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>
45 #define DEBUG_SUBSYSTEM S_OSD
47 #include <lustre_ver.h>
48 #include <libcfs/libcfs.h>
49 #include <obd_support.h>
50 #include <lustre_net.h>
52 #include <obd_class.h>
53 #include <lustre_disk.h>
54 #include <lustre_fid.h>
56 #include "osd_internal.h"
58 #include <sys/dnode.h>
63 #include <sys/spa_impl.h>
64 #include <sys/zfs_znode.h>
65 #include <sys/dmu_tx.h>
66 #include <sys/dmu_objset.h>
67 #include <sys/dsl_prop.h>
68 #include <sys/sa_impl.h>
71 static char *osd_obj_tag = "osd_object";
73 static struct dt_object_operations osd_obj_ops;
74 static struct lu_object_operations osd_lu_obj_ops;
75 extern struct dt_body_operations osd_body_ops;
76 static struct dt_object_operations osd_obj_otable_it_ops;
78 extern struct kmem_cache *osd_object_kmem;
81 osd_object_sa_fini(struct osd_object *obj)
84 sa_handle_destroy(obj->oo_sa_hdl);
85 obj->oo_sa_hdl = NULL;
90 osd_object_sa_init(struct osd_object *obj, udmu_objset_t *uos)
94 LASSERT(obj->oo_sa_hdl == NULL);
95 LASSERT(obj->oo_db != NULL);
97 rc = -sa_handle_get(uos->os, obj->oo_db->db_object, obj,
98 SA_HDL_PRIVATE, &obj->oo_sa_hdl);
102 /* Cache the xattr object id, valid for the life of the object */
103 rc = -sa_lookup(obj->oo_sa_hdl, SA_ZPL_XATTR(uos), &obj->oo_xattr, 8);
105 obj->oo_xattr = ZFS_NO_OBJECT;
108 osd_object_sa_fini(obj);
115 * Add object to list of dirty objects in tx handle.
118 osd_object_sa_dirty_add(struct osd_object *obj, struct osd_thandle *oh)
120 if (!list_empty(&obj->oo_sa_linkage))
123 down(&oh->ot_sa_lock);
124 write_lock(&obj->oo_attr_lock);
125 if (likely(list_empty(&obj->oo_sa_linkage)))
126 list_add(&obj->oo_sa_linkage, &oh->ot_sa_list);
127 write_unlock(&obj->oo_attr_lock);
132 * Release spill block dbuf hold for all dirty SAs.
134 void osd_object_sa_dirty_rele(struct osd_thandle *oh)
136 struct osd_object *obj;
138 down(&oh->ot_sa_lock);
139 while (!list_empty(&oh->ot_sa_list)) {
140 obj = list_entry(oh->ot_sa_list.next,
141 struct osd_object, oo_sa_linkage);
142 sa_spill_rele(obj->oo_sa_hdl);
143 write_lock(&obj->oo_attr_lock);
144 list_del_init(&obj->oo_sa_linkage);
145 write_unlock(&obj->oo_attr_lock);
151 * Update the SA and add the object to the dirty list.
153 int osd_object_sa_update(struct osd_object *obj, sa_attr_type_t type,
154 void *buf, uint32_t buflen, struct osd_thandle *oh)
158 LASSERT(obj->oo_sa_hdl != NULL);
159 LASSERT(oh->ot_tx != NULL);
161 rc = -sa_update(obj->oo_sa_hdl, type, buf, buflen, oh->ot_tx);
162 osd_object_sa_dirty_add(obj, oh);
168 * Bulk update the SA and add the object to the dirty list.
171 osd_object_sa_bulk_update(struct osd_object *obj, sa_bulk_attr_t *attrs,
172 int count, struct osd_thandle *oh)
176 LASSERT(obj->oo_sa_hdl != NULL);
177 LASSERT(oh->ot_tx != NULL);
179 rc = -sa_bulk_update(obj->oo_sa_hdl, attrs, count, oh->ot_tx);
180 osd_object_sa_dirty_add(obj, oh);
186 * Retrieve the attributes of a DMU object
188 int __osd_object_attr_get(const struct lu_env *env, udmu_objset_t *uos,
189 struct osd_object *obj, struct lu_attr *la)
191 struct osa_attr *osa = &osd_oti_get(env)->oti_osa;
193 sa_bulk_attr_t *bulk;
198 LASSERT(obj->oo_db != NULL);
200 rc = -sa_handle_get(uos->os, obj->oo_db->db_object, NULL,
201 SA_HDL_PRIVATE, &sa_hdl);
205 OBD_ALLOC(bulk, sizeof(sa_bulk_attr_t) * 9);
207 GOTO(out_sa, rc = -ENOMEM);
209 la->la_valid |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE | LA_TYPE |
210 LA_SIZE | LA_UID | LA_GID | LA_FLAGS | LA_NLINK;
212 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(uos), NULL, osa->atime, 16);
213 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(uos), NULL, osa->mtime, 16);
214 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(uos), NULL, osa->ctime, 16);
215 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(uos), NULL, &osa->mode, 8);
216 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(uos), NULL, &osa->size, 8);
217 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(uos), NULL, &osa->nlink, 8);
218 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(uos), NULL, &osa->uid, 8);
219 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(uos), NULL, &osa->gid, 8);
220 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(uos), NULL, &osa->flags, 8);
222 rc = -sa_bulk_lookup(sa_hdl, bulk, cnt);
226 la->la_atime = osa->atime[0];
227 la->la_mtime = osa->mtime[0];
228 la->la_ctime = osa->ctime[0];
229 la->la_mode = osa->mode;
230 la->la_uid = osa->uid;
231 la->la_gid = osa->gid;
232 la->la_nlink = osa->nlink;
233 la->la_flags = attrs_zfs2fs(osa->flags);
234 la->la_size = osa->size;
236 if (S_ISCHR(la->la_mode) || S_ISBLK(la->la_mode)) {
237 rc = -sa_lookup(sa_hdl, SA_ZPL_RDEV(uos), &osa->rdev, 8);
240 la->la_rdev = osa->rdev;
241 la->la_valid |= LA_RDEV;
244 OBD_FREE(bulk, sizeof(sa_bulk_attr_t) * 9);
246 sa_handle_destroy(sa_hdl);
251 int __osd_obj2dbuf(const struct lu_env *env, objset_t *os,
252 uint64_t oid, dmu_buf_t **dbp, void *tag)
254 dmu_object_info_t *doi = &osd_oti_get(env)->oti_doi;
259 rc = -sa_buf_hold(os, oid, tag, dbp);
263 dmu_object_info_from_db(*dbp, doi);
264 if (unlikely (oid != DMU_USERUSED_OBJECT &&
265 oid != DMU_GROUPUSED_OBJECT && doi->doi_bonus_type != DMU_OT_SA)) {
266 sa_buf_rele(*dbp, tag);
272 LASSERT((*dbp)->db_object == oid);
273 LASSERT((*dbp)->db_offset == -1);
274 LASSERT((*dbp)->db_data != NULL);
280 * Concurrency: no concurrent access is possible that early in object
283 struct lu_object *osd_object_alloc(const struct lu_env *env,
284 const struct lu_object_header *hdr,
287 struct osd_object *mo;
289 OBD_SLAB_ALLOC_PTR_GFP(mo, osd_object_kmem, GFP_NOFS);
293 l = &mo->oo_dt.do_lu;
294 dt_object_init(&mo->oo_dt, NULL, d);
295 mo->oo_dt.do_ops = &osd_obj_ops;
296 l->lo_ops = &osd_lu_obj_ops;
297 INIT_LIST_HEAD(&mo->oo_sa_linkage);
298 init_rwsem(&mo->oo_sem);
299 sema_init(&mo->oo_guard, 1);
300 rwlock_init(&mo->oo_attr_lock);
308 * Concurrency: shouldn't matter.
