4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2009, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2012, 2015, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/osd-zfs/osd_index.c
38 * Author: Alex Zhuravlev <bzzz@whamcloud.com>
39 * Author: Mike Pershin <tappro@whamcloud.com>
42 #define DEBUG_SUBSYSTEM S_OSD
44 #include <lustre_ver.h>
45 #include <libcfs/libcfs.h>
46 #include <obd_support.h>
47 #include <lustre_net.h>
49 #include <obd_class.h>
50 #include <lustre_disk.h>
51 #include <lustre_fid.h>
53 #include "osd_internal.h"
55 #include <sys/dnode.h>
60 #include <sys/spa_impl.h>
61 #include <sys/zfs_znode.h>
62 #include <sys/dmu_tx.h>
63 #include <sys/dmu_objset.h>
64 #include <sys/dsl_prop.h>
65 #include <sys/sa_impl.h>
68 static inline int osd_object_is_zap(dmu_buf_t *db)
70 dmu_buf_impl_t *dbi = (dmu_buf_impl_t *) db;
76 rc = (dn->dn_type == DMU_OT_DIRECTORY_CONTENTS ||
77 dn->dn_type == DMU_OT_USERGROUP_USED);
83 /* We don't actually have direct access to the zap_hashbits() function
84 * so just pretend like we do for now. If this ever breaks we can look at
86 #define zap_hashbits(zc) 48
89 * | cd (16 bits) | hash (48 bits) |
90 * we need it in other form:
91 * |0| hash (48 bit) | cd (15 bit) |
92 * to be a full 64-bit ordered hash so that Lustre readdir can use it to merge
93 * the readdir hashes from multiple directory stripes uniformly on the client.
94 * Another point is sign bit, the hash range should be in [0, 2^63-1] because
95 * loff_t (for llseek) needs to be a positive value. This means the "cd" field
96 * should only be the low 15 bits.
98 uint64_t osd_zap_cursor_serialize(zap_cursor_t *zc)
100 uint64_t zfs_hash = zap_cursor_serialize(zc) & (~0ULL >> 1);
102 return (zfs_hash >> zap_hashbits(zc)) |
103 (zfs_hash << (63 - zap_hashbits(zc)));
106 void osd_zap_cursor_init_serialized(zap_cursor_t *zc, struct objset *os,
107 uint64_t id, uint64_t dirhash)
109 uint64_t zfs_hash = ((dirhash << zap_hashbits(zc)) & (~0ULL >> 1)) |
110 (dirhash >> (63 - zap_hashbits(zc)));
112 zap_cursor_init_serialized(zc, os, id, zfs_hash);
115 int osd_zap_cursor_init(zap_cursor_t **zc, struct objset *os,
116 uint64_t id, uint64_t dirhash)
121 if (unlikely(t == NULL))
124 osd_zap_cursor_init_serialized(t, os, id, dirhash);
130 void osd_zap_cursor_fini(zap_cursor_t *zc)
136 static inline void osd_obj_cursor_init_serialized(zap_cursor_t *zc,
137 struct osd_object *o,
140 struct osd_device *d = osd_obj2dev(o);
141 osd_zap_cursor_init_serialized(zc, d->od_os,
142 o->oo_db->db_object, dirhash);
145 static inline int osd_obj_cursor_init(zap_cursor_t **zc, struct osd_object *o,
148 struct osd_device *d = osd_obj2dev(o);
149 return osd_zap_cursor_init(zc, d->od_os, o->oo_db->db_object, dirhash);
152 static struct dt_it *osd_index_it_init(const struct lu_env *env,
153 struct dt_object *dt,
156 struct osd_thread_info *info = osd_oti_get(env);
157 struct osd_zap_it *it;
158 struct osd_object *obj = osd_dt_obj(dt);
159 struct lu_object *lo = &dt->do_lu;
163 if (obj->oo_destroyed)
164 RETURN(ERR_PTR(-ENOENT));
166 LASSERT(lu_object_exists(lo));
170 OBD_SLAB_ALLOC_PTR_GFP(it, osd_zapit_cachep, GFP_NOFS);
172 RETURN(ERR_PTR(-ENOMEM));
174 rc = osd_obj_cursor_init(&it->ozi_zc, obj, 0);
176 OBD_SLAB_FREE_PTR(it, osd_zapit_cachep);
184 RETURN((struct dt_it *)it);
187 static void osd_index_it_fini(const struct lu_env *env, struct dt_it *di)
189 struct osd_zap_it *it = (struct osd_zap_it *)di;
190 struct osd_object *obj;
194 LASSERT(it->ozi_obj);
198 osd_zap_cursor_fini(it->ozi_zc);
199 lu_object_put(env, &obj->oo_dt.do_lu);
200 OBD_SLAB_FREE_PTR(it, osd_zapit_cachep);
206 static void osd_index_it_put(const struct lu_env *env, struct dt_it *di)
208 /* PBS: do nothing : ref are incremented at retrive and decreamented
212 static inline void osd_it_append_attrs(struct lu_dirent *ent, __u32 attr,
215 const unsigned align = sizeof(struct luda_type) - 1;
216 struct luda_type *lt;
218 /* check if file type is required */
219 if (attr & LUDA_TYPE) {
220 len = (len + align) & ~align;
222 lt = (void *)ent->lde_name + len;
223 lt->lt_type = cpu_to_le16(DTTOIF(type));
224 ent->lde_attrs |= LUDA_TYPE;
227 ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
231 * Get the object's FID from its LMA EA.
