4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2009, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
31 * Copyright (c) 2012, 2013, Intel Corporation.
32 * Use is subject to license terms.
35 * This file is part of Lustre, http://www.lustre.org/
36 * Lustre is a trademark of Sun Microsystems, Inc.
38 * lustre/osd-zfs/osd_index.c
40 * Author: Alex Zhuravlev <bzzz@whamcloud.com>
41 * Author: Mike Pershin <tappro@whamcloud.com>
44 #define DEBUG_SUBSYSTEM S_OSD
46 #include <lustre_ver.h>
47 #include <libcfs/libcfs.h>
48 #include <obd_support.h>
49 #include <lustre_net.h>
51 #include <obd_class.h>
52 #include <lustre_disk.h>
53 #include <lustre_fid.h>
55 #include "osd_internal.h"
57 #include <sys/dnode.h>
62 #include <sys/spa_impl.h>
63 #include <sys/zfs_znode.h>
64 #include <sys/dmu_tx.h>
65 #include <sys/dmu_objset.h>
66 #include <sys/dsl_prop.h>
67 #include <sys/sa_impl.h>
70 static struct dt_it *osd_index_it_init(const struct lu_env *env,
73 struct lustre_capa *capa)
75 struct osd_thread_info *info = osd_oti_get(env);
76 struct osd_zap_it *it;
77 struct osd_object *obj = osd_dt_obj(dt);
78 struct osd_device *osd = osd_obj2dev(obj);
79 struct lu_object *lo = &dt->do_lu;
82 /* XXX: check capa ? */
84 LASSERT(lu_object_exists(lo));
86 LASSERT(udmu_object_is_zap(obj->oo_db));
89 it = &info->oti_it_zap;
91 if (udmu_zap_cursor_init(&it->ozi_zc, &osd->od_objset,
92 obj->oo_db->db_object, 0))
93 RETURN(ERR_PTR(-ENOMEM));
100 RETURN((struct dt_it *)it);
103 static void osd_index_it_fini(const struct lu_env *env, struct dt_it *di)
105 struct osd_zap_it *it = (struct osd_zap_it *)di;
106 struct osd_object *obj;
110 LASSERT(it->ozi_obj);
114 udmu_zap_cursor_fini(it->ozi_zc);
115 lu_object_put(env, &obj->oo_dt.do_lu);
121 static void osd_index_it_put(const struct lu_env *env, struct dt_it *di)
123 /* PBS: do nothing : ref are incremented at retrive and decreamented
127 int udmu_zap_cursor_retrieve_key(const struct lu_env *env,
128 zap_cursor_t *zc, char *key, int max)
130 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
133 if ((err = zap_cursor_retrieve(zc, za)))
137 strcpy(key, za->za_name);
143 * zap_cursor_retrieve read from current record.
144 * to read bytes we need to call zap_lookup explicitly.
146 int udmu_zap_cursor_retrieve_value(const struct lu_env *env,
147 zap_cursor_t *zc, char *buf,
148 int buf_size, int *bytes_read)
150 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
151 int err, actual_size;
153 if ((err = zap_cursor_retrieve(zc, za)))
156 if (za->za_integer_length <= 0)
159 actual_size = za->za_integer_length * za->za_num_integers;
161 if (actual_size > buf_size) {
162 actual_size = buf_size;
163 buf_size = actual_size / za->za_integer_length;
165 buf_size = za->za_num_integers;
168 err = -zap_lookup(zc->zc_objset, zc->zc_zapobj,
169 za->za_name, za->za_integer_length,
173 *bytes_read = actual_size;
178 static inline void osd_it_append_attrs(struct lu_dirent *ent, __u32 attr,
181 const unsigned align = sizeof(struct luda_type) - 1;
182 struct luda_type *lt;
184 /* check if file type is required */
185 if (attr & LUDA_TYPE) {
186 len = (len + align) & ~align;
188 lt = (void *)ent->lde_name + len;
189 lt->lt_type = cpu_to_le16(DTTOIF(type));
190 ent->lde_attrs |= LUDA_TYPE;
193 ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
197 * as we don't know FID, we can't use LU object, so this function
198 * partially duplicate __osd_xattr_get() which is built around
199 * LU-object and uses it to cache data like regular EA dnode, etc
201 static int osd_find_parent_by_dnode(const struct lu_env *env,
205 struct lustre_mdt_attrs *lma;
206 udmu_objset_t *uos = &osd_obj2dev(osd_dt_obj(o))->od_objset;
209 nvlist_t *nvbuf = NULL;
215 /* first of all, get parent dnode from own attributes */
216 LASSERT(osd_dt_obj(o)->oo_db);
217 rc = -sa_handle_get(uos->os, osd_dt_obj(o)->oo_db->db_object,
218 NULL, SA_HDL_PRIVATE, &sa_hdl);
222 dnode = ZFS_NO_OBJECT;
223 rc = -sa_lookup(sa_hdl, SA_ZPL_PARENT(uos), &dnode, 8);
224 sa_handle_destroy(sa_hdl);