310 int osd_object_init0(const struct lu_env *env, struct osd_object *obj)
312 struct osd_device *osd = osd_obj2dev(obj);
313 udmu_objset_t *uos = &osd->od_objset;
314 const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
318 if (obj->oo_db == NULL)
323 rc = osd_object_sa_init(obj, uos);
327 /* cache attrs in object */
328 rc = __osd_object_attr_get(env, &osd->od_objset,
333 if (likely(!fid_is_acct(fid)))
334 /* no body operations for accounting objects */
335 obj->oo_dt.do_body_ops = &osd_body_ops;
338 * initialize object before marking it existing
340 obj->oo_dt.do_lu.lo_header->loh_attr |= obj->oo_attr.la_mode & S_IFMT;
343 obj->oo_dt.do_lu.lo_header->loh_attr |= LOHA_EXISTS;
348 static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
350 struct osd_thread_info *info = osd_oti_get(env);
353 struct lustre_mdt_attrs *lma;
356 CLASSERT(sizeof(info->oti_buf) >= sizeof(*lma));
357 lma = (struct lustre_mdt_attrs *)info->oti_buf;
359 buf.lb_len = sizeof(info->oti_buf);
361 rc = osd_xattr_get(env, &obj->oo_dt, &buf, XATTR_NAME_LMA, BYPASS_CAPA);
364 lustre_lma_swab(lma);
365 if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
366 CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT))) {
367 CWARN("%s: unsupported incompat LMA feature(s) %#x for "
368 "fid = "DFID"\n", osd_obj2dev(obj)->od_svname,
369 lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
370 PFID(lu_object_fid(&obj->oo_dt.do_lu)));
373 } else if (rc == -ENODATA) {
374 /* haven't initialize LMA xattr */
382 * Concurrency: no concurrent access is possible that early in object
385 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
386 const struct lu_object_conf *conf)
388 struct osd_object *obj = osd_obj(l);
389 struct osd_device *osd = osd_obj2dev(obj);
394 LASSERT(osd_invariant(obj));
396 if (fid_is_otable_it(&l->lo_header->loh_fid)) {
397 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
398 l->lo_header->loh_attr |= LOHA_EXISTS;
402 rc = osd_fid_lookup(env, osd, lu_object_fid(l), &oid);
404 LASSERT(obj->oo_db == NULL);
405 rc = __osd_obj2dbuf(env, osd->od_objset.os, oid,
406 &obj->oo_db, osd_obj_tag);
408 CERROR("%s: lookup "DFID"/"LPX64" failed: rc = %d\n",
409 osd->od_svname, PFID(lu_object_fid(l)), oid, rc);
413 rc = osd_object_init0(env, obj);
417 rc = osd_check_lma(env, obj);
420 } else if (rc == -ENOENT) {
423 LASSERT(osd_invariant(obj));
429 * Concurrency: no concurrent access is possible that late in object
432 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
434 struct osd_object *obj = osd_obj(l);
436 LASSERT(osd_invariant(obj));
438 dt_object_fini(&obj->oo_dt);
439 OBD_SLAB_FREE_PTR(obj, osd_object_kmem);
442 static void __osd_declare_object_destroy(const struct lu_env *env,
443 struct osd_object *obj,
444 struct osd_thandle *oh)
446 struct osd_device *osd = osd_obj2dev(obj);
447 udmu_objset_t *uos = &osd->od_objset;
448 dmu_buf_t *db = obj->oo_db;
449 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
450 uint64_t oid = db->db_object, xid;
451 dmu_tx_t *tx = oh->ot_tx;
455 dmu_tx_hold_free(tx, oid, 0, DMU_OBJECT_END);
457 /* zap holding xattrs */
458 if (obj->oo_xattr != ZFS_NO_OBJECT) {
461 dmu_tx_hold_free(tx, oid, 0, DMU_OBJECT_END);
463 rc = -udmu_zap_cursor_init(&zc, uos, oid, 0);
467 while ((rc = -zap_cursor_retrieve(zc, za)) == 0) {
468 BUG_ON(za->za_integer_length != sizeof(uint64_t));
469 BUG_ON(za->za_num_integers != 1);
471 rc = -zap_lookup(uos->os, obj->oo_xattr, za->za_name,
472 sizeof(uint64_t), 1, &xid);
474 CERROR("%s: xattr lookup failed: rc = %d\n",
478 dmu_tx_hold_free(tx, xid, 0, DMU_OBJECT_END);
480 zap_cursor_advance(zc);
485 udmu_zap_cursor_fini(zc);
488 if (rc && tx->tx_err == 0)
492 static int osd_declare_object_destroy(const struct lu_env *env,
493 struct dt_object *dt,
496 char *buf = osd_oti_get(env)->oti_str;
497 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
498 struct osd_object *obj = osd_dt_obj(dt);
499 struct osd_device *osd = osd_obj2dev(obj);
500 struct osd_thandle *oh;
506 LASSERT(dt_object_exists(dt));
508 oh = container_of0(th, struct osd_thandle, ot_super);
509 LASSERT(oh->ot_tx != NULL);
511 /* declare that we'll destroy the object */
512 __osd_declare_object_destroy(env, obj, oh);
514 /* declare that we'll remove object from fid-dnode mapping */
515 zapid = osd_get_name_n_idx(env, osd, fid, buf);
516 dmu_tx_hold_bonus(oh->ot_tx, zapid);
517 dmu_tx_hold_zap(oh->ot_tx, zapid, 0, buf);
519 /* declare that we'll remove object from inode accounting ZAPs */
520 dmu_tx_hold_bonus(oh->ot_tx, osd->od_iusr_oid);
521 dmu_tx_hold_zap(oh->ot_tx, osd->od_iusr_oid, 0, buf);
522 dmu_tx_hold_bonus(oh->ot_tx, osd->od_igrp_oid);
523 dmu_tx_hold_zap(oh->ot_tx, osd->od_igrp_oid, 0, buf);
526 rc = osd_declare_quota(env, osd, obj->oo_attr.la_uid,
527 obj->oo_attr.la_gid, -1, oh, false, NULL, false);
531 /* data to be truncated */
532 rc = osd_declare_quota(env, osd, obj->oo_attr.la_uid,
533 obj->oo_attr.la_gid, 0, oh, true, NULL, false);
537 int __osd_object_free(udmu_objset_t *uos, uint64_t oid, dmu_tx_t *tx)
539 LASSERT(uos->objects != 0);
540 spin_lock(&uos->lock);
542 spin_unlock(&uos->lock);
544 return -dmu_object_free(uos->os, oid, tx);
548 * Delete a DMU object
550 * The transaction passed to this routine must have
551 * dmu_tx_hold_free(tx, oid, 0, DMU_OBJECT_END) called
552 * and then assigned to a transaction group.