233 * \param[in] env pointer to the thread context
234 * \param[in] osd pointer to the OSD device
235 * \param[in] oid the object's local identifier
236 * \param[out] fid the buffer to hold the object's FID
238 * \retval 0 for success
239 * \retval negative error number on failure
241 static int osd_get_fid_by_oid(const struct lu_env *env, struct osd_device *osd,
242 uint64_t oid, struct lu_fid *fid)
244 struct objset *os = osd->od_os;
245 struct osd_thread_info *oti = osd_oti_get(env);
246 struct lustre_mdt_attrs *lma =
247 (struct lustre_mdt_attrs *)oti->oti_buf;
249 nvlist_t *sa_xattr = NULL;
250 sa_handle_t *sa_hdl = NULL;
251 uchar_t *nv_value = NULL;
252 uint64_t xattr = ZFS_NO_OBJECT;
257 rc = __osd_xattr_load(osd, oid, &sa_xattr);
264 rc = -nvlist_lookup_byte_array(sa_xattr, XATTR_NAME_LMA, &nv_value,
272 if (unlikely(size > sizeof(oti->oti_buf)))
273 GOTO(out, rc = -ERANGE);
275 memcpy(lma, nv_value, size);
280 rc = -sa_handle_get(os, oid, NULL, SA_HDL_PRIVATE, &sa_hdl);
284 rc = -sa_lookup(sa_hdl, SA_ZPL_XATTR(osd), &xattr, 8);
285 sa_handle_destroy(sa_hdl);
290 buf.lb_len = sizeof(oti->oti_buf);
291 rc = __osd_xattr_get_large(env, osd, xattr, &buf,
292 XATTR_NAME_LMA, &size);
297 if (size < sizeof(*lma))
298 GOTO(out, rc = -EIO);
300 lustre_lma_swab(lma);
301 if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
302 CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT))) {
303 CWARN("%s: unsupported incompat LMA feature(s) %#x for "
304 "oid = %#llx\n", osd->od_svname,
305 lma->lma_incompat & ~LMA_INCOMPAT_SUPP, oid);
306 GOTO(out, rc = -EOPNOTSUPP);
308 *fid = lma->lma_self_fid;
313 if (sa_xattr != NULL)
314 nvlist_free(sa_xattr);
319 * As we don't know FID, we can't use LU object, so this function
320 * partially duplicate __osd_xattr_get() which is built around
321 * LU-object and uses it to cache data like regular EA dnode, etc
323 static int osd_find_parent_by_dnode(const struct lu_env *env,
327 struct osd_device *osd = osd_obj2dev(osd_dt_obj(o));
329 uint64_t dnode = ZFS_NO_OBJECT;
333 /* first of all, get parent dnode from own attributes */
334 LASSERT(osd_dt_obj(o)->oo_db);
335 rc = -sa_handle_get(osd->od_os, osd_dt_obj(o)->oo_db->db_object,
336 NULL, SA_HDL_PRIVATE, &sa_hdl);
340 rc = -sa_lookup(sa_hdl, SA_ZPL_PARENT(osd), &dnode, 8);
341 sa_handle_destroy(sa_hdl);
343 rc = osd_get_fid_by_oid(env, osd, dnode, fid);
348 static int osd_find_parent_fid(const struct lu_env *env, struct dt_object *o,
351 struct link_ea_header *leh;
352 struct link_ea_entry *lee;
357 buf.lb_buf = osd_oti_get(env)->oti_buf;
358 buf.lb_len = sizeof(osd_oti_get(env)->oti_buf);
360 rc = osd_xattr_get(env, o, &buf, XATTR_NAME_LINK);
362 rc = osd_xattr_get(env, o, &LU_BUF_NULL, XATTR_NAME_LINK);
366 OBD_ALLOC(buf.lb_buf, rc);
367 if (buf.lb_buf == NULL)
370 rc = osd_xattr_get(env, o, &buf, XATTR_NAME_LINK);
374 if (rc < sizeof(*leh) + sizeof(*lee))
375 GOTO(out, rc = -EINVAL);
378 if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) {
379 leh->leh_magic = LINK_EA_MAGIC;
380 leh->leh_reccount = __swab32(leh->leh_reccount);
381 leh->leh_len = __swab64(leh->leh_len);
383 if (leh->leh_magic != LINK_EA_MAGIC)
384 GOTO(out, rc = -EINVAL);
385 if (leh->leh_reccount == 0)
386 GOTO(out, rc = -ENODATA);
388 lee = (struct link_ea_entry *)(leh + 1);
389 fid_be_to_cpu(fid, (const struct lu_fid *)&lee->lee_parent_fid);
393 if (buf.lb_buf != osd_oti_get(env)->oti_buf)
394 OBD_FREE(buf.lb_buf, buf.lb_len);
397 /* this block can be enabled for additional verification
398 * it's trying to match FID from LinkEA vs. FID from LMA */
402 rc2 = osd_find_parent_by_dnode(env, o, &fid2);
404 if (lu_fid_eq(fid, &fid2) == 0)
405 CERROR("wrong parent: "DFID" != "DFID"\n",
406 PFID(fid), PFID(&fid2));
410 /* no LinkEA is found, let's try to find the fid in parent's LMA */
411 if (unlikely(rc != 0))
412 rc = osd_find_parent_by_dnode(env, o, fid);
417 static int osd_dir_lookup(const struct lu_env *env, struct dt_object *dt,
418 struct dt_rec *rec, const struct dt_key *key)
420 struct osd_thread_info *oti = osd_oti_get(env);
421 struct osd_object *obj = osd_dt_obj(dt);
422 struct osd_device *osd = osd_obj2dev(obj);
423 char *name = (char *)key;
427 if (name[0] == '.') {
429 const struct lu_fid *f = lu_object_fid(&dt->do_lu);
430 memcpy(rec, f, sizeof(*f));
432 } else if (name[1] == '.' && name[2] == 0) {
433 rc = osd_find_parent_fid(env, dt, (struct lu_fid *)rec);
434 RETURN(rc == 0 ? 1 : rc);
438 memset(&oti->oti_zde.lzd_fid, 0, sizeof(struct lu_fid));
439 rc = -zap_lookup(osd->od_os, obj->oo_db->db_object,
440 (char *)key, 8, sizeof(oti->oti_zde) / 8,
441 (void *)&oti->oti_zde);
445 if (likely(fid_is_sane(&oti->oti_zde.lzd_fid))) {
446 memcpy(rec, &oti->oti_zde.lzd_fid, sizeof(struct lu_fid));
450 rc = osd_get_fid_by_oid(env, osd, oti->oti_zde.lzd_reg.zde_dnode,
451 (struct lu_fid *)rec);
453 RETURN(rc == 0 ? 1 : (rc == -ENOENT ? -ENODATA : rc));
456 static int osd_declare_dir_insert(const struct lu_env *env,
457 struct dt_object *dt,
458 const struct dt_rec *rec,
459 const struct dt_key *key,
462 struct osd_object *obj = osd_dt_obj(dt);
463 struct osd_thandle *oh;
468 oh = container_of0(th, struct osd_thandle, ot_super);
470 /* This is for inserting dot/dotdot for new created dir. */
471 if (obj->oo_db == NULL)
472 object = DMU_NEW_OBJECT;
474 object = obj->oo_db->db_object;
476 /* do not specify the key as then DMU is trying to look it up
477 * which is very expensive. usually the layers above lookup
478 * before insertion */
479 dmu_tx_hold_zap(oh->ot_tx, object, TRUE, NULL);
485 * Find the osd object for given fid.
