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.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2009, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2012, 2016, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
32 * lustre/osd-zfs/osd_index.c
34 * Author: Alex Zhuravlev <bzzz@whamcloud.com>
35 * Author: Mike Pershin <tappro@whamcloud.com>
38 #define DEBUG_SUBSYSTEM S_OSD
40 #include <lustre_ver.h>
41 #include <libcfs/libcfs.h>
42 #include <obd_support.h>
43 #include <lustre_net.h>
45 #include <obd_class.h>
46 #include <lustre_disk.h>
47 #include <lustre_fid.h>
49 #include "osd_internal.h"
51 #include <sys/dnode.h>
56 #include <sys/spa_impl.h>
57 #include <sys/zfs_znode.h>
58 #include <sys/dmu_tx.h>
59 #include <sys/dmu_objset.h>
60 #include <sys/dsl_prop.h>
61 #include <sys/sa_impl.h>
64 static inline int osd_object_is_zap(dmu_buf_t *db)
66 dmu_buf_impl_t *dbi = (dmu_buf_impl_t *) db;
72 rc = (dn->dn_type == DMU_OT_DIRECTORY_CONTENTS ||
73 dn->dn_type == DMU_OT_USERGROUP_USED);
79 /* We don't actually have direct access to the zap_hashbits() function
80 * so just pretend like we do for now. If this ever breaks we can look at
82 #define zap_hashbits(zc) 48
85 * | cd (16 bits) | hash (48 bits) |
86 * we need it in other form:
87 * |0| hash (48 bit) | cd (15 bit) |
88 * to be a full 64-bit ordered hash so that Lustre readdir can use it to merge
89 * the readdir hashes from multiple directory stripes uniformly on the client.
90 * Another point is sign bit, the hash range should be in [0, 2^63-1] because
91 * loff_t (for llseek) needs to be a positive value. This means the "cd" field
92 * should only be the low 15 bits.
94 uint64_t osd_zap_cursor_serialize(zap_cursor_t *zc)
96 uint64_t zfs_hash = zap_cursor_serialize(zc) & (~0ULL >> 1);
98 return (zfs_hash >> zap_hashbits(zc)) |
99 (zfs_hash << (63 - zap_hashbits(zc)));
102 void osd_zap_cursor_init_serialized(zap_cursor_t *zc, struct objset *os,
103 uint64_t id, uint64_t dirhash)
105 uint64_t zfs_hash = ((dirhash << zap_hashbits(zc)) & (~0ULL >> 1)) |
106 (dirhash >> (63 - zap_hashbits(zc)));
108 zap_cursor_init_serialized(zc, os, id, zfs_hash);
111 int osd_zap_cursor_init(zap_cursor_t **zc, struct objset *os,
112 uint64_t id, uint64_t dirhash)
117 if (unlikely(t == NULL))
120 osd_zap_cursor_init_serialized(t, os, id, dirhash);
126 void osd_zap_cursor_fini(zap_cursor_t *zc)
132 static inline void osd_obj_cursor_init_serialized(zap_cursor_t *zc,
133 struct osd_object *o,
136 struct osd_device *d = osd_obj2dev(o);
137 osd_zap_cursor_init_serialized(zc, d->od_os,
138 o->oo_db->db_object, dirhash);
141 static inline int osd_obj_cursor_init(zap_cursor_t **zc, struct osd_object *o,
144 struct osd_device *d = osd_obj2dev(o);
145 return osd_zap_cursor_init(zc, d->od_os, o->oo_db->db_object, dirhash);
148 static struct dt_it *osd_index_it_init(const struct lu_env *env,
149 struct dt_object *dt,
152 struct osd_thread_info *info = osd_oti_get(env);
153 struct osd_zap_it *it;
154 struct osd_object *obj = osd_dt_obj(dt);
155 struct lu_object *lo = &dt->do_lu;
159 if (obj->oo_destroyed)
160 RETURN(ERR_PTR(-ENOENT));
162 LASSERT(lu_object_exists(lo));
166 OBD_SLAB_ALLOC_PTR_GFP(it, osd_zapit_cachep, GFP_NOFS);
168 RETURN(ERR_PTR(-ENOMEM));
170 rc = osd_obj_cursor_init(&it->ozi_zc, obj, 0);
172 OBD_SLAB_FREE_PTR(it, osd_zapit_cachep);
180 RETURN((struct dt_it *)it);
183 static void osd_index_it_fini(const struct lu_env *env, struct dt_it *di)
185 struct osd_zap_it *it = (struct osd_zap_it *)di;
186 struct osd_object *obj;
190 LASSERT(it->ozi_obj);
194 osd_zap_cursor_fini(it->ozi_zc);
195 lu_object_put(env, &obj->oo_dt.do_lu);
196 OBD_SLAB_FREE_PTR(it, osd_zapit_cachep);
202 static void osd_index_it_put(const struct lu_env *env, struct dt_it *di)
204 /* PBS: do nothing : ref are incremented at retrive and decreamented
208 static inline void osd_it_append_attrs(struct lu_dirent *ent, __u32 attr,
211 const unsigned align = sizeof(struct luda_type) - 1;
212 struct luda_type *lt;
214 /* check if file type is required */
215 if (attr & LUDA_TYPE) {
216 len = (len + align) & ~align;
218 lt = (void *)ent->lde_name + len;
219 lt->lt_type = cpu_to_le16(DTTOIF(type));
220 ent->lde_attrs |= LUDA_TYPE;
223 ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
227 * Get the object's FID from its LMA EA.
