4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2009, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2012, 2014, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/osd-zfs/osd_object.c
38 * Author: Alex Zhuravlev <bzzz@whamcloud.com>
39 * Author: Mike Pershin <tappro@whamcloud.com>
40 * Author: Johann Lombardi <johann@whamcloud.com>
43 #define DEBUG_SUBSYSTEM S_OSD
45 #include <lustre_ver.h>
46 #include <libcfs/libcfs.h>
47 #include <obd_support.h>
48 #include <lustre_net.h>
50 #include <obd_class.h>
51 #include <lustre_disk.h>
52 #include <lustre_fid.h>
54 #include "osd_internal.h"
56 #include <sys/dnode.h>
61 #include <sys/spa_impl.h>
62 #include <sys/zfs_znode.h>
63 #include <sys/dmu_tx.h>
64 #include <sys/dmu_objset.h>
65 #include <sys/dsl_prop.h>
66 #include <sys/sa_impl.h>
69 char *osd_obj_tag = "osd_object";
71 static struct dt_object_operations osd_obj_ops;
72 static struct lu_object_operations osd_lu_obj_ops;
73 extern struct dt_body_operations osd_body_ops;
74 static struct dt_object_operations osd_obj_otable_it_ops;
76 extern struct kmem_cache *osd_object_kmem;
79 osd_object_sa_fini(struct osd_object *obj)
82 sa_handle_destroy(obj->oo_sa_hdl);
83 obj->oo_sa_hdl = NULL;
88 osd_object_sa_init(struct osd_object *obj, struct osd_device *o)
92 LASSERT(obj->oo_sa_hdl == NULL);
93 LASSERT(obj->oo_db != NULL);
95 rc = -sa_handle_get(o->od_os, obj->oo_db->db_object, obj,
96 SA_HDL_PRIVATE, &obj->oo_sa_hdl);
100 /* Cache the xattr object id, valid for the life of the object */
101 rc = -sa_lookup(obj->oo_sa_hdl, SA_ZPL_XATTR(o), &obj->oo_xattr, 8);
103 obj->oo_xattr = ZFS_NO_OBJECT;
106 osd_object_sa_fini(obj);
113 * Add object to list of dirty objects in tx handle.
116 osd_object_sa_dirty_add(struct osd_object *obj, struct osd_thandle *oh)
118 if (!list_empty(&obj->oo_sa_linkage))
121 down(&oh->ot_sa_lock);
122 write_lock(&obj->oo_attr_lock);
123 if (likely(list_empty(&obj->oo_sa_linkage)))
124 list_add(&obj->oo_sa_linkage, &oh->ot_sa_list);
125 write_unlock(&obj->oo_attr_lock);
130 * Release spill block dbuf hold for all dirty SAs.
132 void osd_object_sa_dirty_rele(struct osd_thandle *oh)
134 struct osd_object *obj;
136 down(&oh->ot_sa_lock);
137 while (!list_empty(&oh->ot_sa_list)) {
138 obj = list_entry(oh->ot_sa_list.next,
139 struct osd_object, oo_sa_linkage);
140 sa_spill_rele(obj->oo_sa_hdl);
141 write_lock(&obj->oo_attr_lock);
142 list_del_init(&obj->oo_sa_linkage);
143 write_unlock(&obj->oo_attr_lock);
149 * Update the SA and add the object to the dirty list.
151 int osd_object_sa_update(struct osd_object *obj, sa_attr_type_t type,
152 void *buf, uint32_t buflen, struct osd_thandle *oh)
156 LASSERT(obj->oo_sa_hdl != NULL);
157 LASSERT(oh->ot_tx != NULL);
159 rc = -sa_update(obj->oo_sa_hdl, type, buf, buflen, oh->ot_tx);
160 osd_object_sa_dirty_add(obj, oh);
166 * Bulk update the SA and add the object to the dirty list.
169 osd_object_sa_bulk_update(struct osd_object *obj, sa_bulk_attr_t *attrs,
170 int count, struct osd_thandle *oh)
174 LASSERT(obj->oo_sa_hdl != NULL);
175 LASSERT(oh->ot_tx != NULL);
177 rc = -sa_bulk_update(obj->oo_sa_hdl, attrs, count, oh->ot_tx);
178 osd_object_sa_dirty_add(obj, oh);
184 * Retrieve the attributes of a DMU object
186 int __osd_object_attr_get(const struct lu_env *env, struct osd_device *o,
187 struct osd_object *obj, struct lu_attr *la)
189 struct osa_attr *osa = &osd_oti_get(env)->oti_osa;
191 sa_bulk_attr_t *bulk;
196 LASSERT(obj->oo_db != NULL);
198 rc = -sa_handle_get(o->od_os, obj->oo_db->db_object, NULL,
199 SA_HDL_PRIVATE, &sa_hdl);
203 OBD_ALLOC(bulk, sizeof(sa_bulk_attr_t) * 9);
205 GOTO(out_sa, rc = -ENOMEM);
207 la->la_valid |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE | LA_TYPE |
208 LA_SIZE | LA_UID | LA_GID | LA_FLAGS | LA_NLINK;
210 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(o), NULL, osa->atime, 16);
211 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(o), NULL, osa->mtime, 16);
212 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(o), NULL, osa->ctime, 16);
213 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(o), NULL, &osa->mode, 8);
214 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(o), NULL, &osa->size, 8);
215 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(o), NULL, &osa->nlink, 8);
216 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(o), NULL, &osa->uid, 8);
217 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(o), NULL, &osa->gid, 8);
218 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(o), NULL, &osa->flags, 8);
220 rc = -sa_bulk_lookup(sa_hdl, bulk, cnt);
224 la->la_atime = osa->atime[0];
225 la->la_mtime = osa->mtime[0];
226 la->la_ctime = osa->ctime[0];
227 la->la_mode = osa->mode;
228 la->la_uid = osa->uid;
229 la->la_gid = osa->gid;
230 la->la_nlink = osa->nlink;
231 la->la_flags = attrs_zfs2fs(osa->flags);
232 la->la_size = osa->size;
234 if (S_ISCHR(la->la_mode) || S_ISBLK(la->la_mode)) {
235 rc = -sa_lookup(sa_hdl, SA_ZPL_RDEV(o), &osa->rdev, 8);
238 la->la_rdev = osa->rdev;
239 la->la_valid |= LA_RDEV;
242 OBD_FREE(bulk, sizeof(sa_bulk_attr_t) * 9);
244 sa_handle_destroy(sa_hdl);
249 int __osd_obj2dbuf(const struct lu_env *env, objset_t *os,
250 uint64_t oid, dmu_buf_t **dbp)
252 dmu_object_info_t *doi = &osd_oti_get(env)->oti_doi;
255 rc = -sa_buf_hold(os, oid, osd_obj_tag, dbp);
259 dmu_object_info_from_db(*dbp, doi);
260 if (unlikely (oid != DMU_USERUSED_OBJECT &&
261 oid != DMU_GROUPUSED_OBJECT && doi->doi_bonus_type != DMU_OT_SA)) {
262 sa_buf_rele(*dbp, osd_obj_tag);
268 LASSERT((*dbp)->db_object == oid);
269 LASSERT((*dbp)->db_offset == -1);
270 LASSERT((*dbp)->db_data != NULL);
276 * Concurrency: no concurrent access is possible that early in object
279 struct lu_object *osd_object_alloc(const struct lu_env *env,
280 const struct lu_object_header *hdr,
283 struct osd_object *mo;
285 OBD_SLAB_ALLOC_PTR_GFP(mo, osd_object_kmem, GFP_NOFS);
289 l = &mo->oo_dt.do_lu;
290 dt_object_init(&mo->oo_dt, NULL, d);
291 mo->oo_dt.do_ops = &osd_obj_ops;
292 l->lo_ops = &osd_lu_obj_ops;
293 INIT_LIST_HEAD(&mo->oo_sa_linkage);
294 init_rwsem(&mo->oo_sem);
295 sema_init(&mo->oo_guard, 1);
296 rwlock_init(&mo->oo_attr_lock);
304 * Concurrency: shouldn't matter.
