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 /* FID seqs not in FLDB, must be local seq */
502 if (unlikely(!fid_seq_in_fldb(fid_seq(fid))))
505 if (osd_seq_exists(env, osd, fid_seq(fid)))
512 * Inserts (key, value) pair in \a directory object.
514 * \param dt osd index object
515 * \param key key for index
516 * \param rec record reference
517 * \param th transaction handler
518 * \param capa capability descriptor
519 * \param ignore_quota update should not affect quota
522 * \retval -ve failure
524 static int osd_dir_insert(const struct lu_env *env, struct dt_object *dt,
525 const struct dt_rec *rec, const struct dt_key *key,
526 struct thandle *th, struct lustre_capa *capa,
529 struct osd_thread_info *oti = osd_oti_get(env);
530 struct osd_object *parent = osd_dt_obj(dt);
531 struct osd_device *osd = osd_obj2dev(parent);
532 struct lu_fid *fid = (struct lu_fid *)rec;
533 struct osd_thandle *oh;
534 struct osd_object *child = NULL;
536 char *name = (char *)key;
540 LASSERT(parent->oo_db);
541 LASSERT(udmu_object_is_zap(parent->oo_db));
543 LASSERT(dt_object_exists(dt));
544 LASSERT(osd_invariant(parent));
547 oh = container_of0(th, struct osd_thandle, ot_super);
549 rc = osd_remote_fid(env, osd, fid);
551 CERROR("%s: Can not find object "DFID": rc = %d\n",
552 osd->od_svname, PFID(fid), rc);
556 if (unlikely(rc == 1)) {
557 /* Insert remote entry */
558 memset(&oti->oti_zde.lzd_reg, 0, sizeof(oti->oti_zde.lzd_reg));
559 oti->oti_zde.lzd_reg.zde_type = IFTODT(S_IFDIR & S_IFMT);
562 * To simulate old Orion setups with ./.. stored in the
565 /* Insert local entry */
566 child = osd_object_find(env, dt, fid);
568 RETURN(PTR_ERR(child));
570 LASSERT(child->oo_db);
571 if (name[0] == '.') {
573 /* do not store ".", instead generate it
574 * during iteration */
576 } else if (name[1] == '.' && name[2] == 0) {
577 /* update parent dnode in the child.
578 * later it will be used to generate ".." */
579 udmu_objset_t *uos = &osd->od_objset;
580 rc = osd_object_sa_update(parent,
582 &child->oo_db->db_object,
587 CLASSERT(sizeof(oti->oti_zde.lzd_reg) == 8);
588 CLASSERT(sizeof(oti->oti_zde) % 8 == 0);
589 attr = child->oo_dt.do_lu.lo_header ->loh_attr;
590 oti->oti_zde.lzd_reg.zde_type = IFTODT(attr & S_IFMT);
591 oti->oti_zde.lzd_reg.zde_dnode = child->oo_db->db_object;
594 oti->oti_zde.lzd_fid = *fid;
595 /* Insert (key,oid) into ZAP */
596 rc = -zap_add(osd->od_objset.os, parent->oo_db->db_object,
597 (char *)key, 8, sizeof(oti->oti_zde) / 8,
598 (void *)&oti->oti_zde, oh->ot_tx);
602 osd_object_put(env, child);
607 static int osd_declare_dir_delete(const struct lu_env *env,
608 struct dt_object *dt,
609 const struct dt_key *key,
612 struct osd_object *obj = osd_dt_obj(dt);
613 struct osd_thandle *oh;
616 LASSERT(dt_object_exists(dt));
617 LASSERT(osd_invariant(obj));
620 oh = container_of0(th, struct osd_thandle, ot_super);
623 LASSERT(udmu_object_is_zap(obj->oo_db));
625 dmu_tx_hold_zap(oh->ot_tx, obj->oo_db->db_object, TRUE, (char *)key);
630 static int osd_dir_delete(const struct lu_env *env, struct dt_object *dt,
631 const struct dt_key *key, struct thandle *th,
632 struct lustre_capa *capa)
634 struct osd_object *obj = osd_dt_obj(dt);
635 struct osd_device *osd = osd_obj2dev(obj);
636 struct osd_thandle *oh;
637 dmu_buf_t *zap_db = obj->oo_db;
638 char *name = (char *)key;
643 LASSERT(udmu_object_is_zap(obj->oo_db));
