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 <libcfs/libcfs.h>
41 #include <obd_support.h>
42 #include <lustre_net.h>
44 #include <obd_class.h>
45 #include <lustre_disk.h>
46 #include <lustre_fid.h>
48 #include "osd_internal.h"
50 #include <sys/dnode.h>
54 #include <sys/spa_impl.h>
55 #include <sys/zfs_znode.h>
56 #include <sys/dmu_tx.h>
57 #include <sys/dmu_objset.h>
58 #include <sys/dsl_prop.h>
59 #include <sys/sa_impl.h>
62 static inline int osd_object_is_zap(dnode_t *dn)
64 return (dn->dn_type == DMU_OT_DIRECTORY_CONTENTS ||
65 dn->dn_type == DMU_OT_USERGROUP_USED);
68 /* We don't actually have direct access to the zap_hashbits() function
69 * so just pretend like we do for now. If this ever breaks we can look at
71 #define zap_hashbits(zc) 48
74 * | cd (16 bits) | hash (48 bits) |
75 * we need it in other form:
76 * |0| hash (48 bit) | cd (15 bit) |
77 * to be a full 64-bit ordered hash so that Lustre readdir can use it to merge
78 * the readdir hashes from multiple directory stripes uniformly on the client.
79 * Another point is sign bit, the hash range should be in [0, 2^63-1] because
80 * loff_t (for llseek) needs to be a positive value. This means the "cd" field
81 * should only be the low 15 bits.
83 uint64_t osd_zap_cursor_serialize(zap_cursor_t *zc)
85 uint64_t zfs_hash = zap_cursor_serialize(zc) & (~0ULL >> 1);
87 return (zfs_hash >> zap_hashbits(zc)) |
88 (zfs_hash << (63 - zap_hashbits(zc)));
91 void osd_zap_cursor_init_serialized(zap_cursor_t *zc, struct objset *os,
92 uint64_t id, uint64_t dirhash)
94 uint64_t zfs_hash = ((dirhash << zap_hashbits(zc)) & (~0ULL >> 1)) |
95 (dirhash >> (63 - zap_hashbits(zc)));
97 zap_cursor_init_serialized(zc, os, id, zfs_hash);
100 int osd_zap_cursor_init(zap_cursor_t **zc, struct objset *os,
101 uint64_t id, uint64_t dirhash)
106 if (unlikely(t == NULL))
109 osd_zap_cursor_init_serialized(t, os, id, dirhash);
115 void osd_zap_cursor_fini(zap_cursor_t *zc)
121 static inline void osd_obj_cursor_init_serialized(zap_cursor_t *zc,
122 struct osd_object *o,
125 struct osd_device *d = osd_obj2dev(o);
126 osd_zap_cursor_init_serialized(zc, d->od_os,
127 o->oo_dn->dn_object, dirhash);
130 static inline int osd_obj_cursor_init(zap_cursor_t **zc, struct osd_object *o,
133 struct osd_device *d = osd_obj2dev(o);
134 return osd_zap_cursor_init(zc, d->od_os, o->oo_dn->dn_object, dirhash);
137 static struct dt_it *osd_index_it_init(const struct lu_env *env,
138 struct dt_object *dt,
141 struct osd_thread_info *info = osd_oti_get(env);
142 struct osd_zap_it *it;
143 struct osd_object *obj = osd_dt_obj(dt);
144 struct lu_object *lo = &dt->do_lu;
148 if (obj->oo_destroyed)
149 RETURN(ERR_PTR(-ENOENT));
151 LASSERT(lu_object_exists(lo));
155 OBD_SLAB_ALLOC_PTR_GFP(it, osd_zapit_cachep, GFP_NOFS);
157 RETURN(ERR_PTR(-ENOMEM));
159 rc = osd_obj_cursor_init(&it->ozi_zc, obj, 0);
161 OBD_SLAB_FREE_PTR(it, osd_zapit_cachep);
169 RETURN((struct dt_it *)it);
172 static void osd_index_it_fini(const struct lu_env *env, struct dt_it *di)
174 struct osd_zap_it *it = (struct osd_zap_it *)di;
175 struct osd_object *obj;
179 LASSERT(it->ozi_obj);
183 osd_zap_cursor_fini(it->ozi_zc);
184 osd_object_put(env, obj);
185 OBD_SLAB_FREE_PTR(it, osd_zapit_cachep);
191 static void osd_index_it_put(const struct lu_env *env, struct dt_it *di)
193 /* PBS: do nothing : ref are incremented at retrive and decreamented
197 static inline void osd_it_append_attrs(struct lu_dirent *ent, __u32 attr,
200 const unsigned align = sizeof(struct luda_type) - 1;
201 struct luda_type *lt;
203 /* check if file type is required */
204 if (attr & LUDA_TYPE) {
205 len = (len + align) & ~align;
207 lt = (void *)ent->lde_name + len;
208 lt->lt_type = cpu_to_le16(DTTOIF(type));
209 ent->lde_attrs |= LUDA_TYPE;
212 ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
215 int __osd_xattr_load_by_oid(struct osd_device *osd, uint64_t oid, nvlist_t **sa)
221 rc = -dmu_bonus_hold(osd->od_os, oid, osd_obj_tag, &db);
223 CERROR("%s: can't get bonus, rc = %d\n", osd->od_svname, rc);
227 rc = -sa_handle_get_from_db(osd->od_os, db, NULL, SA_HDL_PRIVATE, &hdl);
229 dmu_buf_rele(db, osd_obj_tag);
233 rc = __osd_xattr_load(osd, hdl, sa);
235 sa_handle_destroy(hdl);
240 * Get the object's FID from its LMA EA.
