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_remote_fid(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);
483 if (!fid_is_norm(fid) && !fid_is_root(fid))
486 rc = osd_fld_lookup(env, osd, fid, range);
488 CERROR("%s: Can not lookup fld for "DFID"\n",
489 osd_name(osd), PFID(fid));
493 RETURN(ss->ss_node_id != range->lsr_index);
497 * Inserts (key, value) pair in \a directory object.
499 * \param dt osd index object
500 * \param key key for index
501 * \param rec record reference
502 * \param th transaction handler
503 * \param capa capability descriptor
504 * \param ignore_quota update should not affect quota
507 * \retval -ve failure
509 static int osd_dir_insert(const struct lu_env *env, struct dt_object *dt,
510 const struct dt_rec *rec, const struct dt_key *key,
511 struct thandle *th, struct lustre_capa *capa,
514 struct osd_thread_info *oti = osd_oti_get(env);
515 struct osd_object *parent = osd_dt_obj(dt);
516 struct osd_device *osd = osd_obj2dev(parent);
517 struct lu_fid *fid = (struct lu_fid *)rec;
518 struct osd_thandle *oh;
519 struct osd_object *child = NULL;
521 char *name = (char *)key;
525 LASSERT(parent->oo_db);
526 LASSERT(udmu_object_is_zap(parent->oo_db));
528 LASSERT(dt_object_exists(dt));
529 LASSERT(osd_invariant(parent));
532 oh = container_of0(th, struct osd_thandle, ot_super);
534 rc = osd_remote_fid(env, osd, fid);
536 CERROR("%s: Can not find object "DFID": rc = %d\n",
537 osd->od_svname, PFID(fid), rc);
541 if (unlikely(rc == 1)) {
542 /* Insert remote entry */
543 memset(&oti->oti_zde.lzd_reg, 0, sizeof(oti->oti_zde.lzd_reg));
544 oti->oti_zde.lzd_reg.zde_type = IFTODT(S_IFDIR & S_IFMT);
547 * To simulate old Orion setups with ./.. stored in the
550 /* Insert local entry */
551 child = osd_object_find(env, dt, fid);
553 RETURN(PTR_ERR(child));
555 LASSERT(child->oo_db);
556 if (name[0] == '.') {
558 /* do not store ".", instead generate it
559 * during iteration */
561 } else if (name[1] == '.' && name[2] == 0) {
562 /* update parent dnode in the child.
563 * later it will be used to generate ".." */
564 udmu_objset_t *uos = &osd->od_objset;
565 rc = osd_object_sa_update(parent,
567 &child->oo_db->db_object,
572 CLASSERT(sizeof(oti->oti_zde.lzd_reg) == 8);
573 CLASSERT(sizeof(oti->oti_zde) % 8 == 0);
574 attr = child->oo_dt.do_lu.lo_header ->loh_attr;
575 oti->oti_zde.lzd_reg.zde_type = IFTODT(attr & S_IFMT);
576 oti->oti_zde.lzd_reg.zde_dnode = child->oo_db->db_object;
579 oti->oti_zde.lzd_fid = *fid;
580 /* Insert (key,oid) into ZAP */
581 rc = -zap_add(osd->od_objset.os, parent->oo_db->db_object,
582 (char *)key, 8, sizeof(oti->oti_zde) / 8,
583 (void *)&oti->oti_zde, oh->ot_tx);
587 osd_object_put(env, child);
592 static int osd_declare_dir_delete(const struct lu_env *env,
593 struct dt_object *dt,
594 const struct dt_key *key,
597 struct osd_object *obj = osd_dt_obj(dt);
598 struct osd_thandle *oh;
601 LASSERT(dt_object_exists(dt));
602 LASSERT(osd_invariant(obj));
605 oh = container_of0(th, struct osd_thandle, ot_super);
608 LASSERT(udmu_object_is_zap(obj->oo_db));
610 dmu_tx_hold_zap(oh->ot_tx, obj->oo_db->db_object, TRUE, (char *)key);
615 static int osd_dir_delete(const struct lu_env *env, struct dt_object *dt,
616 const struct dt_key *key, struct thandle *th,
617 struct lustre_capa *capa)
619 struct osd_object *obj = osd_dt_obj(dt);
620 struct osd_device *osd = osd_obj2dev(obj);
621 struct osd_thandle *oh;
622 dmu_buf_t *zap_db = obj->oo_db;