646 oh = container_of0(th, struct osd_thandle, ot_super);
649 * In Orion . and .. were stored in the directory (not generated upon
650 * request as now). we preserve them for backward compatibility
652 if (name[0] == '.') {
655 } else if (name[1] == '.' && name[2] == 0) {
660 /* Remove key from the ZAP */
661 rc = -zap_remove(osd->od_objset.os, zap_db->db_object,
662 (char *) key, oh->ot_tx);
664 #if LUSTRE_VERSION_CODE <= OBD_OCD_VERSION(2, 4, 53, 0)
665 if (unlikely(rc == -ENOENT && name[0] == '.' &&
666 (name[1] == 0 || (name[1] == '.' && name[2] == 0))))
669 if (unlikely(rc && rc != -ENOENT))
670 CERROR("%s: zap_remove failed: rc = %d\n", osd->od_svname, rc);
675 static struct dt_it *osd_dir_it_init(const struct lu_env *env,
676 struct dt_object *dt,
678 struct lustre_capa *capa)
680 struct osd_zap_it *it;
682 it = (struct osd_zap_it *)osd_index_it_init(env, dt, unused, capa);
686 RETURN((struct dt_it *)it);
690 * Move Iterator to record specified by \a key
692 * \param di osd iterator
693 * \param key key for index
695 * \retval +ve di points to record with least key not larger than key
696 * \retval 0 di points to exact matched key
697 * \retval -ve failure
699 static int osd_dir_it_get(const struct lu_env *env,
700 struct dt_it *di, const struct dt_key *key)
702 struct osd_zap_it *it = (struct osd_zap_it *)di;
703 struct osd_object *obj = it->ozi_obj;
704 struct osd_device *osd = osd_obj2dev(obj);
705 char *name = (char *)key;
712 udmu_zap_cursor_fini(it->ozi_zc);
714 if (udmu_zap_cursor_init(&it->ozi_zc, &osd->od_objset,
715 obj->oo_db->db_object, 0))
718 /* XXX: implementation of the API is broken at the moment */
719 LASSERT(((const char *)key)[0] == 0);
726 if (name[0] == '.') {
730 } else if (name[1] == '.' && name[2] == 0) {
736 /* neither . nor .. - some real record */
744 static void osd_dir_it_put(const struct lu_env *env, struct dt_it *di)
746 /* PBS: do nothing : ref are incremented at retrive and decreamented
751 * in Orion . and .. were stored in the directory, while ZPL
752 * and current osd-zfs generate them up on request. so, we
753 * need to ignore previously stored . and ..
755 static int osd_index_retrieve_skip_dots(struct osd_zap_it *it,
761 rc = -zap_cursor_retrieve(it->ozi_zc, za);
764 if (unlikely(rc == 0 && za->za_name[0] == '.')) {
765 if (za->za_name[1] == 0) {
767 } else if (za->za_name[1] == '.' &&
768 za->za_name[2] == 0) {
772 zap_cursor_advance(it->ozi_zc);
774 } while (unlikely(rc == 0 && isdot));
780 * to load a directory entry at a time and stored it in
781 * iterator's in-memory data structure.
783 * \param di, struct osd_it_ea, iterator's in memory structure
785 * \retval +ve, iterator reached to end
786 * \retval 0, iterator not reached to end
787 * \retval -ve, on error
789 static int osd_dir_it_next(const struct lu_env *env, struct dt_it *di)
791 struct osd_zap_it *it = (struct osd_zap_it *)di;
792 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
795 /* temp. storage should be enough for any key supported by ZFS */
796 CLASSERT(sizeof(za->za_name) <= sizeof(it->ozi_name));
799 * the first ->next() moves the cursor to .
800 * the second ->next() moves the cursor to ..
801 * then we get to the real records and have to verify any exist
803 if (it->ozi_pos <= 2) {
809 zap_cursor_advance(it->ozi_zc);
812 * According to current API we need to return error if its last entry.