306 int osd_object_init0(const struct lu_env *env, struct osd_object *obj)
308 struct osd_device *osd = osd_obj2dev(obj);
309 const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
313 if (obj->oo_db == NULL)
318 rc = osd_object_sa_init(obj, osd);
322 /* cache attrs in object */
323 rc = __osd_object_attr_get(env, osd, obj, &obj->oo_attr);
327 if (likely(!fid_is_acct(fid)))
328 /* no body operations for accounting objects */
329 obj->oo_dt.do_body_ops = &osd_body_ops;
332 * initialize object before marking it existing
334 obj->oo_dt.do_lu.lo_header->loh_attr |= obj->oo_attr.la_mode & S_IFMT;
337 obj->oo_dt.do_lu.lo_header->loh_attr |= LOHA_EXISTS;
342 static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
344 struct osd_thread_info *info = osd_oti_get(env);
347 struct lustre_mdt_attrs *lma;
350 CLASSERT(sizeof(info->oti_buf) >= sizeof(*lma));
351 lma = (struct lustre_mdt_attrs *)info->oti_buf;
353 buf.lb_len = sizeof(info->oti_buf);
355 rc = osd_xattr_get(env, &obj->oo_dt, &buf, XATTR_NAME_LMA, BYPASS_CAPA);
358 lustre_lma_swab(lma);
359 if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
360 CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT))) {
361 CWARN("%s: unsupported incompat LMA feature(s) %#x for "
362 "fid = "DFID"\n", osd_obj2dev(obj)->od_svname,
363 lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
364 PFID(lu_object_fid(&obj->oo_dt.do_lu)));
367 } else if (rc == -ENODATA) {
368 /* haven't initialize LMA xattr */
376 * Concurrency: no concurrent access is possible that early in object
379 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
380 const struct lu_object_conf *conf)
382 struct osd_object *obj = osd_obj(l);
383 struct osd_device *osd = osd_obj2dev(obj);
388 LASSERT(osd_invariant(obj));
390 if (fid_is_otable_it(&l->lo_header->loh_fid)) {
391 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
392 l->lo_header->loh_attr |= LOHA_EXISTS;
396 rc = osd_fid_lookup(env, osd, lu_object_fid(l), &oid);
398 LASSERT(obj->oo_db == NULL);
399 rc = __osd_obj2dbuf(env, osd->od_os, oid, &obj->oo_db);
401 CERROR("%s: lookup "DFID"/"LPX64" failed: rc = %d\n",
402 osd->od_svname, PFID(lu_object_fid(l)), oid, rc);
406 rc = osd_object_init0(env, obj);
410 rc = osd_check_lma(env, obj);
413 } else if (rc == -ENOENT) {
416 LASSERT(osd_invariant(obj));
422 * Concurrency: no concurrent access is possible that late in object
425 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
427 struct osd_object *obj = osd_obj(l);
429 LASSERT(osd_invariant(obj));
431 dt_object_fini(&obj->oo_dt);
432 OBD_SLAB_FREE_PTR(obj, osd_object_kmem);
435 static void __osd_declare_object_destroy(const struct lu_env *env,
436 struct osd_object *obj,
437 struct osd_thandle *oh)
439 struct osd_device *osd = osd_obj2dev(obj);
440 dmu_buf_t *db = obj->oo_db;
441 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
442 uint64_t oid = db->db_object, xid;
443 dmu_tx_t *tx = oh->ot_tx;
447 dmu_tx_hold_free(tx, oid, 0, DMU_OBJECT_END);
449 /* zap holding xattrs */
450 if (obj->oo_xattr != ZFS_NO_OBJECT) {
453 dmu_tx_hold_free(tx, oid, 0, DMU_OBJECT_END);
455 rc = osd_zap_cursor_init(&zc, osd->od_os, oid, 0);
459 while ((rc = -zap_cursor_retrieve(zc, za)) == 0) {
460 BUG_ON(za->za_integer_length != sizeof(uint64_t));
461 BUG_ON(za->za_num_integers != 1);
463 rc = -zap_lookup(osd->od_os, obj->oo_xattr, za->za_name,
464 sizeof(uint64_t), 1, &xid);
466 CERROR("%s: xattr lookup failed: rc = %d\n",
470 dmu_tx_hold_free(tx, xid, 0, DMU_OBJECT_END);
472 zap_cursor_advance(zc);
477 osd_zap_cursor_fini(zc);
480 if (rc && tx->tx_err == 0)
484 static int osd_declare_object_destroy(const struct lu_env *env,
485 struct dt_object *dt,
488 char *buf = osd_oti_get(env)->oti_str;
489 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
490 struct osd_object *obj = osd_dt_obj(dt);
491 struct osd_device *osd = osd_obj2dev(obj);
492 struct osd_thandle *oh;
498 LASSERT(dt_object_exists(dt));
500 oh = container_of0(th, struct osd_thandle, ot_super);
501 LASSERT(oh->ot_tx != NULL);
503 /* declare that we'll destroy the object */
504 __osd_declare_object_destroy(env, obj, oh);
506 /* declare that we'll remove object from fid-dnode mapping */
507 zapid = osd_get_name_n_idx(env, osd, fid, buf);
508 dmu_tx_hold_bonus(oh->ot_tx, zapid);
509 dmu_tx_hold_zap(oh->ot_tx, zapid, 0, buf);
511 /* declare that we'll remove object from inode accounting ZAPs */
512 dmu_tx_hold_bonus(oh->ot_tx, osd->od_iusr_oid);
513 dmu_tx_hold_zap(oh->ot_tx, osd->od_iusr_oid, 0, buf);
514 dmu_tx_hold_bonus(oh->ot_tx, osd->od_igrp_oid);
515 dmu_tx_hold_zap(oh->ot_tx, osd->od_igrp_oid, 0, buf);
518 rc = osd_declare_quota(env, osd, obj->oo_attr.la_uid,
519 obj->oo_attr.la_gid, -1, oh, false, NULL, false);
523 /* data to be truncated */
524 rc = osd_declare_quota(env, osd, obj->oo_attr.la_uid,
525 obj->oo_attr.la_gid, 0, oh, true, NULL, false);
530 * Delete a DMU object
532 * The transaction passed to this routine must have
533 * dmu_tx_hold_free(tx, oid, 0, DMU_OBJECT_END) called
534 * and then assigned to a transaction group.