623 char *name = (char *)key;
628 LASSERT(udmu_object_is_zap(obj->oo_db));
631 oh = container_of0(th, struct osd_thandle, ot_super);
634 * In Orion . and .. were stored in the directory (not generated upon
635 * request as now). we preserve them for backward compatibility
637 if (name[0] == '.') {
640 } else if (name[1] == '.' && name[2] == 0) {
645 /* Remove key from the ZAP */
646 rc = -zap_remove(osd->od_objset.os, zap_db->db_object,
647 (char *) key, oh->ot_tx);
649 #if LUSTRE_VERSION_CODE <= OBD_OCD_VERSION(2, 4, 53, 0)
650 if (unlikely(rc == -ENOENT && name[0] == '.' &&
651 (name[1] == 0 || (name[1] == '.' && name[2] == 0))))
654 if (unlikely(rc && rc != -ENOENT))
655 CERROR("%s: zap_remove failed: rc = %d\n", osd->od_svname, rc);
660 static struct dt_it *osd_dir_it_init(const struct lu_env *env,
661 struct dt_object *dt,
663 struct lustre_capa *capa)
665 struct osd_zap_it *it;
667 it = (struct osd_zap_it *)osd_index_it_init(env, dt, unused, capa);
671 RETURN((struct dt_it *)it);
675 * Move Iterator to record specified by \a key
677 * \param di osd iterator
678 * \param key key for index
680 * \retval +ve di points to record with least key not larger than key
681 * \retval 0 di points to exact matched key
682 * \retval -ve failure
684 static int osd_dir_it_get(const struct lu_env *env,
685 struct dt_it *di, const struct dt_key *key)
687 struct osd_zap_it *it = (struct osd_zap_it *)di;
688 struct osd_object *obj = it->ozi_obj;
689 struct osd_device *osd = osd_obj2dev(obj);
690 char *name = (char *)key;
697 udmu_zap_cursor_fini(it->ozi_zc);
699 if (udmu_zap_cursor_init(&it->ozi_zc, &osd->od_objset,
700 obj->oo_db->db_object, 0))
703 /* XXX: implementation of the API is broken at the moment */
704 LASSERT(((const char *)key)[0] == 0);
711 if (name[0] == '.') {
715 } else if (name[1] == '.' && name[2] == 0) {
721 /* neither . nor .. - some real record */
729 static void osd_dir_it_put(const struct lu_env *env, struct dt_it *di)
731 /* PBS: do nothing : ref are incremented at retrive and decreamented
736 * in Orion . and .. were stored in the directory, while ZPL
737 * and current osd-zfs generate them up on request. so, we
738 * need to ignore previously stored . and ..
740 static int osd_index_retrieve_skip_dots(struct osd_zap_it *it,
746 rc = -zap_cursor_retrieve(it->ozi_zc, za);
749 if (unlikely(rc == 0 && za->za_name[0] == '.')) {
750 if (za->za_name[1] == 0) {
752 } else if (za->za_name[1] == '.' &&
753 za->za_name[2] == 0) {
757 zap_cursor_advance(it->ozi_zc);
759 } while (unlikely(rc == 0 && isdot));
765 * to load a directory entry at a time and stored it in
766 * iterator's in-memory data structure.
768 * \param di, struct osd_it_ea, iterator's in memory structure
770 * \retval +ve, iterator reached to end
771 * \retval 0, iterator not reached to end
772 * \retval -ve, on error
774 static int osd_dir_it_next(const struct lu_env *env, struct dt_it *di)
776 struct osd_zap_it *it = (struct osd_zap_it *)di;
777 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
780 /* temp. storage should be enough for any key supported by ZFS */
781 CLASSERT(sizeof(za->za_name) <= sizeof(it->ozi_name));
784 * the first ->next() moves the cursor to .
785 * the second ->next() moves the cursor to ..
786 * then we get to the real records and have to verify any exist
788 if (it->ozi_pos <= 2) {
794 zap_cursor_advance(it->ozi_zc);
797 * According to current API we need to return error if its last entry.