242 * \param[in] env pointer to the thread context
243 * \param[in] osd pointer to the OSD device
244 * \param[in] oid the object's local identifier
245 * \param[out] fid the buffer to hold the object's FID
247 * \retval 0 for success
248 * \retval negative error number on failure
250 static int osd_get_fid_by_oid(const struct lu_env *env, struct osd_device *osd,
251 uint64_t oid, struct lu_fid *fid)
253 struct objset *os = osd->od_os;
254 struct osd_thread_info *oti = osd_oti_get(env);
255 struct lustre_mdt_attrs *lma =
256 (struct lustre_mdt_attrs *)oti->oti_buf;
258 nvlist_t *sa_xattr = NULL;
259 sa_handle_t *sa_hdl = NULL;
260 uchar_t *nv_value = NULL;
261 uint64_t xattr = ZFS_NO_OBJECT;
266 rc = __osd_xattr_load_by_oid(osd, oid, &sa_xattr);
273 rc = -nvlist_lookup_byte_array(sa_xattr, XATTR_NAME_LMA, &nv_value,
281 if (unlikely(size > sizeof(oti->oti_buf)))
282 GOTO(out, rc = -ERANGE);
284 memcpy(lma, nv_value, size);
289 rc = -sa_handle_get(os, oid, NULL, SA_HDL_PRIVATE, &sa_hdl);
293 rc = -sa_lookup(sa_hdl, SA_ZPL_XATTR(osd), &xattr, 8);
294 sa_handle_destroy(sa_hdl);
299 buf.lb_len = sizeof(oti->oti_buf);
300 rc = __osd_xattr_get_large(env, osd, xattr, &buf,
301 XATTR_NAME_LMA, &size);
306 if (size < sizeof(*lma))
307 GOTO(out, rc = -EIO);
309 lustre_lma_swab(lma);
310 if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
311 CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT))) {
312 CWARN("%s: unsupported incompat LMA feature(s) %#x for "
313 "oid = %#llx\n", osd->od_svname,
314 lma->lma_incompat & ~LMA_INCOMPAT_SUPP, oid);
315 GOTO(out, rc = -EOPNOTSUPP);
317 *fid = lma->lma_self_fid;
322 if (sa_xattr != NULL)
323 nvlist_free(sa_xattr);
328 * As we don't know FID, we can't use LU object, so this function
329 * partially duplicate __osd_xattr_get() which is built around
330 * LU-object and uses it to cache data like regular EA dnode, etc
332 static int osd_find_parent_by_dnode(const struct lu_env *env,
336 struct osd_object *obj = osd_dt_obj(o);
337 struct osd_device *osd = osd_obj2dev(obj);
338 uint64_t dnode = ZFS_NO_OBJECT;
342 /* first of all, get parent dnode from own attributes */
343 rc = osd_sa_handle_get(obj);
346 rc = -sa_lookup(obj->oo_sa_hdl, SA_ZPL_PARENT(osd), &dnode, 8);
348 rc = osd_get_fid_by_oid(env, osd, dnode, fid);
353 static int osd_find_parent_fid(const struct lu_env *env, struct dt_object *o,
356 struct link_ea_header *leh;
357 struct link_ea_entry *lee;
362 buf.lb_buf = osd_oti_get(env)->oti_buf;
363 buf.lb_len = sizeof(osd_oti_get(env)->oti_buf);
365 rc = osd_xattr_get(env, o, &buf, XATTR_NAME_LINK);
367 rc = osd_xattr_get(env, o, &LU_BUF_NULL, XATTR_NAME_LINK);
371 OBD_ALLOC(buf.lb_buf, rc);
372 if (buf.lb_buf == NULL)
375 rc = osd_xattr_get(env, o, &buf, XATTR_NAME_LINK);
379 if (rc < sizeof(*leh) + sizeof(*lee))
380 GOTO(out, rc = -EINVAL);
383 if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) {
384 leh->leh_magic = LINK_EA_MAGIC;
385 leh->leh_reccount = __swab32(leh->leh_reccount);
386 leh->leh_len = __swab64(leh->leh_len);
388 if (leh->leh_magic != LINK_EA_MAGIC)
389 GOTO(out, rc = -EINVAL);
390 if (leh->leh_reccount == 0)
391 GOTO(out, rc = -ENODATA);
393 lee = (struct link_ea_entry *)(leh + 1);
394 fid_be_to_cpu(fid, (const struct lu_fid *)&lee->lee_parent_fid);
398 if (buf.lb_buf != osd_oti_get(env)->oti_buf)
399 OBD_FREE(buf.lb_buf, buf.lb_len);
402 /* this block can be enabled for additional verification
403 * it's trying to match FID from LinkEA vs. FID from LMA */
407 rc2 = osd_find_parent_by_dnode(env, o, &fid2);
409 if (lu_fid_eq(fid, &fid2) == 0)
410 CERROR("wrong parent: "DFID" != "DFID"\n",
411 PFID(fid), PFID(&fid2));
415 /* no LinkEA is found, let's try to find the fid in parent's LMA */
416 if (unlikely(rc != 0))
417 rc = osd_find_parent_by_dnode(env, o, fid);
422 static int osd_dir_lookup(const struct lu_env *env, struct dt_object *dt,
423 struct dt_rec *rec, const struct dt_key *key)
425 struct osd_thread_info *oti = osd_oti_get(env);
426 struct osd_object *obj = osd_dt_obj(dt);
427 struct osd_device *osd = osd_obj2dev(obj);
428 char *name = (char *)key;
432 if (name[0] == '.') {
434 const struct lu_fid *f = lu_object_fid(&dt->do_lu);
435 memcpy(rec, f, sizeof(*f));
437 } else if (name[1] == '.' && name[2] == 0) {
438 rc = osd_find_parent_fid(env, dt, (struct lu_fid *)rec);
439 RETURN(rc == 0 ? 1 : rc);
443 memset(&oti->oti_zde.lzd_fid, 0, sizeof(struct lu_fid));