228 /* now get EA buffer */
229 rc = __osd_xattr_load(uos, dnode, &nvbuf);
233 /* XXX: if we get that far.. should we cache the result? */
235 /* try to find LMA attribute */
236 LASSERT(nvbuf != NULL);
237 rc = -nvlist_lookup_byte_array(nvbuf, XATTR_NAME_LMA, &value, &size);
238 if (rc == 0 && size >= sizeof(*lma)) {
239 lma = (struct lustre_mdt_attrs *)value;
240 lustre_lma_swab(lma);
241 *fid = lma->lma_self_fid;
246 /* no LMA attribute in SA, let's try regular EA */
248 /* first of all, get parent dnode storing regular EA */
249 rc = -sa_handle_get(uos->os, dnode, NULL, SA_HDL_PRIVATE, &sa_hdl);
253 dnode = ZFS_NO_OBJECT;
254 rc = -sa_lookup(sa_hdl, SA_ZPL_XATTR(uos), &dnode, 8);
255 sa_handle_destroy(sa_hdl);
259 CLASSERT(sizeof(*lma) <= sizeof(osd_oti_get(env)->oti_buf));
260 buf.lb_buf = osd_oti_get(env)->oti_buf;
261 buf.lb_len = sizeof(osd_oti_get(env)->oti_buf);
263 /* now try to find LMA */
264 rc = __osd_xattr_get_large(env, uos, dnode, &buf,
265 XATTR_NAME_LMA, &size);
266 if (rc == 0 && size >= sizeof(*lma)) {
268 lustre_lma_swab(lma);
269 *fid = lma->lma_self_fid;
274 GOTO(out, rc = -EIO);
283 static int osd_find_parent_fid(const struct lu_env *env, struct dt_object *o,
286 struct link_ea_header *leh;
287 struct link_ea_entry *lee;
292 buf.lb_buf = osd_oti_get(env)->oti_buf;
293 buf.lb_len = sizeof(osd_oti_get(env)->oti_buf);
295 rc = osd_xattr_get(env, o, &buf, XATTR_NAME_LINK, BYPASS_CAPA);
297 rc = osd_xattr_get(env, o, &LU_BUF_NULL,
298 XATTR_NAME_LINK, BYPASS_CAPA);
302 OBD_ALLOC(buf.lb_buf, rc);
303 if (buf.lb_buf == NULL)
306 rc = osd_xattr_get(env, o, &buf, XATTR_NAME_LINK, BYPASS_CAPA);
310 if (rc < sizeof(*leh) + sizeof(*lee))
311 GOTO(out, rc = -EINVAL);
314 if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) {
315 leh->leh_magic = LINK_EA_MAGIC;
316 leh->leh_reccount = __swab32(leh->leh_reccount);
317 leh->leh_len = __swab64(leh->leh_len);
319 if (leh->leh_magic != LINK_EA_MAGIC)
320 GOTO(out, rc = -EINVAL);
321 if (leh->leh_reccount == 0)
322 GOTO(out, rc = -ENODATA);
324 lee = (struct link_ea_entry *)(leh + 1);
325 fid_be_to_cpu(fid, (const struct lu_fid *)&lee->lee_parent_fid);
329 if (buf.lb_buf != osd_oti_get(env)->oti_buf)
330 OBD_FREE(buf.lb_buf, buf.lb_len);
333 /* this block can be enabled for additional verification
334 * it's trying to match FID from LinkEA vs. FID from LMA */
338 rc2 = osd_find_parent_by_dnode(env, o, &fid2);
340 if (lu_fid_eq(fid, &fid2) == 0)
341 CERROR("wrong parent: "DFID" != "DFID"\n",
342 PFID(fid), PFID(&fid2));
346 /* no LinkEA is found, let's try to find the fid in parent's LMA */
347 if (unlikely(rc != 0))
348 rc = osd_find_parent_by_dnode(env, o, fid);
353 static int osd_dir_lookup(const struct lu_env *env, struct dt_object *dt,
354 struct dt_rec *rec, const struct dt_key *key,
355 struct lustre_capa *capa)
357 struct osd_thread_info *oti = osd_oti_get(env);
358 struct osd_object *obj = osd_dt_obj(dt);
359 struct osd_device *osd = osd_obj2dev(obj);
360 char *name = (char *)key;
364 LASSERT(udmu_object_is_zap(obj->oo_db));
366 if (name[0] == '.') {
368 const struct lu_fid *f = lu_object_fid(&dt->do_lu);
369 memcpy(rec, f, sizeof(*f));
371 } else if (name[1] == '.' && name[2] == 0) {
372 rc = osd_find_parent_fid(env, dt, (struct lu_fid *)rec);
373 RETURN(rc == 0 ? 1 : rc);
377 rc = -zap_lookup(osd->od_objset.os, obj->oo_db->db_object,
378 (char *)key, 8, sizeof(oti->oti_zde) / 8,
379 (void *)&oti->oti_zde);
380 memcpy(rec, &oti->oti_zde.lzd_fid, sizeof(struct lu_fid));
382 RETURN(rc == 0 ? 1 : rc);
385 static int osd_declare_dir_insert(const struct lu_env *env,
386 struct dt_object *dt,
387 const struct dt_rec *rec,
388 const struct dt_key *key,
391 struct osd_object *obj = osd_dt_obj(dt);
392 struct osd_thandle *oh;
396 oh = container_of0(th, struct osd_thandle, ot_super);
399 LASSERT(udmu_object_is_zap(obj->oo_db));
401 dmu_tx_hold_bonus(oh->ot_tx, obj->oo_db->db_object);
402 dmu_tx_hold_zap(oh->ot_tx, obj->oo_db->db_object, TRUE, (char *)key);
408 * Find the osd object for given fid.