798 * zap_cursor_advance() does not return any value. So we need to call
799 * retrieve to check if there is any record. We should make
800 * changes to Iterator API to not return status for this API
802 rc = osd_index_retrieve_skip_dots(it, za);
804 if (rc == -ENOENT) /* end of dir */
810 static struct dt_key *osd_dir_it_key(const struct lu_env *env,
811 const struct dt_it *di)
813 struct osd_zap_it *it = (struct osd_zap_it *)di;
814 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
818 if (it->ozi_pos <= 1) {
820 RETURN((struct dt_key *)".");
821 } else if (it->ozi_pos == 2) {
822 RETURN((struct dt_key *)"..");
825 if ((rc = -zap_cursor_retrieve(it->ozi_zc, za)))
828 strcpy(it->ozi_name, za->za_name);
830 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 3, 90, 0)
831 if (za->za_name[0] == '.') {
832 if (za->za_name[1] == 0 || (za->za_name[1] == '.' &&
833 za->za_name[2] == 0)) {
834 /* we should not get onto . and ..
835 * stored in the directory. ->next() and
836 * other methods should prevent this
843 RETURN((struct dt_key *)it->ozi_name);
846 static int osd_dir_it_key_size(const struct lu_env *env, const struct dt_it *di)
848 struct osd_zap_it *it = (struct osd_zap_it *)di;
849 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
853 if (it->ozi_pos <= 1) {
856 } else if (it->ozi_pos == 2) {
860 if ((rc = -zap_cursor_retrieve(it->ozi_zc, za)) == 0)
861 rc = strlen(za->za_name);
863 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 3, 90, 0)
864 if (rc == 0 && za->za_name[0] == '.') {
865 if (za->za_name[1] == 0 || (za->za_name[1] == '.' &&
866 za->za_name[2] == 0)) {
867 /* we should not get onto . and ..
868 * stored in the directory. ->next() and
869 * other methods should prevent this
878 static int osd_dir_it_rec(const struct lu_env *env, const struct dt_it *di,
879 struct dt_rec *dtrec, __u32 attr)
881 struct osd_zap_it *it = (struct osd_zap_it *)di;
882 struct lu_dirent *lde = (struct lu_dirent *)dtrec;
883 struct luz_direntry *zde = &osd_oti_get(env)->oti_zde;
884 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
888 if (it->ozi_pos <= 1) {
889 lde->lde_hash = cpu_to_le64(1);
890 strcpy(lde->lde_name, ".");
891 lde->lde_namelen = cpu_to_le16(1);
892 lde->lde_fid = *lu_object_fid(&it->ozi_obj->oo_dt.do_lu);
893 lde->lde_attrs = LUDA_FID;
894 /* append lustre attributes */
895 osd_it_append_attrs(lde, attr, 1, IFTODT(S_IFDIR));
896 lde->lde_reclen = cpu_to_le16(lu_dirent_calc_size(1, attr));
900 } else if (it->ozi_pos == 2) {
901 lde->lde_hash = cpu_to_le64(2);
902 strcpy(lde->lde_name, "..");
903 lde->lde_namelen = cpu_to_le16(2);
904 lde->lde_attrs = LUDA_FID;
905 /* append lustre attributes */
906 osd_it_append_attrs(lde, attr, 2, IFTODT(S_IFDIR));
907 lde->lde_reclen = cpu_to_le16(lu_dirent_calc_size(2, attr));
908 rc = osd_find_parent_fid(env, &it->ozi_obj->oo_dt, &lde->lde_fid);
910 * early Orion code was not setting LinkEA, so it's possible
911 * some setups still have objects with no LinkEA set.