444 rc = osd_zap_lookup(osd, obj->oo_dn->dn_object, obj->oo_dn,
445 (char *)key, 8, sizeof(oti->oti_zde) / 8,
446 (void *)&oti->oti_zde);
450 if (likely(fid_is_sane(&oti->oti_zde.lzd_fid))) {
451 memcpy(rec, &oti->oti_zde.lzd_fid, sizeof(struct lu_fid));
455 rc = osd_get_fid_by_oid(env, osd, oti->oti_zde.lzd_reg.zde_dnode,
456 (struct lu_fid *)rec);
458 RETURN(rc == 0 ? 1 : (rc == -ENOENT ? -ENODATA : rc));
461 static int osd_declare_dir_insert(const struct lu_env *env,
462 struct dt_object *dt,
463 const struct dt_rec *rec,
464 const struct dt_key *key,
467 struct osd_object *obj = osd_dt_obj(dt);
468 struct osd_device *osd = osd_obj2dev(obj);
469 const struct dt_insert_rec *rec1;
470 const struct lu_fid *fid;
471 struct osd_thandle *oh;
475 rec1 = (struct dt_insert_rec *)rec;
477 LASSERT(fid != NULL);
478 LASSERT(rec1->rec_type != 0);
481 oh = container_of0(th, struct osd_thandle, ot_super);
483 /* This is for inserting dot/dotdot for new created dir. */
484 if (obj->oo_dn == NULL)
485 object = DMU_NEW_OBJECT;
487 object = obj->oo_dn->dn_object;
489 /* do not specify the key as then DMU is trying to look it up
490 * which is very expensive. usually the layers above lookup
491 * before insertion */
492 osd_tx_hold_zap(oh->ot_tx, object, obj->oo_dn, TRUE, NULL);
494 osd_idc_find_or_init(env, osd, fid);
499 static int osd_seq_exists(const struct lu_env *env, struct osd_device *osd,
502 struct lu_seq_range *range = &osd_oti_get(env)->oti_seq_range;
503 struct seq_server_site *ss = osd_seq_site(osd);
508 LASSERT(ss->ss_server_fld != NULL);
510 rc = osd_fld_lookup(env, osd, seq, range);
513 CERROR("%s: Can not lookup fld for %#llx\n",
518 RETURN(ss->ss_node_id == range->lsr_index);
521 int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
522 const struct lu_fid *fid)
524 struct seq_server_site *ss = osd_seq_site(osd);
527 /* FID seqs not in FLDB, must be local seq */
528 if (unlikely(!fid_seq_in_fldb(fid_seq(fid))))
531 /* If FLD is not being initialized yet, it only happens during the
532 * initialization, likely during mgs initialization, and we assume
533 * this is local FID. */
534 if (ss == NULL || ss->ss_server_fld == NULL)
537 /* Only check the local FLDB here */
538 if (osd_seq_exists(env, osd, fid_seq(fid)))
545 * Inserts (key, value) pair in \a directory object.
547 * \param dt osd index object
548 * \param key key for index
549 * \param rec record reference
550 * \param th transaction handler
551 * \param ignore_quota update should not affect quota
554 * \retval -ve failure
556 static int osd_dir_insert(const struct lu_env *env, struct dt_object *dt,
557 const struct dt_rec *rec, const struct dt_key *key,
558 struct thandle *th, int ignore_quota)
560 struct osd_thread_info *oti = osd_oti_get(env);
561 struct osd_object *parent = osd_dt_obj(dt);
562 struct osd_device *osd = osd_obj2dev(parent);
563 struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec;
564 const struct lu_fid *fid = rec1->rec_fid;
565 struct osd_thandle *oh;
566 struct osd_idmap_cache *idc;
567 char *name = (char *)key;
569 int num = sizeof(oti->oti_zde) / 8;
572 LASSERT(parent->oo_dn);
574 LASSERT(dt_object_exists(dt));
575 LASSERT(osd_invariant(parent));
578 oh = container_of0(th, struct osd_thandle, ot_super);
580 idc = osd_idc_find(env, osd, fid);
581 if (unlikely(idc == NULL)) {
582 /* this dt_insert() wasn't declared properly, so
583 * FID is missing in OI cache. we better do not
584 * lookup FID in FLDB/OI and don't risk to deadlock,
585 * but in some special cases (lfsck testing, etc)
586 * it's much simpler than fixing a caller */
587 CERROR("%s: "DFID" wasn't declared for insert\n",
588 osd_name(osd), PFID(fid));
589 idc = osd_idc_find_or_init(env, osd, fid);
591 RETURN(PTR_ERR(idc));
594 if (idc->oic_remote) {
595 /* Insert remote entry */
596 memset(&oti->oti_zde.lzd_reg, 0, sizeof(oti->oti_zde.lzd_reg));
597 oti->oti_zde.lzd_reg.zde_type = IFTODT(rec1->rec_type & S_IFMT);
599 if (unlikely(idc->oic_dnode == 0)) {
600 /* for a reason OI cache wasn't filled properly */
601 CERROR("%s: OIC for "DFID" isn't filled\n",
602 osd_name(osd), PFID(fid));
605 if (name[0] == '.') {
607 /* do not store ".", instead generate it
608 * during iteration */
610 } else if (name[1] == '.' && name[2] == 0) {
611 uint64_t dnode = idc->oic_dnode;
612 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT))
615 /* update parent dnode in the child.