410 * \param fid need to find the osd object having this fid
412 * \retval osd_object on success
413 * \retval -ve on error
415 struct osd_object *osd_object_find(const struct lu_env *env,
416 struct dt_object *dt,
417 const struct lu_fid *fid)
419 struct lu_device *ludev = dt->do_lu.lo_dev;
420 struct osd_object *child = NULL;
421 struct lu_object *luch;
422 struct lu_object *lo;
425 * at this point topdev might not exist yet
426 * (i.e. MGS is preparing profiles). so we can
427 * not rely on topdev and instead lookup with
428 * our device passed as topdev. this can't work
429 * if the object isn't cached yet (as osd doesn't
430 * allocate lu_header). IOW, the object must be
431 * in the cache, otherwise lu_object_alloc() crashes
434 luch = lu_object_find_at(env, ludev, fid, NULL);
438 if (lu_object_exists(luch)) {
439 lo = lu_object_locate(luch->lo_header, ludev->ld_type);
443 LU_OBJECT_DEBUG(D_ERROR, env, luch,
444 "%s: object can't be located "DFID"\n",
445 osd_dev(ludev)->od_svname, PFID(fid));
448 lu_object_put(env, luch);
449 CERROR("%s: Unable to get osd_object "DFID"\n",
450 osd_dev(ludev)->od_svname, PFID(fid));
451 child = ERR_PTR(-ENOENT);
454 LU_OBJECT_DEBUG(D_ERROR, env, luch,
455 "%s: lu_object does not exists "DFID"\n",
456 osd_dev(ludev)->od_svname, PFID(fid));
457 lu_object_put(env, luch);
458 child = ERR_PTR(-ENOENT);
465 * Put the osd object once done with it.
467 * \param obj osd object that needs to be put
469 static inline void osd_object_put(const struct lu_env *env,
470 struct osd_object *obj)
472 lu_object_put(env, &obj->oo_dt.do_lu);
475 static int osd_seq_exists(const struct lu_env *env, struct osd_device *osd,
478 struct lu_seq_range *range = &osd_oti_get(env)->oti_seq_range;
479 struct seq_server_site *ss = osd_seq_site(osd);
486 rc = osd_fld_lookup(env, osd, seq, range);
488 CERROR("%s: Can not lookup fld for "LPX64"\n",
493 RETURN(ss->ss_node_id == range->lsr_index);
496 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
501 if (!fid_is_norm(fid) && !fid_is_root(fid))
504 if (osd_seq_exists(env, osd, fid_seq(fid)))
511 * Inserts (key, value) pair in \a directory object.
513 * \param dt osd index object
514 * \param key key for index
515 * \param rec record reference
516 * \param th transaction handler
517 * \param capa capability descriptor
518 * \param ignore_quota update should not affect quota
521 * \retval -ve failure
523 static int osd_dir_insert(const struct lu_env *env, struct dt_object *dt,
524 const struct dt_rec *rec, const struct dt_key *key,
525 struct thandle *th, struct lustre_capa *capa,
528 struct osd_thread_info *oti = osd_oti_get(env);
529 struct osd_object *parent = osd_dt_obj(dt);
530 struct osd_device *osd = osd_obj2dev(parent);
531 struct lu_fid *fid = (struct lu_fid *)rec;
532 struct osd_thandle *oh;
533 struct osd_object *child = NULL;
535 char *name = (char *)key;
539 LASSERT(parent->oo_db);
540 LASSERT(udmu_object_is_zap(parent->oo_db));
542 LASSERT(dt_object_exists(dt));
543 LASSERT(osd_invariant(parent));
546 oh = container_of0(th, struct osd_thandle, ot_super);
548 rc = osd_remote_fid(env, osd, fid);
550 CERROR("%s: Can not find object "DFID": rc = %d\n",
551 osd->od_svname, PFID(fid), rc);
555 if (unlikely(rc == 1)) {
556 /* Insert remote entry */
557 memset(&oti->oti_zde.lzd_reg, 0, sizeof(oti->oti_zde.lzd_reg));
558 oti->oti_zde.lzd_reg.zde_type = IFTODT(S_IFDIR & S_IFMT);
561 * To simulate old Orion setups with ./.. stored in the
564 /* Insert local entry */
565 child = osd_object_find(env, dt, fid);
567 RETURN(PTR_ERR(child));
569 LASSERT(child->oo_db);
570 if (name[0] == '.') {
572 /* do not store ".", instead generate it
573 * during iteration */
575 } else if (name[1] == '.' && name[2] == 0) {
576 /* update parent dnode in the child.