912 * but at that time .. was a real record in the directory
913 * so we should try to lookup .. in ZAP
921 lde->lde_hash = cpu_to_le64(udmu_zap_cursor_serialize(it->ozi_zc));
923 if ((rc = -zap_cursor_retrieve(it->ozi_zc, za)))
926 namelen = strlen(za->za_name);
927 if (namelen > NAME_MAX)
928 GOTO(out, rc = -EOVERFLOW);
929 strcpy(lde->lde_name, za->za_name);
930 lde->lde_namelen = cpu_to_le16(namelen);
932 if (za->za_integer_length != 8 || za->za_num_integers < 3) {
933 CERROR("%s: unsupported direntry format: %d %d\n",
934 osd_obj2dev(it->ozi_obj)->od_svname,
935 za->za_integer_length, (int)za->za_num_integers);
937 GOTO(out, rc = -EIO);
940 rc = -zap_lookup(it->ozi_zc->zc_objset, it->ozi_zc->zc_zapobj,
941 za->za_name, za->za_integer_length, 3, zde);
945 lde->lde_fid = zde->lzd_fid;
946 lde->lde_attrs = LUDA_FID;
948 /* append lustre attributes */
949 osd_it_append_attrs(lde, attr, namelen, zde->lzd_reg.zde_type);
951 lde->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
957 static __u64 osd_dir_it_store(const struct lu_env *env, const struct dt_it *di)
959 struct osd_zap_it *it = (struct osd_zap_it *)di;
963 if (it->ozi_pos <= 2)
966 pos = udmu_zap_cursor_serialize(it->ozi_zc);
973 * rc == 0 -> end of directory.
974 * rc > 0 -> ok, proceed.
975 * rc < 0 -> error. ( EOVERFLOW can be masked.)
977 static int osd_dir_it_load(const struct lu_env *env,
978 const struct dt_it *di, __u64 hash)
980 struct osd_zap_it *it = (struct osd_zap_it *)di;
981 struct osd_object *obj = it->ozi_obj;
982 struct osd_device *osd = osd_obj2dev(obj);
983 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
987 udmu_zap_cursor_fini(it->ozi_zc);
988 if (udmu_zap_cursor_init(&it->ozi_zc, &osd->od_objset,
989 obj->oo_db->db_object, hash))
997 /* to return whether the end has been reached */
998 rc = osd_index_retrieve_skip_dots(it, za);
1001 else if (rc == -ENOENT)
1008 static struct dt_index_operations osd_dir_ops = {
1009 .dio_lookup = osd_dir_lookup,
1010 .dio_declare_insert = osd_declare_dir_insert,
1011 .dio_insert = osd_dir_insert,
1012 .dio_declare_delete = osd_declare_dir_delete,
1013 .dio_delete = osd_dir_delete,
1015 .init = osd_dir_it_init,
1016 .fini = osd_index_it_fini,
1017 .get = osd_dir_it_get,
1018 .put = osd_dir_it_put,
1019 .next = osd_dir_it_next,
1020 .key = osd_dir_it_key,
1021 .key_size = osd_dir_it_key_size,
1022 .rec = osd_dir_it_rec,
1023 .store = osd_dir_it_store,
1024 .load = osd_dir_it_load
1029 * Primitives for index files using binary keys.
1030 * XXX: only 64-bit keys are supported for now.
1033 static int osd_index_lookup(const struct lu_env *env, struct dt_object *dt,
1034 struct dt_rec *rec, const struct dt_key *key,
1035 struct lustre_capa *capa)
1037 struct osd_object *obj = osd_dt_obj(dt);
1038 struct osd_device *osd = osd_obj2dev(obj);
1042 rc = -zap_lookup_uint64(osd->od_objset.os, obj->oo_db->db_object,
1043 (const __u64 *)key, 1, 8, obj->oo_recsize,
1045 RETURN(rc == 0 ? 1 : rc);
1048 static int osd_declare_index_insert(const struct lu_env *env,
1049 struct dt_object *dt,
1050 const struct dt_rec *rec,
