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);
484 LASSERT(ss->ss_server_fld != NULL);
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,
499 struct seq_server_site *ss = osd_seq_site(osd);
502 /* FID seqs not in FLDB, must be local seq */
503 if (unlikely(!fid_seq_in_fldb(fid_seq(fid))))
506 /* If FLD is not being initialized yet, it only happens during the
507 * initialization, likely during mgs initialization, and we assume
508 * this is local FID. */
509 if (ss == NULL || ss->ss_server_fld == NULL)
512 /* Only check the local FLDB here */
513 if (osd_seq_exists(env, osd, fid_seq(fid)))
520 * Inserts (key, value) pair in \a directory object.
522 * \param dt osd index object
523 * \param key key for index
524 * \param rec record reference
525 * \param th transaction handler
526 * \param capa capability descriptor
527 * \param ignore_quota update should not affect quota
530 * \retval -ve failure
532 static int osd_dir_insert(const struct lu_env *env, struct dt_object *dt,
533 const struct dt_rec *rec, const struct dt_key *key,
534 struct thandle *th, struct lustre_capa *capa,
537 struct osd_thread_info *oti = osd_oti_get(env);
538 struct osd_object *parent = osd_dt_obj(dt);
539 struct osd_device *osd = osd_obj2dev(parent);
540 struct lu_fid *fid = (struct lu_fid *)rec;
541 struct osd_thandle *oh;
542 struct osd_object *child = NULL;
544 char *name = (char *)key;
548 LASSERT(parent->oo_db);
549 LASSERT(udmu_object_is_zap(parent->oo_db));
551 LASSERT(dt_object_exists(dt));
552 LASSERT(osd_invariant(parent));
555 oh = container_of0(th, struct osd_thandle, ot_super);
557 rc = osd_remote_fid(env, osd, fid);
559 CERROR("%s: Can not find object "DFID": rc = %d\n",
560 osd->od_svname, PFID(fid), rc);
564 if (unlikely(rc == 1)) {
565 /* Insert remote entry */
566 memset(&oti->oti_zde.lzd_reg, 0, sizeof(oti->oti_zde.lzd_reg));
567 oti->oti_zde.lzd_reg.zde_type = IFTODT(S_IFDIR & S_IFMT);
570 * To simulate old Orion setups with ./.. stored in the
573 /* Insert local entry */
574 child = osd_object_find(env, dt, fid);
576 RETURN(PTR_ERR(child));
578 LASSERT(child->oo_db);
579 if (name[0] == '.') {
581 /* do not store ".", instead generate it
582 * during iteration */
584 } else if (name[1] == '.' && name[2] == 0) {
585 /* update parent dnode in the child.
586 * later it will be used to generate ".." */
587 udmu_objset_t *uos = &osd->od_objset;
588 rc = osd_object_sa_update(parent,
590 &child->oo_db->db_object,
595 CLASSERT(sizeof(oti->oti_zde.lzd_reg) == 8);
596 CLASSERT(sizeof(oti->oti_zde) % 8 == 0);
597 attr = child->oo_dt.do_lu.lo_header ->loh_attr;
598 oti->oti_zde.lzd_reg.zde_type = IFTODT(attr & S_IFMT);
599 oti->oti_zde.lzd_reg.zde_dnode = child->oo_db->db_object;
602 oti->oti_zde.lzd_fid = *fid;
603 /* Insert (key,oid) into ZAP */
604 rc = -zap_add(osd->od_objset.os, parent->oo_db->db_object,
605 (char *)key, 8, sizeof(oti->oti_zde) / 8,
606 (void *)&oti->oti_zde, oh->ot_tx);
610 osd_object_put(env, child);
615 static int osd_declare_dir_delete(const struct lu_env *env,
616 struct dt_object *dt,
617 const struct dt_key *key,
620 struct osd_object *obj = osd_dt_obj(dt);
621 struct osd_thandle *oh;
624 LASSERT(dt_object_exists(dt));