813 * zap_cursor_advance() does not return any value. So we need to call
814 * retrieve to check if there is any record. We should make
815 * changes to Iterator API to not return status for this API
817 rc = osd_index_retrieve_skip_dots(it, za);
819 if (rc == -ENOENT) /* end of dir */
825 static struct dt_key *osd_dir_it_key(const struct lu_env *env,
826 const struct dt_it *di)
828 struct osd_zap_it *it = (struct osd_zap_it *)di;
829 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
833 if (it->ozi_pos <= 1) {
835 RETURN((struct dt_key *)".");
836 } else if (it->ozi_pos == 2) {
837 RETURN((struct dt_key *)"..");
840 if ((rc = -zap_cursor_retrieve(it->ozi_zc, za)))
843 strcpy(it->ozi_name, za->za_name);
845 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 3, 90, 0)
846 if (za->za_name[0] == '.') {
847 if (za->za_name[1] == 0 || (za->za_name[1] == '.' &&
848 za->za_name[2] == 0)) {
849 /* we should not get onto . and ..
850 * stored in the directory. ->next() and
851 * other methods should prevent this
858 RETURN((struct dt_key *)it->ozi_name);
861 static int osd_dir_it_key_size(const struct lu_env *env, const struct dt_it *di)
863 struct osd_zap_it *it = (struct osd_zap_it *)di;
864 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
868 if (it->ozi_pos <= 1) {
871 } else if (it->ozi_pos == 2) {
875 if ((rc = -zap_cursor_retrieve(it->ozi_zc, za)) == 0)
876 rc = strlen(za->za_name);
878 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 3, 90, 0)
879 if (rc == 0 && za->za_name[0] == '.') {
880 if (za->za_name[1] == 0 || (za->za_name[1] == '.' &&
881 za->za_name[2] == 0)) {
882 /* we should not get onto . and ..
883 * stored in the directory. ->next() and
884 * other methods should prevent this
893 static int osd_dir_it_rec(const struct lu_env *env, const struct dt_it *di,
894 struct dt_rec *dtrec, __u32 attr)
896 struct osd_zap_it *it = (struct osd_zap_it *)di;
897 struct lu_dirent *lde = (struct lu_dirent *)dtrec;
898 struct luz_direntry *zde = &osd_oti_get(env)->oti_zde;
899 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
903 if (it->ozi_pos <= 1) {
904 lde->lde_hash = cpu_to_le64(1);
905 strcpy(lde->lde_name, ".");
906 lde->lde_namelen = cpu_to_le16(1);
907 lde->lde_fid = *lu_object_fid(&it->ozi_obj->oo_dt.do_lu);
908 lde->lde_attrs = LUDA_FID;
909 /* append lustre attributes */
910 osd_it_append_attrs(lde, attr, 1, IFTODT(S_IFDIR));
911 lde->lde_reclen = cpu_to_le16(lu_dirent_calc_size(1, attr));
915 } else if (it->ozi_pos == 2) {
916 lde->lde_hash = cpu_to_le64(2);
917 strcpy(lde->lde_name, "..");
918 lde->lde_namelen = cpu_to_le16(2);
919 lde->lde_attrs = LUDA_FID;
920 /* append lustre attributes */
921 osd_it_append_attrs(lde, attr, 2, IFTODT(S_IFDIR));
922 lde->lde_reclen = cpu_to_le16(lu_dirent_calc_size(2, attr));
923 rc = osd_find_parent_fid(env, &it->ozi_obj->oo_dt, &lde->lde_fid);
925 * early Orion code was not setting LinkEA, so it's possible
926 * some setups still have objects with no LinkEA set.
