->oo_attr.la_valid |= la->la_valid;
1006 write_unlock(&obj->oo_attr_lock);
1008 rc = osd_object_sa_bulk_update(obj, bulk, cnt, oh);
1010 OBD_FREE(bulk, sizeof(sa_bulk_attr_t) * 10);
1017 * XXX temporary solution.
1020 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
1021 struct dt_object *parent, struct dt_object *child,
1022 cfs_umode_t child_mode)
1026 memset(ah, 0, sizeof(*ah));
1027 ah->dah_parent = parent;
1028 ah->dah_mode = child_mode;
1031 static int osd_declare_object_create(const struct lu_env *env,
1032 struct dt_object *dt,
1033 struct lu_attr *attr,
1034 struct dt_allocation_hint *hint,
1035 struct dt_object_format *dof,
1036 struct thandle *handle)
1038 char *buf = osd_oti_get(env)->oti_str;
1039 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1040 struct osd_object *obj = osd_dt_obj(dt);
1041 struct osd_device *osd = osd_obj2dev(obj);
1042 struct osd_thandle *oh;
1049 switch (dof->dof_type) {
1053 if (obj->oo_dt.do_body_ops == NULL)
1054 obj->oo_dt.do_body_ops = &osd_body_ops;
1060 LASSERT(handle != NULL);
1061 oh = container_of0(handle, struct osd_thandle, ot_super);
1062 LASSERT(oh->ot_tx != NULL);
1064 switch (dof->dof_type) {
1067 /* for zap create */
1068 dmu_tx_hold_zap(oh->ot_tx, DMU_NEW_OBJECT, 1, NULL);
1073 /* first, we'll create new object */
1074 dmu_tx_hold_bonus(oh->ot_tx, DMU_NEW_OBJECT);
1082 /* and we'll add it to some mapping */
1083 zapid = osd_get_name_n_idx(env, osd, fid, buf);
1084 dmu_tx_hold_bonus(oh->ot_tx, zapid);
1085 dmu_tx_hold_zap(oh->ot_tx, zapid, TRUE, buf);
1087 /* we will also update inode accounting ZAPs */
1088 dmu_tx_hold_bonus(oh->ot_tx, osd->od_iusr_oid);
1089 dmu_tx_hold_zap(oh->ot_tx, osd->od_iusr_oid, TRUE, buf);
1090 dmu_tx_hold_bonus(oh->ot_tx, osd->od_igrp_oid);
1091 dmu_tx_hold_zap(oh->ot_tx, osd->od_igrp_oid, TRUE, buf);
1093 dmu_tx_hold_sa_create(oh->ot_tx, ZFS_SA_BASE_ATTR_SIZE);
1095 __osd_xattr_declare_set(env, obj, sizeof(struct lustre_mdt_attrs),
1096 XATTR_NAME_LMA, oh);
1098 rc = osd_declare_quota(env, osd, attr->la_uid, attr->la_gid, 1, oh,
1099 false, NULL, false);
1103 int __osd_attr_init(const struct lu_env *env, udmu_objset_t *uos, uint64_t oid,
1104 dmu_tx_t *tx, struct lu_attr *la, uint64_t parent)
1106 sa_bulk_attr_t *bulk;
1107 sa_handle_t *sa_hdl;
1108 struct osa_attr *osa = &osd_oti_get(env)->oti_osa;
1116 gen = dmu_tx_get_txg(tx);
1118 ZFS_TIME_ENCODE(&now, crtime);
1120 osa->atime[0] = la->la_atime;
1121 osa->ctime[0] = la->la_ctime;
1122 osa->mtime[0] = la->la_mtime;
1123 osa->mode = la->la_mode;
1124 osa->uid = la->la_uid;
1125 osa->gid = la->la_gid;
1126 osa->rdev = la->la_rdev;
1127 osa->nlink = la->la_nlink;
1128 osa->flags = la->la_flags;
1129 osa->size = la->la_size;
1131 /* Now add in all of the "SA" attributes */
1132 rc = -sa_handle_get(uos->os, oid, NULL, SA_HDL_PRIVATE, &sa_hdl);
1136 OBD_ALLOC(bulk, sizeof(sa_bulk_attr_t) * 13);
1142 * we need to create all SA below upon object create.