229 * \param[in] env pointer to the thread context
230 * \param[in] osd pointer to the OSD device
231 * \param[in] oid the object's local identifier
232 * \param[out] fid the buffer to hold the object's FID
234 * \retval 0 for success
235 * \retval negative error number on failure
237 static int osd_get_fid_by_oid(const struct lu_env *env, struct osd_device *osd,
238 uint64_t oid, struct lu_fid *fid)
240 struct objset *os = osd->od_os;
241 struct osd_thread_info *oti = osd_oti_get(env);
242 struct lustre_mdt_attrs *lma =
243 (struct lustre_mdt_attrs *)oti->oti_buf;
245 nvlist_t *sa_xattr = NULL;
246 sa_handle_t *sa_hdl = NULL;
247 uchar_t *nv_value = NULL;
248 uint64_t xattr = ZFS_NO_OBJECT;
253 rc = __osd_xattr_load(osd, oid, &sa_xattr);
260 rc = -nvlist_lookup_byte_array(sa_xattr, XATTR_NAME_LMA, &nv_value,
268 if (unlikely(size > sizeof(oti->oti_buf)))
269 GOTO(out, rc = -ERANGE);
271 memcpy(lma, nv_value, size);
276 rc = -sa_handle_get(os, oid, NULL, SA_HDL_PRIVATE, &sa_hdl);
280 rc = -sa_lookup(sa_hdl, SA_ZPL_XATTR(osd), &xattr, 8);
281 sa_handle_destroy(sa_hdl);
286 buf.lb_len = sizeof(oti->oti_buf);
287 rc = __osd_xattr_get_large(env, osd, xattr, &buf,
288 XATTR_NAME_LMA, &size);
293 if (size < sizeof(*lma))
294 GOTO(out, rc = -EIO);
296 lustre_lma_swab(lma);
297 if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
298 CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT))) {
299 CWARN("%s: unsupported incompat LMA feature(s) %#x for "
300 "oid = %#llx\n", osd->od_svname,
301 lma->lma_incompat & ~LMA_INCOMPAT_SUPP, oid);
302 GOTO(out, rc = -EOPNOTSUPP);
304 *fid = lma->lma_self_fid;
309 if (sa_xattr != NULL)
310 nvlist_free(sa_xattr);
315 * As we don't know FID, we can't use LU object, so this function
316 * partially duplicate __osd_xattr_get() which is built around
317 * LU-object and uses it to cache data like regular EA dnode, etc
319 static int osd_find_parent_by_dnode(const struct lu_env *env,
323 struct osd_device *osd = osd_obj2dev(osd_dt_obj(o));
325 uint64_t dnode = ZFS_NO_OBJECT;
329 /* first of all, get parent dnode from own attributes */
330 LASSERT(osd_dt_obj(o)->oo_db);
331 rc = -sa_handle_get(osd->od_os, osd_dt_obj(o)->oo_db->db_object,
332 NULL, SA_HDL_PRIVATE, &sa_hdl);
336 rc = -sa_lookup(sa_hdl, SA_ZPL_PARENT(osd), &dnode, 8);
337 sa_handle_destroy(sa_hdl);
339 rc = osd_get_fid_by_oid(env, osd, dnode, fid);
344 static int osd_find_parent_fid(const struct lu_env *env, struct dt_object *o,
347 struct link_ea_header *leh;
348 struct link_ea_entry *lee;
353 buf.lb_buf = osd_oti_get(env)->oti_buf;
354 buf.lb_len = sizeof(osd_oti_get(env)->oti_buf);
356 rc = osd_xattr_get(env, o, &buf, XATTR_NAME_LINK);
358 rc = osd_xattr_get(env, o, &LU_BUF_NULL, XATTR_NAME_LINK);
362 OBD_ALLOC(buf.lb_buf, rc);
363 if (buf.lb_buf == NULL)
366 rc = osd_xattr_get(env, o, &buf, XATTR_NAME_LINK);
370 if (rc < sizeof(*leh) + sizeof(*lee))
371 GOTO(out, rc = -EINVAL);
374 if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) {
375 leh->leh_magic = LINK_EA_MAGIC;
376 leh->leh_reccount = __swab32(leh->leh_reccount);
377 leh->leh_len = __swab64(leh->leh_len);
379 if (leh->leh_magic != LINK_EA_MAGIC)
380 GOTO(out, rc = -EINVAL);
381 if (leh->leh_reccount == 0)
382 GOTO(out, rc = -ENODATA);
384 lee = (struct link_ea_entry *)(leh + 1);
385 fid_be_to_cpu(fid, (const struct lu_fid *)&lee->lee_parent_fid);
389 if (buf.lb_buf != osd_oti_get(env)->oti_buf)
390 OBD_FREE(buf.lb_buf, buf.lb_len);
393 /* this block can be enabled for additional verification
394 * it's trying to match FID from LinkEA vs. FID from LMA */
398 rc2 = osd_find_parent_by_dnode(env, o, &fid2);
400 if (lu_fid_eq(fid, &fid2) == 0)
401 CERROR("wrong parent: "DFID" != "DFID"\n",
402 PFID(fid), PFID(&fid2));
406 /* no LinkEA is found, let's try to find the fid in parent's LMA */
407 if (unlikely(rc != 0))
408 rc = osd_find_parent_by_dnode(env, o, fid);
413 static int osd_dir_lookup(const struct lu_env *env, struct dt_object *dt,
414 struct dt_rec *rec, const struct dt_key *key)
416 struct osd_thread_info *oti = osd_oti_get(env);
417 struct osd_object *obj = osd_dt_obj(dt);
418 struct osd_device *osd = osd_obj2dev(obj);
419 char *name = (char *)key;
423 if (name[0] == '.') {
425 const struct lu_fid *f = lu_object_fid(&dt->do_lu);
426 memcpy(rec, f, sizeof(*f));
428 } else if (name[1] == '.' && name[2] == 0) {
429 rc = osd_find_parent_fid(env, dt, (struct lu_fid *)rec);
430 RETURN(rc == 0 ? 1 : rc);
434 memset(&oti->oti_zde.lzd_fid, 0, sizeof(struct lu_fid));
435 rc = -zap_lookup(osd->od_os, obj->oo_db->db_object,
436 (char *)key, 8, sizeof(oti->oti_zde) / 8,
437 (void *)&oti->oti_zde);
441 if (likely(fid_is_sane(&oti->oti_zde.lzd_fid))) {
442 memcpy(rec, &oti->oti_zde.lzd_fid, sizeof(struct lu_fid));
446 rc = osd_get_fid_by_oid(env, osd, oti->oti_zde.lzd_reg.zde_dnode,
447 (struct lu_fid *)rec);
449 RETURN(rc == 0 ? 1 : (rc == -ENOENT ? -ENODATA : rc));
452 static int osd_declare_dir_insert(const struct lu_env *env,
453 struct dt_object *dt,
454 const struct dt_rec *rec,
455 const struct dt_key *key,
458 struct osd_object *obj = osd_dt_obj(dt);
459 struct osd_thandle *oh;
464 oh = container_of0(th, struct osd_thandle, ot_super);
466 /* This is for inserting dot/dotdot for new created dir. */
467 if (obj->oo_db == NULL)
468 object = DMU_NEW_OBJECT;
470 object = obj->oo_db->db_object;
472 /* do not specify the key as then DMU is trying to look it up
473 * which is very expensive. usually the layers above lookup
474 * before insertion */
475 dmu_tx_hold_zap(oh->ot_tx, object, TRUE, NULL);
481 * Find the osd object for given fid.