554 * This will release db and set it to NULL to prevent further dbuf releases.
556 static int __osd_object_destroy(const struct lu_env *env,
557 struct osd_object *obj,
558 dmu_tx_t *tx, void *tag)
560 struct osd_device *osd = osd_obj2dev(obj);
561 udmu_objset_t *uos = &osd->od_objset;
563 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
567 /* Assert that the transaction has been assigned to a
568 transaction group. */
569 LASSERT(tx->tx_txg != 0);
571 /* zap holding xattrs */
572 if (obj->oo_xattr != ZFS_NO_OBJECT) {
573 rc = -udmu_zap_cursor_init(&zc, uos, obj->oo_xattr, 0);
576 while ((rc = -zap_cursor_retrieve(zc, za)) == 0) {
577 BUG_ON(za->za_integer_length != sizeof(uint64_t));
578 BUG_ON(za->za_num_integers != 1);
580 rc = -zap_lookup(uos->os, obj->oo_xattr, za->za_name,
581 sizeof(uint64_t), 1, &xid);
583 CERROR("%s: lookup xattr %s failed: rc = %d\n",
584 osd->od_svname, za->za_name, rc);
587 rc = __osd_object_free(uos, xid, tx);
589 CERROR("%s: fetch xattr %s failed: rc = %d\n",
590 osd->od_svname, za->za_name, rc);
591 zap_cursor_advance(zc);
593 udmu_zap_cursor_fini(zc);
595 rc = __osd_object_free(uos, obj->oo_xattr, tx);
597 CERROR("%s: freeing xattr failed: rc = %d\n",
601 return __osd_object_free(uos, obj->oo_db->db_object, tx);
604 static int osd_object_destroy(const struct lu_env *env,
605 struct dt_object *dt,
608 char *buf = osd_oti_get(env)->oti_str;
609 struct osd_object *obj = osd_dt_obj(dt);
610 struct osd_device *osd = osd_obj2dev(obj);
611 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
612 struct osd_thandle *oh;
617 LASSERT(obj->oo_db != NULL);
618 LASSERT(dt_object_exists(dt));
619 LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
621 oh = container_of0(th, struct osd_thandle, ot_super);
623 LASSERT(oh->ot_tx != NULL);
625 zapid = osd_get_name_n_idx(env, osd, fid, buf);
627 /* remove obj ref from index dir (it depends) */
628 rc = -zap_remove(osd->od_objset.os, zapid, buf, oh->ot_tx);
630 CERROR("%s: zap_remove() failed: rc = %d\n",
635 /* Remove object from inode accounting. It is not fatal for the destroy
636 * operation if something goes wrong while updating accounting, but we
637 * still log an error message to notify the administrator */
638 rc = -zap_increment_int(osd->od_objset.os, osd->od_iusr_oid,
639 obj->oo_attr.la_uid, -1, oh->ot_tx);
641 CERROR("%s: failed to remove "DFID" from accounting ZAP for usr"
642 " %d: rc = %d\n", osd->od_svname, PFID(fid),
643 obj->oo_attr.la_uid, rc);
644 rc = -zap_increment_int(osd->od_objset.os, osd->od_igrp_oid,
645 obj->oo_attr.la_gid, -1, oh->ot_tx);
647 CERROR("%s: failed to remove "DFID" from accounting ZAP for grp"
648 " %d: rc = %d\n", osd->od_svname, PFID(fid),
649 obj->oo_attr.la_gid, rc);
652 rc = __osd_object_destroy(env, obj, oh->ot_tx, osd_obj_tag);
654 CERROR("%s: __osd_object_destroy() failed: rc = %d\n",
660 /* not needed in the cache anymore */
661 set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
666 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
668 struct osd_object *obj = osd_obj(l);
670 if (obj->oo_db != NULL) {
671 osd_object_sa_fini(obj);
672 if (obj->oo_sa_xattr) {
673 nvlist_free(obj->oo_sa_xattr);
674 obj->oo_sa_xattr = NULL;
676 sa_buf_rele(obj->oo_db, osd_obj_tag);
677 list_del(&obj->oo_sa_linkage);
683 * Concurrency: ->loo_object_release() is called under site spin-lock.
685 static void osd_object_release(const struct lu_env *env,
691 * Concurrency: shouldn't matter.