616 * later it will be used to generate ".." */
617 rc = osd_object_sa_update(parent,
624 CLASSERT(sizeof(oti->oti_zde.lzd_reg) == 8);
625 CLASSERT(sizeof(oti->oti_zde) % 8 == 0);
626 oti->oti_zde.lzd_reg.zde_type = IFTODT(rec1->rec_type & S_IFMT);
627 oti->oti_zde.lzd_reg.zde_dnode = idc->oic_dnode;
630 oti->oti_zde.lzd_fid = *fid;
631 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INDIR))
632 oti->oti_zde.lzd_fid.f_ver = ~0;
633 if (OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF))
635 /* Insert (key,oid) into ZAP */
636 rc = osd_zap_add(osd, parent->oo_dn->dn_object, parent->oo_dn,
637 (char *)key, 8, num, (void *)&oti->oti_zde, oh->ot_tx);
638 if (unlikely(rc == -EEXIST &&
639 name[0] == '.' && name[1] == '.' && name[2] == 0))
640 /* Update (key,oid) in ZAP */
641 rc = -zap_update(osd->od_os, parent->oo_dn->dn_object,
642 (char *)key, 8, sizeof(oti->oti_zde) / 8,
643 (void *)&oti->oti_zde, oh->ot_tx);
650 static int osd_declare_dir_delete(const struct lu_env *env,
651 struct dt_object *dt,
652 const struct dt_key *key,
655 struct osd_object *obj = osd_dt_obj(dt);
656 struct osd_thandle *oh;
660 LASSERT(dt_object_exists(dt));
661 LASSERT(osd_invariant(obj));
664 oh = container_of0(th, struct osd_thandle, ot_super);
666 if (dt_object_exists(dt)) {
668 dnode = obj->oo_dn->dn_object;
670 dnode = DMU_NEW_OBJECT;
673 /* do not specify the key as then DMU is trying to look it up
674 * which is very expensive. usually the layers above lookup
676 osd_tx_hold_zap(oh->ot_tx, dnode, obj->oo_dn, FALSE, NULL);
681 static int osd_dir_delete(const struct lu_env *env, struct dt_object *dt,
682 const struct dt_key *key, struct thandle *th)
684 struct osd_object *obj = osd_dt_obj(dt);
685 struct osd_device *osd = osd_obj2dev(obj);
686 struct osd_thandle *oh;
687 dnode_t *zap_dn = obj->oo_dn;
688 char *name = (char *)key;
695 oh = container_of0(th, struct osd_thandle, ot_super);
698 * In Orion . and .. were stored in the directory (not generated upon
699 * request as now). we preserve them for backward compatibility
701 if (name[0] == '.') {
704 } else if (name[1] == '.' && name[2] == 0) {
709 /* Remove key from the ZAP */
710 rc = osd_zap_remove(osd, zap_dn->dn_object, zap_dn,
711 (char *)key, oh->ot_tx);
713 if (unlikely(rc && rc != -ENOENT))
714 CERROR("%s: zap_remove failed: rc = %d\n", osd->od_svname, rc);
719 static struct dt_it *osd_dir_it_init(const struct lu_env *env,
720 struct dt_object *dt,
723 struct osd_zap_it *it;
725 it = (struct osd_zap_it *)osd_index_it_init(env, dt, unused);
729 RETURN((struct dt_it *)it);
733 * Move Iterator to record specified by \a key
735 * \param di osd iterator
736 * \param key key for index
738 * \retval +ve di points to record with least key not larger than key
739 * \retval 0 di points to exact matched key
740 * \retval -ve failure
742 static int osd_dir_it_get(const struct lu_env *env,
743 struct dt_it *di, const struct dt_key *key)
745 struct osd_zap_it *it = (struct osd_zap_it *)di;
746 struct osd_object *obj = it->ozi_obj;
747 char *name = (char *)key;
754 /* reset the cursor */
755 zap_cursor_fini(it->ozi_zc);
756 osd_obj_cursor_init_serialized(it->ozi_zc, obj, 0);
758 /* XXX: implementation of the API is broken at the moment */
759 LASSERT(((const char *)key)[0] == 0);
766 if (name[0] == '.') {
770 } else if (name[1] == '.' && name[2] == 0) {
776 /* neither . nor .. - some real record */
784 static void osd_dir_it_put(const struct lu_env *env, struct dt_it *di)
786 /* PBS: do nothing : ref are incremented at retrive and decreamented
791 * in Orion . and .. were stored in the directory, while ZPL
792 * and current osd-zfs generate them up on request. so, we
793 * need to ignore previously stored . and ..
795 static int osd_index_retrieve_skip_dots(struct osd_zap_it *it,
801 rc = -zap_cursor_retrieve(it->ozi_zc, za);
804 if (unlikely(rc == 0 && za->za_name[0] == '.')) {
805 if (za->za_name[1] == 0) {
807 } else if (za->za_name[1] == '.' &&
808 za->za_name[2] == 0) {
812 zap_cursor_advance(it->ozi_zc);
814 } while (unlikely(rc == 0 && isdot));
820 * to load a directory entry at a time and stored it in
821 * iterator's in-memory data structure.
823 * \param di, struct osd_it_ea, iterator's in memory structure
825 * \retval +ve, iterator reached to end
826 * \retval 0, iterator not reached to end
827 * \retval -ve, on error
829 static int osd_dir_it_next(const struct lu_env *env, struct dt_it *di)
831 struct osd_zap_it *it = (struct osd_zap_it *)di;
832 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
837 /* temp. storage should be enough for any key supported by ZFS */
838 CLASSERT(sizeof(za->za_name) <= sizeof(it->ozi_name));
841 * the first ->next() moves the cursor to .
842 * the second ->next() moves the cursor to ..
843 * then we get to the real records and have to verify any exist
845 if (it->ozi_pos <= 2) {
851 zap_cursor_advance(it->ozi_zc);
855 * According to current API we need to return error if its last entry.
856 * zap_cursor_advance() does not return any value. So we need to call
857 * retrieve to check if there is any record. We should make
858 * changes to Iterator API to not return status for this API
860 rc = osd_index_retrieve_skip_dots(it, za);
862 if (rc == -ENOENT) /* end of dir */
868 static struct dt_key *osd_dir_it_key(const struct lu_env *env,
869 const struct dt_it *di)
871 struct osd_zap_it *it = (struct osd_zap_it *)di;
872 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
876 if (it->ozi_pos <= 1) {
878 RETURN((struct dt_key *)".");
879 } else if (it->ozi_pos == 2) {
880 RETURN((struct dt_key *)"..");
883 if ((rc = -zap_cursor_retrieve(it->ozi_zc, za)))
886 strcpy(it->ozi_name, za->za_name);
888 RETURN((struct dt_key *)it->ozi_name);
891 static int osd_dir_it_key_size(const struct lu_env *env, const struct dt_it *di)