927 * but at that time .. was a real record in the directory
928 * so we should try to lookup .. in ZAP
936 rc = -zap_cursor_retrieve(it->ozi_zc, za);
937 if (unlikely(rc != 0))
940 lde->lde_hash = cpu_to_le64(udmu_zap_cursor_serialize(it->ozi_zc));
941 namelen = strlen(za->za_name);
942 if (namelen > NAME_MAX)
943 GOTO(out, rc = -EOVERFLOW);
944 strcpy(lde->lde_name, za->za_name);
945 lde->lde_namelen = cpu_to_le16(namelen);
947 if (za->za_integer_length != 8 || za->za_num_integers < 3) {
948 CERROR("%s: unsupported direntry format: %d %d\n",
949 osd_obj2dev(it->ozi_obj)->od_svname,
950 za->za_integer_length, (int)za->za_num_integers);
952 GOTO(out, rc = -EIO);
955 rc = -zap_lookup(it->ozi_zc->zc_objset, it->ozi_zc->zc_zapobj,
956 za->za_name, za->za_integer_length, 3, zde);
960 lde->lde_fid = zde->lzd_fid;
961 lde->lde_attrs = LUDA_FID;
963 /* append lustre attributes */
964 osd_it_append_attrs(lde, attr, namelen, zde->lzd_reg.zde_type);
966 lde->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
972 static int osd_dir_it_rec_size(const struct lu_env *env, const struct dt_it *di,
975 struct osd_zap_it *it = (struct osd_zap_it *)di;
976 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
980 if (it->ozi_pos <= 1)
982 else if (it->ozi_pos == 2)
986 rc = lu_dirent_calc_size(namelen, attr);
990 rc = -zap_cursor_retrieve(it->ozi_zc, za);
991 if (unlikely(rc != 0))
994 if (za->za_integer_length != 8 || za->za_num_integers < 3) {
995 CERROR("%s: unsupported direntry format: %d %d\n",
996 osd_obj2dev(it->ozi_obj)->od_svname,
997 za->za_integer_length, (int)za->za_num_integers);
1001 namelen = strlen(za->za_name);
1002 if (namelen > NAME_MAX)
1005 rc = lu_dirent_calc_size(namelen, attr);
1010 static __u64 osd_dir_it_store(const struct lu_env *env, const struct dt_it *di)
1012 struct osd_zap_it *it = (struct osd_zap_it *)di;
1016 if (it->ozi_pos <= 2)
1019 pos = udmu_zap_cursor_serialize(it->ozi_zc);
1026 * rc == 0 -> end of directory.
1027 * rc > 0 -> ok, proceed.
1028 * rc < 0 -> error. ( EOVERFLOW can be masked.)
1030 static int osd_dir_it_load(const struct lu_env *env,
1031 const struct dt_it *di, __u64 hash)
1033 struct osd_zap_it *it = (struct osd_zap_it *)di;
1034 struct osd_object *obj = it->ozi_obj;
1035 struct osd_device *osd = osd_obj2dev(obj);
1036 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
1040 udmu_zap_cursor_fini(it->ozi_zc);
1041 if (udmu_zap_cursor_init(&it->ozi_zc, &osd->od_objset,
1042 obj->oo_db->db_object, hash))
1050 /* to return whether the end has been reached */
1051 rc = osd_index_retrieve_skip_dots(it, za);
1054 else if (rc == -ENOENT)
1061 static struct dt_index_operations osd_dir_ops = {
1062 .dio_lookup = osd_dir_lookup,
1063 .dio_declare_insert = osd_declare_dir_insert,
1064 .dio_insert = osd_dir_insert,
1065 .dio_declare_delete = osd_declare_dir_delete,
1066 .dio_delete = osd_dir_delete,
1068 .init = osd_dir_it_init,
1069 .fini = osd_index_it_fini,
1070 .get = osd_dir_it_get,
1071 .put = osd_dir_it_put,
1072 .next = osd_dir_it_next,
1073 .key = osd_dir_it_key,
1074 .key_size = osd_dir_it_key_size,
1075 .rec = osd_dir_it_rec,
1076 .rec_size = osd_dir_it_rec_size,
1077 .store = osd_dir_it_store,
1078 .load = osd_dir_it_load
1083 * Primitives for index files using binary keys.