693 static int osd_object_print(const struct lu_env *env, void *cookie,
694 lu_printer_t p, const struct lu_object *l)
696 struct osd_object *o = osd_obj(l);
698 return (*p)(env, cookie, LUSTRE_OSD_ZFS_NAME"-object@%p", o);
701 static void osd_object_read_lock(const struct lu_env *env,
702 struct dt_object *dt, unsigned role)
704 struct osd_object *obj = osd_dt_obj(dt);
706 LASSERT(osd_invariant(obj));
708 down_read(&obj->oo_sem);
711 static void osd_object_write_lock(const struct lu_env *env,
712 struct dt_object *dt, unsigned role)
714 struct osd_object *obj = osd_dt_obj(dt);
716 LASSERT(osd_invariant(obj));
718 down_write(&obj->oo_sem);
721 static void osd_object_read_unlock(const struct lu_env *env,
722 struct dt_object *dt)
724 struct osd_object *obj = osd_dt_obj(dt);
726 LASSERT(osd_invariant(obj));
727 up_read(&obj->oo_sem);
730 static void osd_object_write_unlock(const struct lu_env *env,
731 struct dt_object *dt)
733 struct osd_object *obj = osd_dt_obj(dt);
735 LASSERT(osd_invariant(obj));
736 up_write(&obj->oo_sem);
739 static int osd_object_write_locked(const struct lu_env *env,
740 struct dt_object *dt)
742 struct osd_object *obj = osd_dt_obj(dt);
745 LASSERT(osd_invariant(obj));
747 if (down_write_trylock(&obj->oo_sem)) {
749 up_write(&obj->oo_sem);
754 static int osd_attr_get(const struct lu_env *env,
755 struct dt_object *dt,
756 struct lu_attr *attr,
757 struct lustre_capa *capa)
759 struct osd_object *obj = osd_dt_obj(dt);
763 LASSERT(dt_object_exists(dt));
764 LASSERT(osd_invariant(obj));
767 read_lock(&obj->oo_attr_lock);
768 *attr = obj->oo_attr;
769 read_unlock(&obj->oo_attr_lock);
771 /* with ZFS_DEBUG zrl_add_debug() called by DB_DNODE_ENTER()
772 * from within sa_object_size() can block on a mutex, so
773 * we can't call sa_object_size() holding rwlock */
774 sa_object_size(obj->oo_sa_hdl, &blksize, &blocks);
775 /* we do not control size of indices, so always calculate
776 * it from number of blocks reported by DMU */
777 if (S_ISDIR(attr->la_mode))
778 attr->la_size = 512 * blocks;
779 /* Block size may be not set; suggest maximal I/O transfers. */
781 blksize = 1ULL << SPA_MAXBLOCKSHIFT;
783 attr->la_blksize = blksize;
784 attr->la_blocks = blocks;
785 attr->la_valid |= LA_BLOCKS | LA_BLKSIZE;
790 /* Simple wrapper on top of qsd API which implement quota transfer for osd
791 * setattr needs. As a reminder, only the root user can change ownership of
792 * a file, that's why EDQUOT & EINPROGRESS errors are discarded */
793 static inline int qsd_transfer(const struct lu_env *env,
794 struct qsd_instance *qsd,
795 struct lquota_trans *trans, int qtype,
796 __u64 orig_id, __u64 new_id, __u64 bspace,
797 struct lquota_id_info *qi)
801 if (unlikely(qsd == NULL))
804 LASSERT(qtype >= 0 && qtype < MAXQUOTAS);
805 qi->lqi_type = qtype;
807 /* inode accounting */
808 qi->lqi_is_blk = false;
810 /* one more inode for the new owner ... */
811 qi->lqi_id.qid_uid = new_id;
813 rc = qsd_op_begin(env, qsd, trans, qi, NULL);
814 if (rc == -EDQUOT || rc == -EINPROGRESS)
819 /* and one less inode for the current id */
820 qi->lqi_id.qid_uid = orig_id;;
822 /* can't get EDQUOT when reducing usage */
823 rc = qsd_op_begin(env, qsd, trans, qi, NULL);
824 if (rc == -EINPROGRESS)
829 /* block accounting */
830 qi->lqi_is_blk = true;
832 /* more blocks for the new owner ... */
833 qi->lqi_id.qid_uid = new_id;
834 qi->lqi_space = bspace;
835 rc = qsd_op_begin(env, qsd, trans, qi, NULL);
836 if (rc == -EDQUOT || rc == -EINPROGRESS)
841 /* and finally less blocks for the current owner */
842 qi->lqi_id.qid_uid = orig_id;
843 qi->lqi_space = -bspace;
844 rc = qsd_op_begin(env, qsd, trans, qi, NULL);
845 /* can't get EDQUOT when reducing usage */
846 if (rc == -EINPROGRESS)
851 static int osd_declare_attr_set(const struct lu_env *env,
852 struct dt_object *dt,
853 const struct lu_attr *attr,
854 struct thandle *handle)
856 struct osd_thread_info *info = osd_oti_get(env);
857 char *buf = osd_oti_get(env)->oti_str;
858 struct osd_object *obj = osd_dt_obj(dt);
859 struct osd_device *osd = osd_obj2dev(obj);
860 struct osd_thandle *oh;
866 if (!dt_object_exists(dt)) {
867 /* XXX: sanity check that object creation is declared */
871 LASSERT(handle != NULL);
872 LASSERT(osd_invariant(obj));
874 oh = container_of0(handle, struct osd_thandle, ot_super);
876 LASSERT(obj->oo_sa_hdl != NULL);
877 LASSERT(oh->ot_tx != NULL);
878 dmu_tx_hold_sa(oh->ot_tx, obj->oo_sa_hdl, 0);
879 if (oh->ot_tx->tx_err != 0)
880 RETURN(-oh->ot_tx->tx_err);
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;
936 __u64 valid = la->la_valid;
941 LASSERT(handle != NULL);
942 LASSERT(dt_object_exists(dt));
943 LASSERT(osd_invariant(obj));
944 LASSERT(obj->oo_sa_hdl);
946 oh = container_of0(handle, struct osd_thandle, ot_super);
947 /* Assert that the transaction has been assigned to a
948 transaction group. */
949 LASSERT(oh->ot_tx->tx_txg != 0);
951 /* Only allow set size for regular file */
952 if (!S_ISREG(dt->do_lu.lo_header->loh_attr))
953 valid &= ~(LA_SIZE | LA_BLOCKS);
958 OBD_ALLOC(bulk, sizeof(sa_bulk_attr_t) * 10);
962 /* do both accounting updates outside oo_attr_lock below */
963 if ((valid & LA_UID) && (la->la_uid != obj->oo_attr.la_uid)) {
964 /* Update user accounting. Failure isn't fatal, but we still
965 * log an error message */
966 rc = -zap_increment_int(osd->od_objset.os, osd->od_iusr_oid,
967 la->la_uid, 1, oh->ot_tx);
969 CERROR("%s: failed to update accounting ZAP for user "
970 "%d (%d)\n", osd->od_svname, la->la_uid, rc);
971 rc = -zap_increment_int(osd->od_objset.os, osd->od_iusr_oid,