625 LASSERT(osd_invariant(obj));
628 oh = container_of0(th, struct osd_thandle, ot_super);
631 LASSERT(udmu_object_is_zap(obj->oo_db));
633 dmu_tx_hold_zap(oh->ot_tx, obj->oo_db->db_object, TRUE, (char *)key);
638 static int osd_dir_delete(const struct lu_env *env, struct dt_object *dt,
639 const struct dt_key *key, struct thandle *th,
640 struct lustre_capa *capa)
642 struct osd_object *obj = osd_dt_obj(dt);
643 struct osd_device *osd = osd_obj2dev(obj);
644 struct osd_thandle *oh;
645 dmu_buf_t *zap_db = obj->oo_db;
646 char *name = (char *)key;
651 LASSERT(udmu_object_is_zap(obj->oo_db));
654 oh = container_of0(th, struct osd_thandle, ot_super);
657 * In Orion . and .. were stored in the directory (not generated upon
658 * request as now). we preserve them for backward compatibility
660 if (name[0] == '.') {
663 } else if (name[1] == '.' && name[2] == 0) {
668 /* Remove key from the ZAP */
669 rc = -zap_remove(osd->od_objset.os, zap_db->db_object,
670 (char *) key, oh->ot_tx);
672 #if LUSTRE_VERSION_CODE <= OBD_OCD_VERSION(2, 4, 53, 0)
673 if (unlikely(rc == -ENOENT && name[0] == '.' &&
674 (name[1] == 0 || (name[1] == '.' && name[2] == 0))))
677 if (unlikely(rc && rc != -ENOENT))
678 CERROR("%s: zap_remove failed: rc = %d\n", osd->od_svname, rc);
683 static struct dt_it *osd_dir_it_init(const struct lu_env *env,
684 struct dt_object *dt,
686 struct lustre_capa *capa)
688 struct osd_zap_it *it;
690 it = (struct osd_zap_it *)osd_index_it_init(env, dt, unused, capa);
694 RETURN((struct dt_it *)it);
698 * Move Iterator to record specified by \a key
700 * \param di osd iterator
701 * \param key key for index
703 * \retval +ve di points to record with least key not larger than key
704 * \retval 0 di points to exact matched key
705 * \retval -ve failure
707 static int osd_dir_it_get(const struct lu_env *env,
708 struct dt_it *di, const struct dt_key *key)
710 struct osd_zap_it *it = (struct osd_zap_it *)di;
711 struct osd_object *obj = it->ozi_obj;
712 struct osd_device *osd = osd_obj2dev(obj);
713 char *name = (char *)key;
720 udmu_zap_cursor_fini(it->ozi_zc);
722 if (udmu_zap_cursor_init(&it->ozi_zc, &osd->od_objset,
723 obj->oo_db->db_object, 0))
726 /* XXX: implementation of the API is broken at the moment */
727 LASSERT(((const char *)key)[0] == 0);
734 if (name[0] == '.') {
738 } else if (name[1] == '.' && name[2] == 0) {
744 /* neither . nor .. - some real record */
752 static void osd_dir_it_put(const struct lu_env *env, struct dt_it *di)
754 /* PBS: do nothing : ref are incremented at retrive and decreamented
759 * in Orion . and .. were stored in the directory, while ZPL
760 * and current osd-zfs generate them up on request. so, we
761 * need to ignore previously stored . and ..
763 static int osd_index_retrieve_skip_dots(struct osd_zap_it *it,
769 rc = -zap_cursor_retrieve(it->ozi_zc, za);
772 if (unlikely(rc == 0 && za->za_name[0] == '.')) {
773 if (za->za_name[1] == 0) {
775 } else if (za->za_name[1] == '.' &&
776 za->za_name[2] == 0) {
780 zap_cursor_advance(it->ozi_zc);
782 } while (unlikely(rc == 0 && isdot));
788 * to load a directory entry at a time and stored it in
789 * iterator's in-memory data structure.