483 * \param fid need to find the osd object having this fid
485 * \retval osd_object on success
486 * \retval -ve on error
488 struct osd_object *osd_object_find(const struct lu_env *env,
489 struct dt_object *dt,
490 const struct lu_fid *fid)
492 struct lu_device *ludev = dt->do_lu.lo_dev;
493 struct osd_object *child = NULL;
494 struct lu_object *luch;
495 struct lu_object *lo;
498 * at this point topdev might not exist yet
499 * (i.e. MGS is preparing profiles). so we can
500 * not rely on topdev and instead lookup with
501 * our device passed as topdev. this can't work
502 * if the object isn't cached yet (as osd doesn't
503 * allocate lu_header). IOW, the object must be
504 * in the cache, otherwise lu_object_alloc() crashes
507 luch = lu_object_find_at(env, ludev, fid, NULL);
511 if (lu_object_exists(luch)) {
512 lo = lu_object_locate(luch->lo_header, ludev->ld_type);
516 LU_OBJECT_DEBUG(D_ERROR, env, luch,
517 "%s: object can't be located "DFID,
518 osd_dev(ludev)->od_svname, PFID(fid));
521 lu_object_put(env, luch);
522 CERROR("%s: Unable to get osd_object "DFID"\n",
523 osd_dev(ludev)->od_svname, PFID(fid));
524 child = ERR_PTR(-ENOENT);
527 LU_OBJECT_DEBUG(D_ERROR, env, luch,
528 "%s: lu_object does not exists "DFID,
529 osd_dev(ludev)->od_svname, PFID(fid));
530 lu_object_put(env, luch);
531 child = ERR_PTR(-ENOENT);
538 * Put the osd object once done with it.
540 * \param obj osd object that needs to be put
542 static inline void osd_object_put(const struct lu_env *env,
543 struct osd_object *obj)
545 lu_object_put(env, &obj->oo_dt.do_lu);
548 static int osd_seq_exists(const struct lu_env *env, struct osd_device *osd,
551 struct lu_seq_range *range = &osd_oti_get(env)->oti_seq_range;
552 struct seq_server_site *ss = osd_seq_site(osd);
557 LASSERT(ss->ss_server_fld != NULL);
559 rc = osd_fld_lookup(env, osd, seq, range);
562 CERROR("%s: Can not lookup fld for %#llx\n",
567 RETURN(ss->ss_node_id == range->lsr_index);
570 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
571 const struct lu_fid *fid)
573 struct seq_server_site *ss = osd_seq_site(osd);
576 /* FID seqs not in FLDB, must be local seq */
577 if (unlikely(!fid_seq_in_fldb(fid_seq(fid))))
580 /* If FLD is not being initialized yet, it only happens during the
581 * initialization, likely during mgs initialization, and we assume
582 * this is local FID. */
583 if (ss == NULL || ss->ss_server_fld == NULL)
586 /* Only check the local FLDB here */
587 if (osd_seq_exists(env, osd, fid_seq(fid)))
594 * Inserts (key, value) pair in \a directory object.
596 * \param dt osd index object
597 * \param key key for index
598 * \param rec record reference
599 * \param th transaction handler
600 * \param ignore_quota update should not affect quota
603 * \retval -ve failure
605 static int osd_dir_insert(const struct lu_env *env, struct dt_object *dt,
606 const struct dt_rec *rec, const struct dt_key *key,
607 struct thandle *th, int ignore_quota)
609 struct osd_thread_info *oti = osd_oti_get(env);
610 struct osd_object *parent = osd_dt_obj(dt);
611 struct osd_device *osd = osd_obj2dev(parent);
612 struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec;
613 const struct lu_fid *fid = rec1->rec_fid;
614 struct osd_thandle *oh;
615 struct osd_object *child = NULL;
617 char *name = (char *)key;
621 LASSERT(parent->oo_db);
623 LASSERT(dt_object_exists(dt));
624 LASSERT(osd_invariant(parent));
627 oh = container_of0(th, struct osd_thandle, ot_super);
629 rc = osd_remote_fid(env, osd, fid);
631 CERROR("%s: Can not find object "DFID": rc = %d\n",
632 osd->od_svname, PFID(fid), rc);
636 if (unlikely(rc == 1)) {
637 /* Insert remote entry */
638 memset(&oti->oti_zde.lzd_reg, 0, sizeof(oti->oti_zde.lzd_reg));
639 oti->oti_zde.lzd_reg.zde_type = IFTODT(rec1->rec_type & S_IFMT);
642 * To simulate old Orion setups with ./.. stored in the
645 /* Insert local entry */
646 child = osd_object_find(env, dt, fid);
648 RETURN(PTR_ERR(child));
650 LASSERT(child->oo_db);
651 if (name[0] == '.') {
653 /* do not store ".", instead generate it
654 * during iteration */
656 } else if (name[1] == '.' && name[2] == 0) {
657 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT)) {
658 struct lu_fid tfid = *fid;
660 osd_object_put(env, child);
662 child = osd_object_find(env, dt, &tfid);
664 RETURN(PTR_ERR(child));
666 LASSERT(child->oo_db);
669 /* update parent dnode in the child.