536 * This will release db and set it to NULL to prevent further dbuf releases.
538 static int __osd_object_destroy(const struct lu_env *env,
539 struct osd_object *obj,
540 dmu_tx_t *tx, void *tag)
542 struct osd_device *osd = osd_obj2dev(obj);
544 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
548 /* Assert that the transaction has been assigned to a
549 transaction group. */
550 LASSERT(tx->tx_txg != 0);
552 /* zap holding xattrs */
553 if (obj->oo_xattr != ZFS_NO_OBJECT) {
554 rc = osd_zap_cursor_init(&zc, osd->od_os, obj->oo_xattr, 0);
557 while ((rc = -zap_cursor_retrieve(zc, za)) == 0) {
558 BUG_ON(za->za_integer_length != sizeof(uint64_t));
559 BUG_ON(za->za_num_integers != 1);
561 rc = -zap_lookup(osd->od_os, obj->oo_xattr, za->za_name,
562 sizeof(uint64_t), 1, &xid);
564 CERROR("%s: lookup xattr %s failed: rc = %d\n",
565 osd->od_svname, za->za_name, rc);
568 rc = -dmu_object_free(osd->od_os, xid, tx);
570 CERROR("%s: fetch xattr %s failed: rc = %d\n",
571 osd->od_svname, za->za_name, rc);
572 zap_cursor_advance(zc);
574 osd_zap_cursor_fini(zc);
576 rc = -dmu_object_free(osd->od_os, obj->oo_xattr, tx);
578 CERROR("%s: freeing xattr failed: rc = %d\n",
582 return -dmu_object_free(osd->od_os, obj->oo_db->db_object, tx);
585 static int osd_object_destroy(const struct lu_env *env,
586 struct dt_object *dt,
589 char *buf = osd_oti_get(env)->oti_str;
590 struct osd_object *obj = osd_dt_obj(dt);
591 struct osd_device *osd = osd_obj2dev(obj);
592 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
593 struct osd_thandle *oh;
598 LASSERT(obj->oo_db != NULL);
599 LASSERT(dt_object_exists(dt));
600 LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
602 oh = container_of0(th, struct osd_thandle, ot_super);
604 LASSERT(oh->ot_tx != NULL);
606 zapid = osd_get_name_n_idx(env, osd, fid, buf);
608 /* remove obj ref from index dir (it depends) */
609 rc = -zap_remove(osd->od_os, zapid, buf, oh->ot_tx);
611 CERROR("%s: zap_remove() failed: rc = %d\n",
616 /* Remove object from inode accounting. It is not fatal for the destroy
617 * operation if something goes wrong while updating accounting, but we
618 * still log an error message to notify the administrator */
619 rc = -zap_increment_int(osd->od_os, osd->od_iusr_oid,
620 obj->oo_attr.la_uid, -1, oh->ot_tx);
622 CERROR("%s: failed to remove "DFID" from accounting ZAP for usr"
623 " %d: rc = %d\n", osd->od_svname, PFID(fid),
624 obj->oo_attr.la_uid, rc);
625 rc = -zap_increment_int(osd->od_os, osd->od_igrp_oid,
626 obj->oo_attr.la_gid, -1, oh->ot_tx);
628 CERROR("%s: failed to remove "DFID" from accounting ZAP for grp"
629 " %d: rc = %d\n", osd->od_svname, PFID(fid),
630 obj->oo_attr.la_gid, rc);
633 rc = __osd_object_destroy(env, obj, oh->ot_tx, osd_obj_tag);
635 CERROR("%s: __osd_object_destroy() failed: rc = %d\n",
641 /* not needed in the cache anymore */
642 set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
647 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
649 struct osd_object *obj = osd_obj(l);
651 if (obj->oo_db != NULL) {
652 osd_object_sa_fini(obj);
653 if (obj->oo_sa_xattr) {
654 nvlist_free(obj->oo_sa_xattr);
655 obj->oo_sa_xattr = NULL;
657 sa_buf_rele(obj->oo_db, osd_obj_tag);
658 list_del(&obj->oo_sa_linkage);
664 * Concurrency: ->loo_object_release() is called under site spin-lock.
666 static void osd_object_release(const struct lu_env *env,
672 * Concurrency: shouldn't matter.