487 * \param fid need to find the osd object having this fid
489 * \retval osd_object on success
490 * \retval -ve on error
492 struct osd_object *osd_object_find(const struct lu_env *env,
493 struct dt_object *dt,
494 const struct lu_fid *fid)
496 struct lu_device *ludev = dt->do_lu.lo_dev;
497 struct osd_object *child = NULL;
498 struct lu_object *luch;
499 struct lu_object *lo;
502 * at this point topdev might not exist yet
503 * (i.e. MGS is preparing profiles). so we can
504 * not rely on topdev and instead lookup with
505 * our device passed as topdev. this can't work
506 * if the object isn't cached yet (as osd doesn't
507 * allocate lu_header). IOW, the object must be
508 * in the cache, otherwise lu_object_alloc() crashes
511 luch = lu_object_find_at(env, ludev, fid, NULL);
515 if (lu_object_exists(luch)) {
516 lo = lu_object_locate(luch->lo_header, ludev->ld_type);
520 LU_OBJECT_DEBUG(D_ERROR, env, luch,
521 "%s: object can't be located "DFID,
522 osd_dev(ludev)->od_svname, PFID(fid));
525 lu_object_put(env, luch);
526 CERROR("%s: Unable to get osd_object "DFID"\n",
527 osd_dev(ludev)->od_svname, PFID(fid));
528 child = ERR_PTR(-ENOENT);
531 LU_OBJECT_DEBUG(D_ERROR, env, luch,
532 "%s: lu_object does not exists "DFID,
533 osd_dev(ludev)->od_svname, PFID(fid));
534 lu_object_put(env, luch);
535 child = ERR_PTR(-ENOENT);
542 * Put the osd object once done with it.
544 * \param obj osd object that needs to be put
546 static inline void osd_object_put(const struct lu_env *env,
547 struct osd_object *obj)
549 lu_object_put(env, &obj->oo_dt.do_lu);
552 static int osd_seq_exists(const struct lu_env *env, struct osd_device *osd,
555 struct lu_seq_range *range = &osd_oti_get(env)->oti_seq_range;
556 struct seq_server_site *ss = osd_seq_site(osd);
561 LASSERT(ss->ss_server_fld != NULL);
563 rc = osd_fld_lookup(env, osd, seq, range);
566 CERROR("%s: Can not lookup fld for %#llx\n",
571 RETURN(ss->ss_node_id == range->lsr_index);
574 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
575 const struct lu_fid *fid)
577 struct seq_server_site *ss = osd_seq_site(osd);
580 /* FID seqs not in FLDB, must be local seq */
581 if (unlikely(!fid_seq_in_fldb(fid_seq(fid))))
584 /* If FLD is not being initialized yet, it only happens during the
585 * initialization, likely during mgs initialization, and we assume
586 * this is local FID. */
587 if (ss == NULL || ss->ss_server_fld == NULL)
590 /* Only check the local FLDB here */
591 if (osd_seq_exists(env, osd, fid_seq(fid)))
598 * Inserts (key, value) pair in \a directory object.
600 * \param dt osd index object
601 * \param key key for index
602 * \param rec record reference
603 * \param th transaction handler
604 * \param ignore_quota update should not affect quota
607 * \retval -ve failure
609 static int osd_dir_insert(const struct lu_env *env, struct dt_object *dt,
610 const struct dt_rec *rec, const struct dt_key *key,
611 struct thandle *th, int ignore_quota)
613 struct osd_thread_info *oti = osd_oti_get(env);
614 struct osd_object *parent = osd_dt_obj(dt);
615 struct osd_device *osd = osd_obj2dev(parent);
616 struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec;
617 const struct lu_fid *fid = rec1->rec_fid;
618 struct osd_thandle *oh;
619 struct osd_object *child = NULL;
621 char *name = (char *)key;
625 LASSERT(parent->oo_db);
627 LASSERT(dt_object_exists(dt));
628 LASSERT(osd_invariant(parent));
631 oh = container_of0(th, struct osd_thandle, ot_super);
633 rc = osd_remote_fid(env, osd, fid);
635 CERROR("%s: Can not find object "DFID": rc = %d\n",
636 osd->od_svname, PFID(fid), rc);
640 if (unlikely(rc == 1)) {
641 /* Insert remote entry */
642 memset(&oti->oti_zde.lzd_reg, 0, sizeof(oti->oti_zde.lzd_reg));
643 oti->oti_zde.lzd_reg.zde_type = IFTODT(rec1->rec_type & S_IFMT);
646 * To simulate old Orion setups with ./.. stored in the
649 /* Insert local entry */
650 child = osd_object_find(env, dt, fid);
652 RETURN(PTR_ERR(child));
654 LASSERT(child->oo_db);
655 if (name[0] == '.') {
657 /* do not store ".", instead generate it
658 * during iteration */
660 } else if (name[1] == '.' && name[2] == 0) {
661 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT)) {
662 struct lu_fid tfid = *fid;
664 osd_object_put(env, child);
666 child = osd_object_find(env, dt, &tfid);
668 RETURN(PTR_ERR(child));
670 LASSERT(child->oo_db);
673 /* update parent dnode in the child.