670 * later it will be used to generate ".." */
671 rc = osd_object_sa_update(parent,
673 &child->oo_db->db_object,
679 CLASSERT(sizeof(oti->oti_zde.lzd_reg) == 8);
680 CLASSERT(sizeof(oti->oti_zde) % 8 == 0);
681 attr = child->oo_dt.do_lu.lo_header ->loh_attr;
682 oti->oti_zde.lzd_reg.zde_type = IFTODT(attr & S_IFMT);
683 oti->oti_zde.lzd_reg.zde_dnode = child->oo_db->db_object;
686 oti->oti_zde.lzd_fid = *fid;
687 /* Insert (key,oid) into ZAP */
688 rc = -zap_add(osd->od_os, parent->oo_db->db_object,
689 (char *)key, 8, sizeof(oti->oti_zde) / 8,
690 (void *)&oti->oti_zde, oh->ot_tx);
691 if (unlikely(rc == -EEXIST &&
692 name[0] == '.' && name[1] == '.' && name[2] == 0))
693 /* Update (key,oid) in ZAP */
694 rc = -zap_update(osd->od_os, parent->oo_db->db_object,
695 (char *)key, 8, sizeof(oti->oti_zde) / 8,
696 (void *)&oti->oti_zde, oh->ot_tx);
700 osd_object_put(env, child);
705 static int osd_declare_dir_delete(const struct lu_env *env,
706 struct dt_object *dt,
707 const struct dt_key *key,
710 struct osd_object *obj = osd_dt_obj(dt);
711 struct osd_thandle *oh;
715 LASSERT(dt_object_exists(dt));
716 LASSERT(osd_invariant(obj));
719 oh = container_of0(th, struct osd_thandle, ot_super);
721 if (dt_object_exists(dt)) {
723 dnode = obj->oo_db->db_object;
725 dnode = DMU_NEW_OBJECT;
728 /* do not specify the key as then DMU is trying to look it up
729 * which is very expensive. usually the layers above lookup
731 dmu_tx_hold_zap(oh->ot_tx, dnode, FALSE, NULL);
736 static int osd_dir_delete(const struct lu_env *env, struct dt_object *dt,
737 const struct dt_key *key, struct thandle *th)
739 struct osd_object *obj = osd_dt_obj(dt);
740 struct osd_device *osd = osd_obj2dev(obj);
741 struct osd_thandle *oh;
742 dmu_buf_t *zap_db = obj->oo_db;
743 char *name = (char *)key;
750 oh = container_of0(th, struct osd_thandle, ot_super);
753 * In Orion . and .. were stored in the directory (not generated upon
754 * request as now). we preserve them for backward compatibility
756 if (name[0] == '.') {
759 } else if (name[1] == '.' && name[2] == 0) {
764 /* Remove key from the ZAP */
765 rc = -zap_remove(osd->od_os, zap_db->db_object,
766 (char *) key, oh->ot_tx);
768 if (unlikely(rc && rc != -ENOENT))
769 CERROR("%s: zap_remove failed: rc = %d\n", osd->od_svname, rc);
774 static struct dt_it *osd_dir_it_init(const struct lu_env *env,
775 struct dt_object *dt,
778 struct osd_zap_it *it;
780 it = (struct osd_zap_it *)osd_index_it_init(env, dt, unused);
784 RETURN((struct dt_it *)it);
788 * Move Iterator to record specified by \a key
790 * \param di osd iterator
791 * \param key key for index
793 * \retval +ve di points to record with least key not larger than key
794 * \retval 0 di points to exact matched key
795 * \retval -ve failure
797 static int osd_dir_it_get(const struct lu_env *env,
798 struct dt_it *di, const struct dt_key *key)
800 struct osd_zap_it *it = (struct osd_zap_it *)di;
801 struct osd_object *obj = it->ozi_obj;
802 char *name = (char *)key;
809 /* reset the cursor */
810 zap_cursor_fini(it->ozi_zc);
811 osd_obj_cursor_init_serialized(it->ozi_zc, obj, 0);
813 /* XXX: implementation of the API is broken at the moment */
814 LASSERT(((const char *)key)[0] == 0);
821 if (name[0] == '.') {
825 } else if (name[1] == '.' && name[2] == 0) {
831 /* neither . nor .. - some real record */
839 static void osd_dir_it_put(const struct lu_env *env, struct dt_it *di)
841 /* PBS: do nothing : ref are incremented at retrive and decreamented
846 * in Orion . and .. were stored in the directory, while ZPL
847 * and current osd-zfs generate them up on request. so, we
848 * need to ignore previously stored . and ..
850 static int osd_index_retrieve_skip_dots(struct osd_zap_it *it,
856 rc = -zap_cursor_retrieve(it->ozi_zc, za);
859 if (unlikely(rc == 0 && za->za_name[0] == '.')) {
860 if (za->za_name[1] == 0) {
862 } else if (za->za_name[1] == '.' &&
863 za->za_name[2] == 0) {
867 zap_cursor_advance(it->ozi_zc);
869 } while (unlikely(rc == 0 && isdot));
875 * to load a directory entry at a time and stored it in
876 * iterator's in-memory data structure.
878 * \param di, struct osd_it_ea, iterator's in memory structure
880 * \retval +ve, iterator reached to end
881 * \retval 0, iterator not reached to end
882 * \retval -ve, on error
884 static int osd_dir_it_next(const struct lu_env *env, struct dt_it *di)
886 struct osd_zap_it *it = (struct osd_zap_it *)di;
887 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
892 /* temp. storage should be enough for any key supported by ZFS */
893 CLASSERT(sizeof(za->za_name) <= sizeof(it->ozi_name));
896 * the first ->next() moves the cursor to .
897 * the second ->next() moves the cursor to ..
898 * then we get to the real records and have to verify any exist
900 if (it->ozi_pos <= 2) {
906 zap_cursor_advance(it->ozi_zc);
910 * According to current API we need to return error if its last entry.