791 * \param di, struct osd_it_ea, iterator's in memory structure
793 * \retval +ve, iterator reached to end
794 * \retval 0, iterator not reached to end
795 * \retval -ve, on error
797 static int osd_dir_it_next(const struct lu_env *env, struct dt_it *di)
799 struct osd_zap_it *it = (struct osd_zap_it *)di;
800 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
803 /* temp. storage should be enough for any key supported by ZFS */
804 CLASSERT(sizeof(za->za_name) <= sizeof(it->ozi_name));
807 * the first ->next() moves the cursor to .
808 * the second ->next() moves the cursor to ..
809 * then we get to the real records and have to verify any exist
811 if (it->ozi_pos <= 2) {
817 zap_cursor_advance(it->ozi_zc);
820 * According to current API we need to return error if its last entry.
821 * zap_cursor_advance() does not return any value. So we need to call
822 * retrieve to check if there is any record. We should make
823 * changes to Iterator API to not return status for this API
825 rc = osd_index_retrieve_skip_dots(it, za);
827 if (rc == -ENOENT) /* end of dir */
833 static struct dt_key *osd_dir_it_key(const struct lu_env *env,
834 const struct dt_it *di)
836 struct osd_zap_it *it = (struct osd_zap_it *)di;
837 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
841 if (it->ozi_pos <= 1) {
843 RETURN((struct dt_key *)".");
844 } else if (it->ozi_pos == 2) {
845 RETURN((struct dt_key *)"..");
848 if ((rc = -zap_cursor_retrieve(it->ozi_zc, za)))
851 strcpy(it->ozi_name, za->za_name);
853 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 3, 90, 0)
854 if (za->za_name[0] == '.') {
855 if (za->za_name[1] == 0 || (za->za_name[1] == '.' &&
856 za->za_name[2] == 0)) {
857 /* we should not get onto . and ..
858 * stored in the directory. ->next() and
859 * other methods should prevent this
866 RETURN((struct dt_key *)it->ozi_name);
869 static int osd_dir_it_key_size(const struct lu_env *env, const struct dt_it *di)
871 struct osd_zap_it *it = (struct osd_zap_it *)di;
872 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
876 if (it->ozi_pos <= 1) {
879 } else if (it->ozi_pos == 2) {
883 if ((rc = -zap_cursor_retrieve(it->ozi_zc, za)) == 0)
884 rc = strlen(za->za_name);
886 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 3, 90, 0)
887 if (rc == 0 && za->za_name[0] == '.') {
888 if (za->za_name[1] == 0 || (za->za_name[1] == '.' &&
889 za->za_name[2] == 0)) {
890 /* we should not get onto . and ..
891 * stored in the directory. ->next() and
892 * other methods should prevent this
901 static int osd_dir_it_rec(const struct lu_env *env, const struct dt_it *di,
902 struct dt_rec *dtrec, __u32 attr)
904 struct osd_zap_it *it = (struct osd_zap_it *)di;
905 struct lu_dirent *lde = (struct lu_dirent *)dtrec;
906 struct luz_direntry *zde = &osd_oti_get(env)->oti_zde;
907 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
911 if (it->ozi_pos <= 1) {
912 lde->lde_hash = cpu_to_le64(1);
913 strcpy(lde->lde_name, ".");
914 lde->lde_namelen = cpu_to_le16(1);
915 lde->lde_fid = *lu_object_fid(&it->ozi_obj->oo_dt.do_lu);
916 lde->lde_attrs = LUDA_FID;
917 /* append lustre attributes */
918 osd_it_append_attrs(lde, attr, 1, IFTODT(S_IFDIR));
919 lde->lde_reclen = cpu_to_le16(lu_dirent_calc_size(1, attr));
923 } else if (it->ozi_pos == 2) {
924 lde->lde_hash = cpu_to_le64(2);
925 strcpy(lde->lde_name, "..");
926 lde->lde_namelen = cpu_to_le16(2);
927 lde->lde_attrs = LUDA_FID;
928 /* append lustre attributes */
929 osd_it_append_attrs(lde, attr, 2, IFTODT(S_IFDIR));
930 lde->lde_reclen = cpu_to_le16(lu_dirent_calc_size(2, attr));
931 rc = osd_find_parent_fid(env, &it->ozi_obj->oo_dt, &lde->lde_fid);
933 * early Orion code was not setting LinkEA, so it's possible
934 * some setups still have objects with no LinkEA set.