674 * later it will be used to generate ".." */
675 rc = osd_object_sa_update(parent,
677 &child->oo_db->db_object,
683 CLASSERT(sizeof(oti->oti_zde.lzd_reg) == 8);
684 CLASSERT(sizeof(oti->oti_zde) % 8 == 0);
685 attr = child->oo_dt.do_lu.lo_header ->loh_attr;
686 oti->oti_zde.lzd_reg.zde_type = IFTODT(attr & S_IFMT);
687 oti->oti_zde.lzd_reg.zde_dnode = child->oo_db->db_object;
690 oti->oti_zde.lzd_fid = *fid;
691 /* Insert (key,oid) into ZAP */
692 rc = -zap_add(osd->od_os, parent->oo_db->db_object,
693 (char *)key, 8, sizeof(oti->oti_zde) / 8,
694 (void *)&oti->oti_zde, oh->ot_tx);
695 if (unlikely(rc == -EEXIST &&
696 name[0] == '.' && name[1] == '.' && name[2] == 0))
697 /* Update (key,oid) in ZAP */
698 rc = -zap_update(osd->od_os, parent->oo_db->db_object,
699 (char *)key, 8, sizeof(oti->oti_zde) / 8,
700 (void *)&oti->oti_zde, oh->ot_tx);
704 osd_object_put(env, child);
709 static int osd_declare_dir_delete(const struct lu_env *env,
710 struct dt_object *dt,
711 const struct dt_key *key,
714 struct osd_object *obj = osd_dt_obj(dt);
715 struct osd_thandle *oh;
719 LASSERT(dt_object_exists(dt));
720 LASSERT(osd_invariant(obj));
723 oh = container_of0(th, struct osd_thandle, ot_super);
725 if (dt_object_exists(dt)) {
727 dnode = obj->oo_db->db_object;
729 dnode = DMU_NEW_OBJECT;
732 /* do not specify the key as then DMU is trying to look it up
733 * which is very expensive. usually the layers above lookup
735 dmu_tx_hold_zap(oh->ot_tx, dnode, FALSE, NULL);
740 static int osd_dir_delete(const struct lu_env *env, struct dt_object *dt,
741 const struct dt_key *key, struct thandle *th)
743 struct osd_object *obj = osd_dt_obj(dt);
744 struct osd_device *osd = osd_obj2dev(obj);
745 struct osd_thandle *oh;
746 dmu_buf_t *zap_db = obj->oo_db;
747 char *name = (char *)key;
754 oh = container_of0(th, struct osd_thandle, ot_super);
757 * In Orion . and .. were stored in the directory (not generated upon
758 * request as now). we preserve them for backward compatibility
760 if (name[0] == '.') {
763 } else if (name[1] == '.' && name[2] == 0) {
768 /* Remove key from the ZAP */
769 rc = -zap_remove(osd->od_os, zap_db->db_object,
770 (char *) key, oh->ot_tx);
772 if (unlikely(rc && rc != -ENOENT))
773 CERROR("%s: zap_remove failed: rc = %d\n", osd->od_svname, rc);
778 static struct dt_it *osd_dir_it_init(const struct lu_env *env,
779 struct dt_object *dt,
782 struct osd_zap_it *it;
784 it = (struct osd_zap_it *)osd_index_it_init(env, dt, unused);
788 RETURN((struct dt_it *)it);
792 * Move Iterator to record specified by \a key
794 * \param di osd iterator
795 * \param key key for index
797 * \retval +ve di points to record with least key not larger than key
798 * \retval 0 di points to exact matched key
799 * \retval -ve failure
801 static int osd_dir_it_get(const struct lu_env *env,
802 struct dt_it *di, const struct dt_key *key)
804 struct osd_zap_it *it = (struct osd_zap_it *)di;
805 struct osd_object *obj = it->ozi_obj;
806 char *name = (char *)key;
813 /* reset the cursor */
814 zap_cursor_fini(it->ozi_zc);
815 osd_obj_cursor_init_serialized(it->ozi_zc, obj, 0);
817 /* XXX: implementation of the API is broken at the moment */
818 LASSERT(((const char *)key)[0] == 0);
825 if (name[0] == '.') {
829 } else if (name[1] == '.' && name[2] == 0) {
835 /* neither . nor .. - some real record */
843 static void osd_dir_it_put(const struct lu_env *env, struct dt_it *di)
845 /* PBS: do nothing : ref are incremented at retrive and decreamented
850 * in Orion . and .. were stored in the directory, while ZPL
851 * and current osd-zfs generate them up on request. so, we
852 * need to ignore previously stored . and ..
854 static int osd_index_retrieve_skip_dots(struct osd_zap_it *it,
860 rc = -zap_cursor_retrieve(it->ozi_zc, za);
863 if (unlikely(rc == 0 && za->za_name[0] == '.')) {
864 if (za->za_name[1] == 0) {
866 } else if (za->za_name[1] == '.' &&
867 za->za_name[2] == 0) {
871 zap_cursor_advance(it->ozi_zc);
873 } while (unlikely(rc == 0 && isdot));
879 * to load a directory entry at a time and stored it in
880 * iterator's in-memory data structure.
882 * \param di, struct osd_it_ea, iterator's in memory structure
884 * \retval +ve, iterator reached to end
885 * \retval 0, iterator not reached to end
886 * \retval -ve, on error
888 static int osd_dir_it_next(const struct lu_env *env, struct dt_it *di)
890 struct osd_zap_it *it = (struct osd_zap_it *)di;
891 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
896 /* temp. storage should be enough for any key supported by ZFS */
897 CLASSERT(sizeof(za->za_name) <= sizeof(it->ozi_name));
900 * the first ->next() moves the cursor to .
901 * the second ->next() moves the cursor to ..
902 * then we get to the real records and have to verify any exist
904 if (it->ozi_pos <= 2) {
910 zap_cursor_advance(it->ozi_zc);
914 * According to current API we need to return error if its last entry.