1086 /* key integer_size is 8 */
1087 static int osd_prepare_key_uint64(struct osd_object *o, __u64 *dst,
1088 const struct dt_key *src)
1095 /* align keysize to 64bit */
1096 size = (o->oo_keysize + sizeof(__u64) - 1) / sizeof(__u64);
1097 size *= sizeof(__u64);
1099 LASSERT(size <= MAXNAMELEN);
1101 if (unlikely(size > o->oo_keysize))
1102 memset(dst + o->oo_keysize, 0, size - o->oo_keysize);
1103 memcpy(dst, (const char *)src, o->oo_keysize);
1105 return (size/sizeof(__u64));
1108 static int osd_index_lookup(const struct lu_env *env, struct dt_object *dt,
1109 struct dt_rec *rec, const struct dt_key *key,
1110 struct lustre_capa *capa)
1112 struct osd_object *obj = osd_dt_obj(dt);
1113 struct osd_device *osd = osd_obj2dev(obj);
1114 __u64 *k = osd_oti_get(env)->oti_key64;
1118 rc = osd_prepare_key_uint64(obj, k, key);
1120 rc = -zap_lookup_uint64(osd->od_objset.os, obj->oo_db->db_object,
1121 k, rc, obj->oo_recusize, obj->oo_recsize,
1123 RETURN(rc == 0 ? 1 : rc);
1126 static int osd_declare_index_insert(const struct lu_env *env,
1127 struct dt_object *dt,
1128 const struct dt_rec *rec,
1129 const struct dt_key *key,
1132 struct osd_object *obj = osd_dt_obj(dt);
1133 struct osd_thandle *oh;
1136 LASSERT(th != NULL);
1137 oh = container_of0(th, struct osd_thandle, ot_super);
1139 LASSERT(obj->oo_db);
1141 dmu_tx_hold_bonus(oh->ot_tx, obj->oo_db->db_object);
1143 /* It is not clear what API should be used for binary keys, so we pass
1144 * a null name which has the side effect of over-reserving space,
1145 * accounting for the worst case. See zap_count_write() */
1146 dmu_tx_hold_zap(oh->ot_tx, obj->oo_db->db_object, TRUE, NULL);
1151 static int osd_index_insert(const struct lu_env *env, struct dt_object *dt,
1152 const struct dt_rec *rec, const struct dt_key *key,
1153 struct thandle *th, struct lustre_capa *capa,
1156 struct osd_object *obj = osd_dt_obj(dt);
1157 struct osd_device *osd = osd_obj2dev(obj);
1158 struct osd_thandle *oh;
1159 __u64 *k = osd_oti_get(env)->oti_key64;
1163 LASSERT(obj->oo_db);
1164 LASSERT(dt_object_exists(dt));
1165 LASSERT(osd_invariant(obj));
1166 LASSERT(th != NULL);
1168 oh = container_of0(th, struct osd_thandle, ot_super);
1170 rc = osd_prepare_key_uint64(obj, k, key);
1172 /* Insert (key,oid) into ZAP */
1173 rc = -zap_add_uint64(osd->od_objset.os, obj->oo_db->db_object,
1174 k, rc, obj->oo_recusize, obj->oo_recsize,
1175 (void *)rec, oh->ot_tx);
1179 static int osd_declare_index_delete(const struct lu_env *env,
1180 struct dt_object *dt,
1181 const struct dt_key *key,
1184 struct osd_object *obj = osd_dt_obj(dt);
1185 struct osd_thandle *oh;
1188 LASSERT(dt_object_exists(dt));
1189 LASSERT(osd_invariant(obj));
1190 LASSERT(th != NULL);
1191 LASSERT(obj->oo_db);
1193 oh = container_of0(th, struct osd_thandle, ot_super);
1194 dmu_tx_hold_zap(oh->ot_tx, obj->oo_db->db_object, TRUE, NULL);
1199 static int osd_index_delete(const struct lu_env *env, struct dt_object *dt,
1200 const struct dt_key *key, struct thandle *th,
1201 struct lustre_capa *capa)
1203 struct osd_object *obj = osd_dt_obj(dt);
1204 struct osd_device *osd = osd_obj2dev(obj);
1205 struct osd_thandle *oh;
1206 __u64 *k = osd_oti_get(env)->oti_key64;
1210 LASSERT(obj->oo_db);
1211 LASSERT(th != NULL);
1212 oh = container_of0(th, struct osd_thandle, ot_super);
1214 rc = osd_prepare_key_uint64(obj, k, key);
1216 /* Remove binary key from the ZAP */
1217 rc = -zap_remove_uint64(osd->od_objset.os, obj->oo_db->db_object,
1222 static int osd_index_it_get(const struct lu_env *env, struct dt_it *di,
1223 const struct dt_key *key)
1225 struct osd_zap_it *it = (struct osd_zap_it *)di;
1226 struct osd_object *obj = it->ozi_obj;
1227 struct osd_device *osd = osd_obj2dev(obj);
1231 LASSERT(it->ozi_zc);
1234 * XXX: we need a binary version of zap_cursor_move_to_key()
1235 * to implement this API */
1236 if (*((const __u64 *)key) != 0)
1237 CERROR("NOT IMPLEMETED YET (move to "LPX64")\n",
1240 zap_cursor_fini(it->ozi_zc);
1241 memset(it->ozi_zc, 0, sizeof(*it->ozi_zc));
1242 zap_cursor_init(it->ozi_zc, osd->od_objset.os, obj->oo_db->db_object);
1248 static int osd_index_it_next(const struct lu_env *env, struct dt_it *di)
1250 struct osd_zap_it *it = (struct osd_zap_it *)di;
1251 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
1255 if (it->ozi_reset == 0)
1256 zap_cursor_advance(it->ozi_zc);
1260 * According to current API we need to return error if it's last entry.