972 obj->oo_attr.la_uid, -1, oh->ot_tx);
974 CERROR("%s: failed to update accounting ZAP for user "
975 "%d (%d)\n", osd->od_svname,
976 obj->oo_attr.la_uid, rc);
978 if ((valid & LA_GID) && (la->la_gid != obj->oo_attr.la_gid)) {
979 /* Update group accounting. Failure isn't fatal, but we still
980 * log an error message */
981 rc = -zap_increment_int(osd->od_objset.os, osd->od_igrp_oid,
982 la->la_gid, 1, oh->ot_tx);
984 CERROR("%s: failed to update accounting ZAP for user "
985 "%d (%d)\n", osd->od_svname, la->la_gid, rc);
986 rc = -zap_increment_int(osd->od_objset.os, osd->od_igrp_oid,
987 obj->oo_attr.la_gid, -1, oh->ot_tx);
989 CERROR("%s: failed to update accounting ZAP for user "
990 "%d (%d)\n", osd->od_svname,
991 obj->oo_attr.la_gid, rc);
994 write_lock(&obj->oo_attr_lock);
996 if (valid & LA_ATIME) {
997 osa->atime[0] = obj->oo_attr.la_atime = la->la_atime;
998 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(uos), NULL,
1001 if (valid & LA_MTIME) {
1002 osa->mtime[0] = obj->oo_attr.la_mtime = la->la_mtime;
1003 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(uos), NULL,
1006 if (valid & LA_CTIME) {
1007 osa->ctime[0] = obj->oo_attr.la_ctime = la->la_ctime;
1008 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(uos), NULL,
1011 if (valid & LA_MODE) {
1012 /* mode is stored along with type, so read it first */
1013 obj->oo_attr.la_mode = (obj->oo_attr.la_mode & S_IFMT) |
1014 (la->la_mode & ~S_IFMT);
1015 osa->mode = obj->oo_attr.la_mode;
1016 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(uos), NULL,
1019 if (valid & LA_SIZE) {
1020 osa->size = obj->oo_attr.la_size = la->la_size;
1021 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(uos), NULL,
1024 if (valid & LA_NLINK) {
1025 osa->nlink = obj->oo_attr.la_nlink = la->la_nlink;
1026 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(uos), NULL,
1029 if (valid & LA_RDEV) {
1030 osa->rdev = obj->oo_attr.la_rdev = la->la_rdev;
1031 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_RDEV(uos), NULL,
1034 if (valid & LA_FLAGS) {
1035 osa->flags = attrs_fs2zfs(la->la_flags);
1036 /* many flags are not supported by zfs, so ensure a good cached
1038 obj->oo_attr.la_flags = attrs_zfs2fs(osa->flags);
1039 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(uos), NULL,
1042 if (valid & LA_UID) {
1043 osa->uid = obj->oo_attr.la_uid = la->la_uid;
1044 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(uos), NULL,
1047 if (valid & LA_GID) {
1048 osa->gid = obj->oo_attr.la_gid = la->la_gid;
1049 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(uos), NULL,
1052 obj->oo_attr.la_valid |= valid;
1053 write_unlock(&obj->oo_attr_lock);
1055 rc = osd_object_sa_bulk_update(obj, bulk, cnt, oh);
1057 OBD_FREE(bulk, sizeof(sa_bulk_attr_t) * 10);
1064 * XXX temporary solution.
1067 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
1068 struct dt_object *parent, struct dt_object *child,
1073 memset(ah, 0, sizeof(*ah));
1074 ah->dah_parent = parent;
1075 ah->dah_mode = child_mode;
1078 static int osd_declare_object_create(const struct lu_env *env,
1079 struct dt_object *dt,
1080 struct lu_attr *attr,
1081 struct dt_allocation_hint *hint,
1082 struct dt_object_format *dof,
1083 struct thandle *handle)
1085 char *buf = osd_oti_get(env)->oti_str;
1086 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1087 struct osd_object *obj = osd_dt_obj(dt);
1088 struct osd_device *osd = osd_obj2dev(obj);
1089 struct osd_thandle *oh;
1096 switch (dof->dof_type) {
1100 if (obj->oo_dt.do_body_ops == NULL)
1101 obj->oo_dt.do_body_ops = &osd_body_ops;
1107 LASSERT(handle != NULL);
1108 oh = container_of0(handle, struct osd_thandle, ot_super);
1109 LASSERT(oh->ot_tx != NULL);
1111 switch (dof->dof_type) {
1114 /* for zap create */
1115 dmu_tx_hold_zap(oh->ot_tx, DMU_NEW_OBJECT, 1, NULL);
1120 /* first, we'll create new object */
1121 dmu_tx_hold_bonus(oh->ot_tx, DMU_NEW_OBJECT);
1129 /* and we'll add it to some mapping */
1130 zapid = osd_get_name_n_idx(env, osd, fid, buf);
1131 dmu_tx_hold_bonus(oh->ot_tx, zapid);
1132 dmu_tx_hold_zap(oh->ot_tx, zapid, TRUE, buf);
1134 /* we will also update inode accounting ZAPs */
1135 dmu_tx_hold_bonus(oh->ot_tx, osd->od_iusr_oid);
1136 dmu_tx_hold_zap(oh->ot_tx, osd->od_iusr_oid, TRUE, buf);
1137 dmu_tx_hold_bonus(oh->ot_tx, osd->od_igrp_oid);
1138 dmu_tx_hold_zap(oh->ot_tx, osd->od_igrp_oid, TRUE, buf);
1140 dmu_tx_hold_sa_create(oh->ot_tx, ZFS_SA_BASE_ATTR_SIZE);
1142 __osd_xattr_declare_set(env, obj, sizeof(struct lustre_mdt_attrs),
1143 XATTR_NAME_LMA, oh);
1145 rc = osd_declare_quota(env, osd, attr->la_uid, attr->la_gid, 1, oh,
1146 false, NULL, false);
1150 int __osd_attr_init(const struct lu_env *env, udmu_objset_t *uos, uint64_t oid,
1151 dmu_tx_t *tx, struct lu_attr *la, uint64_t parent)
1153 sa_bulk_attr_t *bulk;
1154 sa_handle_t *sa_hdl;
1155 struct osa_attr *osa = &osd_oti_get(env)->oti_osa;
1163 gen = dmu_tx_get_txg(tx);
1165 ZFS_TIME_ENCODE(&now, crtime);
1167 osa->atime[0] = la->la_atime;
1168 osa->ctime[0] = la->la_ctime;
1169 osa->mtime[0] = la->la_mtime;
1170 osa->mode = la->la_mode;
1171 osa->uid = la->la_uid;
1172 osa->gid = la->la_gid;
1173 osa->rdev = la->la_rdev;
1174 osa->nlink = la->la_nlink;
1175 osa->flags = attrs_fs2zfs(la->la_flags);
1176 osa->size = la->la_size;
1178 /* Now add in all of the "SA" attributes */
1179 rc = -sa_handle_get(uos->os, oid, NULL, SA_HDL_PRIVATE, &sa_hdl);
1183 OBD_ALLOC(bulk, sizeof(sa_bulk_attr_t) * 13);
1189 * we need to create all SA below upon object create.