915 * zap_cursor_advance() does not return any value. So we need to call
916 * retrieve to check if there is any record. We should make
917 * changes to Iterator API to not return status for this API
919 rc = osd_index_retrieve_skip_dots(it, za);
921 if (rc == -ENOENT) /* end of dir */
927 static struct dt_key *osd_dir_it_key(const struct lu_env *env,
928 const struct dt_it *di)
930 struct osd_zap_it *it = (struct osd_zap_it *)di;
931 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
935 if (it->ozi_pos <= 1) {
937 RETURN((struct dt_key *)".");
938 } else if (it->ozi_pos == 2) {
939 RETURN((struct dt_key *)"..");
942 if ((rc = -zap_cursor_retrieve(it->ozi_zc, za)))
945 strcpy(it->ozi_name, za->za_name);
947 RETURN((struct dt_key *)it->ozi_name);
950 static int osd_dir_it_key_size(const struct lu_env *env, const struct dt_it *di)
952 struct osd_zap_it *it = (struct osd_zap_it *)di;
953 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
957 if (it->ozi_pos <= 1) {
960 } else if (it->ozi_pos == 2) {
964 if ((rc = -zap_cursor_retrieve(it->ozi_zc, za)) == 0)
965 rc = strlen(za->za_name);
970 static int osd_dir_it_rec(const struct lu_env *env, const struct dt_it *di,
971 struct dt_rec *dtrec, __u32 attr)
973 struct osd_zap_it *it = (struct osd_zap_it *)di;
974 struct lu_dirent *lde = (struct lu_dirent *)dtrec;
975 struct luz_direntry *zde = &osd_oti_get(env)->oti_zde;
976 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
980 if (it->ozi_pos <= 1) {
981 lde->lde_hash = cpu_to_le64(1);
982 strcpy(lde->lde_name, ".");
983 lde->lde_namelen = cpu_to_le16(1);
984 lde->lde_fid = *lu_object_fid(&it->ozi_obj->oo_dt.do_lu);
985 lde->lde_attrs = LUDA_FID;
986 /* append lustre attributes */
987 osd_it_append_attrs(lde, attr, 1, IFTODT(S_IFDIR));
988 lde->lde_reclen = cpu_to_le16(lu_dirent_calc_size(1, attr));
992 } else if (it->ozi_pos == 2) {
993 lde->lde_hash = cpu_to_le64(2);
994 strcpy(lde->lde_name, "..");
995 lde->lde_namelen = cpu_to_le16(2);
996 lde->lde_attrs = LUDA_FID;
997 /* append lustre attributes */
998 osd_it_append_attrs(lde, attr, 2, IFTODT(S_IFDIR));
999 lde->lde_reclen = cpu_to_le16(lu_dirent_calc_size(2, attr));
1000 rc = osd_find_parent_fid(env, &it->ozi_obj->oo_dt, &lde->lde_fid);
1002 /* ENOENT happens at the root of filesystem so ignore it */
1010 rc = -zap_cursor_retrieve(it->ozi_zc, za);
1011 if (unlikely(rc != 0))
1014 lde->lde_hash = cpu_to_le64(osd_zap_cursor_serialize(it->ozi_zc));
1015 namelen = strlen(za->za_name);
1016 if (namelen > NAME_MAX)
1017 GOTO(out, rc = -EOVERFLOW);
1018 strcpy(lde->lde_name, za->za_name);
1019 lde->lde_namelen = cpu_to_le16(namelen);
1021 if (za->za_integer_length != 8 || za->za_num_integers < 3) {
1022 CERROR("%s: unsupported direntry format: %d %d\n",
1023 osd_obj2dev(it->ozi_obj)->od_svname,
1024 za->za_integer_length, (int)za->za_num_integers);
1026 GOTO(out, rc = -EIO);
1029 rc = -zap_lookup(it->ozi_zc->zc_objset, it->ozi_zc->zc_zapobj,
1030 za->za_name, za->za_integer_length, 3, zde);
1034 lde->lde_fid = zde->lzd_fid;
1035 lde->lde_attrs = LUDA_FID;
1037 /* append lustre attributes */
1038 osd_it_append_attrs(lde, attr, namelen, zde->lzd_reg.zde_type);
1040 lde->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
1046 static int osd_dir_it_rec_size(const struct lu_env *env, const struct dt_it *di,
1049 struct osd_zap_it *it = (struct osd_zap_it *)di;
1050 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
1055 if (it->ozi_pos <= 1)
1057 else if (it->ozi_pos == 2)
1061 rc = lu_dirent_calc_size(namelen, attr);
1065 rc = -zap_cursor_retrieve(it->ozi_zc, za);
1066 if (unlikely(rc != 0))
1069 if (za->za_integer_length != 8 || za->za_num_integers < 3) {
1070 CERROR("%s: unsupported direntry format: %d %d\n",
1071 osd_obj2dev(it->ozi_obj)->od_svname,
1072 za->za_integer_length, (int)za->za_num_integers);
1076 namelen = strlen(za->za_name);
1077 if (namelen > NAME_MAX)
1080 rc = lu_dirent_calc_size(namelen, attr);
1085 static __u64 osd_dir_it_store(const struct lu_env *env, const struct dt_it *di)
1087 struct osd_zap_it *it = (struct osd_zap_it *)di;
1091 if (it->ozi_pos <= 2)
1094 pos = osd_zap_cursor_serialize(it->ozi_zc);
1101 * rc == 0 -> end of directory.
1102 * rc > 0 -> ok, proceed.
1103 * rc < 0 -> error. ( EOVERFLOW can be masked.)
1105 static int osd_dir_it_load(const struct lu_env *env,
1106 const struct dt_it *di, __u64 hash)
1108 struct osd_zap_it *it = (struct osd_zap_it *)di;
1109 struct osd_object *obj = it->ozi_obj;
1110 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
1114 /* reset the cursor */
1115 zap_cursor_fini(it->ozi_zc);
1116 osd_obj_cursor_init_serialized(it->ozi_zc, obj, hash);
1123 /* to return whether the end has been reached */
1124 rc = osd_index_retrieve_skip_dots(it, za);
1127 else if (rc == -ENOENT)
1134 struct dt_index_operations osd_dir_ops = {
1135 .dio_lookup = osd_dir_lookup,
1136 .dio_declare_insert = osd_declare_dir_insert,
1137 .dio_insert = osd_dir_insert,
1138 .dio_declare_delete = osd_declare_dir_delete,
1139 .dio_delete = osd_dir_delete,
1141 .init = osd_dir_it_init,
1142 .fini = osd_index_it_fini,
1143 .get = osd_dir_it_get,
1144 .put = osd_dir_it_put,
1145 .next = osd_dir_it_next,
1146 .key = osd_dir_it_key,
1147 .key_size = osd_dir_it_key_size,
1148 .rec = osd_dir_it_rec,
1149 .rec_size = osd_dir_it_rec_size,
1150 .store = osd_dir_it_store,
1151 .load = osd_dir_it_load
1156 * Primitives for index files using binary keys.