577 * later it will be used to generate ".." */
578 udmu_objset_t *uos = &osd->od_objset;
579 rc = osd_object_sa_update(parent,
581 &child->oo_db->db_object,
586 CLASSERT(sizeof(oti->oti_zde.lzd_reg) == 8);
587 CLASSERT(sizeof(oti->oti_zde) % 8 == 0);
588 attr = child->oo_dt.do_lu.lo_header ->loh_attr;
589 oti->oti_zde.lzd_reg.zde_type = IFTODT(attr & S_IFMT);
590 oti->oti_zde.lzd_reg.zde_dnode = child->oo_db->db_object;
593 oti->oti_zde.lzd_fid = *fid;
594 /* Insert (key,oid) into ZAP */
595 rc = -zap_add(osd->od_objset.os, parent->oo_db->db_object,
596 (char *)key, 8, sizeof(oti->oti_zde) / 8,
597 (void *)&oti->oti_zde, oh->ot_tx);
601 osd_object_put(env, child);
606 static int osd_declare_dir_delete(const struct lu_env *env,
607 struct dt_object *dt,
608 const struct dt_key *key,
611 struct osd_object *obj = osd_dt_obj(dt);
612 struct osd_thandle *oh;
615 LASSERT(dt_object_exists(dt));
616 LASSERT(osd_invariant(obj));
619 oh = container_of0(th, struct osd_thandle, ot_super);
622 LASSERT(udmu_object_is_zap(obj->oo_db));
624 dmu_tx_hold_zap(oh->ot_tx, obj->oo_db->db_object, TRUE, (char *)key);
629 static int osd_dir_delete(const struct lu_env *env, struct dt_object *dt,
630 const struct dt_key *key, struct thandle *th,
631 struct lustre_capa *capa)
633 struct osd_object *obj = osd_dt_obj(dt);
634 struct osd_device *osd = osd_obj2dev(obj);
635 struct osd_thandle *oh;
636 dmu_buf_t *zap_db = obj->oo_db;
637 char *name = (char *)key;
642 LASSERT(udmu_object_is_zap(obj->oo_db));
645 oh = container_of0(th, struct osd_thandle, ot_super);
648 * In Orion . and .. were stored in the directory (not generated upon
649 * request as now). we preserve them for backward compatibility
651 if (name[0] == '.') {
654 } else if (name[1] == '.' && name[2] == 0) {
659 /* Remove key from the ZAP */
660 rc = -zap_remove(osd->od_objset.os, zap_db->db_object,
661 (char *) key, oh->ot_tx);
663 #if LUSTRE_VERSION_CODE <= OBD_OCD_VERSION(2, 4, 53, 0)
664 if (unlikely(rc == -ENOENT && name[0] == '.' &&
665 (name[1] == 0 || (name[1] == '.' && name[2] == 0))))
668 if (unlikely(rc && rc != -ENOENT))
669 CERROR("%s: zap_remove failed: rc = %d\n", osd->od_svname, rc);
674 static struct dt_it *osd_dir_it_init(const struct lu_env *env,
675 struct dt_object *dt,
677 struct lustre_capa *capa)
679 struct osd_zap_it *it;
681 it = (struct osd_zap_it *)osd_index_it_init(env, dt, unused, capa);
685 RETURN((struct dt_it *)it);
689 * Move Iterator to record specified by \a key
691 * \param di osd iterator
692 * \param key key for index
694 * \retval +ve di points to record with least key not larger than key
695 * \retval 0 di points to exact matched key
696 * \retval -ve failure
698 static int osd_dir_it_get(const struct lu_env *env,
699 struct dt_it *di, const struct dt_key *key)
701 struct osd_zap_it *it = (struct osd_zap_it *)di;
702 struct osd_object *obj = it->ozi_obj;
703 struct osd_device *osd = osd_obj2dev(obj);
704 char *name = (char *)key;
711 udmu_zap_cursor_fini(it->ozi_zc);
713 if (udmu_zap_cursor_init(&it->ozi_zc, &osd->od_objset,
714 obj->oo_db->db_object, 0))
717 /* XXX: implementation of the API is broken at the moment */
718 LASSERT(((const char *)key)[0] == 0);
725 if (name[0] == '.') {
729 } else if (name[1] == '.' && name[2] == 0) {
735 /* neither . nor .. - some real record */
743 static void osd_dir_it_put(const struct lu_env *env, struct dt_it *di)
745 /* PBS: do nothing : ref are incremented at retrive and decreamented
750 * in Orion . and .. were stored in the directory, while ZPL
751 * and current osd-zfs generate them up on request. so, we
752 * need to ignore previously stored . and ..