1144 * XXX The attribute order matters since the accounting callback relies
1145 * on static offsets (i.e. SA_*_OFFSET, see zfs_space_delta_cb()) to
1146 * look up the UID/GID attributes. Moreover, the callback does not seem
1147 * to support the spill block.
1148 * We define attributes in the same order as SA_*_OFFSET in order to
1149 * work around the problem. See ORI-610.
1152 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(uos), NULL, &osa->mode, 8);
1153 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(uos), NULL, &osa->size, 8);
1154 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GEN(uos), NULL, &gen, 8);
1155 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(uos), NULL, &osa->uid, 8);
1156 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(uos), NULL, &osa->gid, 8);
1157 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_PARENT(uos), NULL, &parent, 8);
1158 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(uos), NULL, &osa->flags, 8);
1159 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(uos), NULL, osa->atime, 16);
1160 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(uos), NULL, osa->mtime, 16);
1161 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(uos), NULL, osa->ctime, 16);
1162 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CRTIME(uos), NULL, crtime, 16);
1163 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(uos), NULL, &osa->nlink, 8);
1164 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_RDEV(uos), NULL, &osa->rdev, 8);
1166 rc = -sa_replace_all_by_template(sa_hdl, bulk, cnt, tx);
1168 OBD_FREE(bulk, sizeof(sa_bulk_attr_t) * 13);
1170 sa_handle_destroy(sa_hdl);
1175 * The transaction passed to this routine must have
1176 * dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT) called and then assigned
1177 * to a transaction group.
1179 int __osd_object_create(const struct lu_env *env, udmu_objset_t *uos,
1180 dmu_buf_t **dbp, dmu_tx_t *tx, struct lu_attr *la,
1181 uint64_t parent, void *tag)
1187 spin_lock(&uos->lock);
1189 spin_unlock(&uos->lock);
1191 /* Assert that the transaction has been assigned to a
1192 transaction group. */
1193 LASSERT(tx->tx_txg != 0);
1195 /* Create a new DMU object. */
1196 oid = dmu_object_alloc(uos->os, DMU_OT_PLAIN_FILE_CONTENTS, 0,
1197 DMU_OT_SA, DN_MAX_BONUSLEN, tx);
1198 rc = -sa_buf_hold(uos->os, oid, tag, dbp);
1202 LASSERT(la->la_valid & LA_MODE);
1206 return __osd_attr_init(env, uos, oid, tx, la, parent);
1210 * The transaction passed to this routine must have
1211 * dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, ...) called and then assigned
1212 * to a transaction group.
1214 * Using ZAP_FLAG_HASH64 will force the ZAP to always be a FAT ZAP.
1215 * This is fine for directories today, because storing the FID in the dirent
1216 * will also require a FAT ZAP. If there is a new type of micro ZAP created
1217 * then we might need to re-evaluate the use of this flag and instead do
1218 * a conversion from the different internal ZAP hash formats being used. */
1219 int __osd_zap_create(const struct lu_env *env, udmu_objset_t *uos,
1220 dmu_buf_t **zap_dbp, dmu_tx_t *tx,
1221 struct lu_attr *la, uint64_t parent,
1222 void *tag, zap_flags_t flags)
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 oid = zap_create_flags(uos->os, 0, flags | ZAP_FLAG_HASH64,
1238 DMU_OT_DIRECTORY_CONTENTS, 12, 12,
1239 DMU_OT_SA, DN_MAX_BONUSLEN, tx);
1241 rc = -sa_buf_hold(uos->os, oid, tag, zap_dbp);
1245 LASSERT(la->la_valid & LA_MODE);
1249 return __osd_attr_init(env, uos, oid, tx, la, parent);
1252 static dmu_buf_t *osd_mkidx(const struct lu_env *env, struct osd_device *osd,
1253 struct lu_attr *la, uint64_t parent,
1254 struct osd_thandle *oh)
1259 /* Index file should be created as regular file in order not to confuse
1260 * ZPL which could interpret them as directory.