1159 /* key integer_size is 8 */
1160 static int osd_prepare_key_uint64(struct osd_object *o, __u64 *dst,
1161 const struct dt_key *src)
1168 /* align keysize to 64bit */
1169 size = (o->oo_keysize + sizeof(__u64) - 1) / sizeof(__u64);
1170 size *= sizeof(__u64);
1172 LASSERT(size <= MAXNAMELEN);
1174 if (unlikely(size > o->oo_keysize))
1175 memset(dst + o->oo_keysize, 0, size - o->oo_keysize);
1176 memcpy(dst, (const char *)src, o->oo_keysize);
1178 return (size/sizeof(__u64));
1181 static int osd_index_lookup(const struct lu_env *env, struct dt_object *dt,
1182 struct dt_rec *rec, const struct dt_key *key)
1184 struct osd_object *obj = osd_dt_obj(dt);
1185 struct osd_device *osd = osd_obj2dev(obj);
1186 __u64 *k = osd_oti_get(env)->oti_key64;
1190 rc = osd_prepare_key_uint64(obj, k, key);
1192 rc = -zap_lookup_uint64(osd->od_os, obj->oo_db->db_object,
1193 k, rc, obj->oo_recusize, obj->oo_recsize,
1195 RETURN(rc == 0 ? 1 : rc);
1198 static int osd_declare_index_insert(const struct lu_env *env,
1199 struct dt_object *dt,
1200 const struct dt_rec *rec,
1201 const struct dt_key *key,
1204 struct osd_object *obj = osd_dt_obj(dt);
1205 struct osd_thandle *oh;
1208 LASSERT(th != NULL);
1209 oh = container_of0(th, struct osd_thandle, ot_super);
1211 LASSERT(obj->oo_db);
1213 dmu_tx_hold_bonus(oh->ot_tx, obj->oo_db->db_object);
1215 /* do not specify the key as then DMU is trying to look it up
1216 * which is very expensive. usually the layers above lookup
1217 * before insertion */
1218 dmu_tx_hold_zap(oh->ot_tx, obj->oo_db->db_object, TRUE, NULL);
1223 static int osd_index_insert(const struct lu_env *env, struct dt_object *dt,
1224 const struct dt_rec *rec, const struct dt_key *key,
1225 struct thandle *th, int ignore_quota)
1227 struct osd_object *obj = osd_dt_obj(dt);
1228 struct osd_device *osd = osd_obj2dev(obj);
1229 struct osd_thandle *oh;
1230 __u64 *k = osd_oti_get(env)->oti_key64;
1234 LASSERT(obj->oo_db);
1235 LASSERT(dt_object_exists(dt));
1236 LASSERT(osd_invariant(obj));
1237 LASSERT(th != NULL);
1239 oh = container_of0(th, struct osd_thandle, ot_super);
1241 rc = osd_prepare_key_uint64(obj, k, key);
1243 /* Insert (key,oid) into ZAP */
1244 rc = -zap_add_uint64(osd->od_os, obj->oo_db->db_object,
1245 k, rc, obj->oo_recusize, obj->oo_recsize,
1246 (void *)rec, oh->ot_tx);
1250 static int osd_declare_index_delete(const struct lu_env *env,
1251 struct dt_object *dt,
1252 const struct dt_key *key,
1255 struct osd_object *obj = osd_dt_obj(dt);
1256 struct osd_thandle *oh;
1259 LASSERT(dt_object_exists(dt));
1260 LASSERT(osd_invariant(obj));
1261 LASSERT(th != NULL);
1262 LASSERT(obj->oo_db);
1264 oh = container_of0(th, struct osd_thandle, ot_super);
1266 /* do not specify the key as then DMU is trying to look it up
1267 * which is very expensive. usually the layers above lookup
1268 * before deletion */
1269 dmu_tx_hold_zap(oh->ot_tx, obj->oo_db->db_object, FALSE, NULL);
1274 static int osd_index_delete(const struct lu_env *env, struct dt_object *dt,
1275 const struct dt_key *key, struct thandle *th)
1277 struct osd_object *obj = osd_dt_obj(dt);
1278 struct osd_device *osd = osd_obj2dev(obj);
1279 struct osd_thandle *oh;
1280 __u64 *k = osd_oti_get(env)->oti_key64;
1284 LASSERT(obj->oo_db);
1285 LASSERT(th != NULL);
1286 oh = container_of0(th, struct osd_thandle, ot_super);
1288 rc = osd_prepare_key_uint64(obj, k, key);
1290 /* Remove binary key from the ZAP */
1291 rc = -zap_remove_uint64(osd->od_os, obj->oo_db->db_object,
1296 static int osd_index_it_get(const struct lu_env *env, struct dt_it *di,
1297 const struct dt_key *key)
1299 struct osd_zap_it *it = (struct osd_zap_it *)di;
1300 struct osd_object *obj = it->ozi_obj;
1301 struct osd_device *osd = osd_obj2dev(obj);
1305 LASSERT(it->ozi_zc);
1308 * XXX: we need a binary version of zap_cursor_move_to_key()
1309 * to implement this API */
1310 if (*((const __u64 *)key) != 0)
1311 CERROR("NOT IMPLEMETED YET (move to %#llx)\n",
1314 zap_cursor_fini(it->ozi_zc);
1315 zap_cursor_init(it->ozi_zc, osd->od_os, obj->oo_db->db_object);
1321 static int osd_index_it_next(const struct lu_env *env, struct dt_it *di)
1323 struct osd_zap_it *it = (struct osd_zap_it *)di;
1324 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
1328 if (it->ozi_reset == 0)
1329 zap_cursor_advance(it->ozi_zc);
1333 * According to current API we need to return error if it's last entry.