1261 * zap_cursor_advance() does not return any value. So we need to call
1262 * retrieve to check if there is any record. We should make
1263 * changes to Iterator API to not return status for this API
1265 rc = -zap_cursor_retrieve(it->ozi_zc, za);
1272 static struct dt_key *osd_index_it_key(const struct lu_env *env,
1273 const struct dt_it *di)
1275 struct osd_zap_it *it = (struct osd_zap_it *)di;
1276 struct osd_object *obj = it->ozi_obj;
1277 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
1282 rc = -zap_cursor_retrieve(it->ozi_zc, za);
1284 RETURN(ERR_PTR(rc));
1286 /* the binary key is stored in the name */
1287 memcpy(&it->ozi_key, za->za_name, obj->oo_keysize);
1289 RETURN((struct dt_key *)&it->ozi_key);
1292 static int osd_index_it_key_size(const struct lu_env *env,
1293 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 RETURN(obj->oo_keysize);
1300 static int osd_index_it_rec(const struct lu_env *env, const struct dt_it *di,
1301 struct dt_rec *rec, __u32 attr)
1303 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
1304 struct osd_zap_it *it = (struct osd_zap_it *)di;
1305 struct osd_object *obj = it->ozi_obj;
1306 struct osd_device *osd = osd_obj2dev(obj);
1307 __u64 *k = osd_oti_get(env)->oti_key64;
1312 rc = -zap_cursor_retrieve(it->ozi_zc, za);
1316 rc = osd_prepare_key_uint64(obj, k, (const struct dt_key *)za->za_name);
1318 rc = -zap_lookup_uint64(osd->od_objset.os, obj->oo_db->db_object,
1319 k, rc, obj->oo_recusize, obj->oo_recsize,
1324 static __u64 osd_index_it_store(const struct lu_env *env,
1325 const struct dt_it *di)
1327 struct osd_zap_it *it = (struct osd_zap_it *)di;
1330 RETURN((__u64)zap_cursor_serialize(it->ozi_zc));
1333 static int osd_index_it_load(const struct lu_env *env, const struct dt_it *di,
1336 struct osd_zap_it *it = (struct osd_zap_it *)di;
1337 struct osd_object *obj = it->ozi_obj;
1338 struct osd_device *osd = osd_obj2dev(obj);
1339 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
1343 /* close the current cursor */
1344 zap_cursor_fini(it->ozi_zc);
1346 /* create a new one starting at hash */
1347 memset(it->ozi_zc, 0, sizeof(*it->ozi_zc));
1348 zap_cursor_init_serialized(it->ozi_zc, osd->od_objset.os,
1349 obj->oo_db->db_object, hash);
1352 rc = -zap_cursor_retrieve(it->ozi_zc, za);
1355 else if (rc == -ENOENT)
1361 static struct dt_index_operations osd_index_ops = {
1362 .dio_lookup = osd_index_lookup,
1363 .dio_declare_insert = osd_declare_index_insert,
1364 .dio_insert = osd_index_insert,
1365 .dio_declare_delete = osd_declare_index_delete,
1366 .dio_delete = osd_index_delete,
1368 .init = osd_index_it_init,
1369 .fini = osd_index_it_fini,
1370 .get = osd_index_it_get,