1261 * We set ZAP_FLAG_UINT64_KEY to let ZFS know than we are going to use
1263 LASSERT(S_ISREG(la->la_mode));
1264 rc = __osd_zap_create(env, &osd->od_objset, &db, oh->ot_tx, la,
1265 parent, osd_obj_tag, ZAP_FLAG_UINT64_KEY);
1271 static dmu_buf_t *osd_mkdir(const struct lu_env *env, struct osd_device *osd,
1272 struct lu_attr *la, uint64_t parent,
1273 struct osd_thandle *oh)
1278 LASSERT(S_ISDIR(la->la_mode));
1279 rc = __osd_zap_create(env, &osd->od_objset, &db, oh->ot_tx, la,
1280 parent, osd_obj_tag, 0);
1286 static dmu_buf_t* osd_mkreg(const struct lu_env *env, struct osd_device *osd,
1287 struct lu_attr *la, uint64_t parent,
1288 struct osd_thandle *oh)
1293 LASSERT(S_ISREG(la->la_mode));
1294 rc = __osd_object_create(env, &osd->od_objset, &db, oh->ot_tx, la,
1295 parent, osd_obj_tag);
1300 * XXX: a hack, OST to use bigger blocksize. we need
1301 * a method in OSD API to control this from OFD/MDD
1303 if (!lu_device_is_md(osd2lu_dev(osd))) {
1304 rc = -dmu_object_set_blocksize(osd->od_objset.os,
1306 128 << 10, 0, oh->ot_tx);
1308 CERROR("%s: can't change blocksize: %d\n",
1309 osd->od_svname, rc);
1317 static dmu_buf_t *osd_mksym(const struct lu_env *env, struct osd_device *osd,
1318 struct lu_attr *la, uint64_t parent,
1319 struct osd_thandle *oh)
1324 LASSERT(S_ISLNK(la->la_mode));
1325 rc = __osd_object_create(env, &osd->od_objset, &db, oh->ot_tx, la,
1326 parent, osd_obj_tag);
1332 static dmu_buf_t *osd_mknod(const struct lu_env *env, struct osd_device *osd,
1333 struct lu_attr *la, uint64_t parent,
1334 struct osd_thandle *oh)
1339 la->la_valid = LA_MODE;
1340 if (S_ISCHR(la->la_mode) || S_ISBLK(la->la_mode))
1341 la->la_valid |= LA_RDEV;
1343 rc = __osd_object_create(env, &osd->od_objset, &db, oh->ot_tx, la,
1344 parent, osd_obj_tag);
1350 typedef dmu_buf_t *(*osd_obj_type_f)(const struct lu_env *env,
1351 struct osd_device *osd,
1354 struct osd_thandle *oh);
1356 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
1358 osd_obj_type_f result;
1384 * Primitives for directory (i.e. ZAP) handling
1386 static inline int osd_init_lma(const struct lu_env *env, struct osd_object *obj,
1387 const struct lu_fid *fid, struct osd_thandle *oh)
1389 struct osd_thread_info *info = osd_oti_get(env);
1390 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
1394 lustre_lma_init(lma, fid, 0);
1395 lustre_lma_swab(lma);
1397 buf.lb_len = sizeof(*lma);
1399 rc = osd_xattr_set_internal(env, obj, &buf, XATTR_NAME_LMA,
1400 LU_XATTR_CREATE, oh, BYPASS_CAPA);
1406 * Concurrency: @dt is write locked.