1334 * zap_cursor_advance() does not return any value. So we need to call
1335 * retrieve to check if there is any record. We should make
1336 * changes to Iterator API to not return status for this API
1338 rc = -zap_cursor_retrieve(it->ozi_zc, za);
1345 static struct dt_key *osd_index_it_key(const struct lu_env *env,
1346 const struct dt_it *di)
1348 struct osd_zap_it *it = (struct osd_zap_it *)di;
1349 struct osd_object *obj = it->ozi_obj;
1350 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
1355 rc = -zap_cursor_retrieve(it->ozi_zc, za);
1357 RETURN(ERR_PTR(rc));
1359 /* the binary key is stored in the name */
1360 memcpy(&it->ozi_key, za->za_name, obj->oo_keysize);
1362 RETURN((struct dt_key *)&it->ozi_key);
1365 static int osd_index_it_key_size(const struct lu_env *env,
1366 const struct dt_it *di)
1368 struct osd_zap_it *it = (struct osd_zap_it *)di;
1369 struct osd_object *obj = it->ozi_obj;
1370 RETURN(obj->oo_keysize);
1373 static int osd_index_it_rec(const struct lu_env *env, const struct dt_it *di,
1374 struct dt_rec *rec, __u32 attr)
1376 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
1377 struct osd_zap_it *it = (struct osd_zap_it *)di;
1378 struct osd_object *obj = it->ozi_obj;
1379 struct osd_device *osd = osd_obj2dev(obj);
1380 __u64 *k = osd_oti_get(env)->oti_key64;
1385 rc = -zap_cursor_retrieve(it->ozi_zc, za);
1389 rc = osd_prepare_key_uint64(obj, k, (const struct dt_key *)za->za_name);
1391 rc = -zap_lookup_uint64(osd->od_os, obj->oo_db->db_object,
1392 k, rc, obj->oo_recusize, obj->oo_recsize,
1397 static __u64 osd_index_it_store(const struct lu_env *env,
1398 const struct dt_it *di)
1400 struct osd_zap_it *it = (struct osd_zap_it *)di;
1403 RETURN((__u64)zap_cursor_serialize(it->ozi_zc));
1406 static int osd_index_it_load(const struct lu_env *env, const struct dt_it *di,
1409 struct osd_zap_it *it = (struct osd_zap_it *)di;
1410 struct osd_object *obj = it->ozi_obj;
1411 struct osd_device *osd = osd_obj2dev(obj);
1412 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
1416 /* reset the cursor */
1417 zap_cursor_fini(it->ozi_zc);
1418 zap_cursor_init_serialized(it->ozi_zc, osd->od_os,
1419 obj->oo_db->db_object, hash);
1422 rc = -zap_cursor_retrieve(it->ozi_zc, za);
1425 else if (rc == -ENOENT)
1431 static struct dt_index_operations osd_index_ops = {
1432 .dio_lookup = osd_index_lookup,
1433 .dio_declare_insert = osd_declare_index_insert,
1434 .dio_insert = osd_index_insert,
1435 .dio_declare_delete = osd_declare_index_delete,
1436 .dio_delete = osd_index_delete,
1438 .init = osd_index_it_init,
1439 .fini = osd_index_it_fini,
1440 .get = osd_index_it_get,
1441 .put = osd_index_it_put,
1442 .next = osd_index_it_next,
1443 .key = osd_index_it_key,
1444 .key_size = osd_index_it_key_size,
1445 .rec = osd_index_it_rec,
1446 .store = osd_index_it_store,
1447 .load = osd_index_it_load
1451 struct osd_metadnode_it {
1452 struct osd_device *mit_dev;
1454 struct lu_fid mit_fid;
1456 __u64 mit_prefetched_dnode;
1459 static struct dt_it *osd_zfs_otable_it_init(const struct lu_env *env,
1460 struct dt_object *dt, __u32 attr)
1462 struct osd_device *dev = osd_dev(dt->do_lu.lo_dev);
1463 struct osd_metadnode_it *it;
1467 if (unlikely(it == NULL))
1468 RETURN(ERR_PTR(-ENOMEM));
1472 /* XXX: dmu_object_next() does NOT find dnodes allocated
1473 * in the current non-committed txg, so we force txg
1474 * commit to find all existing dnodes ... */
1475 txg_wait_synced(dmu_objset_pool(dev->od_os), 0ULL);
1477 RETURN((struct dt_it *)it);
1480 static void osd_zfs_otable_it_fini(const struct lu_env *env, struct dt_it *di)
1482 struct osd_metadnode_it *it = (struct osd_metadnode_it *)di;
1487 static int osd_zfs_otable_it_get(const struct lu_env *env,
1488 struct dt_it *di, const struct dt_key *key)
1493 static void osd_zfs_otable_it_put(const struct lu_env *env, struct dt_it *di)
1497 #define OTABLE_PREFETCH 256
1499 static void osd_zfs_otable_prefetch(const struct lu_env *env,
1500 struct osd_metadnode_it *it)
1502 struct osd_device *dev = it->mit_dev;
1505 /* can go negative on the very first access to the iterator
1506 * or if some non-Lustre objects were found */
1507 if (unlikely(it->mit_prefetched < 0))
1508 it->mit_prefetched = 0;
1510 if (it->mit_prefetched >= (OTABLE_PREFETCH >> 1))
1513 if (it->mit_prefetched_dnode == 0)
1514 it->mit_prefetched_dnode = it->mit_pos;
1516 while (it->mit_prefetched < OTABLE_PREFETCH) {
1517 rc = -dmu_object_next(dev->od_os, &it->mit_prefetched_dnode,
1519 if (unlikely(rc != 0))
1522 osd_dmu_prefetch(dev->od_os, it->mit_prefetched_dnode,
1523 0, 0, 0, ZIO_PRIORITY_ASYNC_READ);
1525 it->mit_prefetched++;
1529 static int osd_zfs_otable_it_next(const struct lu_env *env, struct dt_it *di)
1531 struct osd_metadnode_it *it = (struct osd_metadnode_it *)di;
1532 struct lustre_mdt_attrs *lma;
1533 struct osd_device *dev = it->mit_dev;
1534 nvlist_t *nvbuf = NULL;
1539 memset(&it->mit_fid, 0, sizeof(it->mit_fid));
1541 dnode = it->mit_pos;
1543 rc = -dmu_object_next(dev->od_os, &it->mit_pos, B_FALSE, 0);
1544 if (unlikely(rc != 0))
1546 it->mit_prefetched--;
1548 /* LMA is required for this to be a Lustre object.