935 * but at that time .. was a real record in the directory
936 * so we should try to lookup .. in ZAP
944 rc = -zap_cursor_retrieve(it->ozi_zc, za);
945 if (unlikely(rc != 0))
948 lde->lde_hash = cpu_to_le64(udmu_zap_cursor_serialize(it->ozi_zc));
949 namelen = strlen(za->za_name);
950 if (namelen > NAME_MAX)
951 GOTO(out, rc = -EOVERFLOW);
952 strcpy(lde->lde_name, za->za_name);
953 lde->lde_namelen = cpu_to_le16(namelen);
955 if (za->za_integer_length != 8 || za->za_num_integers < 3) {
956 CERROR("%s: unsupported direntry format: %d %d\n",
957 osd_obj2dev(it->ozi_obj)->od_svname,
958 za->za_integer_length, (int)za->za_num_integers);
960 GOTO(out, rc = -EIO);
963 rc = -zap_lookup(it->ozi_zc->zc_objset, it->ozi_zc->zc_zapobj,
964 za->za_name, za->za_integer_length, 3, zde);
968 lde->lde_fid = zde->lzd_fid;
969 lde->lde_attrs = LUDA_FID;
971 /* append lustre attributes */
972 osd_it_append_attrs(lde, attr, namelen, zde->lzd_reg.zde_type);
974 lde->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
980 static int osd_dir_it_rec_size(const struct lu_env *env, const struct dt_it *di,
983 struct osd_zap_it *it = (struct osd_zap_it *)di;
984 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
988 if (it->ozi_pos <= 1)
990 else if (it->ozi_pos == 2)
994 rc = lu_dirent_calc_size(namelen, attr);
998 rc = -zap_cursor_retrieve(it->ozi_zc, za);
999 if (unlikely(rc != 0))
1002 if (za->za_integer_length != 8 || za->za_num_integers < 3) {
1003 CERROR("%s: unsupported direntry format: %d %d\n",
1004 osd_obj2dev(it->ozi_obj)->od_svname,
1005 za->za_integer_length, (int)za->za_num_integers);
1009 namelen = strlen(za->za_name);
1010 if (namelen > NAME_MAX)
1013 rc = lu_dirent_calc_size(namelen, attr);
1018 static __u64 osd_dir_it_store(const struct lu_env *env, const struct dt_it *di)
1020 struct osd_zap_it *it = (struct osd_zap_it *)di;
1024 if (it->ozi_pos <= 2)
1027 pos = udmu_zap_cursor_serialize(it->ozi_zc);
1034 * rc == 0 -> end of directory.
1035 * rc > 0 -> ok, proceed.
1036 * rc < 0 -> error. ( EOVERFLOW can be masked.)