1191 * XXX The attribute order matters since the accounting callback relies
1192 * on static offsets (i.e. SA_*_OFFSET, see zfs_space_delta_cb()) to
1193 * look up the UID/GID attributes. Moreover, the callback does not seem
1194 * to support the spill block.
1195 * We define attributes in the same order as SA_*_OFFSET in order to
1196 * work around the problem. See ORI-610.
1199 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(uos), NULL, &osa->mode, 8);
1200 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(uos), NULL, &osa->size, 8);
1201 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GEN(uos), NULL, &gen, 8);
1202 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(uos), NULL, &osa->uid, 8);
1203 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(uos), NULL, &osa->gid, 8);
1204 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_PARENT(uos), NULL, &parent, 8);
1205 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(uos), NULL, &osa->flags, 8);
1206 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(uos), NULL, osa->atime, 16);
1207 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(uos), NULL, osa->mtime, 16);
1208 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(uos), NULL, osa->ctime, 16);
1209 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CRTIME(uos), NULL, crtime, 16);
1210 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(uos), NULL, &osa->nlink, 8);
1211 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_RDEV(uos), NULL, &osa->rdev, 8);
1213 rc = -sa_replace_all_by_template(sa_hdl, bulk, cnt, tx);
1215 OBD_FREE(bulk, sizeof(sa_bulk_attr_t) * 13);
1217 sa_handle_destroy(sa_hdl);
1222 * The transaction passed to this routine must have
1223 * dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT) called and then assigned
1224 * to a transaction group.
1226 int __osd_object_create(const struct lu_env *env, udmu_objset_t *uos,
1227 dmu_buf_t **dbp, dmu_tx_t *tx, struct lu_attr *la,
1228 uint64_t parent, void *tag)
1234 /* Assert that the transaction has been assigned to a
1235 transaction group. */
1236 LASSERT(tx->tx_txg != 0);
1238 /* Create a new DMU object. */
1239 oid = dmu_object_alloc(uos->os, DMU_OT_PLAIN_FILE_CONTENTS, 0,
1240 DMU_OT_SA, DN_MAX_BONUSLEN, tx);
1241 rc = -sa_buf_hold(uos->os, oid, tag, dbp);
1242 LASSERTF(rc == 0, "sa_buf_hold "LPU64" failed: %d\n", oid, rc);
1244 LASSERT(la->la_valid & LA_MODE);
1248 rc = __osd_attr_init(env, uos, oid, tx, la, parent);
1250 sa_buf_rele(*dbp, tag);
1252 dmu_object_free(uos->os, oid, tx);
1256 spin_lock(&uos->lock);
1258 spin_unlock(&uos->lock);
1263 * The transaction passed to this routine must have
1264 * dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, ...) called and then assigned
1265 * to a transaction group.
1267 * Using ZAP_FLAG_HASH64 will force the ZAP to always be a FAT ZAP.
1268 * This is fine for directories today, because storing the FID in the dirent
1269 * will also require a FAT ZAP. If there is a new type of micro ZAP created
1270 * then we might need to re-evaluate the use of this flag and instead do
1271 * a conversion from the different internal ZAP hash formats being used. */
1272 int __osd_zap_create(const struct lu_env *env, udmu_objset_t *uos,
1273 dmu_buf_t **zap_dbp, dmu_tx_t *tx,
1274 struct lu_attr *la, uint64_t parent,
1275 void *tag, zap_flags_t flags)
1282 spin_lock(&uos->lock);
1284 spin_unlock(&uos->lock);
1286 /* Assert that the transaction has been assigned to a
1287 transaction group. */
1288 LASSERT(tx->tx_txg != 0);
1290 oid = zap_create_flags(uos->os, 0, flags | ZAP_FLAG_HASH64,
1291 DMU_OT_DIRECTORY_CONTENTS,
1292 14, /* == ZFS fzap_default_block_shift */
1293 DN_MAX_INDBLKSHIFT, /* indirect block shift */
1294 DMU_OT_SA, DN_MAX_BONUSLEN, tx);
1296 rc = -sa_buf_hold(uos->os, oid, tag, zap_dbp);
1300 LASSERT(la->la_valid & LA_MODE);
1304 return __osd_attr_init(env, uos, oid, tx, la, parent);
1307 static dmu_buf_t *osd_mkidx(const struct lu_env *env, struct osd_device *osd,
1308 struct lu_attr *la, uint64_t parent,
1309 struct osd_thandle *oh)
1314 /* Index file should be created as regular file in order not to confuse
1315 * ZPL which could interpret them as directory.
1316 * We set ZAP_FLAG_UINT64_KEY to let ZFS know than we are going to use
1318 LASSERT(S_ISREG(la->la_mode));
1319 rc = __osd_zap_create(env, &osd->od_objset, &db, oh->ot_tx, la,
1320 parent, osd_obj_tag, ZAP_FLAG_UINT64_KEY);
1326 static dmu_buf_t *osd_mkdir(const struct lu_env *env, struct osd_device *osd,
1327 struct lu_attr *la, uint64_t parent,
1328 struct osd_thandle *oh)
1333 LASSERT(S_ISDIR(la->la_mode));
1334 rc = __osd_zap_create(env, &osd->od_objset, &db, oh->ot_tx, la,
1335 parent, osd_obj_tag, 0);
1341 static dmu_buf_t* osd_mkreg(const struct lu_env *env, struct osd_device *osd,
1342 struct lu_attr *la, uint64_t parent,
1343 struct osd_thandle *oh)
1348 LASSERT(S_ISREG(la->la_mode));
1349 rc = __osd_object_create(env, &osd->od_objset, &db, oh->ot_tx, la,
1350 parent, osd_obj_tag);
1355 * XXX: a hack, OST to use bigger blocksize. we need
1356 * a method in OSD API to control this from OFD/MDD
1358 if (!lu_device_is_md(osd2lu_dev(osd))) {
1359 rc = -dmu_object_set_blocksize(osd->od_objset.os,
1361 128 << 10, 0, oh->ot_tx);
1363 CERROR("%s: can't change blocksize: %d\n",
1364 osd->od_svname, rc);
1372 static dmu_buf_t *osd_mksym(const struct lu_env *env, struct osd_device *osd,
1373 struct lu_attr *la, uint64_t parent,
1374 struct osd_thandle *oh)
1379 LASSERT(S_ISLNK(la->la_mode));
1380 rc = __osd_object_create(env, &osd->od_objset, &db, oh->ot_tx, la,
1381 parent, osd_obj_tag);
1387 static dmu_buf_t *osd_mknod(const struct lu_env *env, struct osd_device *osd,
1388 struct lu_attr *la, uint64_t parent,
1389 struct osd_thandle *oh)
1394 la->la_valid = LA_MODE;
1395 if (S_ISCHR(la->la_mode) || S_ISBLK(la->la_mode))
1396 la->la_valid |= LA_RDEV;
1398 rc = __osd_object_create(env, &osd->od_objset, &db, oh->ot_tx, la,
1399 parent, osd_obj_tag);
1405 typedef dmu_buf_t *(*osd_obj_type_f)(const struct lu_env *env,
1406 struct osd_device *osd,
1409 struct osd_thandle *oh);
1411 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
1413 osd_obj_type_f result;
1439 * Primitives for directory (i.e. ZAP) handling
1441 static inline int osd_init_lma(const struct lu_env *env, struct osd_object *obj,
1442 const struct lu_fid *fid, struct osd_thandle *oh)
1444 struct osd_thread_info *info = osd_oti_get(env);
1445 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
1449 lustre_lma_init(lma, fid, 0, 0);
1450 lustre_lma_swab(lma);
1452 buf.lb_len = sizeof(*lma);
1454 rc = osd_xattr_set_internal(env, obj, &buf, XATTR_NAME_LMA,
1455 LU_XATTR_CREATE, oh, BYPASS_CAPA);
1461 * Concurrency: @dt is write locked.