1371 .put = osd_index_it_put,
1372 .next = osd_index_it_next,
1373 .key = osd_index_it_key,
1374 .key_size = osd_index_it_key_size,
1375 .rec = osd_index_it_rec,
1376 .store = osd_index_it_store,
1377 .load = osd_index_it_load
1381 struct osd_metadnode_it {
1382 struct osd_device *mit_dev;
1384 struct lu_fid mit_fid;
1386 __u64 mit_prefetched_dnode;
1389 static struct dt_it *osd_zfs_otable_it_init(const struct lu_env *env,
1390 struct dt_object *dt, __u32 attr,
1391 struct lustre_capa *capa)
1393 struct osd_device *dev = osd_dev(dt->do_lu.lo_dev);
1394 struct osd_metadnode_it *it;
1398 if (unlikely(it == NULL))
1399 RETURN(ERR_PTR(-ENOMEM));
1403 /* XXX: dmu_object_next() does NOT find dnodes allocated
1404 * in the current non-committed txg, so we force txg
1405 * commit to find all existing dnodes ... */
1406 txg_wait_synced(dmu_objset_pool(dev->od_objset.os), 0ULL);
1408 RETURN((struct dt_it *)it);
1411 static void osd_zfs_otable_it_fini(const struct lu_env *env, struct dt_it *di)
1413 struct osd_metadnode_it *it = (struct osd_metadnode_it *)di;
1418 static int osd_zfs_otable_it_get(const struct lu_env *env,
1419 struct dt_it *di, const struct dt_key *key)
1424 static void osd_zfs_otable_it_put(const struct lu_env *env, struct dt_it *di)
1428 #define OTABLE_PREFETCH 256
1430 static void osd_zfs_otable_prefetch(const struct lu_env *env,
1431 struct osd_metadnode_it *it)
1433 struct osd_device *dev = it->mit_dev;
1434 udmu_objset_t *uos = &dev->od_objset;
1437 /* can go negative on the very first access to the iterator
1438 * or if some non-Lustre objects were found */
1439 if (unlikely(it->mit_prefetched < 0))
1440 it->mit_prefetched = 0;
1442 if (it->mit_prefetched >= (OTABLE_PREFETCH >> 1))
1445 if (it->mit_prefetched_dnode == 0)
1446 it->mit_prefetched_dnode = it->mit_pos;
1448 while (it->mit_prefetched < OTABLE_PREFETCH) {
1449 rc = -dmu_object_next(uos->os, &it->mit_prefetched_dnode,
1451 if (unlikely(rc != 0))
1454 /* dmu_prefetch() was exported in 0.6.2, if you use with
1455 * an older release, just comment it out - this is an
1457 dmu_prefetch(uos->os, it->mit_prefetched_dnode, 0, 0);
1459 it->mit_prefetched++;
1463 static int osd_zfs_otable_it_next(const struct lu_env *env, struct dt_it *di)
1465 struct osd_metadnode_it *it = (struct osd_metadnode_it *)di;
1466 struct lustre_mdt_attrs *lma;
1467 struct osd_device *dev = it->mit_dev;
1468 udmu_objset_t *uos = &dev->od_objset;
1469 nvlist_t *nvbuf = NULL;
1474 memset(&it->mit_fid, 0, sizeof(it->mit_fid));
1476 dnode = it->mit_pos;
1478 rc = -dmu_object_next(uos->os, &it->mit_pos, B_FALSE, 0);
1479 if (unlikely(rc != 0))
1481 it->mit_prefetched--;
1483 /* LMA is required for this to be a Lustre object.