754 static int osd_index_retrieve_skip_dots(struct osd_zap_it *it,
760 rc = -zap_cursor_retrieve(it->ozi_zc, za);
763 if (unlikely(rc == 0 && za->za_name[0] == '.')) {
764 if (za->za_name[1] == 0) {
766 } else if (za->za_name[1] == '.' &&
767 za->za_name[2] == 0) {
771 zap_cursor_advance(it->ozi_zc);
773 } while (unlikely(rc == 0 && isdot));
779 * to load a directory entry at a time and stored it in
780 * iterator's in-memory data structure.
782 * \param di, struct osd_it_ea, iterator's in memory structure
784 * \retval +ve, iterator reached to end
785 * \retval 0, iterator not reached to end
786 * \retval -ve, on error
788 static int osd_dir_it_next(const struct lu_env *env, struct dt_it *di)
790 struct osd_zap_it *it = (struct osd_zap_it *)di;
791 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
794 /* temp. storage should be enough for any key supported by ZFS */
795 CLASSERT(sizeof(za->za_name) <= sizeof(it->ozi_name));
798 * the first ->next() moves the cursor to .
799 * the second ->next() moves the cursor to ..
800 * then we get to the real records and have to verify any exist
802 if (it->ozi_pos <= 2) {
808 zap_cursor_advance(it->ozi_zc);
811 * According to current API we need to return error if its last entry.
812 * zap_cursor_advance() does not return any value. So we need to call
813 * retrieve to check if there is any record. We should make
814 * changes to Iterator API to not return status for this API
816 rc = osd_index_retrieve_skip_dots(it, za);
818 if (rc == -ENOENT) /* end of dir */
824 static struct dt_key *osd_dir_it_key(const struct lu_env *env,
825 const struct dt_it *di)
827 struct osd_zap_it *it = (struct osd_zap_it *)di;
828 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
832 if (it->ozi_pos <= 1) {
834 RETURN((struct dt_key *)".");
835 } else if (it->ozi_pos == 2) {
836 RETURN((struct dt_key *)"..");
839 if ((rc = -zap_cursor_retrieve(it->ozi_zc, za)))
842 strcpy(it->ozi_name, za->za_name);
844 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 3, 90, 0)
845 if (za->za_name[0] == '.') {
846 if (za->za_name[1] == 0 || (za->za_name[1] == '.' &&
847 za->za_name[2] == 0)) {
848 /* we should not get onto . and ..
849 * stored in the directory. ->next() and
850 * other methods should prevent this
857 RETURN((struct dt_key *)it->ozi_name);
860 static int osd_dir_it_key_size(const struct lu_env *env, const struct dt_it *di)
862 struct osd_zap_it *it = (struct osd_zap_it *)di;
863 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
867 if (it->ozi_pos <= 1) {
870 } else if (it->ozi_pos == 2) {
874 if ((rc = -zap_cursor_retrieve(it->ozi_zc, za)) == 0)
875 rc = strlen(za->za_name);
877 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 3, 90, 0)
878 if (rc == 0 && za->za_name[0] == '.') {
879 if (za->za_name[1] == 0 || (za->za_name[1] == '.' &&
880 za->za_name[2] == 0)) {
881 /* we should not get onto . and ..
882 * stored in the directory. ->next() and
883 * other methods should prevent this
892 static int osd_dir_it_rec(const struct lu_env *env, const struct dt_it *di,
893 struct dt_rec *dtrec, __u32 attr)
895 struct osd_zap_it *it = (struct osd_zap_it *)di;
896 struct lu_dirent *lde = (struct lu_dirent *)dtrec;
897 struct luz_direntry *zde = &osd_oti_get(env)->oti_zde;
898 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
902 if (it->ozi_pos <= 1) {
903 lde->lde_hash = cpu_to_le64(1);
904 strcpy(lde->lde_name, ".");
905 lde->lde_namelen = cpu_to_le16(1);
906 lde->lde_fid = *lu_object_fid(&it->ozi_obj->oo_dt.do_lu);
907 lde->lde_attrs = LUDA_FID;
908 /* append lustre attributes */
909 osd_it_append_attrs(lde, attr, 1, IFTODT(S_IFDIR));
910 lde->lde_reclen = cpu_to_le16(lu_dirent_calc_size(1, attr));
914 } else if (it->ozi_pos == 2) {
915 lde->lde_hash = cpu_to_le64(2);
916 strcpy(lde->lde_name, "..");
917 lde->lde_namelen = cpu_to_le16(2);
918 lde->lde_attrs = LUDA_FID;
919 /* append lustre attributes */
920 osd_it_append_attrs(lde, attr, 2, IFTODT(S_IFDIR));
921 lde->lde_reclen = cpu_to_le16(lu_dirent_calc_size(2, attr));
922 rc = osd_find_parent_fid(env, &it->ozi_obj->oo_dt, &lde->lde_fid);
924 * early Orion code was not setting LinkEA, so it's possible
925 * some setups still have objects with no LinkEA set.