1463 static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
1464 struct lu_attr *attr,
1465 struct dt_allocation_hint *hint,
1466 struct dt_object_format *dof,
1469 struct zpl_direntry *zde = &osd_oti_get(env)->oti_zde.lzd_reg;
1470 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1471 struct osd_object *obj = osd_dt_obj(dt);
1472 struct osd_device *osd = osd_obj2dev(obj);
1473 char *buf = osd_oti_get(env)->oti_str;
1474 struct osd_thandle *oh;
1481 /* concurrent create declarations should not see
1482 * the object inconsistent (db, attr, etc).
1483 * in regular cases acquisition should be cheap */
1484 down(&obj->oo_guard);
1486 LASSERT(osd_invariant(obj));
1487 LASSERT(!dt_object_exists(dt));
1488 LASSERT(dof != NULL);
1490 LASSERT(th != NULL);
1491 oh = container_of0(th, struct osd_thandle, ot_super);
1494 * XXX missing: Quote handling.
1497 LASSERT(obj->oo_db == NULL);
1499 /* to follow ZFS on-disk format we need
1500 * to initialize parent dnode properly */
1502 if (hint && hint->dah_parent)
1503 zapid = osd_dt_obj(hint->dah_parent)->oo_db->db_object;
1505 db = osd_create_type_f(dof->dof_type)(env, osd, attr, zapid, oh);
1507 GOTO(out, rc = PTR_ERR(db));
1510 zde->zde_dnode = db->db_object;
1511 zde->zde_type = IFTODT(attr->la_mode & S_IFMT);
1513 zapid = osd_get_name_n_idx(env, osd, fid, buf);
1515 rc = -zap_add(osd->od_objset.os, zapid, buf, 8, 1, zde, oh->ot_tx);
1519 /* Add new object to inode accounting.
1520 * Errors are not considered as fatal */
1521 rc = -zap_increment_int(osd->od_objset.os, osd->od_iusr_oid,
1522 (attr->la_valid & LA_UID) ? attr->la_uid : 0, 1,
1525 CERROR("%s: failed to add "DFID" to accounting ZAP for usr %d "
1526 "(%d)\n", osd->od_svname, PFID(fid), attr->la_uid, rc);
1527 rc = -zap_increment_int(osd->od_objset.os, osd->od_igrp_oid,
1528 (attr->la_valid & LA_GID) ? attr->la_gid : 0, 1,
1531 CERROR("%s: failed to add "DFID" to accounting ZAP for grp %d "
1532 "(%d)\n", osd->od_svname, PFID(fid), attr->la_gid, rc);
1534 /* configure new osd object */
1536 rc = osd_object_init0(env, obj);
1537 LASSERT(ergo(rc == 0, dt_object_exists(dt)));
1538 LASSERT(osd_invariant(obj));
1540 rc = osd_init_lma(env, obj, fid, oh);
1542 CERROR("%s: can not set LMA on "DFID": rc = %d\n",
1543 osd->od_svname, PFID(fid), rc);
1544 /* ignore errors during LMA initialization */
1553 static int osd_declare_object_ref_add(const struct lu_env *env,
1554 struct dt_object *dt,
1557 return osd_declare_attr_set(env, dt, NULL, th);
1561 * Concurrency: @dt is write locked.
1563 static int osd_object_ref_add(const struct lu_env *env,
1564 struct dt_object *dt,
1565 struct thandle *handle)
1567 struct osd_object *obj = osd_dt_obj(dt);
1568 struct osd_thandle *oh;
1569 struct osd_device *osd = osd_obj2dev(obj);
1570 udmu_objset_t *uos = &osd->od_objset;
1576 LASSERT(osd_invariant(obj));
1577 LASSERT(dt_object_exists(dt));
1578 LASSERT(obj->oo_sa_hdl != NULL);
1580 oh = container_of0(handle, struct osd_thandle, ot_super);
1582 write_lock(&obj->oo_attr_lock);
1583 nlink = ++obj->oo_attr.la_nlink;
1584 write_unlock(&obj->oo_attr_lock);
1586 rc = osd_object_sa_update(obj, SA_ZPL_LINKS(uos), &nlink, 8, oh);
1590 static int osd_declare_object_ref_del(const struct lu_env *env,
1591 struct dt_object *dt,
1592 struct thandle *handle)
1594 return osd_declare_attr_set(env, dt, NULL, handle);
1598 * Concurrency: @dt is write locked.