674 static int osd_object_print(const struct lu_env *env, void *cookie,
675 lu_printer_t p, const struct lu_object *l)
677 struct osd_object *o = osd_obj(l);
679 return (*p)(env, cookie, LUSTRE_OSD_ZFS_NAME"-object@%p", o);
682 static void osd_object_read_lock(const struct lu_env *env,
683 struct dt_object *dt, unsigned role)
685 struct osd_object *obj = osd_dt_obj(dt);
687 LASSERT(osd_invariant(obj));
689 down_read(&obj->oo_sem);
692 static void osd_object_write_lock(const struct lu_env *env,
693 struct dt_object *dt, unsigned role)
695 struct osd_object *obj = osd_dt_obj(dt);
697 LASSERT(osd_invariant(obj));
699 down_write(&obj->oo_sem);
702 static void osd_object_read_unlock(const struct lu_env *env,
703 struct dt_object *dt)
705 struct osd_object *obj = osd_dt_obj(dt);
707 LASSERT(osd_invariant(obj));
708 up_read(&obj->oo_sem);
711 static void osd_object_write_unlock(const struct lu_env *env,
712 struct dt_object *dt)
714 struct osd_object *obj = osd_dt_obj(dt);
716 LASSERT(osd_invariant(obj));
717 up_write(&obj->oo_sem);
720 static int osd_object_write_locked(const struct lu_env *env,
721 struct dt_object *dt)
723 struct osd_object *obj = osd_dt_obj(dt);
726 LASSERT(osd_invariant(obj));
728 if (down_write_trylock(&obj->oo_sem)) {
730 up_write(&obj->oo_sem);
735 static int osd_attr_get(const struct lu_env *env,
736 struct dt_object *dt,
737 struct lu_attr *attr,
738 struct lustre_capa *capa)
740 struct osd_object *obj = osd_dt_obj(dt);
744 LASSERT(dt_object_exists(dt));
745 LASSERT(osd_invariant(obj));
748 read_lock(&obj->oo_attr_lock);
749 *attr = obj->oo_attr;
750 read_unlock(&obj->oo_attr_lock);
752 /* with ZFS_DEBUG zrl_add_debug() called by DB_DNODE_ENTER()
753 * from within sa_object_size() can block on a mutex, so
754 * we can't call sa_object_size() holding rwlock */
755 sa_object_size(obj->oo_sa_hdl, &blksize, &blocks);
756 /* we do not control size of indices, so always calculate
757 * it from number of blocks reported by DMU */
758 if (S_ISDIR(attr->la_mode))
759 attr->la_size = 512 * blocks;
760 /* Block size may be not set; suggest maximal I/O transfers. */
762 blksize = 1ULL << SPA_MAXBLOCKSHIFT;
764 attr->la_blksize = blksize;
765 attr->la_blocks = blocks;
766 attr->la_valid |= LA_BLOCKS | LA_BLKSIZE;
771 /* Simple wrapper on top of qsd API which implement quota transfer for osd
772 * setattr needs. As a reminder, only the root user can change ownership of
773 * a file, that's why EDQUOT & EINPROGRESS errors are discarded */
774 static inline int qsd_transfer(const struct lu_env *env,
775 struct qsd_instance *qsd,
776 struct lquota_trans *trans, int qtype,
777 __u64 orig_id, __u64 new_id, __u64 bspace,
778 struct lquota_id_info *qi)
782 if (unlikely(qsd == NULL))
785 LASSERT(qtype >= 0 && qtype < MAXQUOTAS);
786 qi->lqi_type = qtype;
788 /* inode accounting */
789 qi->lqi_is_blk = false;
791 /* one more inode for the new owner ... */
792 qi->lqi_id.qid_uid = new_id;
794 rc = qsd_op_begin(env, qsd, trans, qi, NULL);
795 if (rc == -EDQUOT || rc == -EINPROGRESS)
800 /* and one less inode for the current id */
801 qi->lqi_id.qid_uid = orig_id;;
803 /* can't get EDQUOT when reducing usage */
804 rc = qsd_op_begin(env, qsd, trans, qi, NULL);
805 if (rc == -EINPROGRESS)
810 /* block accounting */
811 qi->lqi_is_blk = true;
813 /* more blocks for the new owner ... */
814 qi->lqi_id.qid_uid = new_id;
815 qi->lqi_space = bspace;
816 rc = qsd_op_begin(env, qsd, trans, qi, NULL);
817 if (rc == -EDQUOT || rc == -EINPROGRESS)
822 /* and finally less blocks for the current owner */
823 qi->lqi_id.qid_uid = orig_id;
824 qi->lqi_space = -bspace;
825 rc = qsd_op_begin(env, qsd, trans, qi, NULL);
826 /* can't get EDQUOT when reducing usage */
827 if (rc == -EINPROGRESS)
832 static int osd_declare_attr_set(const struct lu_env *env,
833 struct dt_object *dt,
834 const struct lu_attr *attr,
835 struct thandle *handle)
837 struct osd_thread_info *info = osd_oti_get(env);
838 char *buf = osd_oti_get(env)->oti_str;
839 struct osd_object *obj = osd_dt_obj(dt);
840 struct osd_device *osd = osd_obj2dev(obj);
841 struct osd_thandle *oh;
847 if (!dt_object_exists(dt)) {
848 /* XXX: sanity check that object creation is declared */
852 LASSERT(handle != NULL);
853 LASSERT(osd_invariant(obj));
855 oh = container_of0(handle, struct osd_thandle, ot_super);
857 LASSERT(obj->oo_sa_hdl != NULL);
858 LASSERT(oh->ot_tx != NULL);
859 dmu_tx_hold_sa(oh->ot_tx, obj->oo_sa_hdl, 0);
860 if (oh->ot_tx->tx_err != 0)
861 RETURN(-oh->ot_tx->tx_err);
863 sa_object_size(obj->oo_sa_hdl, &blksize, &bspace);
864 bspace = toqb(bspace * blksize);
866 if (attr && attr->la_valid & LA_UID) {
867 /* account for user inode tracking ZAP update */
868 dmu_tx_hold_bonus(oh->ot_tx, osd->od_iusr_oid);
869 dmu_tx_hold_zap(oh->ot_tx, osd->od_iusr_oid, TRUE, buf);
871 /* quota enforcement for user */
872 if (attr->la_uid != obj->oo_attr.la_uid) {
873 rc = qsd_transfer(env, osd->od_quota_slave,
874 &oh->ot_quota_trans, USRQUOTA,
875 obj->oo_attr.la_uid, attr->la_uid,
876 bspace, &info->oti_qi);
881 if (attr && attr->la_valid & LA_GID) {
882 /* account for user inode tracking ZAP update */
883 dmu_tx_hold_bonus(oh->ot_tx, osd->od_igrp_oid);
884 dmu_tx_hold_zap(oh->ot_tx, osd->od_igrp_oid, TRUE, buf);
886 /* quota enforcement for group */
887 if (attr->la_gid != obj->oo_attr.la_gid) {
888 rc = qsd_transfer(env, osd->od_quota_slave,
889 &oh->ot_quota_trans, GRPQUOTA,
890 obj->oo_attr.la_gid, attr->la_gid,
891 bspace, &info->oti_qi);
901 * Set the attributes of an object
903 * The transaction passed to this routine must have
904 * dmu_tx_hold_bonus(tx, oid) called and then assigned
905 * to a transaction group.