926 * but at that time .. was a real record in the directory
927 * so we should try to lookup .. in ZAP
935 lde->lde_hash = cpu_to_le64(udmu_zap_cursor_serialize(it->ozi_zc));
937 if ((rc = -zap_cursor_retrieve(it->ozi_zc, za)))
940 namelen = strlen(za->za_name);
941 if (namelen > NAME_MAX)
942 GOTO(out, rc = -EOVERFLOW);
943 strcpy(lde->lde_name, za->za_name);
944 lde->lde_namelen = cpu_to_le16(namelen);
946 if (za->za_integer_length != 8 || za->za_num_integers < 3) {
947 CERROR("%s: unsupported direntry format: %d %d\n",
948 osd_obj2dev(it->ozi_obj)->od_svname,
949 za->za_integer_length, (int)za->za_num_integers);
951 GOTO(out, rc = -EIO);
954 rc = -zap_lookup(it->ozi_zc->zc_objset, it->ozi_zc->zc_zapobj,
955 za->za_name, za->za_integer_length, 3, zde);
959 lde->lde_fid = zde->lzd_fid;
960 lde->lde_attrs = LUDA_FID;
962 /* append lustre attributes */
963 osd_it_append_attrs(lde, attr, namelen, zde->lzd_reg.zde_type);
965 lde->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
971 static __u64 osd_dir_it_store(const struct lu_env *env, const struct dt_it *di)
973 struct osd_zap_it *it = (struct osd_zap_it *)di;
977 if (it->ozi_pos <= 2)
980 pos = udmu_zap_cursor_serialize(it->ozi_zc);
987 * rc == 0 -> end of directory.
988 * rc > 0 -> ok, proceed.
989 * rc < 0 -> error. ( EOVERFLOW can be masked.)
991 static int osd_dir_it_load(const struct lu_env *env,
992 const struct dt_it *di, __u64 hash)
994 struct osd_zap_it *it = (struct osd_zap_it *)di;
995 struct osd_object *obj = it->ozi_obj;
996 struct osd_device *osd = osd_obj2dev(obj);
997 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
1001 udmu_zap_cursor_fini(it->ozi_zc);
1002 if (udmu_zap_cursor_init(&it->ozi_zc, &osd->od_objset,
1003 obj->oo_db->db_object, hash))
1011 /* to return whether the end has been reached */
1012 rc = osd_index_retrieve_skip_dots(it, za);
1015 else if (rc == -ENOENT)
1022 static struct dt_index_operations osd_dir_ops = {
1023 .dio_lookup = osd_dir_lookup,
1024 .dio_declare_insert = osd_declare_dir_insert,
1025 .dio_insert = osd_dir_insert,
1026 .dio_declare_delete = osd_declare_dir_delete,
1027 .dio_delete = osd_dir_delete,
1029 .init = osd_dir_it_init,
1030 .fini = osd_index_it_fini,
1031 .get = osd_dir_it_get,
1032 .put = osd_dir_it_put,
1033 .next = osd_dir_it_next,
1034 .key = osd_dir_it_key,
1035 .key_size = osd_dir_it_key_size,
1036 .rec = osd_dir_it_rec,
1037 .store = osd_dir_it_store,
1038 .load = osd_dir_it_load
1043 * Primitives for index files using binary keys.
1046 /* key integer_size is 8 */
1047 static int osd_prepare_key_uint64(struct osd_object *o, __u64 *dst,
1048 const struct dt_key *src)
1055 /* align keysize to 64bit */
1056 size = (o->oo_keysize + sizeof(__u64) - 1) / sizeof(__u64);
1057 size *= sizeof(__u64);
1059 LASSERT(size <= MAXNAMELEN);
1061 if (unlikely(size > o->oo_keysize))
1062 memset(dst + o->oo_keysize, 0, size - o->oo_keysize);
1063 memcpy(dst, (const char *)src, o->oo_keysize);
1065 return (size/sizeof(__u64));
1068 static int osd_index_lookup(const struct lu_env *env, struct dt_object *dt,
1069 struct dt_rec *rec, const struct dt_key *key,
1070 struct lustre_capa *capa)
1072 struct osd_object *obj = osd_dt_obj(dt);
1073 struct osd_device *osd = osd_obj2dev(obj);
1074 __u64 *k = osd_oti_get(env)->oti_key64;
1078 rc = osd_prepare_key_uint64(obj, k, key);
1080 rc = -zap_lookup_uint64(osd->od_objset.os, obj->oo_db->db_object,
1081 k, rc, obj->oo_recusize, obj->oo_recsize,
1083 RETURN(rc == 0 ? 1 : rc);