1600 static int osd_object_ref_del(const struct lu_env *env,
1601 struct dt_object *dt,
1602 struct thandle *handle)
1604 struct osd_object *obj = osd_dt_obj(dt);
1605 struct osd_thandle *oh;
1606 struct osd_device *osd = osd_obj2dev(obj);
1607 udmu_objset_t *uos = &osd->od_objset;
1613 LASSERT(osd_invariant(obj));
1614 LASSERT(dt_object_exists(dt));
1615 LASSERT(obj->oo_sa_hdl != NULL);
1617 oh = container_of0(handle, struct osd_thandle, ot_super);
1618 LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
1620 write_lock(&obj->oo_attr_lock);
1621 nlink = --obj->oo_attr.la_nlink;
1622 write_unlock(&obj->oo_attr_lock);
1624 rc = osd_object_sa_update(obj, SA_ZPL_LINKS(uos), &nlink, 8, oh);
1628 static int capa_is_sane(const struct lu_env *env, struct osd_device *dev,
1629 struct lustre_capa *capa, struct lustre_capa_key *keys)
1631 struct osd_thread_info *oti = osd_oti_get(env);
1632 struct obd_capa *oc;
1636 oc = capa_lookup(dev->od_capa_hash, capa, 0);
1638 if (capa_is_expired(oc)) {
1639 DEBUG_CAPA(D_ERROR, capa, "expired");
1646 spin_lock(&capa_lock);
1647 for (i = 0; i < 2; i++) {
1648 if (keys[i].lk_keyid == capa->lc_keyid) {
1649 oti->oti_capa_key = keys[i];
1653 spin_unlock(&capa_lock);
1656 DEBUG_CAPA(D_ERROR, capa, "no matched capa key");
1660 rc = capa_hmac(oti->oti_capa.lc_hmac, capa, oti->oti_capa_key.lk_key);
1663 if (memcmp(oti->oti_capa.lc_hmac, capa->lc_hmac, sizeof(capa->lc_hmac)))
1665 DEBUG_CAPA(D_ERROR, capa, "HMAC mismatch");
1669 oc = capa_add(dev->od_capa_hash, capa);
1675 static int osd_object_auth(const struct lu_env *env, struct dt_object *dt,
1676 struct lustre_capa *capa, __u64 opc)
1678 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1679 struct osd_device *dev = osd_dev(dt->do_lu.lo_dev);
1682 if (!dev->od_fl_capa)
1685 if (capa == BYPASS_CAPA)
1689 CERROR("no capability is provided for fid "DFID"\n", PFID(fid));
1693 if (!lu_fid_eq(fid, &capa->lc_fid)) {
1694 DEBUG_CAPA(D_ERROR, capa, "fid "DFID" mismatch with",PFID(fid));
1698 if (!capa_opc_supported(capa, opc)) {
1699 DEBUG_CAPA(D_ERROR, capa, "opc "LPX64" not supported by", opc);
1703 if ((rc = capa_is_sane(env, dev, capa, dev->od_capa_keys))) {
1704 DEBUG_CAPA(D_ERROR, capa, "insane (rc %d)", rc);
1711 static struct obd_capa *osd_capa_get(const struct lu_env *env,
1712 struct dt_object *dt,
1713 struct lustre_capa *old,
1716 struct osd_thread_info *info = osd_oti_get(env);
1717 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1718 struct osd_object *obj = osd_dt_obj(dt);
1719 struct osd_device *dev = osd_obj2dev(obj);
1720 struct lustre_capa_key *key = &info->oti_capa_key;
1721 struct lustre_capa *capa = &info->oti_capa;
1722 struct obd_capa *oc;
1726 if (!dev->od_fl_capa)
1727 RETURN(ERR_PTR(-ENOENT));
1729 LASSERT(dt_object_exists(dt));
1730 LASSERT(osd_invariant(obj));
1732 /* renewal sanity check */
1733 if (old && osd_object_auth(env, dt, old, opc))
1734 RETURN(ERR_PTR(-EACCES));
1736 capa->lc_fid = *fid;
1739 capa->lc_flags = dev->od_capa_alg << 24;
1740 capa->lc_timeout = dev->od_capa_timeout;
1741 capa->lc_expiry = 0;
1743 oc = capa_lookup(dev->od_capa_hash, capa, 1);
1745 LASSERT(!capa_is_expired(oc));
1749 spin_lock(&capa_lock);
1750 *key = dev->od_capa_keys[1];
1751 spin_unlock(&capa_lock);
1753 capa->lc_keyid = key->lk_keyid;
1754 capa->lc_expiry = cfs_time_current_sec() + dev->od_capa_timeout;
1756 rc = capa_hmac(capa->lc_hmac, capa, key->lk_key);
1758 DEBUG_CAPA(D_ERROR, capa, "HMAC failed: %d for", rc);
1760 RETURN(ERR_PTR(rc));
1763 oc = capa_add(dev->od_capa_hash, capa);
1767 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
1768 __u64 start, __u64 end)
1770 struct osd_device *osd = osd_obj2dev(osd_dt_obj(dt));
1773 /* XXX: no other option than syncing the whole filesystem until we
1774 * support ZIL. If the object tracked the txg that it was last
1775 * modified in, it could pass that txg here instead of "0". Maybe
1776 * the changes are already committed, so no wait is needed at all? */
1777 txg_wait_synced(dmu_objset_pool(osd->od_objset.os), 0ULL);
1782 static struct dt_object_operations osd_obj_ops = {
1783 .do_read_lock = osd_object_read_lock,
1784 .do_write_lock = osd_object_write_lock,
1785 .do_read_unlock = osd_object_read_unlock,
1786 .do_write_unlock = osd_object_write_unlock,
1787 .do_write_locked = osd_object_write_locked,
1788 .do_attr_get = osd_attr_get,
1789 .do_declare_attr_set = osd_declare_attr_set,
1790 .do_attr_set = osd_attr_set,
1791 .do_ah_init = osd_ah_init,
1792 .do_declare_create = osd_declare_object_create,
1793 .do_create = osd_object_create,
1794 .do_declare_destroy = osd_declare_object_destroy,
1795 .do_destroy = osd_object_destroy,
1796 .do_index_try = osd_index_try,
1797 .do_declare_ref_add = osd_declare_object_ref_add,
1798 .do_ref_add = osd_object_ref_add,
1799 .do_declare_ref_del = osd_declare_object_ref_del,
1800 .do_ref_del = osd_object_ref_del,
1801 .do_xattr_get = osd_xattr_get,
1802 .do_declare_xattr_set = osd_declare_xattr_set,
1803 .do_xattr_set = osd_xattr_set,
1804 .do_declare_xattr_del = osd_declare_xattr_del,
1805 .do_xattr_del = osd_xattr_del,
1806 .do_xattr_list = osd_xattr_list,
1807 .do_capa_get = osd_capa_get,
1808 .do_object_sync = osd_object_sync,
1811 static struct lu_object_operations osd_lu_obj_ops = {
1812 .loo_object_init = osd_object_init,
1813 .loo_object_delete = osd_object_delete,
1814 .loo_object_release = osd_object_release,
1815 .loo_object_free = osd_object_free,
1816 .loo_object_print = osd_object_print,
1817 .loo_object_invariant = osd_object_invariant,
1820 static int osd_otable_it_attr_get(const struct lu_env *env,
1821 struct dt_object *dt,
1822 struct lu_attr *attr,
1823 struct lustre_capa *capa)
1829 static struct dt_object_operations osd_obj_otable_it_ops = {
1830 .do_attr_get = osd_otable_it_attr_get,
1831 .do_index_try = osd_index_try,