911 * zap_cursor_advance() does not return any value. So we need to call
912 * retrieve to check if there is any record. We should make
913 * changes to Iterator API to not return status for this API
915 rc = osd_index_retrieve_skip_dots(it, za);
917 if (rc == -ENOENT) /* end of dir */
923 static struct dt_key *osd_dir_it_key(const struct lu_env *env,
924 const struct dt_it *di)
926 struct osd_zap_it *it = (struct osd_zap_it *)di;
927 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
931 if (it->ozi_pos <= 1) {
933 RETURN((struct dt_key *)".");
934 } else if (it->ozi_pos == 2) {
935 RETURN((struct dt_key *)"..");
938 if ((rc = -zap_cursor_retrieve(it->ozi_zc, za)))
941 strcpy(it->ozi_name, za->za_name);
943 RETURN((struct dt_key *)it->ozi_name);
946 static int osd_dir_it_key_size(const struct lu_env *env, const struct dt_it *di)
948 struct osd_zap_it *it = (struct osd_zap_it *)di;
949 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
953 if (it->ozi_pos <= 1) {
956 } else if (it->ozi_pos == 2) {
960 if ((rc = -zap_cursor_retrieve(it->ozi_zc, za)) == 0)
961 rc = strlen(za->za_name);
966 static int osd_dir_it_rec(const struct lu_env *env, const struct dt_it *di,
967 struct dt_rec *dtrec, __u32 attr)
969 struct osd_zap_it *it = (struct osd_zap_it *)di;
970 struct lu_dirent *lde = (struct lu_dirent *)dtrec;
971 struct luz_direntry *zde = &osd_oti_get(env)->oti_zde;
972 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
976 if (it->ozi_pos <= 1) {
977 lde->lde_hash = cpu_to_le64(1);
978 strcpy(lde->lde_name, ".");
979 lde->lde_namelen = cpu_to_le16(1);
980 lde->lde_fid = *lu_object_fid(&it->ozi_obj->oo_dt.do_lu);
981 lde->lde_attrs = LUDA_FID;
982 /* append lustre attributes */
983 osd_it_append_attrs(lde, attr, 1, IFTODT(S_IFDIR));
984 lde->lde_reclen = cpu_to_le16(lu_dirent_calc_size(1, attr));
988 } else if (it->ozi_pos == 2) {
989 lde->lde_hash = cpu_to_le64(2);
990 strcpy(lde->lde_name, "..");
991 lde->lde_namelen = cpu_to_le16(2);
992 lde->lde_attrs = LUDA_FID;
993 /* append lustre attributes */
994 osd_it_append_attrs(lde, attr, 2, IFTODT(S_IFDIR));
995 lde->lde_reclen = cpu_to_le16(lu_dirent_calc_size(2, attr));
996 rc = osd_find_parent_fid(env, &it->ozi_obj->oo_dt, &lde->lde_fid);
998 /* ENOENT happens at the root of filesystem so ignore it */
1006 rc = -zap_cursor_retrieve(it->ozi_zc, za);
1007 if (unlikely(rc != 0))
1010 lde->lde_hash = cpu_to_le64(osd_zap_cursor_serialize(it->ozi_zc));
1011 namelen = strlen(za->za_name);
1012 if (namelen > NAME_MAX)
1013 GOTO(out, rc = -EOVERFLOW);
1014 strcpy(lde->lde_name, za->za_name);
1015 lde->lde_namelen = cpu_to_le16(namelen);
1017 if (za->za_integer_length != 8 || za->za_num_integers < 3) {
1018 CERROR("%s: unsupported direntry format: %d %d\n",
1019 osd_obj2dev(it->ozi_obj)->od_svname,
1020 za->za_integer_length, (int)za->za_num_integers);
1022 GOTO(out, rc = -EIO);
1025 rc = -zap_lookup(it->ozi_zc->zc_objset, it->ozi_zc->zc_zapobj,
1026 za->za_name, za->za_integer_length, 3, zde);
1030 lde->lde_fid = zde->lzd_fid;
1031 lde->lde_attrs = LUDA_FID;
1033 /* append lustre attributes */
1034 osd_it_append_attrs(lde, attr, namelen, zde->lzd_reg.zde_type);
1036 lde->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
1042 static int osd_dir_it_rec_size(const struct lu_env *env, const struct dt_it *di,
1045 struct osd_zap_it *it = (struct osd_zap_it *)di;
1046 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
1051 if (it->ozi_pos <= 1)
1053 else if (it->ozi_pos == 2)
1057 rc = lu_dirent_calc_size(namelen, attr);
1061 rc = -zap_cursor_retrieve(it->ozi_zc, za);
1062 if (unlikely(rc != 0))
1065 if (za->za_integer_length != 8 || za->za_num_integers < 3) {
1066 CERROR("%s: unsupported direntry format: %d %d\n",
1067 osd_obj2dev(it->ozi_obj)->od_svname,
1068 za->za_integer_length, (int)za->za_num_integers);
1072 namelen = strlen(za->za_name);
1073 if (namelen > NAME_MAX)
1076 rc = lu_dirent_calc_size(namelen, attr);
1081 static __u64 osd_dir_it_store(const struct lu_env *env, const struct dt_it *di)
1083 struct osd_zap_it *it = (struct osd_zap_it *)di;
1087 if (it->ozi_pos <= 2)
1090 pos = osd_zap_cursor_serialize(it->ozi_zc);
1097 * rc == 0 -> end of directory.
1098 * rc > 0 -> ok, proceed.
1099 * rc < 0 -> error. ( EOVERFLOW can be masked.)
1101 static int osd_dir_it_load(const struct lu_env *env,
1102 const struct dt_it *di, __u64 hash)
1104 struct osd_zap_it *it = (struct osd_zap_it *)di;
1105 struct osd_object *obj = it->ozi_obj;
1106 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
1110 /* reset the cursor */
1111 zap_cursor_fini(it->ozi_zc);
1112 osd_obj_cursor_init_serialized(it->ozi_zc, obj, hash);
1119 /* to return whether the end has been reached */
1120 rc = osd_index_retrieve_skip_dots(it, za);
1123 else if (rc == -ENOENT)
1130 struct dt_index_operations osd_dir_ops = {
1131 .dio_lookup = osd_dir_lookup,
1132 .dio_declare_insert = osd_declare_dir_insert,
1133 .dio_insert = osd_dir_insert,
1134 .dio_declare_delete = osd_declare_dir_delete,
1135 .dio_delete = osd_dir_delete,
1137 .init = osd_dir_it_init,
1138 .fini = osd_index_it_fini,
1139 .get = osd_dir_it_get,
1140 .put = osd_dir_it_put,
1141 .next = osd_dir_it_next,
1142 .key = osd_dir_it_key,
1143 .key_size = osd_dir_it_key_size,
1144 .rec = osd_dir_it_rec,
1145 .rec_size = osd_dir_it_rec_size,
1146 .store = osd_dir_it_store,
1147 .load = osd_dir_it_load
1152 * Primitives for index files using binary keys.