893 struct osd_zap_it *it = (struct osd_zap_it *)di;
894 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
898 if (it->ozi_pos <= 1) {
901 } else if (it->ozi_pos == 2) {
905 if ((rc = -zap_cursor_retrieve(it->ozi_zc, za)) == 0)
906 rc = strlen(za->za_name);
912 osd_dirent_update(const struct lu_env *env, struct osd_device *dev,
913 uint64_t zap, const char *key, struct luz_direntry *zde)
919 tx = dmu_tx_create(dev->od_os);
923 dmu_tx_hold_zap(tx, zap, TRUE, NULL);
924 rc = -dmu_tx_assign(tx, TXG_WAIT);
926 rc = -zap_update(dev->od_os, zap, key, 8, sizeof(*zde) / 8,
927 (const void *)zde, tx);
936 static int osd_dir_it_rec(const struct lu_env *env, const struct dt_it *di,
937 struct dt_rec *dtrec, __u32 attr)
939 struct osd_zap_it *it = (struct osd_zap_it *)di;
940 struct lu_dirent *lde = (struct lu_dirent *)dtrec;
941 struct osd_thread_info *info = osd_oti_get(env);
942 struct luz_direntry *zde = &info->oti_zde;
943 zap_attribute_t *za = &info->oti_za;
944 struct lu_fid *fid = &info->oti_fid;
945 struct osd_device *osd = osd_obj2dev(it->ozi_obj);
950 if (it->ozi_pos <= 1) {
951 lde->lde_hash = cpu_to_le64(1);
952 strcpy(lde->lde_name, ".");
953 lde->lde_namelen = cpu_to_le16(1);
954 fid_cpu_to_le(&lde->lde_fid,
955 lu_object_fid(&it->ozi_obj->oo_dt.do_lu));
956 lde->lde_attrs = LUDA_FID;
957 /* append lustre attributes */
958 osd_it_append_attrs(lde, attr, 1, IFTODT(S_IFDIR));
959 lde->lde_reclen = cpu_to_le16(lu_dirent_calc_size(1, attr));
962 } else if (it->ozi_pos == 2) {
963 lde->lde_hash = cpu_to_le64(2);
964 strcpy(lde->lde_name, "..");
965 lde->lde_namelen = cpu_to_le16(2);
966 rc = osd_find_parent_fid(env, &it->ozi_obj->oo_dt, fid);
968 fid_cpu_to_le(&lde->lde_fid, fid);
969 lde->lde_attrs = LUDA_FID;
970 } else if (rc != -ENOENT) {
971 /* ENOENT happens at the root of filesystem, ignore */
975 /* append lustre attributes */
976 osd_it_append_attrs(lde, attr, 2, IFTODT(S_IFDIR));
977 lde->lde_reclen = cpu_to_le16(lu_dirent_calc_size(2, attr));
983 rc = -zap_cursor_retrieve(it->ozi_zc, za);
987 lde->lde_hash = cpu_to_le64(osd_zap_cursor_serialize(it->ozi_zc));
988 namelen = strlen(za->za_name);
989 if (namelen > NAME_MAX)
991 strcpy(lde->lde_name, za->za_name);
992 lde->lde_namelen = cpu_to_le16(namelen);
994 if (za->za_integer_length != 8) {
995 CERROR("%s: unsupported direntry format: %d %d\n",
997 za->za_integer_length, (int)za->za_num_integers);
1001 rc = osd_zap_lookup(osd, it->ozi_zc->zc_zapobj, it->ozi_obj->oo_dn,
1002 za->za_name, za->za_integer_length, 3, zde);
1006 if (za->za_num_integers >= 3 && fid_is_sane(&zde->lzd_fid)) {
1007 lde->lde_attrs = LUDA_FID;
1008 fid_cpu_to_le(&lde->lde_fid, &zde->lzd_fid);
1009 GOTO(pack_attr, rc = 0);
1012 if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP))
1015 rc = osd_get_fid_by_oid(env, osd, zde->lzd_reg.zde_dnode, fid);
1017 lde->lde_attrs = LUDA_UNKNOWN;
1018 GOTO(pack_attr, rc = 0);
1021 if (!(attr & LUDA_VERIFY)) {
1022 fid_cpu_to_le(&lde->lde_fid, fid);
1023 lde->lde_attrs = LUDA_FID;
1024 GOTO(pack_attr, rc = 0);
1027 if (attr & LUDA_VERIFY_DRYRUN) {
1028 fid_cpu_to_le(&lde->lde_fid, fid);
1029 lde->lde_attrs = LUDA_FID | LUDA_REPAIR;
1030 GOTO(pack_attr, rc = 0);
1033 fid_cpu_to_le(&lde->lde_fid, fid);
1034 lde->lde_attrs = LUDA_FID;
1035 zde->lzd_fid = *fid;
1036 rc = osd_dirent_update(env, osd, it->ozi_zc->zc_zapobj,
1039 lde->lde_attrs |= LUDA_UNKNOWN;
1040 GOTO(pack_attr, rc = 0);
1043 lde->lde_attrs |= LUDA_REPAIR;
1045 GOTO(pack_attr, rc = 0);
1048 osd_it_append_attrs(lde, attr, namelen, zde->lzd_reg.zde_type);
1049 lde->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
1053 static int osd_dir_it_rec_size(const struct lu_env *env, const struct dt_it *di,
1056 struct osd_zap_it *it = (struct osd_zap_it *)di;
1057 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
1062 if (it->ozi_pos <= 1)
1064 else if (it->ozi_pos == 2)
1068 rc = lu_dirent_calc_size(namelen, attr);
1072 rc = -zap_cursor_retrieve(it->ozi_zc, za);
1073 if (unlikely(rc != 0))
1076 if (za->za_integer_length != 8 || za->za_num_integers < 3) {
1077 CERROR("%s: unsupported direntry format: %d %d\n",
1078 osd_obj2dev(it->ozi_obj)->od_svname,
1079 za->za_integer_length, (int)za->za_num_integers);
1083 namelen = strlen(za->za_name);
1084 if (namelen > NAME_MAX)
1087 rc = lu_dirent_calc_size(namelen, attr);
1092 static __u64 osd_dir_it_store(const struct lu_env *env, const struct dt_it *di)
1094 struct osd_zap_it *it = (struct osd_zap_it *)di;
1098 if (it->ozi_pos <= 2)
1101 pos = osd_zap_cursor_serialize(it->ozi_zc);
1108 * rc == 0 -> end of directory.
1109 * rc > 0 -> ok, proceed.
1110 * rc < 0 -> error. ( EOVERFLOW can be masked.)
1112 static int osd_dir_it_load(const struct lu_env *env,
1113 const struct dt_it *di, __u64 hash)
1115 struct osd_zap_it *it = (struct osd_zap_it *)di;
1116 struct osd_object *obj = it->ozi_obj;
1117 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
1121 /* reset the cursor */
1122 zap_cursor_fini(it->ozi_zc);
1123 osd_obj_cursor_init_serialized(it->ozi_zc, obj, hash);
1130 /* to return whether the end has been reached */
1131 rc = osd_index_retrieve_skip_dots(it, za);
1134 else if (rc == -ENOENT)
1141 struct dt_index_operations osd_dir_ops = {
1142 .dio_lookup = osd_dir_lookup,
1143 .dio_declare_insert = osd_declare_dir_insert,
1144 .dio_insert = osd_dir_insert,
1145 .dio_declare_delete = osd_declare_dir_delete,
1146 .dio_delete = osd_dir_delete,
1148 .init = osd_dir_it_init,
1149 .fini = osd_index_it_fini,
1150 .get = osd_dir_it_get,
1151 .put = osd_dir_it_put,
1152 .next = osd_dir_it_next,
1153 .key = osd_dir_it_key,
1154 .key_size = osd_dir_it_key_size,
1155 .rec = osd_dir_it_rec,
1156 .rec_size = osd_dir_it_rec_size,
1157 .store = osd_dir_it_store,
1158 .load = osd_dir_it_load
1163 * Primitives for index files using binary keys.