1086 static int osd_declare_index_insert(const struct lu_env *env,
1087 struct dt_object *dt,
1088 const struct dt_rec *rec,
1089 const struct dt_key *key,
1092 struct osd_object *obj = osd_dt_obj(dt);
1093 struct osd_thandle *oh;
1096 LASSERT(th != NULL);
1097 oh = container_of0(th, struct osd_thandle, ot_super);
1099 LASSERT(obj->oo_db);
1101 dmu_tx_hold_bonus(oh->ot_tx, obj->oo_db->db_object);
1103 /* It is not clear what API should be used for binary keys, so we pass
1104 * a null name which has the side effect of over-reserving space,
1105 * accounting for the worst case. See zap_count_write() */
1106 dmu_tx_hold_zap(oh->ot_tx, obj->oo_db->db_object, TRUE, NULL);
1111 static int osd_index_insert(const struct lu_env *env, struct dt_object *dt,
1112 const struct dt_rec *rec, const struct dt_key *key,
1113 struct thandle *th, struct lustre_capa *capa,
1116 struct osd_object *obj = osd_dt_obj(dt);
1117 struct osd_device *osd = osd_obj2dev(obj);
1118 struct osd_thandle *oh;
1119 __u64 *k = osd_oti_get(env)->oti_key64;
1123 LASSERT(obj->oo_db);
1124 LASSERT(dt_object_exists(dt));
1125 LASSERT(osd_invariant(obj));
1126 LASSERT(th != NULL);
1128 oh = container_of0(th, struct osd_thandle, ot_super);
1130 rc = osd_prepare_key_uint64(obj, k, key);
1132 /* Insert (key,oid) into ZAP */
1133 rc = -zap_add_uint64(osd->od_objset.os, obj->oo_db->db_object,
1134 k, rc, obj->oo_recusize, obj->oo_recsize,
1135 (void *)rec, oh->ot_tx);
1139 static int osd_declare_index_delete(const struct lu_env *env,
1140 struct dt_object *dt,
1141 const struct dt_key *key,
1144 struct osd_object *obj = osd_dt_obj(dt);
1145 struct osd_thandle *oh;
1148 LASSERT(dt_object_exists(dt));
1149 LASSERT(osd_invariant(obj));
1150 LASSERT(th != NULL);
1151 LASSERT(obj->oo_db);
1153 oh = container_of0(th, struct osd_thandle, ot_super);
1154 dmu_tx_hold_zap(oh->ot_tx, obj->oo_db->db_object, TRUE, NULL);
1159 static int osd_index_delete(const struct lu_env *env, struct dt_object *dt,
1160 const struct dt_key *key, struct thandle *th,
1161 struct lustre_capa *capa)
1163 struct osd_object *obj = osd_dt_obj(dt);
1164 struct osd_device *osd = osd_obj2dev(obj);
1165 struct osd_thandle *oh;
1166 __u64 *k = osd_oti_get(env)->oti_key64;
1170 LASSERT(obj->oo_db);
1171 LASSERT(th != NULL);
1172 oh = container_of0(th, struct osd_thandle, ot_super);
1174 rc = osd_prepare_key_uint64(obj, k, key);
1176 /* Remove binary key from the ZAP */
1177 rc = -zap_remove_uint64(osd->od_objset.os, obj->oo_db->db_object,
1182 static int osd_index_it_get(const struct lu_env *env, struct dt_it *di,
1183 const struct dt_key *key)
1185 struct osd_zap_it *it = (struct osd_zap_it *)di;
1186 struct osd_object *obj = it->ozi_obj;
1187 struct osd_device *osd = osd_obj2dev(obj);
1191 LASSERT(it->ozi_zc);
1194 * XXX: we need a binary version of zap_cursor_move_to_key()
1195 * to implement this API */
1196 if (*((const __u64 *)key) != 0)
1197 CERROR("NOT IMPLEMETED YET (move to %Lx)\n", *((__u64 *)key));
1199 zap_cursor_fini(it->ozi_zc);
1200 memset(it->ozi_zc, 0, sizeof(*it->ozi_zc));
1201 zap_cursor_init(it->ozi_zc, osd->od_objset.os, obj->oo_db->db_object);
1207 static int osd_index_it_next(const struct lu_env *env, struct dt_it *di)
1209 struct osd_zap_it *it = (struct osd_zap_it *)di;
1210 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
1214 if (it->ozi_reset == 0)
1215 zap_cursor_advance(it->ozi_zc);
1219 * According to current API we need to return error if it's last entry.