1155 /* key integer_size is 8 */
1156 static int osd_prepare_key_uint64(struct osd_object *o, __u64 *dst,
1157 const struct dt_key *src)
1164 /* align keysize to 64bit */
1165 size = (o->oo_keysize + sizeof(__u64) - 1) / sizeof(__u64);
1166 size *= sizeof(__u64);
1168 LASSERT(size <= MAXNAMELEN);
1170 if (unlikely(size > o->oo_keysize))
1171 memset(dst + o->oo_keysize, 0, size - o->oo_keysize);
1172 memcpy(dst, (const char *)src, o->oo_keysize);
1174 return (size/sizeof(__u64));
1177 static int osd_index_lookup(const struct lu_env *env, struct dt_object *dt,
1178 struct dt_rec *rec, const struct dt_key *key)
1180 struct osd_object *obj = osd_dt_obj(dt);
1181 struct osd_device *osd = osd_obj2dev(obj);
1182 __u64 *k = osd_oti_get(env)->oti_key64;
1186 rc = osd_prepare_key_uint64(obj, k, key);
1188 rc = -zap_lookup_uint64(osd->od_os, obj->oo_db->db_object,
1189 k, rc, obj->oo_recusize, obj->oo_recsize,
1191 RETURN(rc == 0 ? 1 : rc);
1194 static int osd_declare_index_insert(const struct lu_env *env,
1195 struct dt_object *dt,
1196 const struct dt_rec *rec,
1197 const struct dt_key *key,
1200 struct osd_object *obj = osd_dt_obj(dt);
1201 struct osd_thandle *oh;
1204 LASSERT(th != NULL);
1205 oh = container_of0(th, struct osd_thandle, ot_super);
1207 LASSERT(obj->oo_db);
1209 dmu_tx_hold_bonus(oh->ot_tx, obj->oo_db->db_object);
1211 /* do not specify the key as then DMU is trying to look it up
1212 * which is very expensive. usually the layers above lookup
1213 * before insertion */
1214 dmu_tx_hold_zap(oh->ot_tx, obj->oo_db->db_object, TRUE, NULL);
1219 static int osd_index_insert(const struct lu_env *env, struct dt_object *dt,
1220 const struct dt_rec *rec, const struct dt_key *key,
1221 struct thandle *th, int ignore_quota)
1223 struct osd_object *obj = osd_dt_obj(dt);
1224 struct osd_device *osd = osd_obj2dev(obj);
1225 struct osd_thandle *oh;
1226 __u64 *k = osd_oti_get(env)->oti_key64;
1230 LASSERT(obj->oo_db);
1231 LASSERT(dt_object_exists(dt));
1232 LASSERT(osd_invariant(obj));
1233 LASSERT(th != NULL);
1235 oh = container_of0(th, struct osd_thandle, ot_super);
1237 rc = osd_prepare_key_uint64(obj, k, key);
1239 /* Insert (key,oid) into ZAP */
1240 rc = -zap_add_uint64(osd->od_os, obj->oo_db->db_object,
1241 k, rc, obj->oo_recusize, obj->oo_recsize,
1242 (void *)rec, oh->ot_tx);
1246 static int osd_declare_index_delete(const struct lu_env *env,
1247 struct dt_object *dt,
1248 const struct dt_key *key,
1251 struct osd_object *obj = osd_dt_obj(dt);
1252 struct osd_thandle *oh;
1255 LASSERT(dt_object_exists(dt));
1256 LASSERT(osd_invariant(obj));
1257 LASSERT(th != NULL);
1258 LASSERT(obj->oo_db);
1260 oh = container_of0(th, struct osd_thandle, ot_super);
1262 /* do not specify the key as then DMU is trying to look it up
1263 * which is very expensive. usually the layers above lookup
1264 * before deletion */
1265 dmu_tx_hold_zap(oh->ot_tx, obj->oo_db->db_object, FALSE, NULL);
1270 static int osd_index_delete(const struct lu_env *env, struct dt_object *dt,
1271 const struct dt_key *key, struct thandle *th)
1273 struct osd_object *obj = osd_dt_obj(dt);
1274 struct osd_device *osd = osd_obj2dev(obj);
1275 struct osd_thandle *oh;
1276 __u64 *k = osd_oti_get(env)->oti_key64;
1280 LASSERT(obj->oo_db);
1281 LASSERT(th != NULL);
1282 oh = container_of0(th, struct osd_thandle, ot_super);
1284 rc = osd_prepare_key_uint64(obj, k, key);
1286 /* Remove binary key from the ZAP */
1287 rc = -zap_remove_uint64(osd->od_os, obj->oo_db->db_object,
1292 static int osd_index_it_get(const struct lu_env *env, struct dt_it *di,
1293 const struct dt_key *key)
1295 struct osd_zap_it *it = (struct osd_zap_it *)di;
1296 struct osd_object *obj = it->ozi_obj;
1297 struct osd_device *osd = osd_obj2dev(obj);
1301 LASSERT(it->ozi_zc);
1304 * XXX: we need a binary version of zap_cursor_move_to_key()
1305 * to implement this API */
1306 if (*((const __u64 *)key) != 0)
1307 CERROR("NOT IMPLEMETED YET (move to %#llx)\n",
1310 zap_cursor_fini(it->ozi_zc);
1311 zap_cursor_init(it->ozi_zc, osd->od_os, obj->oo_db->db_object);
1317 static int osd_index_it_next(const struct lu_env *env, struct dt_it *di)
1319 struct osd_zap_it *it = (struct osd_zap_it *)di;
1320 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
1324 if (it->ozi_reset == 0)
1325 zap_cursor_advance(it->ozi_zc);
1329 * According to current API we need to return error if it's last entry.