1166 /* key integer_size is 8 */
1167 static int osd_prepare_key_uint64(struct osd_object *o, __u64 *dst,
1168 const struct dt_key *src)
1175 /* align keysize to 64bit */
1176 size = (o->oo_keysize + sizeof(__u64) - 1) / sizeof(__u64);
1177 size *= sizeof(__u64);
1179 LASSERT(size <= MAXNAMELEN);
1181 if (unlikely(size > o->oo_keysize))
1182 memset(dst + o->oo_keysize, 0, size - o->oo_keysize);
1183 memcpy(dst, (const char *)src, o->oo_keysize);
1185 return (size/sizeof(__u64));
1188 static int osd_index_lookup(const struct lu_env *env, struct dt_object *dt,
1189 struct dt_rec *rec, const struct dt_key *key)
1191 struct osd_object *obj = osd_dt_obj(dt);
1192 struct osd_device *osd = osd_obj2dev(obj);
1193 __u64 *k = osd_oti_get(env)->oti_key64;
1197 rc = osd_prepare_key_uint64(obj, k, key);
1199 rc = -zap_lookup_uint64(osd->od_os, obj->oo_dn->dn_object,
1200 k, rc, obj->oo_recusize, obj->oo_recsize,
1202 RETURN(rc == 0 ? 1 : rc);
1205 static int osd_declare_index_insert(const struct lu_env *env,
1206 struct dt_object *dt,
1207 const struct dt_rec *rec,
1208 const struct dt_key *key,
1211 struct osd_object *obj = osd_dt_obj(dt);
1212 struct osd_thandle *oh;
1215 LASSERT(th != NULL);
1216 oh = container_of0(th, struct osd_thandle, ot_super);
1218 LASSERT(obj->oo_dn);
1220 /* do not specify the key as then DMU is trying to look it up
1221 * which is very expensive. usually the layers above lookup
1222 * before insertion */
1223 osd_tx_hold_zap(oh->ot_tx, obj->oo_dn->dn_object, obj->oo_dn,
1229 static int osd_index_insert(const struct lu_env *env, struct dt_object *dt,
1230 const struct dt_rec *rec, const struct dt_key *key,
1231 struct thandle *th, int ignore_quota)
1233 struct osd_object *obj = osd_dt_obj(dt);
1234 struct osd_device *osd = osd_obj2dev(obj);
1235 struct osd_thandle *oh;
1236 __u64 *k = osd_oti_get(env)->oti_key64;
1240 LASSERT(obj->oo_dn);
1241 LASSERT(dt_object_exists(dt));
1242 LASSERT(osd_invariant(obj));
1243 LASSERT(th != NULL);
1245 oh = container_of0(th, struct osd_thandle, ot_super);
1247 rc = osd_prepare_key_uint64(obj, k, key);
1249 /* Insert (key,oid) into ZAP */
1250 rc = -zap_add_uint64(osd->od_os, obj->oo_dn->dn_object,
1251 k, rc, obj->oo_recusize, obj->oo_recsize,
1252 (void *)rec, oh->ot_tx);
1256 static int osd_declare_index_delete(const struct lu_env *env,
1257 struct dt_object *dt,
1258 const struct dt_key *key,
1261 struct osd_object *obj = osd_dt_obj(dt);
1262 struct osd_thandle *oh;
1265 LASSERT(dt_object_exists(dt));
1266 LASSERT(osd_invariant(obj));
1267 LASSERT(th != NULL);
1268 LASSERT(obj->oo_dn);
1270 oh = container_of0(th, struct osd_thandle, ot_super);
1272 /* do not specify the key as then DMU is trying to look it up
1273 * which is very expensive. usually the layers above lookup
1274 * before deletion */
1275 osd_tx_hold_zap(oh->ot_tx, obj->oo_dn->dn_object, obj->oo_dn,
1281 static int osd_index_delete(const struct lu_env *env, struct dt_object *dt,
1282 const struct dt_key *key, struct thandle *th)
1284 struct osd_object *obj = osd_dt_obj(dt);
1285 struct osd_device *osd = osd_obj2dev(obj);
1286 struct osd_thandle *oh;
1287 __u64 *k = osd_oti_get(env)->oti_key64;
1291 LASSERT(obj->oo_dn);
1292 LASSERT(th != NULL);
1293 oh = container_of0(th, struct osd_thandle, ot_super);
1295 rc = osd_prepare_key_uint64(obj, k, key);
1297 /* Remove binary key from the ZAP */
1298 rc = -zap_remove_uint64(osd->od_os, obj->oo_dn->dn_object,
1303 static int osd_index_it_get(const struct lu_env *env, struct dt_it *di,
1304 const struct dt_key *key)
1306 struct osd_zap_it *it = (struct osd_zap_it *)di;
1307 struct osd_object *obj = it->ozi_obj;
1308 struct osd_device *osd = osd_obj2dev(obj);
1312 LASSERT(it->ozi_zc);
1315 * XXX: we need a binary version of zap_cursor_move_to_key()
1316 * to implement this API */
1317 if (*((const __u64 *)key) != 0)
1318 CERROR("NOT IMPLEMETED YET (move to %#llx)\n",
1321 zap_cursor_fini(it->ozi_zc);
1322 zap_cursor_init(it->ozi_zc, osd->od_os, obj->oo_dn->dn_object);
1328 static int osd_index_it_next(const struct lu_env *env, struct dt_it *di)
1330 struct osd_zap_it *it = (struct osd_zap_it *)di;
1331 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
1335 if (it->ozi_reset == 0)
1336 zap_cursor_advance(it->ozi_zc);
1340 * According to current API we need to return error if it's last entry.