1408 static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
1409 struct lu_attr *attr,
1410 struct dt_allocation_hint *hint,
1411 struct dt_object_format *dof,
1414 struct zpl_direntry *zde = &osd_oti_get(env)->oti_zde.lzd_reg;
1415 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1416 struct osd_object *obj = osd_dt_obj(dt);
1417 struct osd_device *osd = osd_obj2dev(obj);
1418 char *buf = osd_oti_get(env)->oti_str;
1419 struct osd_thandle *oh;
1426 /* concurrent create declarations should not see
1427 * the object inconsistent (db, attr, etc).
1428 * in regular cases acquisition should be cheap */
1429 down(&obj->oo_guard);
1431 LASSERT(osd_invariant(obj));
1432 LASSERT(!dt_object_exists(dt));
1433 LASSERT(dof != NULL);
1435 LASSERT(th != NULL);
1436 oh = container_of0(th, struct osd_thandle, ot_super);
1439 * XXX missing: Quote handling.
1442 LASSERT(obj->oo_db == NULL);
1444 /* to follow ZFS on-disk format we need
1445 * to initialize parent dnode properly */
1447 if (hint && hint->dah_parent)
1448 zapid = osd_dt_obj(hint->dah_parent)->oo_db->db_object;
1450 db = osd_create_type_f(dof->dof_type)(env, osd, attr, zapid, oh);
1452 GOTO(out, rc = PTR_ERR(th));
1455 zde->zde_dnode = db->db_object;
1456 zde->zde_type = IFTODT(attr->la_mode & S_IFMT);
1458 zapid = osd_get_name_n_idx(env, osd, fid, buf);
1460 rc = -zap_add(osd->od_objset.os, zapid, buf, 8, 1, zde, oh->ot_tx);
1464 /* Add new object to inode accounting.
1465 * Errors are not considered as fatal */
1466 rc = -zap_increment_int(osd->od_objset.os, osd->od_iusr_oid,
1467 (attr->la_valid & LA_UID) ? attr->la_uid : 0, 1,
1470 CERROR("%s: failed to add "DFID" to accounting ZAP for usr %d "
1471 "(%d)\n", osd->od_svname, PFID(fid), attr->la_uid, rc);
1472 rc = -zap_increment_int(osd->od_objset.os, osd->od_igrp_oid,
1473 (attr->la_valid & LA_GID) ? attr->la_gid : 0, 1,
1476 CERROR("%s: failed to add "DFID" to accounting ZAP for grp %d "
1477 "(%d)\n", osd->od_svname, PFID(fid), attr->la_gid, rc);
1479 /* configure new osd object */
1481 rc = osd_object_init0(env, obj);
1482 LASSERT(ergo(rc == 0, dt_object_exists(dt)));
1483 LASSERT(osd_invariant(obj));
1485 rc = osd_init_lma(env, obj, fid, oh);
1487 CERROR("%s: can not set LMA on "DFID": rc = %d\n",
1488 osd->od_svname, PFID(fid), rc);
1489 /* ignore errors during LMA initialization */
1498 static int osd_declare_object_ref_add(const struct lu_env *env,
1499 struct dt_object *dt,
1502 return osd_declare_attr_set(env, dt, NULL, th);
1506 * Concurrency: @dt is write locked.
1508 static int osd_object_ref_add(const struct lu_env *env,
1509 struct dt_object *dt,
1510 struct thandle *handle)
1512 struct osd_object *obj = osd_dt_obj(dt);
1513 struct osd_thandle *oh;
1514 struct osd_device *osd = osd_obj2dev(obj);
1515 udmu_objset_t *uos = &osd->od_objset;
1521 LASSERT(osd_invariant(obj));
1522 LASSERT(dt_object_exists(dt));
1523 LASSERT(obj->oo_sa_hdl != NULL);
1525 oh = container_of0(handle, struct osd_thandle, ot_super);
1527 write_lock(&obj->oo_attr_lock);
1528 nlink = ++obj->oo_attr.la_nlink;
1529 write_unlock(&obj->oo_attr_lock);
1531 rc = osd_object_sa_update(obj, SA_ZPL_LINKS(uos), &nlink, 8, oh);
1535 static int osd_declare_object_ref_del(const struct lu_env *env,
1536 struct dt_object *dt,
1537 struct thandle *handle)
1539 return osd_declare_attr_set(env, dt, NULL, handle);
1543 * Concurrency: @dt is write locked.