1330 * zap_cursor_advance() does not return any value. So we need to call
1331 * retrieve to check if there is any record. We should make
1332 * changes to Iterator API to not return status for this API
1334 rc = -zap_cursor_retrieve(it->ozi_zc, za);
1341 static struct dt_key *osd_index_it_key(const struct lu_env *env,
1342 const struct dt_it *di)
1344 struct osd_zap_it *it = (struct osd_zap_it *)di;
1345 struct osd_object *obj = it->ozi_obj;
1346 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
1351 rc = -zap_cursor_retrieve(it->ozi_zc, za);
1353 RETURN(ERR_PTR(rc));
1355 /* the binary key is stored in the name */
1356 memcpy(&it->ozi_key, za->za_name, obj->oo_keysize);
1358 RETURN((struct dt_key *)&it->ozi_key);
1361 static int osd_index_it_key_size(const struct lu_env *env,
1362 const struct dt_it *di)
1364 struct osd_zap_it *it = (struct osd_zap_it *)di;
1365 struct osd_object *obj = it->ozi_obj;
1366 RETURN(obj->oo_keysize);
1369 static int osd_index_it_rec(const struct lu_env *env, const struct dt_it *di,
1370 struct dt_rec *rec, __u32 attr)
1372 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
1373 struct osd_zap_it *it = (struct osd_zap_it *)di;
1374 struct osd_object *obj = it->ozi_obj;
1375 struct osd_device *osd = osd_obj2dev(obj);
1376 __u64 *k = osd_oti_get(env)->oti_key64;
1381 rc = -zap_cursor_retrieve(it->ozi_zc, za);
1385 rc = osd_prepare_key_uint64(obj, k, (const struct dt_key *)za->za_name);
1387 rc = -zap_lookup_uint64(osd->od_os, obj->oo_db->db_object,
1388 k, rc, obj->oo_recusize, obj->oo_recsize,
1393 static __u64 osd_index_it_store(const struct lu_env *env,
1394 const struct dt_it *di)
1396 struct osd_zap_it *it = (struct osd_zap_it *)di;
1399 RETURN((__u64)zap_cursor_serialize(it->ozi_zc));
1402 static int osd_index_it_load(const struct lu_env *env, const struct dt_it *di,
1405 struct osd_zap_it *it = (struct osd_zap_it *)di;
1406 struct osd_object *obj = it->ozi_obj;
1407 struct osd_device *osd = osd_obj2dev(obj);
1408 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
1412 /* reset the cursor */
1413 zap_cursor_fini(it->ozi_zc);
1414 zap_cursor_init_serialized(it->ozi_zc, osd->od_os,
1415 obj->oo_db->db_object, hash);
1418 rc = -zap_cursor_retrieve(it->ozi_zc, za);
1421 else if (rc == -ENOENT)
1427 static struct dt_index_operations osd_index_ops = {
1428 .dio_lookup = osd_index_lookup,
1429 .dio_declare_insert = osd_declare_index_insert,
1430 .dio_insert = osd_index_insert,
1431 .dio_declare_delete = osd_declare_index_delete,
1432 .dio_delete = osd_index_delete,
1434 .init = osd_index_it_init,
1435 .fini = osd_index_it_fini,
1436 .get = osd_index_it_get,
1437 .put = osd_index_it_put,
1438 .next = osd_index_it_next,
1439 .key = osd_index_it_key,
1440 .key_size = osd_index_it_key_size,
1441 .rec = osd_index_it_rec,
1442 .store = osd_index_it_store,
1443 .load = osd_index_it_load
1447 struct osd_metadnode_it {
1448 struct osd_device *mit_dev;
1450 struct lu_fid mit_fid;
1452 __u64 mit_prefetched_dnode;
1455 static struct dt_it *osd_zfs_otable_it_init(const struct lu_env *env,
1456 struct dt_object *dt, __u32 attr)
1458 struct osd_device *dev = osd_dev(dt->do_lu.lo_dev);
1459 struct osd_metadnode_it *it;
1463 if (unlikely(it == NULL))
1464 RETURN(ERR_PTR(-ENOMEM));
1468 /* XXX: dmu_object_next() does NOT find dnodes allocated
1469 * in the current non-committed txg, so we force txg
1470 * commit to find all existing dnodes ... */
1471 txg_wait_synced(dmu_objset_pool(dev->od_os), 0ULL);
1473 RETURN((struct dt_it *)it);
1476 static void osd_zfs_otable_it_fini(const struct lu_env *env, struct dt_it *di)
1478 struct osd_metadnode_it *it = (struct osd_metadnode_it *)di;
1483 static int osd_zfs_otable_it_get(const struct lu_env *env,
1484 struct dt_it *di, const struct dt_key *key)
1489 static void osd_zfs_otable_it_put(const struct lu_env *env, struct dt_it *di)
1493 #define OTABLE_PREFETCH 256
1495 static void osd_zfs_otable_prefetch(const struct lu_env *env,
1496 struct osd_metadnode_it *it)
1498 struct osd_device *dev = it->mit_dev;
1501 /* can go negative on the very first access to the iterator
1502 * or if some non-Lustre objects were found */
1503 if (unlikely(it->mit_prefetched < 0))
1504 it->mit_prefetched = 0;
1506 if (it->mit_prefetched >= (OTABLE_PREFETCH >> 1))
1509 if (it->mit_prefetched_dnode == 0)
1510 it->mit_prefetched_dnode = it->mit_pos;
1512 while (it->mit_prefetched < OTABLE_PREFETCH) {
1513 rc = -dmu_object_next(dev->od_os, &it->mit_prefetched_dnode,
1515 if (unlikely(rc != 0))
1518 osd_dmu_prefetch(dev->od_os, it->mit_prefetched_dnode,
1519 0, 0, 0, ZIO_PRIORITY_ASYNC_READ);
1521 it->mit_prefetched++;
1525 static int osd_zfs_otable_it_next(const struct lu_env *env, struct dt_it *di)
1527 struct osd_metadnode_it *it = (struct osd_metadnode_it *)di;
1528 struct lustre_mdt_attrs *lma;
1529 struct osd_device *dev = it->mit_dev;
1530 nvlist_t *nvbuf = NULL;
1535 memset(&it->mit_fid, 0, sizeof(it->mit_fid));
1537 dnode = it->mit_pos;
1539 rc = -dmu_object_next(dev->od_os, &it->mit_pos, B_FALSE, 0);
1540 if (unlikely(rc != 0))
1542 it->mit_prefetched--;
1544 /* LMA is required for this to be a Lustre object.