1341 * zap_cursor_advance() does not return any value. So we need to call
1342 * retrieve to check if there is any record. We should make
1343 * changes to Iterator API to not return status for this API
1345 rc = -zap_cursor_retrieve(it->ozi_zc, za);
1352 static struct dt_key *osd_index_it_key(const struct lu_env *env,
1353 const struct dt_it *di)
1355 struct osd_zap_it *it = (struct osd_zap_it *)di;
1356 struct osd_object *obj = it->ozi_obj;
1357 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
1362 rc = -zap_cursor_retrieve(it->ozi_zc, za);
1364 RETURN(ERR_PTR(rc));
1366 /* the binary key is stored in the name */
1367 memcpy(&it->ozi_key, za->za_name, obj->oo_keysize);
1369 RETURN((struct dt_key *)&it->ozi_key);
1372 static int osd_index_it_key_size(const struct lu_env *env,
1373 const struct dt_it *di)
1375 struct osd_zap_it *it = (struct osd_zap_it *)di;
1376 struct osd_object *obj = it->ozi_obj;
1377 RETURN(obj->oo_keysize);
1380 static int osd_index_it_rec(const struct lu_env *env, const struct dt_it *di,
1381 struct dt_rec *rec, __u32 attr)
1383 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
1384 struct osd_zap_it *it = (struct osd_zap_it *)di;
1385 struct osd_object *obj = it->ozi_obj;
1386 struct osd_device *osd = osd_obj2dev(obj);
1387 __u64 *k = osd_oti_get(env)->oti_key64;
1392 rc = -zap_cursor_retrieve(it->ozi_zc, za);
1396 rc = osd_prepare_key_uint64(obj, k, (const struct dt_key *)za->za_name);
1398 rc = -zap_lookup_uint64(osd->od_os, obj->oo_dn->dn_object,
1399 k, rc, obj->oo_recusize, obj->oo_recsize,
1404 static __u64 osd_index_it_store(const struct lu_env *env,
1405 const struct dt_it *di)
1407 struct osd_zap_it *it = (struct osd_zap_it *)di;
1410 RETURN((__u64)zap_cursor_serialize(it->ozi_zc));
1413 static int osd_index_it_load(const struct lu_env *env, const struct dt_it *di,
1416 struct osd_zap_it *it = (struct osd_zap_it *)di;
1417 struct osd_object *obj = it->ozi_obj;
1418 struct osd_device *osd = osd_obj2dev(obj);
1419 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
1423 /* reset the cursor */
1424 zap_cursor_fini(it->ozi_zc);
1425 zap_cursor_init_serialized(it->ozi_zc, osd->od_os,
1426 obj->oo_dn->dn_object, hash);
1429 rc = -zap_cursor_retrieve(it->ozi_zc, za);
1432 else if (rc == -ENOENT)
1438 static struct dt_index_operations osd_index_ops = {
1439 .dio_lookup = osd_index_lookup,
1440 .dio_declare_insert = osd_declare_index_insert,
1441 .dio_insert = osd_index_insert,
1442 .dio_declare_delete = osd_declare_index_delete,
1443 .dio_delete = osd_index_delete,
1445 .init = osd_index_it_init,
1446 .fini = osd_index_it_fini,
1447 .get = osd_index_it_get,
1448 .put = osd_index_it_put,
1449 .next = osd_index_it_next,
1450 .key = osd_index_it_key,
1451 .key_size = osd_index_it_key_size,
1452 .rec = osd_index_it_rec,
1453 .store = osd_index_it_store,
1454 .load = osd_index_it_load
1458 struct osd_metadnode_it {
1459 struct osd_device *mit_dev;
1461 struct lu_fid mit_fid;
1463 __u64 mit_prefetched_dnode;
1466 static struct dt_it *osd_zfs_otable_it_init(const struct lu_env *env,
1467 struct dt_object *dt, __u32 attr)
1469 struct osd_device *dev = osd_dev(dt->do_lu.lo_dev);
1470 struct osd_metadnode_it *it;
1474 if (unlikely(it == NULL))
1475 RETURN(ERR_PTR(-ENOMEM));
1479 /* XXX: dmu_object_next() does NOT find dnodes allocated
1480 * in the current non-committed txg, so we force txg
1481 * commit to find all existing dnodes ... */
1482 if (!dev->od_dt_dev.dd_rdonly)
1483 txg_wait_synced(dmu_objset_pool(dev->od_os), 0ULL);
1485 RETURN((struct dt_it *)it);
1488 static void osd_zfs_otable_it_fini(const struct lu_env *env, struct dt_it *di)
1490 struct osd_metadnode_it *it = (struct osd_metadnode_it *)di;
1495 static int osd_zfs_otable_it_get(const struct lu_env *env,
1496 struct dt_it *di, const struct dt_key *key)
1501 static void osd_zfs_otable_it_put(const struct lu_env *env, struct dt_it *di)
1505 #define OTABLE_PREFETCH 256
1507 static void osd_zfs_otable_prefetch(const struct lu_env *env,
1508 struct osd_metadnode_it *it)
1510 struct osd_device *dev = it->mit_dev;
1513 /* can go negative on the very first access to the iterator
1514 * or if some non-Lustre objects were found */
1515 if (unlikely(it->mit_prefetched < 0))
1516 it->mit_prefetched = 0;
1518 if (it->mit_prefetched >= (OTABLE_PREFETCH >> 1))
1521 if (it->mit_prefetched_dnode == 0)
1522 it->mit_prefetched_dnode = it->mit_pos;
1524 while (it->mit_prefetched < OTABLE_PREFETCH) {
1525 rc = -dmu_object_next(dev->od_os, &it->mit_prefetched_dnode,
1527 if (unlikely(rc != 0))
1530 osd_dmu_prefetch(dev->od_os, it->mit_prefetched_dnode,
1531 0, 0, 0, ZIO_PRIORITY_ASYNC_READ);
1533 it->mit_prefetched++;
1537 static int osd_zfs_otable_it_next(const struct lu_env *env, struct dt_it *di)
1539 struct osd_metadnode_it *it = (struct osd_metadnode_it *)di;
1540 struct lustre_mdt_attrs *lma;
1541 struct osd_device *dev = it->mit_dev;
1542 nvlist_t *nvbuf = NULL;
1547 memset(&it->mit_fid, 0, sizeof(it->mit_fid));
1549 dnode = it->mit_pos;
1551 rc = -dmu_object_next(dev->od_os, &it->mit_pos, B_FALSE, 0);
1552 if (unlikely(rc != 0))
1554 it->mit_prefetched--;
1556 /* LMA is required for this to be a Lustre object.