907 static int osd_attr_set(const struct lu_env *env, struct dt_object *dt,
908 const struct lu_attr *la, struct thandle *handle,
909 struct lustre_capa *capa)
911 struct osd_object *obj = osd_dt_obj(dt);
912 struct osd_device *osd = osd_obj2dev(obj);
913 struct osd_thandle *oh;
914 struct osa_attr *osa = &osd_oti_get(env)->oti_osa;
915 sa_bulk_attr_t *bulk;
916 __u64 valid = la->la_valid;
921 LASSERT(handle != NULL);
922 LASSERT(dt_object_exists(dt));
923 LASSERT(osd_invariant(obj));
924 LASSERT(obj->oo_sa_hdl);
926 oh = container_of0(handle, struct osd_thandle, ot_super);
927 /* Assert that the transaction has been assigned to a
928 transaction group. */
929 LASSERT(oh->ot_tx->tx_txg != 0);
931 /* Only allow set size for regular file */
932 if (!S_ISREG(dt->do_lu.lo_header->loh_attr))
933 valid &= ~(LA_SIZE | LA_BLOCKS);
938 OBD_ALLOC(bulk, sizeof(sa_bulk_attr_t) * 10);
942 /* do both accounting updates outside oo_attr_lock below */
943 if ((valid & LA_UID) && (la->la_uid != obj->oo_attr.la_uid)) {
944 /* Update user accounting. Failure isn't fatal, but we still
945 * log an error message */
946 rc = -zap_increment_int(osd->od_os, osd->od_iusr_oid,
947 la->la_uid, 1, oh->ot_tx);
949 CERROR("%s: failed to update accounting ZAP for user "
950 "%d (%d)\n", osd->od_svname, la->la_uid, rc);
951 rc = -zap_increment_int(osd->od_os, osd->od_iusr_oid,
952 obj->oo_attr.la_uid, -1, oh->ot_tx);
954 CERROR("%s: failed to update accounting ZAP for user "
955 "%d (%d)\n", osd->od_svname,
956 obj->oo_attr.la_uid, rc);
958 if ((valid & LA_GID) && (la->la_gid != obj->oo_attr.la_gid)) {
959 /* Update group accounting. Failure isn't fatal, but we still
960 * log an error message */
961 rc = -zap_increment_int(osd->od_os, osd->od_igrp_oid,
962 la->la_gid, 1, oh->ot_tx);
964 CERROR("%s: failed to update accounting ZAP for user "
965 "%d (%d)\n", osd->od_svname, la->la_gid, rc);
966 rc = -zap_increment_int(osd->od_os, osd->od_igrp_oid,
967 obj->oo_attr.la_gid, -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_gid, rc);
974 write_lock(&obj->oo_attr_lock);
976 if (valid & LA_ATIME) {
977 osa->atime[0] = obj->oo_attr.la_atime = la->la_atime;
978 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(osd), NULL,
981 if (valid & LA_MTIME) {
982 osa->mtime[0] = obj->oo_attr.la_mtime = la->la_mtime;
983 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(osd), NULL,
986 if (valid & LA_CTIME) {
987 osa->ctime[0] = obj->oo_attr.la_ctime = la->la_ctime;
988 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(osd), NULL,
991 if (valid & LA_MODE) {
992 /* mode is stored along with type, so read it first */
993 obj->oo_attr.la_mode = (obj->oo_attr.la_mode & S_IFMT) |
994 (la->la_mode & ~S_IFMT);
995 osa->mode = obj->oo_attr.la_mode;
996 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(osd), NULL,
999 if (valid & LA_SIZE) {
1000 osa->size = obj->oo_attr.la_size = la->la_size;
1001 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(osd), NULL,
1004 if (valid & LA_NLINK) {
1005 osa->nlink = obj->oo_attr.la_nlink = la->la_nlink;
1006 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(osd), NULL,
1009 if (valid & LA_RDEV) {
1010 osa->rdev = obj->oo_attr.la_rdev = la->la_rdev;
1011 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_RDEV(osd), NULL,
1014 if (valid & LA_FLAGS) {
1015 osa->flags = attrs_fs2zfs(la->la_flags);
1016 /* many flags are not supported by zfs, so ensure a good cached
1018 obj->oo_attr.la_flags = attrs_zfs2fs(osa->flags);
1019 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(osd), NULL,
1022 if (valid & LA_UID) {
1023 osa->uid = obj->oo_attr.la_uid = la->la_uid;
1024 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(osd), NULL,
1027 if (valid & LA_GID) {
1028 osa->gid = obj->oo_attr.la_gid = la->la_gid;
1029 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(osd), NULL,
1032 obj->oo_attr.la_valid |= valid;
1033 write_unlock(&obj->oo_attr_lock);
1035 rc = osd_object_sa_bulk_update(obj, bulk, cnt, oh);
1037 OBD_FREE(bulk, sizeof(sa_bulk_attr_t) * 10);
1044 * XXX temporary solution.
1047 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
1048 struct dt_object *parent, struct dt_object *child,
1053 memset(ah, 0, sizeof(*ah));
1054 ah->dah_parent = parent;
1055 ah->dah_mode = child_mode;
1058 static int osd_declare_object_create(const struct lu_env *env,
1059 struct dt_object *dt,
1060 struct lu_attr *attr,
1061 struct dt_allocation_hint *hint,
1062 struct dt_object_format *dof,
1063 struct thandle *handle)
1065 char *buf = osd_oti_get(env)->oti_str;
1066 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1067 struct osd_object *obj = osd_dt_obj(dt);
1068 struct osd_device *osd = osd_obj2dev(obj);
1069 struct osd_thandle *oh;
1076 switch (dof->dof_type) {
1080 if (obj->oo_dt.do_body_ops == NULL)
1081 obj->oo_dt.do_body_ops = &osd_body_ops;
1087 LASSERT(handle != NULL);
1088 oh = container_of0(handle, struct osd_thandle, ot_super);
1089 LASSERT(oh->ot_tx != NULL);
1091 switch (dof->dof_type) {
1093 dt->do_index_ops = &osd_dir_ops;
1095 /* for zap create */
1096 dmu_tx_hold_zap(oh->ot_tx, DMU_NEW_OBJECT, 1, NULL);
1101 /* first, we'll create new object */
1102 dmu_tx_hold_bonus(oh->ot_tx, DMU_NEW_OBJECT);
1110 /* and we'll add it to some mapping */
1111 zapid = osd_get_name_n_idx(env, osd, fid, buf);
1112 dmu_tx_hold_bonus(oh->ot_tx, zapid);
1113 dmu_tx_hold_zap(oh->ot_tx, zapid, TRUE, buf);
1115 /* we will also update inode accounting ZAPs */
1116 dmu_tx_hold_bonus(oh->ot_tx, osd->od_iusr_oid);
1117 dmu_tx_hold_zap(oh->ot_tx, osd->od_iusr_oid, TRUE, buf);
1118 dmu_tx_hold_bonus(oh->ot_tx, osd->od_igrp_oid);
1119 dmu_tx_hold_zap(oh->ot_tx, osd->od_igrp_oid, TRUE, buf);
1121 dmu_tx_hold_sa_create(oh->ot_tx, ZFS_SA_BASE_ATTR_SIZE);
1123 __osd_xattr_declare_set(env, obj, sizeof(struct lustre_mdt_attrs),
1124 XATTR_NAME_LMA, oh);
1126 rc = osd_declare_quota(env, osd, attr->la_uid, attr->la_gid, 1, oh,
1127 false, NULL, false);
1131 int __osd_attr_init(const struct lu_env *env, struct osd_device *osd,
1132 uint64_t oid, dmu_tx_t *tx, struct lu_attr *la,
1135 sa_bulk_attr_t *bulk;
1136 sa_handle_t *sa_hdl;
1137 struct osa_attr *osa = &osd_oti_get(env)->oti_osa;
1145 gen = dmu_tx_get_txg(tx);
1147 ZFS_TIME_ENCODE(&now, crtime);
1149 osa->atime[0] = la->la_atime;
1150 osa->ctime[0] = la->la_ctime;
1151 osa->mtime[0] = la->la_mtime;
1152 osa->mode = la->la_mode;
1153 osa->uid = la->la_uid;
1154 osa->gid = la->la_gid;
1155 osa->rdev = la->la_rdev;
1156 osa->nlink = la->la_nlink;
1157 osa->flags = attrs_fs2zfs(la->la_flags);
1158 osa->size = la->la_size;
1160 /* Now add in all of the "SA" attributes */
1161 rc = -sa_handle_get(osd->od_os, oid, NULL, SA_HDL_PRIVATE, &sa_hdl);
1165 OBD_ALLOC(bulk, sizeof(sa_bulk_attr_t) * 13);
1171 * we need to create all SA below upon object create.