1038 static int osd_dir_it_load(const struct lu_env *env,
1039 const struct dt_it *di, __u64 hash)
1041 struct osd_zap_it *it = (struct osd_zap_it *)di;
1042 struct osd_object *obj = it->ozi_obj;
1043 struct osd_device *osd = osd_obj2dev(obj);
1044 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
1048 udmu_zap_cursor_fini(it->ozi_zc);
1049 if (udmu_zap_cursor_init(&it->ozi_zc, &osd->od_objset,
1050 obj->oo_db->db_object, hash))
1058 /* to return whether the end has been reached */
1059 rc = osd_index_retrieve_skip_dots(it, za);
1062 else if (rc == -ENOENT)
1069 static struct dt_index_operations osd_dir_ops = {
1070 .dio_lookup = osd_dir_lookup,
1071 .dio_declare_insert = osd_declare_dir_insert,
1072 .dio_insert = osd_dir_insert,
1073 .dio_declare_delete = osd_declare_dir_delete,
1074 .dio_delete = osd_dir_delete,
1076 .init = osd_dir_it_init,
1077 .fini = osd_index_it_fini,
1078 .get = osd_dir_it_get,
1079 .put = osd_dir_it_put,
1080 .next = osd_dir_it_next,
1081 .key = osd_dir_it_key,
1082 .key_size = osd_dir_it_key_size,
1083 .rec = osd_dir_it_rec,
1084 .rec_size = osd_dir_it_rec_size,
1085 .store = osd_dir_it_store,
1086 .load = osd_dir_it_load
1091 * Primitives for index files using binary keys.
1094 /* key integer_size is 8 */
1095 static int osd_prepare_key_uint64(struct osd_object *o, __u64 *dst,
1096 const struct dt_key *src)
1103 /* align keysize to 64bit */
1104 size = (o->oo_keysize + sizeof(__u64) - 1) / sizeof(__u64);
1105 size *= sizeof(__u64);
1107 LASSERT(size <= MAXNAMELEN);
1109 if (unlikely(size > o->oo_keysize))
1110 memset(dst + o->oo_keysize, 0, size - o->oo_keysize);
1111 memcpy(dst, (const char *)src, o->oo_keysize);
1113 return (size/sizeof(__u64));
1116 static int osd_index_lookup(const struct lu_env *env, struct dt_object *dt,
1117 struct dt_rec *rec, const struct dt_key *key,
1118 struct lustre_capa *capa)
1120 struct osd_object *obj = osd_dt_obj(dt);
1121 struct osd_device *osd = osd_obj2dev(obj);
1122 __u64 *k = osd_oti_get(env)->oti_key64;
1126 rc = osd_prepare_key_uint64(obj, k, key);
1128 rc = -zap_lookup_uint64(osd->od_objset.os, obj->oo_db->db_object,
1129 k, rc, obj->oo_recusize, obj->oo_recsize,
1131 RETURN(rc == 0 ? 1 : rc);
1134 static int osd_declare_index_insert(const struct lu_env *env,
1135 struct dt_object *dt,
1136 const struct dt_rec *rec,
1137 const struct dt_key *key,
1140 struct osd_object *obj = osd_dt_obj(dt);
1141 struct osd_thandle *oh;
1144 LASSERT(th != NULL);
1145 oh = container_of0(th, struct osd_thandle, ot_super);
1147 LASSERT(obj->oo_db);
1149 dmu_tx_hold_bonus(oh->ot_tx, obj->oo_db->db_object);
1151 /* It is not clear what API should be used for binary keys, so we pass
1152 * a null name which has the side effect of over-reserving space,
1153 * accounting for the worst case. See zap_count_write() */
1154 dmu_tx_hold_zap(oh->ot_tx, obj->oo_db->db_object, TRUE, NULL);
1159 static int osd_index_insert(const struct lu_env *env, struct dt_object *dt,
1160 const struct dt_rec *rec, const struct dt_key *key,
1161 struct thandle *th, struct lustre_capa *capa,
1164 struct osd_object *obj = osd_dt_obj(dt);
1165 struct osd_device *osd = osd_obj2dev(obj);
1166 struct osd_thandle *oh;