1484 * If there is no xattr skip it. */
1485 rc = __osd_xattr_load(uos, it->mit_pos, &nvbuf);
1486 if (unlikely(rc != 0))
1489 LASSERT(nvbuf != NULL);
1490 rc = -nvlist_lookup_byte_array(nvbuf, XATTR_NAME_LMA, &v, &s);
1491 if (likely(rc == 0)) {
1493 lma = (struct lustre_mdt_attrs *)v;
1494 lustre_lma_swab(lma);
1495 it->mit_fid = lma->lma_self_fid;
1499 /* not a Lustre object, try next one */
1506 /* we aren't prefetching in the above loop because the number of
1507 * non-Lustre objects is very small and we will be repeating very
1508 * rare. in case we want to use this to iterate over non-Lustre
1509 * objects (i.e. when we convert regular ZFS in Lustre) it makes
1510 * sense to initiate prefetching in the loop */
1512 /* 0 - there are more items, +1 - the end */
1513 if (likely(rc == 0))
1514 osd_zfs_otable_prefetch(env, it);
1516 CDEBUG(D_OTHER, "advance: %llu -> %llu "DFID": %d\n", dnode,
1517 it->mit_pos, PFID(&it->mit_fid), rc);
1523 static struct dt_key *osd_zfs_otable_it_key(const struct lu_env *env,
1524 const struct dt_it *di)
1529 static int osd_zfs_otable_it_key_size(const struct lu_env *env,
1530 const struct dt_it *di)
1532 return sizeof(__u64);
1535 static int osd_zfs_otable_it_rec(const struct lu_env *env,
1536 const struct dt_it *di,
1537 struct dt_rec *rec, __u32 attr)
1539 struct osd_metadnode_it *it = (struct osd_metadnode_it *)di;
1540 struct lu_fid *fid = (struct lu_fid *)rec;
1549 static __u64 osd_zfs_otable_it_store(const struct lu_env *env,
1550 const struct dt_it *di)
1552 struct osd_metadnode_it *it = (struct osd_metadnode_it *)di;
1557 static int osd_zfs_otable_it_load(const struct lu_env *env,
1558 const struct dt_it *di, __u64 hash)
1560 struct osd_metadnode_it *it = (struct osd_metadnode_it *)di;
1563 it->mit_prefetched = 0;
1564 it->mit_prefetched_dnode = 0;
1566 return osd_zfs_otable_it_next(env, (struct dt_it *)di);
1569 static int osd_zfs_otable_it_key_rec(const struct lu_env *env,
1570 const struct dt_it *di, void *key_rec)
1575 const struct dt_index_operations osd_zfs_otable_ops = {
1577 .init = osd_zfs_otable_it_init,
1578 .fini = osd_zfs_otable_it_fini,
1579 .get = osd_zfs_otable_it_get,
1580 .put = osd_zfs_otable_it_put,
1581 .next = osd_zfs_otable_it_next,
1582 .key = osd_zfs_otable_it_key,
1583 .key_size = osd_zfs_otable_it_key_size,
1584 .rec = osd_zfs_otable_it_rec,
1585 .store = osd_zfs_otable_it_store,
1586 .load = osd_zfs_otable_it_load,
1587 .key_rec = osd_zfs_otable_it_key_rec,
1591 int osd_index_try(const struct lu_env *env, struct dt_object *dt,
1592 const struct dt_index_features *feat)
1594 struct osd_object *obj = osd_dt_obj(dt);
1597 LASSERT(dt_object_exists(dt));
1600 * XXX: implement support for fixed-size keys sorted with natural
1601 * numerical way (not using internal hash value)
1603 if (feat->dif_flags & DT_IND_RANGE)
1606 if (unlikely(feat == &dt_otable_features)) {
1607 dt->do_index_ops = &osd_zfs_otable_ops;
1611 LASSERT(obj->oo_db != NULL);
1612 if (likely(feat == &dt_directory_features)) {
1613 if (udmu_object_is_zap(obj->oo_db))
1614 dt->do_index_ops = &osd_dir_ops;
1617 } else if (unlikely(feat == &dt_acct_features)) {
1618 LASSERT(fid_is_acct(lu_object_fid(&dt->do_lu)));
1619 dt->do_index_ops = &osd_acct_index_ops;
1620 } else if (udmu_object_is_zap(obj->oo_db) &&
1621 dt->do_index_ops == NULL) {
1622 /* For index file, we don't support variable key & record sizes
1623 * and the key has to be unique */
1624 if ((feat->dif_flags & ~DT_IND_UPDATE) != 0)
1627 if (feat->dif_keysize_max > ZAP_MAXNAMELEN)
1629 if (feat->dif_keysize_max != feat->dif_keysize_min)
1632 /* As for the record size, it should be a multiple of 8 bytes
1633 * and smaller than the maximum value length supported by ZAP.
1635 if (feat->dif_recsize_max > ZAP_MAXVALUELEN)
1637 if (feat->dif_recsize_max != feat->dif_recsize_min)
1640 obj->oo_keysize = feat->dif_keysize_max;
1641 obj->oo_recsize = feat->dif_recsize_max;
1642 obj->oo_recusize = 1;
1644 /* ZFS prefers to work with array of 64bits */
1645 if ((obj->oo_recsize & 7) == 0) {
1646 obj->oo_recsize >>= 3;
1647 obj->oo_recusize = 8;
1649 dt->do_index_ops = &osd_index_ops;