1173 * XXX The attribute order matters since the accounting callback relies
1174 * on static offsets (i.e. SA_*_OFFSET, see zfs_space_delta_cb()) to
1175 * look up the UID/GID attributes. Moreover, the callback does not seem
1176 * to support the spill block.
1177 * We define attributes in the same order as SA_*_OFFSET in order to
1178 * work around the problem. See ORI-610.
1181 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(osd), NULL, &osa->mode, 8);
1182 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(osd), NULL, &osa->size, 8);
1183 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GEN(osd), NULL, &gen, 8);
1184 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(osd), NULL, &osa->uid, 8);
1185 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(osd), NULL, &osa->gid, 8);
1186 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_PARENT(osd), NULL, &parent, 8);
1187 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(osd), NULL, &osa->flags, 8);
1188 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(osd), NULL, osa->atime, 16);
1189 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(osd), NULL, osa->mtime, 16);
1190 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(osd), NULL, osa->ctime, 16);
1191 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CRTIME(osd), NULL, crtime, 16);
1192 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(osd), NULL, &osa->nlink, 8);
1193 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_RDEV(osd), NULL, &osa->rdev, 8);
1195 rc = -sa_replace_all_by_template(sa_hdl, bulk, cnt, tx);
1197 OBD_FREE(bulk, sizeof(sa_bulk_attr_t) * 13);
1199 sa_handle_destroy(sa_hdl);
1204 * The transaction passed to this routine must have
1205 * dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT) called and then assigned
1206 * to a transaction group.
1208 int __osd_object_create(const struct lu_env *env, struct osd_device *osd,
1209 dmu_buf_t **dbp, dmu_tx_t *tx, struct lu_attr *la,
1215 /* Assert that the transaction has been assigned to a
1216 transaction group. */
1217 LASSERT(tx->tx_txg != 0);
1219 /* Create a new DMU object. */
1220 oid = dmu_object_alloc(osd->od_os, DMU_OT_PLAIN_FILE_CONTENTS, 0,
1221 DMU_OT_SA, DN_MAX_BONUSLEN, tx);
1222 rc = -sa_buf_hold(osd->od_os, oid, osd_obj_tag, dbp);
1223 LASSERTF(rc == 0, "sa_buf_hold "LPU64" failed: %d\n", oid, rc);
1225 LASSERT(la->la_valid & LA_MODE);
1229 rc = __osd_attr_init(env, osd, oid, tx, la, parent);
1231 sa_buf_rele(*dbp, osd_obj_tag);
1233 dmu_object_free(osd->od_os, oid, tx);
1241 * The transaction passed to this routine must have
1242 * dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, ...) called and then assigned
1243 * to a transaction group.
1245 * Using ZAP_FLAG_HASH64 will force the ZAP to always be a FAT ZAP.
1246 * This is fine for directories today, because storing the FID in the dirent
1247 * will also require a FAT ZAP. If there is a new type of micro ZAP created
1248 * then we might need to re-evaluate the use of this flag and instead do
1249 * a conversion from the different internal ZAP hash formats being used. */
1250 int __osd_zap_create(const struct lu_env *env, struct osd_device *osd,
1251 dmu_buf_t **zap_dbp, dmu_tx_t *tx,
1252 struct lu_attr *la, uint64_t parent, zap_flags_t flags)
1257 /* Assert that the transaction has been assigned to a
1258 transaction group. */
1259 LASSERT(tx->tx_txg != 0);
1261 oid = zap_create_flags(osd->od_os, 0, flags | ZAP_FLAG_HASH64,
1262 DMU_OT_DIRECTORY_CONTENTS,
1263 14, /* == ZFS fzap_default_block_shift */
1264 DN_MAX_INDBLKSHIFT, /* indirect block shift */
1265 DMU_OT_SA, DN_MAX_BONUSLEN, tx);
1267 rc = -sa_buf_hold(osd->od_os, oid, osd_obj_tag, zap_dbp);
1271 LASSERT(la->la_valid & LA_MODE);
1275 return __osd_attr_init(env, osd, oid, tx, la, parent);
1278 static dmu_buf_t *osd_mkidx(const struct lu_env *env, struct osd_device *osd,
1279 struct lu_attr *la, uint64_t parent,
1280 struct osd_thandle *oh)
1285 /* Index file should be created as regular file in order not to confuse
1286 * ZPL which could interpret them as directory.