1167 __u64 *k = osd_oti_get(env)->oti_key64;
1171 LASSERT(obj->oo_db);
1172 LASSERT(dt_object_exists(dt));
1173 LASSERT(osd_invariant(obj));
1174 LASSERT(th != NULL);
1176 oh = container_of0(th, struct osd_thandle, ot_super);
1178 rc = osd_prepare_key_uint64(obj, k, key);
1180 /* Insert (key,oid) into ZAP */
1181 rc = -zap_add_uint64(osd->od_objset.os, obj->oo_db->db_object,
1182 k, rc, obj->oo_recusize, obj->oo_recsize,
1183 (void *)rec, oh->ot_tx);
1187 static int osd_declare_index_delete(const struct lu_env *env,
1188 struct dt_object *dt,
1189 const struct dt_key *key,
1192 struct osd_object *obj = osd_dt_obj(dt);
1193 struct osd_thandle *oh;
1196 LASSERT(dt_object_exists(dt));
1197 LASSERT(osd_invariant(obj));
1198 LASSERT(th != NULL);
1199 LASSERT(obj->oo_db);
1201 oh = container_of0(th, struct osd_thandle, ot_super);
1202 dmu_tx_hold_zap(oh->ot_tx, obj->oo_db->db_object, TRUE, NULL);
1207 static int osd_index_delete(const struct lu_env *env, struct dt_object *dt,
1208 const struct dt_key *key, struct thandle *th,
1209 struct lustre_capa *capa)
1211 struct osd_object *obj = osd_dt_obj(dt);
1212 struct osd_device *osd = osd_obj2dev(obj);
1213 struct osd_thandle *oh;
1214 __u64 *k = osd_oti_get(env)->oti_key64;
1218 LASSERT(obj->oo_db);
1219 LASSERT(th != NULL);
1220 oh = container_of0(th, struct osd_thandle, ot_super);
1222 rc = osd_prepare_key_uint64(obj, k, key);
1224 /* Remove binary key from the ZAP */
1225 rc = -zap_remove_uint64(osd->od_objset.os, obj->oo_db->db_object,
1230 static int osd_index_it_get(const struct lu_env *env, struct dt_it *di,
1231 const struct dt_key *key)
1233 struct osd_zap_it *it = (struct osd_zap_it *)di;
1234 struct osd_object *obj = it->ozi_obj;
1235 struct osd_device *osd = osd_obj2dev(obj);
1239 LASSERT(it->ozi_zc);
1242 * XXX: we need a binary version of zap_cursor_move_to_key()
1243 * to implement this API */
1244 if (*((const __u64 *)key) != 0)
1245 CERROR("NOT IMPLEMETED YET (move to "LPX64")\n",
1248 zap_cursor_fini(it->ozi_zc);
1249 memset(it->ozi_zc, 0, sizeof(*it->ozi_zc));
1250 zap_cursor_init(it->ozi_zc, osd->od_objset.os, obj->oo_db->db_object);
1256 static int osd_index_it_next(const struct lu_env *env, struct dt_it *di)
1258 struct osd_zap_it *it = (struct osd_zap_it *)di;
1259 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
1263 if (it->ozi_reset == 0)
1264 zap_cursor_advance(it->ozi_zc);
1268 * According to current API we need to return error if it's last entry.
1269 * zap_cursor_advance() does not return any value. So we need to call
1270 * retrieve to check if there is any record. We should make
1271 * changes to Iterator API to not return status for this API
1273 rc = -zap_cursor_retrieve(it->ozi_zc, za);
1280 static struct dt_key *osd_index_it_key(const struct lu_env *env,
1281 const struct dt_it *di)
1283 struct osd_zap_it *it = (struct osd_zap_it *)di;
1284 struct osd_object *obj = it->ozi_obj;
1285 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
1290 rc = -zap_cursor_retrieve(it->ozi_zc, za);
1292 RETURN(ERR_PTR(rc));
1294 /* the binary key is stored in the name */
1295 memcpy(&it->ozi_key, za->za_name, obj->oo_keysize);
1297 RETURN((struct dt_key *)&it->ozi_key);