1051 const struct dt_key *key,
1054 struct osd_object *obj = osd_dt_obj(dt);
1055 struct osd_thandle *oh;
1058 LASSERT(th != NULL);
1059 oh = container_of0(th, struct osd_thandle, ot_super);
1061 LASSERT(obj->oo_db);
1063 dmu_tx_hold_bonus(oh->ot_tx, obj->oo_db->db_object);
1065 /* It is not clear what API should be used for binary keys, so we pass
1066 * a null name which has the side effect of over-reserving space,
1067 * accounting for the worst case. See zap_count_write() */
1068 dmu_tx_hold_zap(oh->ot_tx, obj->oo_db->db_object, TRUE, NULL);
1073 static int osd_index_insert(const struct lu_env *env, struct dt_object *dt,
1074 const struct dt_rec *rec, const struct dt_key *key,
1075 struct thandle *th, struct lustre_capa *capa,
1078 struct osd_object *obj = osd_dt_obj(dt);
1079 struct osd_device *osd = osd_obj2dev(obj);
1080 struct osd_thandle *oh;
1084 LASSERT(obj->oo_db);
1085 LASSERT(dt_object_exists(dt));
1086 LASSERT(osd_invariant(obj));
1087 LASSERT(th != NULL);
1089 oh = container_of0(th, struct osd_thandle, ot_super);
1091 /* Insert (key,oid) into ZAP */
1092 rc = -zap_add_uint64(osd->od_objset.os, obj->oo_db->db_object,
1093 (const __u64 *)key, 1, 8, obj->oo_recsize,
1094 (void *)rec, oh->ot_tx);
1098 static int osd_declare_index_delete(const struct lu_env *env,
1099 struct dt_object *dt,
1100 const struct dt_key *key,
1103 struct osd_object *obj = osd_dt_obj(dt);
1104 struct osd_thandle *oh;
1107 LASSERT(dt_object_exists(dt));
1108 LASSERT(osd_invariant(obj));
1109 LASSERT(th != NULL);
1110 LASSERT(obj->oo_db);
1112 oh = container_of0(th, struct osd_thandle, ot_super);
1113 dmu_tx_hold_zap(oh->ot_tx, obj->oo_db->db_object, TRUE, NULL);
1118 static int osd_index_delete(const struct lu_env *env, struct dt_object *dt,
1119 const struct dt_key *key, struct thandle *th,
1120 struct lustre_capa *capa)
1122 struct osd_object *obj = osd_dt_obj(dt);
1123 struct osd_device *osd = osd_obj2dev(obj);
1124 struct osd_thandle *oh;
1128 LASSERT(obj->oo_db);
1129 LASSERT(th != NULL);
1130 oh = container_of0(th, struct osd_thandle, ot_super);
1132 /* Remove binary key from the ZAP */
1133 rc = -zap_remove_uint64(osd->od_objset.os, obj->oo_db->db_object,
1134 (const __u64 *)key, 1, oh->ot_tx);
1138 static int osd_index_it_get(const struct lu_env *env, struct dt_it *di,
1139 const struct dt_key *key)
1141 struct osd_zap_it *it = (struct osd_zap_it *)di;
1142 struct osd_object *obj = it->ozi_obj;
1143 struct osd_device *osd = osd_obj2dev(obj);
1147 LASSERT(it->ozi_zc);
1149 /* XXX: API is broken at the moment */
1150 LASSERT(*((const __u64 *)key) == 0);
1152 zap_cursor_fini(it->ozi_zc);
1153 memset(it->ozi_zc, 0, sizeof(*it->ozi_zc));
1154 zap_cursor_init(it->ozi_zc, osd->od_objset.os, obj->oo_db->db_object);
1160 static int osd_index_it_next(const struct lu_env *env, struct dt_it *di)
1162 struct osd_zap_it *it = (struct osd_zap_it *)di;
1163 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
1167 if (it->ozi_reset == 0)
1168 zap_cursor_advance(it->ozi_zc);
1172 * According to current API we need to return error if it's last entry.