1287 * We set ZAP_FLAG_UINT64_KEY to let ZFS know than we are going to use
1289 LASSERT(S_ISREG(la->la_mode));
1290 rc = __osd_zap_create(env, osd, &db, oh->ot_tx, la, parent,
1291 ZAP_FLAG_UINT64_KEY);
1297 static dmu_buf_t *osd_mkdir(const struct lu_env *env, struct osd_device *osd,
1298 struct lu_attr *la, uint64_t parent,
1299 struct osd_thandle *oh)
1304 LASSERT(S_ISDIR(la->la_mode));
1305 rc = __osd_zap_create(env, osd, &db, oh->ot_tx, la, parent, 0);
1311 static dmu_buf_t* osd_mkreg(const struct lu_env *env, struct osd_device *osd,
1312 struct lu_attr *la, uint64_t parent,
1313 struct osd_thandle *oh)
1318 LASSERT(S_ISREG(la->la_mode));
1319 rc = __osd_object_create(env, osd, &db, oh->ot_tx, la, parent);
1324 * XXX: a hack, OST to use bigger blocksize. we need
1325 * a method in OSD API to control this from OFD/MDD
1327 if (!lu_device_is_md(osd2lu_dev(osd))) {
1328 rc = -dmu_object_set_blocksize(osd->od_os,
1330 128 << 10, 0, oh->ot_tx);
1332 CERROR("%s: can't change blocksize: %d\n",
1333 osd->od_svname, rc);
1341 static dmu_buf_t *osd_mksym(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_ISLNK(la->la_mode));
1349 rc = __osd_object_create(env, osd, &db, oh->ot_tx, la, parent);
1355 static dmu_buf_t *osd_mknod(const struct lu_env *env, struct osd_device *osd,
1356 struct lu_attr *la, uint64_t parent,
1357 struct osd_thandle *oh)
1362 la->la_valid = LA_MODE;
1363 if (S_ISCHR(la->la_mode) || S_ISBLK(la->la_mode))
1364 la->la_valid |= LA_RDEV;
1366 rc = __osd_object_create(env, osd, &db, oh->ot_tx, la, parent);
1372 typedef dmu_buf_t *(*osd_obj_type_f)(const struct lu_env *env,
1373 struct osd_device *osd,
1376 struct osd_thandle *oh);
1378 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
1380 osd_obj_type_f result;
1406 * Primitives for directory (i.e. ZAP) handling
1408 static inline int osd_init_lma(const struct lu_env *env, struct osd_object *obj,
1409 const struct lu_fid *fid, struct osd_thandle *oh)
1411 struct osd_thread_info *info = osd_oti_get(env);
1412 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
1416 lustre_lma_init(lma, fid, 0, 0);
1417 lustre_lma_swab(lma);
1419 buf.lb_len = sizeof(*lma);
1421 rc = osd_xattr_set_internal(env, obj, &buf, XATTR_NAME_LMA,
1422 LU_XATTR_CREATE, oh, BYPASS_CAPA);
1428 * Concurrency: @dt is write locked.
1430 static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
1431 struct lu_attr *attr,
1432 struct dt_allocation_hint *hint,
1433 struct dt_object_format *dof,
1436 struct zpl_direntry *zde = &osd_oti_get(env)->oti_zde.lzd_reg;
1437 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1438 struct osd_object *obj = osd_dt_obj(dt);
1439 struct osd_device *osd = osd_obj2dev(obj);
1440 char *buf = osd_oti_get(env)->oti_str;
1441 struct osd_thandle *oh;
1448 /* concurrent create declarations should not see
1449 * the object inconsistent (db, attr, etc).
1450 * in regular cases acquisition should be cheap */
1451 down(&obj->oo_guard);
1453 LASSERT(osd_invariant(obj));
1454 LASSERT(!dt_object_exists(dt));
1455 LASSERT(dof != NULL);
1457 LASSERT(th != NULL);
1458 oh = container_of0(th, struct osd_thandle, ot_super);
1461 * XXX missing: Quote handling.
1464 LASSERT(obj->oo_db == NULL);
1466 /* to follow ZFS on-disk format we need
1467 * to initialize parent dnode properly */
1469 if (hint && hint->dah_parent)
1470 zapid = osd_dt_obj(hint->dah_parent)->oo_db->db_object;
1472 db = osd_create_type_f(dof->dof_type)(env, osd, attr, zapid, oh);
1474 GOTO(out, rc = PTR_ERR(db));
1477 zde->zde_dnode = db->db_object;
1478 zde->zde_type = IFTODT(attr->la_mode & S_IFMT);
1480 zapid = osd_get_name_n_idx(env, osd, fid, buf);
1482 rc = -zap_add(osd->od_os, zapid, buf, 8, 1, zde, oh->ot_tx);
1486 /* Add new object to inode accounting.
1487 * Errors are not considered as fatal */
1488 rc = -zap_increment_int(osd->od_os, osd->od_iusr_oid,
1489 (attr->la_valid & LA_UID) ? attr->la_uid : 0, 1,
1492 CERROR("%s: failed to add "DFID" to accounting ZAP for usr %d "
1493 "(%d)\n", osd->od_svname, PFID(fid), attr->la_uid, rc);
1494 rc = -zap_increment_int(osd->od_os, osd->od_igrp_oid,
1495 (attr->la_valid & LA_GID) ? attr->la_gid : 0, 1,
1498 CERROR("%s: failed to add "DFID" to accounting ZAP for grp %d "
1499 "(%d)\n", osd->od_svname, PFID(fid), attr->la_gid, rc);
1501 /* configure new osd object */
1503 rc = osd_object_init0(env, obj);
1504 LASSERT(ergo(rc == 0, dt_object_exists(dt)));
1505 LASSERT(osd_invariant(obj));
1507 rc = osd_init_lma(env, obj, fid, oh);
1509 CERROR("%s: can not set LMA on "DFID": rc = %d\n",
1510 osd->od_svname, PFID(fid), rc);
1511 /* ignore errors during LMA initialization */
1520 static int osd_declare_object_ref_add(const struct lu_env *env,
1521 struct dt_object *dt,
1524 return osd_declare_attr_set(env, dt, NULL, th);
1528 * Concurrency: @dt is write locked.
1530 static int osd_object_ref_add(const struct lu_env *env,
1531 struct dt_object *dt,
1532 struct thandle *handle)
1534 struct osd_object *obj = osd_dt_obj(dt);
1535 struct osd_thandle *oh;
1536 struct osd_device *osd = osd_obj2dev(obj);
1542 LASSERT(osd_invariant(obj));
1543 LASSERT(dt_object_exists(dt));
1544 LASSERT(obj->oo_sa_hdl != NULL);
1546 oh = container_of0(handle, struct osd_thandle, ot_super);
1548 write_lock(&obj->oo_attr_lock);
1549 nlink = ++obj->oo_attr.la_nlink;
1550 write_unlock(&obj->oo_attr_lock);
1552 rc = osd_object_sa_update(obj, SA_ZPL_LINKS(osd), &nlink, 8, oh);
1556 static int osd_declare_object_ref_del(const struct lu_env *env,
1557 struct dt_object *dt,
1558 struct thandle *handle)
1560 return osd_declare_attr_set(env, dt, NULL, handle);
1564 * Concurrency: @dt is write locked.