1545 static int osd_object_ref_del(const struct lu_env *env,
1546 struct dt_object *dt,
1547 struct thandle *handle)
1549 struct osd_object *obj = osd_dt_obj(dt);
1550 struct osd_thandle *oh;
1551 struct osd_device *osd = osd_obj2dev(obj);
1552 udmu_objset_t *uos = &osd->od_objset;
1558 LASSERT(osd_invariant(obj));
1559 LASSERT(dt_object_exists(dt));
1560 LASSERT(obj->oo_sa_hdl != NULL);
1562 oh = container_of0(handle, struct osd_thandle, ot_super);
1563 LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
1565 write_lock(&obj->oo_attr_lock);
1566 nlink = --obj->oo_attr.la_nlink;
1567 write_unlock(&obj->oo_attr_lock);
1569 rc = osd_object_sa_update(obj, SA_ZPL_LINKS(uos), &nlink, 8, oh);
1573 static int capa_is_sane(const struct lu_env *env, struct osd_device *dev,
1574 struct lustre_capa *capa, struct lustre_capa_key *keys)
1576 struct osd_thread_info *oti = osd_oti_get(env);
1577 struct obd_capa *oc;
1581 oc = capa_lookup(dev->od_capa_hash, capa, 0);
1583 if (capa_is_expired(oc)) {
1584 DEBUG_CAPA(D_ERROR, capa, "expired");
1591 spin_lock(&capa_lock);
1592 for (i = 0; i < 2; i++) {
1593 if (keys[i].lk_keyid == capa->lc_keyid) {
1594 oti->oti_capa_key = keys[i];
1598 spin_unlock(&capa_lock);
1601 DEBUG_CAPA(D_ERROR, capa, "no matched capa key");
1605 rc = capa_hmac(oti->oti_capa.lc_hmac, capa, oti->oti_capa_key.lk_key);
1608 if (memcmp(oti->oti_capa.lc_hmac, capa->lc_hmac, sizeof(capa->lc_hmac)))
1610 DEBUG_CAPA(D_ERROR, capa, "HMAC mismatch");
1614 oc = capa_add(dev->od_capa_hash, capa);
1620 static int osd_object_auth(const struct lu_env *env, struct dt_object *dt,
1621 struct lustre_capa *capa, __u64 opc)
1623 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1624 struct osd_device *dev = osd_dev(dt->do_lu.lo_dev);
1627 if (!dev->od_fl_capa)
1630 if (capa == BYPASS_CAPA)
1634 CERROR("no capability is provided for fid "DFID"\n", PFID(fid));
1638 if (!lu_fid_eq(fid, &capa->lc_fid)) {
1639 DEBUG_CAPA(D_ERROR, capa, "fid "DFID" mismatch with",PFID(fid));
1643 if (!capa_opc_supported(capa, opc)) {
1644 DEBUG_CAPA(D_ERROR, capa, "opc "LPX64" not supported by", opc);
1648 if ((rc = capa_is_sane(env, dev, capa, dev->od_capa_keys))) {
1649 DEBUG_CAPA(D_ERROR, capa, "insane (rc %d)", rc);
1656 static struct obd_capa *osd_capa_get(const struct lu_env *env,
1657 struct dt_object *dt,
1658 struct lustre_capa *old,
1661 struct osd_thread_info *info = osd_oti_get(env);
1662 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1663 struct osd_object *obj = osd_dt_obj(dt);
1664 struct osd_device *dev = osd_obj2dev(obj);
1665 struct lustre_capa_key *key = &info->oti_capa_key;
1666 struct lustre_capa *capa = &info->oti_capa;
1667 struct obd_capa *oc;
1671 if (!dev->od_fl_capa)
1672 RETURN(ERR_PTR(-ENOENT));
1674 LASSERT(dt_object_exists(dt));