1549 * If there is no xattr skip it. */
1550 rc = __osd_xattr_load(dev, it->mit_pos, &nvbuf);
1551 if (unlikely(rc != 0))
1554 LASSERT(nvbuf != NULL);
1555 rc = -nvlist_lookup_byte_array(nvbuf, XATTR_NAME_LMA, &v, &s);
1556 if (likely(rc == 0)) {
1558 lma = (struct lustre_mdt_attrs *)v;
1559 lustre_lma_swab(lma);
1560 it->mit_fid = lma->lma_self_fid;
1564 /* not a Lustre object, try next one */
1571 /* we aren't prefetching in the above loop because the number of
1572 * non-Lustre objects is very small and we will be repeating very
1573 * rare. in case we want to use this to iterate over non-Lustre
1574 * objects (i.e. when we convert regular ZFS in Lustre) it makes
1575 * sense to initiate prefetching in the loop */
1577 /* 0 - there are more items, +1 - the end */
1578 if (likely(rc == 0))
1579 osd_zfs_otable_prefetch(env, it);
1581 CDEBUG(D_OTHER, "advance: %llu -> %llu "DFID": %d\n", dnode,
1582 it->mit_pos, PFID(&it->mit_fid), rc);
1588 static struct dt_key *osd_zfs_otable_it_key(const struct lu_env *env,
1589 const struct dt_it *di)
1594 static int osd_zfs_otable_it_key_size(const struct lu_env *env,
1595 const struct dt_it *di)
1597 return sizeof(__u64);
1600 static int osd_zfs_otable_it_rec(const struct lu_env *env,
1601 const struct dt_it *di,
1602 struct dt_rec *rec, __u32 attr)
1604 struct osd_metadnode_it *it = (struct osd_metadnode_it *)di;
1605 struct lu_fid *fid = (struct lu_fid *)rec;
1614 static __u64 osd_zfs_otable_it_store(const struct lu_env *env,
1615 const struct dt_it *di)
1617 struct osd_metadnode_it *it = (struct osd_metadnode_it *)di;
1622 static int osd_zfs_otable_it_load(const struct lu_env *env,
1623 const struct dt_it *di, __u64 hash)
1625 struct osd_metadnode_it *it = (struct osd_metadnode_it *)di;
1628 it->mit_prefetched = 0;
1629 it->mit_prefetched_dnode = 0;
1631 return osd_zfs_otable_it_next(env, (struct dt_it *)di);
1634 static int osd_zfs_otable_it_key_rec(const struct lu_env *env,
1635 const struct dt_it *di, void *key_rec)
1640 const struct dt_index_operations osd_zfs_otable_ops = {
1642 .init = osd_zfs_otable_it_init,
1643 .fini = osd_zfs_otable_it_fini,
1644 .get = osd_zfs_otable_it_get,
1645 .put = osd_zfs_otable_it_put,
1646 .next = osd_zfs_otable_it_next,
1647 .key = osd_zfs_otable_it_key,
1648 .key_size = osd_zfs_otable_it_key_size,
1649 .rec = osd_zfs_otable_it_rec,
1650 .store = osd_zfs_otable_it_store,
1651 .load = osd_zfs_otable_it_load,
1652 .key_rec = osd_zfs_otable_it_key_rec,
1656 int osd_index_try(const struct lu_env *env, struct dt_object *dt,
1657 const struct dt_index_features *feat)
1659 struct osd_object *obj = osd_dt_obj(dt);
1663 down_read(&obj->oo_guard);
1666 * XXX: implement support for fixed-size keys sorted with natural
1667 * numerical way (not using internal hash value)
1669 if (feat->dif_flags & DT_IND_RANGE)
1670 GOTO(out, rc = -ERANGE);
1672 if (unlikely(feat == &dt_otable_features)) {
1673 dt->do_index_ops = &osd_zfs_otable_ops;
1677 LASSERT(!dt_object_exists(dt) || obj->oo_db != NULL);
1678 if (likely(feat == &dt_directory_features)) {
1679 if (!dt_object_exists(dt) || osd_object_is_zap(obj->oo_db))
1680 dt->do_index_ops = &osd_dir_ops;
1682 GOTO(out, rc = -ENOTDIR);
1683 } else if (unlikely(feat == &dt_acct_features)) {
1684 LASSERT(fid_is_acct(lu_object_fid(&dt->do_lu)));
1685 dt->do_index_ops = &osd_acct_index_ops;
1686 } else if (dt->do_index_ops == NULL) {
1687 /* For index file, we don't support variable key & record sizes
1688 * and the key has to be unique */
1689 if ((feat->dif_flags & ~DT_IND_UPDATE) != 0)
1690 GOTO(out, rc = -EINVAL);
1692 if (feat->dif_keysize_max > ZAP_MAXNAMELEN)
1693 GOTO(out, rc = -E2BIG);
1694 if (feat->dif_keysize_max != feat->dif_keysize_min)
1695 GOTO(out, rc = -EINVAL);
1697 /* As for the record size, it should be a multiple of 8 bytes
1698 * and smaller than the maximum value length supported by ZAP.
1700 if (feat->dif_recsize_max > ZAP_MAXVALUELEN)
1701 GOTO(out, rc = -E2BIG);
1702 if (feat->dif_recsize_max != feat->dif_recsize_min)
1703 GOTO(out, rc = -EINVAL);
1705 obj->oo_keysize = feat->dif_keysize_max;
1706 obj->oo_recsize = feat->dif_recsize_max;
1707 obj->oo_recusize = 1;
1709 /* ZFS prefers to work with array of 64bits */
1710 if ((obj->oo_recsize & 7) == 0) {
1711 obj->oo_recsize >>= 3;
1712 obj->oo_recusize = 8;
1714 dt->do_index_ops = &osd_index_ops;
1718 up_read(&obj->oo_guard);