1300 static int osd_index_it_key_size(const struct lu_env *env,
1301 const struct dt_it *di)
1303 struct osd_zap_it *it = (struct osd_zap_it *)di;
1304 struct osd_object *obj = it->ozi_obj;
1305 RETURN(obj->oo_keysize);
1308 static int osd_index_it_rec(const struct lu_env *env, const struct dt_it *di,
1309 struct dt_rec *rec, __u32 attr)
1311 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
1312 struct osd_zap_it *it = (struct osd_zap_it *)di;
1313 struct osd_object *obj = it->ozi_obj;
1314 struct osd_device *osd = osd_obj2dev(obj);
1315 __u64 *k = osd_oti_get(env)->oti_key64;
1320 rc = -zap_cursor_retrieve(it->ozi_zc, za);
1324 rc = osd_prepare_key_uint64(obj, k, (const struct dt_key *)za->za_name);
1326 rc = -zap_lookup_uint64(osd->od_objset.os, obj->oo_db->db_object,
1327 k, rc, obj->oo_recusize, obj->oo_recsize,
1332 static __u64 osd_index_it_store(const struct lu_env *env,
1333 const struct dt_it *di)
1335 struct osd_zap_it *it = (struct osd_zap_it *)di;
1338 RETURN((__u64)zap_cursor_serialize(it->ozi_zc));
1341 static int osd_index_it_load(const struct lu_env *env, const struct dt_it *di,
1344 struct osd_zap_it *it = (struct osd_zap_it *)di;
1345 struct osd_object *obj = it->ozi_obj;
1346 struct osd_device *osd = osd_obj2dev(obj);
1347 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
1351 /* close the current cursor */
1352 zap_cursor_fini(it->ozi_zc);
1354 /* create a new one starting at hash */
1355 memset(it->ozi_zc, 0, sizeof(*it->ozi_zc));
1356 zap_cursor_init_serialized(it->ozi_zc, osd->od_objset.os,
1357 obj->oo_db->db_object, hash);
1360 rc = -zap_cursor_retrieve(it->ozi_zc, za);
1363 else if (rc == -ENOENT)
1369 static struct dt_index_operations osd_index_ops = {
1370 .dio_lookup = osd_index_lookup,
1371 .dio_declare_insert = osd_declare_index_insert,
1372 .dio_insert = osd_index_insert,
1373 .dio_declare_delete = osd_declare_index_delete,
1374 .dio_delete = osd_index_delete,
1376 .init = osd_index_it_init,
1377 .fini = osd_index_it_fini,
1378 .get = osd_index_it_get,
1379 .put = osd_index_it_put,
1380 .next = osd_index_it_next,
1381 .key = osd_index_it_key,
1382 .key_size = osd_index_it_key_size,
1383 .rec = osd_index_it_rec,
1384 .store = osd_index_it_store,
1385 .load = osd_index_it_load
1389 struct osd_metadnode_it {
1390 struct osd_device *mit_dev;
1392 struct lu_fid mit_fid;
1394 __u64 mit_prefetched_dnode;
1397 static struct dt_it *osd_zfs_otable_it_init(const struct lu_env *env,
1398 struct dt_object *dt, __u32 attr,
1399 struct lustre_capa *capa)
1401 struct osd_device *dev = osd_dev(dt->do_lu.lo_dev);
1402 struct osd_metadnode_it *it;
1406 if (unlikely(it == NULL))
1407 RETURN(ERR_PTR(-ENOMEM));
1411 /* XXX: dmu_object_next() does NOT find dnodes allocated
1412 * in the current non-committed txg, so we force txg
1413 * commit to find all existing dnodes ... */
1414 txg_wait_synced(dmu_objset_pool(dev->od_objset.os), 0ULL);
1416 RETURN((struct dt_it *)it);
1419 static void osd_zfs_otable_it_fini(const struct lu_env *env, struct dt_it *di)
1421 struct osd_metadnode_it *it = (struct osd_metadnode_it *)di;
1426 static int osd_zfs_otable_it_get(const struct lu_env *env,
1427 struct dt_it *di, const struct dt_key *key)