1675 LASSERT(osd_invariant(obj));
1677 /* renewal sanity check */
1678 if (old && osd_object_auth(env, dt, old, opc))
1679 RETURN(ERR_PTR(-EACCES));
1681 capa->lc_fid = *fid;
1684 capa->lc_flags = dev->od_capa_alg << 24;
1685 capa->lc_timeout = dev->od_capa_timeout;
1686 capa->lc_expiry = 0;
1688 oc = capa_lookup(dev->od_capa_hash, capa, 1);
1690 LASSERT(!capa_is_expired(oc));
1694 spin_lock(&capa_lock);
1695 *key = dev->od_capa_keys[1];
1696 spin_unlock(&capa_lock);
1698 capa->lc_keyid = key->lk_keyid;
1699 capa->lc_expiry = cfs_time_current_sec() + dev->od_capa_timeout;
1701 rc = capa_hmac(capa->lc_hmac, capa, key->lk_key);
1703 DEBUG_CAPA(D_ERROR, capa, "HMAC failed: %d for", rc);
1705 RETURN(ERR_PTR(rc));
1708 oc = capa_add(dev->od_capa_hash, capa);
1712 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt)
1714 struct osd_device *osd = osd_obj2dev(osd_dt_obj(dt));
1717 /* XXX: no other option than syncing the whole filesystem until we
1719 txg_wait_synced(dmu_objset_pool(osd->od_objset.os), 0ULL);
1724 static struct dt_object_operations osd_obj_ops = {
1725 .do_read_lock = osd_object_read_lock,
1726 .do_write_lock = osd_object_write_lock,
1727 .do_read_unlock = osd_object_read_unlock,
1728 .do_write_unlock = osd_object_write_unlock,
1729 .do_write_locked = osd_object_write_locked,
1730 .do_attr_get = osd_attr_get,
1731 .do_declare_attr_set = osd_declare_attr_set,
1732 .do_attr_set = osd_attr_set,
1733 .do_ah_init = osd_ah_init,
1734 .do_declare_create = osd_declare_object_create,
1735 .do_create = osd_object_create,
1736 .do_declare_destroy = osd_declare_object_destroy,
1737 .do_destroy = osd_object_destroy,
1738 .do_index_try = osd_index_try,
1739 .do_declare_ref_add = osd_declare_object_ref_add,
1740 .do_ref_add = osd_object_ref_add,
1741 .do_declare_ref_del = osd_declare_object_ref_del,
1742 .do_ref_del = osd_object_ref_del,
1743 .do_xattr_get = osd_xattr_get,
1744 .do_declare_xattr_set = osd_declare_xattr_set,
1745 .do_xattr_set = osd_xattr_set,
1746 .do_declare_xattr_del = osd_declare_xattr_del,
1747 .do_xattr_del = osd_xattr_del,
1748 .do_xattr_list = osd_xattr_list,
1749 .do_capa_get = osd_capa_get,
1750 .do_object_sync = osd_object_sync,
1753 static struct lu_object_operations osd_lu_obj_ops = {
1754 .loo_object_init = osd_object_init,
1755 .loo_object_delete = osd_object_delete,
1756 .loo_object_release = osd_object_release,
1757 .loo_object_free = osd_object_free,
1758 .loo_object_print = osd_object_print,
1759 .loo_object_invariant = osd_object_invariant,
1762 static int osd_otable_it_attr_get(const struct lu_env *env,
1763 struct dt_object *dt,
1764 struct lu_attr *attr,
1765 struct lustre_capa *capa)
1771 static struct dt_object_operations osd_obj_otable_it_ops = {
1772 .do_attr_get = osd_otable_it_attr_get,
1773 .do_index_try = osd_index_try,