1220 * zap_cursor_advance() does not return any value. So we need to call
1221 * retrieve to check if there is any record. We should make
1222 * changes to Iterator API to not return status for this API
1224 rc = -zap_cursor_retrieve(it->ozi_zc, za);
1231 static struct dt_key *osd_index_it_key(const struct lu_env *env,
1232 const struct dt_it *di)
1234 struct osd_zap_it *it = (struct osd_zap_it *)di;
1235 struct osd_object *obj = it->ozi_obj;
1236 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
1241 rc = -zap_cursor_retrieve(it->ozi_zc, za);
1243 RETURN(ERR_PTR(rc));
1245 /* the binary key is stored in the name */
1246 memcpy(&it->ozi_key, za->za_name, obj->oo_keysize);
1248 RETURN((struct dt_key *)&it->ozi_key);
1251 static int osd_index_it_key_size(const struct lu_env *env,
1252 const struct dt_it *di)
1254 struct osd_zap_it *it = (struct osd_zap_it *)di;
1255 struct osd_object *obj = it->ozi_obj;
1256 RETURN(obj->oo_keysize);
1259 static int osd_index_it_rec(const struct lu_env *env, const struct dt_it *di,
1260 struct dt_rec *rec, __u32 attr)
1262 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
1263 struct osd_zap_it *it = (struct osd_zap_it *)di;
1264 struct osd_object *obj = it->ozi_obj;
1265 struct osd_device *osd = osd_obj2dev(obj);
1266 __u64 *k = osd_oti_get(env)->oti_key64;
1271 rc = -zap_cursor_retrieve(it->ozi_zc, za);
1275 rc = osd_prepare_key_uint64(obj, k, (const struct dt_key *)za->za_name);
1277 rc = -zap_lookup_uint64(osd->od_objset.os, obj->oo_db->db_object,
1278 k, rc, obj->oo_recusize, obj->oo_recsize,
1283 static __u64 osd_index_it_store(const struct lu_env *env,
1284 const struct dt_it *di)
1286 struct osd_zap_it *it = (struct osd_zap_it *)di;
1289 RETURN((__u64)zap_cursor_serialize(it->ozi_zc));
1292 static int osd_index_it_load(const struct lu_env *env, const struct dt_it *di,
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);
1298 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
1302 /* close the current cursor */
1303 zap_cursor_fini(it->ozi_zc);
1305 /* create a new one starting at hash */
1306 memset(it->ozi_zc, 0, sizeof(*it->ozi_zc));
1307 zap_cursor_init_serialized(it->ozi_zc, osd->od_objset.os,
1308 obj->oo_db->db_object, hash);
1311 rc = -zap_cursor_retrieve(it->ozi_zc, za);
1314 else if (rc == -ENOENT)
1320 static struct dt_index_operations osd_index_ops = {
1321 .dio_lookup = osd_index_lookup,
1322 .dio_declare_insert = osd_declare_index_insert,
1323 .dio_insert = osd_index_insert,
1324 .dio_declare_delete = osd_declare_index_delete,
1325 .dio_delete = osd_index_delete,
1327 .init = osd_index_it_init,
1328 .fini = osd_index_it_fini,
1329 .get = osd_index_it_get,
1330 .put = osd_index_it_put,
1331 .next = osd_index_it_next,
1332 .key = osd_index_it_key,
1333 .key_size = osd_index_it_key_size,
1334 .rec = osd_index_it_rec,
1335 .store = osd_index_it_store,
1336 .load = osd_index_it_load
1340 int osd_index_try(const struct lu_env *env, struct dt_object *dt,
1341 const struct dt_index_features *feat)
1343 struct osd_object *obj = osd_dt_obj(dt);
1346 LASSERT(dt_object_exists(dt));
1349 * XXX: implement support for fixed-size keys sorted with natural
1350 * numerical way (not using internal hash value)
1352 if (feat->dif_flags & DT_IND_RANGE)
1355 if (unlikely(feat == &dt_otable_features))
1356 /* do not support oi scrub yet. */
1359 LASSERT(obj->oo_db != NULL);
1360 if (likely(feat == &dt_directory_features)) {
1361 if (udmu_object_is_zap(obj->oo_db))
1362 dt->do_index_ops = &osd_dir_ops;
1365 } else if (unlikely(feat == &dt_acct_features)) {
1366 LASSERT(fid_is_acct(lu_object_fid(&dt->do_lu)));
1367 dt->do_index_ops = &osd_acct_index_ops;
1368 } else if (udmu_object_is_zap(obj->oo_db) &&
1369 dt->do_index_ops == NULL) {
1370 /* For index file, we don't support variable key & record sizes
1371 * and the key has to be unique */
1372 if ((feat->dif_flags & ~DT_IND_UPDATE) != 0)
1375 if (feat->dif_keysize_max > ZAP_MAXNAMELEN)
1377 if (feat->dif_keysize_max != feat->dif_keysize_min)
1380 /* As for the record size, it should be a multiple of 8 bytes
1381 * and smaller than the maximum value length supported by ZAP.
1383 if (feat->dif_recsize_max > ZAP_MAXVALUELEN)
1385 if (feat->dif_recsize_max != feat->dif_recsize_min)
1388 obj->oo_keysize = feat->dif_keysize_max;
1389 obj->oo_recsize = feat->dif_recsize_max;
1390 obj->oo_recusize = 1;
1392 /* ZFS prefers to work with array of 64bits */
1393 if ((obj->oo_recsize & 7) == 0) {
1394 obj->oo_recsize >>= 3;
1395 obj->oo_recusize = 8;
1397 dt->do_index_ops = &osd_index_ops;