1545 * If there is no xattr skip it. */
1546 rc = __osd_xattr_load(dev, it->mit_pos, &nvbuf);
1547 if (unlikely(rc != 0))
1550 LASSERT(nvbuf != NULL);
1551 rc = -nvlist_lookup_byte_array(nvbuf, XATTR_NAME_LMA, &v, &s);
1552 if (likely(rc == 0)) {
1554 lma = (struct lustre_mdt_attrs *)v;
1555 lustre_lma_swab(lma);
1556 it->mit_fid = lma->lma_self_fid;
1560 /* not a Lustre object, try next one */
1567 /* we aren't prefetching in the above loop because the number of
1568 * non-Lustre objects is very small and we will be repeating very
1569 * rare. in case we want to use this to iterate over non-Lustre
1570 * objects (i.e. when we convert regular ZFS in Lustre) it makes
1571 * sense to initiate prefetching in the loop */
1573 /* 0 - there are more items, +1 - the end */
1574 if (likely(rc == 0))
1575 osd_zfs_otable_prefetch(env, it);
1577 CDEBUG(D_OTHER, "advance: %llu -> %llu "DFID": %d\n", dnode,
1578 it->mit_pos, PFID(&it->mit_fid), rc);
1584 static struct dt_key *osd_zfs_otable_it_key(const struct lu_env *env,
1585 const struct dt_it *di)
1590 static int osd_zfs_otable_it_key_size(const struct lu_env *env,
1591 const struct dt_it *di)
1593 return sizeof(__u64);
1596 static int osd_zfs_otable_it_rec(const struct lu_env *env,
1597 const struct dt_it *di,
1598 struct dt_rec *rec, __u32 attr)
1600 struct osd_metadnode_it *it = (struct osd_metadnode_it *)di;
1601 struct lu_fid *fid = (struct lu_fid *)rec;
1610 static __u64 osd_zfs_otable_it_store(const struct lu_env *env,
1611 const struct dt_it *di)
1613 struct osd_metadnode_it *it = (struct osd_metadnode_it *)di;
1618 static int osd_zfs_otable_it_load(const struct lu_env *env,
1619 const struct dt_it *di, __u64 hash)
1621 struct osd_metadnode_it *it = (struct osd_metadnode_it *)di;
1624 it->mit_prefetched = 0;
1625 it->mit_prefetched_dnode = 0;
1627 return osd_zfs_otable_it_next(env, (struct dt_it *)di);
1630 static int osd_zfs_otable_it_key_rec(const struct lu_env *env,
1631 const struct dt_it *di, void *key_rec)
1636 const struct dt_index_operations osd_zfs_otable_ops = {
1638 .init = osd_zfs_otable_it_init,
1639 .fini = osd_zfs_otable_it_fini,
1640 .get = osd_zfs_otable_it_get,
1641 .put = osd_zfs_otable_it_put,
1642 .next = osd_zfs_otable_it_next,
1643 .key = osd_zfs_otable_it_key,
1644 .key_size = osd_zfs_otable_it_key_size,
1645 .rec = osd_zfs_otable_it_rec,
1646 .store = osd_zfs_otable_it_store,
1647 .load = osd_zfs_otable_it_load,
1648 .key_rec = osd_zfs_otable_it_key_rec,
1652 int osd_index_try(const struct lu_env *env, struct dt_object *dt,
1653 const struct dt_index_features *feat)
1655 struct osd_object *obj = osd_dt_obj(dt);
1659 down_read(&obj->oo_guard);
1662 * XXX: implement support for fixed-size keys sorted with natural
1663 * numerical way (not using internal hash value)
1665 if (feat->dif_flags & DT_IND_RANGE)
1666 GOTO(out, rc = -ERANGE);
1668 if (unlikely(feat == &dt_otable_features)) {
1669 dt->do_index_ops = &osd_zfs_otable_ops;
1673 LASSERT(!dt_object_exists(dt) || obj->oo_db != NULL);
1674 if (likely(feat == &dt_directory_features)) {
1675 if (!dt_object_exists(dt) || osd_object_is_zap(obj->oo_db))
1676 dt->do_index_ops = &osd_dir_ops;
1678 GOTO(out, rc = -ENOTDIR);
1679 } else if (unlikely(feat == &dt_acct_features)) {
1680 LASSERT(fid_is_acct(lu_object_fid(&dt->do_lu)));
1681 dt->do_index_ops = &osd_acct_index_ops;
1682 } else if (dt->do_index_ops == NULL) {
1683 /* For index file, we don't support variable key & record sizes
1684 * and the key has to be unique */
1685 if ((feat->dif_flags & ~DT_IND_UPDATE) != 0)
1686 GOTO(out, rc = -EINVAL);
1688 if (feat->dif_keysize_max > ZAP_MAXNAMELEN)
1689 GOTO(out, rc = -E2BIG);
1690 if (feat->dif_keysize_max != feat->dif_keysize_min)
1691 GOTO(out, rc = -EINVAL);
1693 /* As for the record size, it should be a multiple of 8 bytes
1694 * and smaller than the maximum value length supported by ZAP.
1696 if (feat->dif_recsize_max > ZAP_MAXVALUELEN)
1697 GOTO(out, rc = -E2BIG);
1698 if (feat->dif_recsize_max != feat->dif_recsize_min)
1699 GOTO(out, rc = -EINVAL);
1701 obj->oo_keysize = feat->dif_keysize_max;
1702 obj->oo_recsize = feat->dif_recsize_max;
1703 obj->oo_recusize = 1;
1705 /* ZFS prefers to work with array of 64bits */
1706 if ((obj->oo_recsize & 7) == 0) {
1707 obj->oo_recsize >>= 3;
1708 obj->oo_recusize = 8;
1710 dt->do_index_ops = &osd_index_ops;
1714 up_read(&obj->oo_guard);