1566 static int osd_object_ref_del(const struct lu_env *env,
1567 struct dt_object *dt,
1568 struct thandle *handle)
1570 struct osd_object *obj = osd_dt_obj(dt);
1571 struct osd_thandle *oh;
1572 struct osd_device *osd = osd_obj2dev(obj);
1578 LASSERT(osd_invariant(obj));
1579 LASSERT(dt_object_exists(dt));
1580 LASSERT(obj->oo_sa_hdl != NULL);
1582 oh = container_of0(handle, struct osd_thandle, ot_super);
1583 LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
1585 write_lock(&obj->oo_attr_lock);
1586 nlink = --obj->oo_attr.la_nlink;
1587 write_unlock(&obj->oo_attr_lock);
1589 rc = osd_object_sa_update(obj, SA_ZPL_LINKS(osd), &nlink, 8, oh);
1593 static int capa_is_sane(const struct lu_env *env, struct osd_device *dev,
1594 struct lustre_capa *capa, struct lustre_capa_key *keys)
1596 struct osd_thread_info *oti = osd_oti_get(env);
1597 struct obd_capa *oc;
1601 oc = capa_lookup(dev->od_capa_hash, capa, 0);
1603 if (capa_is_expired(oc)) {
1604 DEBUG_CAPA(D_ERROR, capa, "expired");
1611 spin_lock(&capa_lock);
1612 for (i = 0; i < 2; i++) {
1613 if (keys[i].lk_keyid == capa->lc_keyid) {
1614 oti->oti_capa_key = keys[i];
1618 spin_unlock(&capa_lock);
1621 DEBUG_CAPA(D_ERROR, capa, "no matched capa key");
1625 rc = capa_hmac(oti->oti_capa.lc_hmac, capa, oti->oti_capa_key.lk_key);
1628 if (memcmp(oti->oti_capa.lc_hmac, capa->lc_hmac, sizeof(capa->lc_hmac)))
1630 DEBUG_CAPA(D_ERROR, capa, "HMAC mismatch");
1634 oc = capa_add(dev->od_capa_hash, capa);
1640 static int osd_object_auth(const struct lu_env *env, struct dt_object *dt,
1641 struct lustre_capa *capa, __u64 opc)
1643 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1644 struct osd_device *dev = osd_dev(dt->do_lu.lo_dev);
1647 if (!dev->od_fl_capa)
1650 if (capa == BYPASS_CAPA)
1654 CERROR("no capability is provided for fid "DFID"\n", PFID(fid));
1658 if (!lu_fid_eq(fid, &capa->lc_fid)) {
1659 DEBUG_CAPA(D_ERROR, capa, "fid "DFID" mismatch with",PFID(fid));
1663 if (!capa_opc_supported(capa, opc)) {
1664 DEBUG_CAPA(D_ERROR, capa, "opc "LPX64" not supported by", opc);
1668 if ((rc = capa_is_sane(env, dev, capa, dev->od_capa_keys))) {
1669 DEBUG_CAPA(D_ERROR, capa, "insane (rc %d)", rc);
1676 static struct obd_capa *osd_capa_get(const struct lu_env *env,
1677 struct dt_object *dt,
1678 struct lustre_capa *old,
1681 struct osd_thread_info *info = osd_oti_get(env);
1682 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1683 struct osd_object *obj = osd_dt_obj(dt);
1684 struct osd_device *dev = osd_obj2dev(obj);
1685 struct lustre_capa_key *key = &info->oti_capa_key;
1686 struct lustre_capa *capa = &info->oti_capa;
1687 struct obd_capa *oc;
1691 if (!dev->od_fl_capa)
1692 RETURN(ERR_PTR(-ENOENT));
1694 LASSERT(dt_object_exists(dt));
1695 LASSERT(osd_invariant(obj));
1697 /* renewal sanity check */
1698 if (old && osd_object_auth(env, dt, old, opc))
1699 RETURN(ERR_PTR(-EACCES));
1701 capa->lc_fid = *fid;
1704 capa->lc_flags = dev->od_capa_alg << 24;
1705 capa->lc_timeout = dev->od_capa_timeout;
1706 capa->lc_expiry = 0;
1708 oc = capa_lookup(dev->od_capa_hash, capa, 1);
1710 LASSERT(!capa_is_expired(oc));
1714 spin_lock(&capa_lock);
1715 *key = dev->od_capa_keys[1];
1716 spin_unlock(&capa_lock);
1718 capa->lc_keyid = key->lk_keyid;
1719 capa->lc_expiry = cfs_time_current_sec() + dev->od_capa_timeout;
1721 rc = capa_hmac(capa->lc_hmac, capa, key->lk_key);
1723 DEBUG_CAPA(D_ERROR, capa, "HMAC failed: %d for", rc);
1725 RETURN(ERR_PTR(rc));
1728 oc = capa_add(dev->od_capa_hash, capa);
1732 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
1733 __u64 start, __u64 end)
1735 struct osd_device *osd = osd_obj2dev(osd_dt_obj(dt));
1738 /* XXX: no other option than syncing the whole filesystem until we
1739 * support ZIL. If the object tracked the txg that it was last
1740 * modified in, it could pass that txg here instead of "0". Maybe
1741 * the changes are already committed, so no wait is needed at all? */
1742 txg_wait_synced(dmu_objset_pool(osd->od_os), 0ULL);
1747 static struct dt_object_operations osd_obj_ops = {
1748 .do_read_lock = osd_object_read_lock,
1749 .do_write_lock = osd_object_write_lock,
1750 .do_read_unlock = osd_object_read_unlock,
1751 .do_write_unlock = osd_object_write_unlock,
1752 .do_write_locked = osd_object_write_locked,
1753 .do_attr_get = osd_attr_get,
1754 .do_declare_attr_set = osd_declare_attr_set,
1755 .do_attr_set = osd_attr_set,
1756 .do_ah_init = osd_ah_init,
1757 .do_declare_create = osd_declare_object_create,
1758 .do_create = osd_object_create,
1759 .do_declare_destroy = osd_declare_object_destroy,
1760 .do_destroy = osd_object_destroy,
1761 .do_index_try = osd_index_try,
1762 .do_declare_ref_add = osd_declare_object_ref_add,
1763 .do_ref_add = osd_object_ref_add,
1764 .do_declare_ref_del = osd_declare_object_ref_del,
1765 .do_ref_del = osd_object_ref_del,
1766 .do_xattr_get = osd_xattr_get,
1767 .do_declare_xattr_set = osd_declare_xattr_set,
1768 .do_xattr_set = osd_xattr_set,
1769 .do_declare_xattr_del = osd_declare_xattr_del,
1770 .do_xattr_del = osd_xattr_del,
1771 .do_xattr_list = osd_xattr_list,
1772 .do_capa_get = osd_capa_get,
1773 .do_object_sync = osd_object_sync,
1776 static struct lu_object_operations osd_lu_obj_ops = {
1777 .loo_object_init = osd_object_init,
1778 .loo_object_delete = osd_object_delete,
1779 .loo_object_release = osd_object_release,
1780 .loo_object_free = osd_object_free,
1781 .loo_object_print = osd_object_print,
1782 .loo_object_invariant = osd_object_invariant,
1785 static int osd_otable_it_attr_get(const struct lu_env *env,
1786 struct dt_object *dt,
1787 struct lu_attr *attr,
1788 struct lustre_capa *capa)
1794 static struct dt_object_operations osd_obj_otable_it_ops = {
1795 .do_attr_get = osd_otable_it_attr_get,
1796 .do_index_try = osd_index_try,