1557 * If there is no xattr skip it. */
1558 rc = __osd_xattr_load_by_oid(dev, it->mit_pos, &nvbuf);
1559 if (unlikely(rc != 0))
1562 LASSERT(nvbuf != NULL);
1563 rc = -nvlist_lookup_byte_array(nvbuf, XATTR_NAME_LMA, &v, &s);
1564 if (likely(rc == 0)) {
1566 lma = (struct lustre_mdt_attrs *)v;
1567 lustre_lma_swab(lma);
1568 it->mit_fid = lma->lma_self_fid;
1572 /* not a Lustre object, try next one */
1579 /* we aren't prefetching in the above loop because the number of
1580 * non-Lustre objects is very small and we will be repeating very
1581 * rare. in case we want to use this to iterate over non-Lustre
1582 * objects (i.e. when we convert regular ZFS in Lustre) it makes
1583 * sense to initiate prefetching in the loop */
1585 /* 0 - there are more items, +1 - the end */
1586 if (likely(rc == 0))
1587 osd_zfs_otable_prefetch(env, it);
1589 CDEBUG(D_OTHER, "advance: %llu -> %llu "DFID": %d\n", dnode,
1590 it->mit_pos, PFID(&it->mit_fid), rc);
1596 static struct dt_key *osd_zfs_otable_it_key(const struct lu_env *env,
1597 const struct dt_it *di)
1602 static int osd_zfs_otable_it_key_size(const struct lu_env *env,
1603 const struct dt_it *di)
1605 return sizeof(__u64);
1608 static int osd_zfs_otable_it_rec(const struct lu_env *env,
1609 const struct dt_it *di,
1610 struct dt_rec *rec, __u32 attr)
1612 struct osd_metadnode_it *it = (struct osd_metadnode_it *)di;
1613 struct lu_fid *fid = (struct lu_fid *)rec;
1622 static __u64 osd_zfs_otable_it_store(const struct lu_env *env,
1623 const struct dt_it *di)
1625 struct osd_metadnode_it *it = (struct osd_metadnode_it *)di;
1630 static int osd_zfs_otable_it_load(const struct lu_env *env,
1631 const struct dt_it *di, __u64 hash)
1633 struct osd_metadnode_it *it = (struct osd_metadnode_it *)di;
1636 it->mit_prefetched = 0;
1637 it->mit_prefetched_dnode = 0;
1639 return osd_zfs_otable_it_next(env, (struct dt_it *)di);
1642 static int osd_zfs_otable_it_key_rec(const struct lu_env *env,
1643 const struct dt_it *di, void *key_rec)
1648 const struct dt_index_operations osd_zfs_otable_ops = {
1650 .init = osd_zfs_otable_it_init,
1651 .fini = osd_zfs_otable_it_fini,
1652 .get = osd_zfs_otable_it_get,
1653 .put = osd_zfs_otable_it_put,
1654 .next = osd_zfs_otable_it_next,
1655 .key = osd_zfs_otable_it_key,
1656 .key_size = osd_zfs_otable_it_key_size,
1657 .rec = osd_zfs_otable_it_rec,
1658 .store = osd_zfs_otable_it_store,
1659 .load = osd_zfs_otable_it_load,
1660 .key_rec = osd_zfs_otable_it_key_rec,
1664 int osd_index_try(const struct lu_env *env, struct dt_object *dt,
1665 const struct dt_index_features *feat)
1667 struct osd_object *obj = osd_dt_obj(dt);
1671 down_read(&obj->oo_guard);
1674 * XXX: implement support for fixed-size keys sorted with natural
1675 * numerical way (not using internal hash value)
1677 if (feat->dif_flags & DT_IND_RANGE)
1678 GOTO(out, rc = -ERANGE);
1680 if (unlikely(feat == &dt_otable_features)) {
1681 dt->do_index_ops = &osd_zfs_otable_ops;
1685 LASSERT(!dt_object_exists(dt) || obj->oo_dn != NULL);
1686 if (likely(feat == &dt_directory_features)) {
1687 if (!dt_object_exists(dt) || osd_object_is_zap(obj->oo_dn))
1688 dt->do_index_ops = &osd_dir_ops;
1690 GOTO(out, rc = -ENOTDIR);
1691 } else if (unlikely(feat == &dt_acct_features)) {
1692 LASSERT(fid_is_acct(lu_object_fid(&dt->do_lu)));
1693 dt->do_index_ops = &osd_acct_index_ops;
1694 } else if (dt->do_index_ops == NULL) {
1695 /* For index file, we don't support variable key & record sizes
1696 * and the key has to be unique */
1697 if ((feat->dif_flags & ~DT_IND_UPDATE) != 0)
1698 GOTO(out, rc = -EINVAL);
1700 if (feat->dif_keysize_max > ZAP_MAXNAMELEN)
1701 GOTO(out, rc = -E2BIG);
1702 if (feat->dif_keysize_max != feat->dif_keysize_min)
1703 GOTO(out, rc = -EINVAL);
1705 /* As for the record size, it should be a multiple of 8 bytes
1706 * and smaller than the maximum value length supported by ZAP.
1708 if (feat->dif_recsize_max > ZAP_MAXVALUELEN)
1709 GOTO(out, rc = -E2BIG);
1710 if (feat->dif_recsize_max != feat->dif_recsize_min)
1711 GOTO(out, rc = -EINVAL);
1713 obj->oo_keysize = feat->dif_keysize_max;
1714 obj->oo_recsize = feat->dif_recsize_max;
1715 obj->oo_recusize = 1;
1717 /* ZFS prefers to work with array of 64bits */
1718 if ((obj->oo_recsize & 7) == 0) {
1719 obj->oo_recsize >>= 3;
1720 obj->oo_recusize = 8;
1722 dt->do_index_ops = &osd_index_ops;
1726 up_read(&obj->oo_guard);