1173 * zap_cursor_advance() does not return any value. So we need to call
1174 * retrieve to check if there is any record. We should make
1175 * changes to Iterator API to not return status for this API
1177 rc = -zap_cursor_retrieve(it->ozi_zc, za);
1184 static struct dt_key *osd_index_it_key(const struct lu_env *env,
1185 const struct dt_it *di)
1187 struct osd_zap_it *it = (struct osd_zap_it *)di;
1188 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
1193 rc = -zap_cursor_retrieve(it->ozi_zc, za);
1195 RETURN(ERR_PTR(rc));
1197 /* the binary key is stored in the name */
1198 it->ozi_key = *((__u64 *)za->za_name);
1200 RETURN((struct dt_key *)&it->ozi_key);
1203 static int osd_index_it_key_size(const struct lu_env *env,
1204 const struct dt_it *di)
1206 /* we only support 64-bit binary keys for the time being */
1207 RETURN(sizeof(__u64));
1210 static int osd_index_it_rec(const struct lu_env *env, const struct dt_it *di,
1211 struct dt_rec *rec, __u32 attr)
1213 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
1214 struct osd_zap_it *it = (struct osd_zap_it *)di;
1215 struct osd_object *obj = it->ozi_obj;
1216 struct osd_device *osd = osd_obj2dev(obj);
1221 rc = -zap_cursor_retrieve(it->ozi_zc, za);
1225 rc = -zap_lookup_uint64(osd->od_objset.os, obj->oo_db->db_object,
1226 (const __u64 *)za->za_name, 1, 8,
1227 obj->oo_recsize, (void *)rec);
1231 static __u64 osd_index_it_store(const struct lu_env *env,
1232 const struct dt_it *di)
1234 struct osd_zap_it *it = (struct osd_zap_it *)di;
1237 RETURN((__u64)zap_cursor_serialize(it->ozi_zc));
1240 static int osd_index_it_load(const struct lu_env *env, const struct dt_it *di,
1243 struct osd_zap_it *it = (struct osd_zap_it *)di;
1244 struct osd_object *obj = it->ozi_obj;
1245 struct osd_device *osd = osd_obj2dev(obj);
1246 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
1250 /* close the current cursor */
1251 zap_cursor_fini(it->ozi_zc);
1253 /* create a new one starting at hash */
1254 memset(it->ozi_zc, 0, sizeof(*it->ozi_zc));
1255 zap_cursor_init_serialized(it->ozi_zc, osd->od_objset.os,
1256 obj->oo_db->db_object, hash);
1259 rc = -zap_cursor_retrieve(it->ozi_zc, za);
1262 else if (rc == -ENOENT)
1268 static struct dt_index_operations osd_index_ops = {
1269 .dio_lookup = osd_index_lookup,
1270 .dio_declare_insert = osd_declare_index_insert,
1271 .dio_insert = osd_index_insert,
1272 .dio_declare_delete = osd_declare_index_delete,
1273 .dio_delete = osd_index_delete,
1275 .init = osd_index_it_init,
1276 .fini = osd_index_it_fini,
1277 .get = osd_index_it_get,
1278 .put = osd_index_it_put,
1279 .next = osd_index_it_next,
1280 .key = osd_index_it_key,
1281 .key_size = osd_index_it_key_size,
1282 .rec = osd_index_it_rec,
1283 .store = osd_index_it_store,
1284 .load = osd_index_it_load
1288 int osd_index_try(const struct lu_env *env, struct dt_object *dt,
1289 const struct dt_index_features *feat)
1291 struct osd_object *obj = osd_dt_obj(dt);
1294 LASSERT(dt_object_exists(dt));
1297 * XXX: implement support for fixed-size keys sorted with natural
1298 * numerical way (not using internal hash value)
1300 if (feat->dif_flags & DT_IND_RANGE)
1303 if (unlikely(feat == &dt_otable_features))
1304 /* do not support oi scrub yet. */
1307 LASSERT(obj->oo_db != NULL);
1308 if (likely(feat == &dt_directory_features)) {
1309 if (udmu_object_is_zap(obj->oo_db))
1310 dt->do_index_ops = &osd_dir_ops;
1313 } else if (unlikely(feat == &dt_acct_features)) {
1314 LASSERT(fid_is_acct(lu_object_fid(&dt->do_lu)));
1315 dt->do_index_ops = &osd_acct_index_ops;
1316 } else if (udmu_object_is_zap(obj->oo_db) &&
1317 dt->do_index_ops == NULL) {
1318 /* For index file, we don't support variable key & record sizes
1319 * and the key has to be unique */
1320 if ((feat->dif_flags & ~DT_IND_UPDATE) != 0)
1323 /* Although the zap_*_uint64() primitives support large keys, we
1324 * limit ourselves to 64-bit keys for now */
1325 if (feat->dif_keysize_max != sizeof(__u64) ||
1326 feat->dif_keysize_min != sizeof(__u64))
1329 /* As for the record size, it should be a multiple of 8 bytes
1330 * and smaller than the maximum value length supported by ZAP.
1332 if (feat->dif_recsize_max > ZAP_MAXVALUELEN)
1334 if (feat->dif_recsize_max != feat->dif_recsize_min ||
1335 (feat->dif_recsize_max & (sizeof(__u64) - 1)))
1338 obj->oo_recsize = feat->dif_recsize_max / sizeof(__u64);
1339 dt->do_index_ops = &osd_index_ops;