1432 static void osd_zfs_otable_it_put(const struct lu_env *env, struct dt_it *di)
1436 #define OTABLE_PREFETCH 256
1438 static void osd_zfs_otable_prefetch(const struct lu_env *env,
1439 struct osd_metadnode_it *it)
1441 struct osd_device *dev = it->mit_dev;
1442 udmu_objset_t *uos = &dev->od_objset;
1445 /* can go negative on the very first access to the iterator
1446 * or if some non-Lustre objects were found */
1447 if (unlikely(it->mit_prefetched < 0))
1448 it->mit_prefetched = 0;
1450 if (it->mit_prefetched >= (OTABLE_PREFETCH >> 1))
1453 if (it->mit_prefetched_dnode == 0)
1454 it->mit_prefetched_dnode = it->mit_pos;
1456 while (it->mit_prefetched < OTABLE_PREFETCH) {
1457 rc = -dmu_object_next(uos->os, &it->mit_prefetched_dnode,
1459 if (unlikely(rc != 0))
1462 /* dmu_prefetch() was exported in 0.6.2, if you use with
1463 * an older release, just comment it out - this is an
1465 dmu_prefetch(uos->os, it->mit_prefetched_dnode, 0, 0);
1467 it->mit_prefetched++;
1471 static int osd_zfs_otable_it_next(const struct lu_env *env, struct dt_it *di)
1473 struct osd_metadnode_it *it = (struct osd_metadnode_it *)di;
1474 struct lustre_mdt_attrs *lma;
1475 struct osd_device *dev = it->mit_dev;
1476 udmu_objset_t *uos = &dev->od_objset;
1477 nvlist_t *nvbuf = NULL;
1482 memset(&it->mit_fid, 0, sizeof(it->mit_fid));
1484 dnode = it->mit_pos;
1486 rc = -dmu_object_next(uos->os, &it->mit_pos, B_FALSE, 0);
1487 if (unlikely(rc != 0))
1489 it->mit_prefetched--;
1491 /* LMA is required for this to be a Lustre object.
1492 * If there is no xattr skip it. */
1493 rc = __osd_xattr_load(uos, it->mit_pos, &nvbuf);
1494 if (unlikely(rc != 0))
1497 LASSERT(nvbuf != NULL);
1498 rc = -nvlist_lookup_byte_array(nvbuf, XATTR_NAME_LMA, &v, &s);
1499 if (likely(rc == 0)) {
1501 lma = (struct lustre_mdt_attrs *)v;
1502 lustre_lma_swab(lma);
1503 it->mit_fid = lma->lma_self_fid;
1507 /* not a Lustre object, try next one */
1514 /* we aren't prefetching in the above loop because the number of
1515 * non-Lustre objects is very small and we will be repeating very
1516 * rare. in case we want to use this to iterate over non-Lustre
1517 * objects (i.e. when we convert regular ZFS in Lustre) it makes
1518 * sense to initiate prefetching in the loop */
1520 /* 0 - there are more items, +1 - the end */
1521 if (likely(rc == 0))
1522 osd_zfs_otable_prefetch(env, it);
1524 CDEBUG(D_OTHER, "advance: %llu -> %llu "DFID": %d\n", dnode,
1525 it->mit_pos, PFID(&it->mit_fid), rc);
1531 static struct dt_key *osd_zfs_otable_it_key(const struct lu_env *env,
1532 const struct dt_it *di)
1537 static int osd_zfs_otable_it_key_size(const struct lu_env *env,
1538 const struct dt_it *di)
1540 return sizeof(__u64);
1543 static int osd_zfs_otable_it_rec(const struct lu_env *env,
1544 const struct dt_it *di,
1545 struct dt_rec *rec, __u32 attr)
1547 struct osd_metadnode_it *it = (struct osd_metadnode_it *)di;
1548 struct lu_fid *fid = (struct lu_fid *)rec;
1557 static __u64 osd_zfs_otable_it_store(const struct lu_env *env,
1558 const struct dt_it *di)
1560 struct osd_metadnode_it *it = (struct osd_metadnode_it *)di;
1565 static int osd_zfs_otable_it_load(const struct lu_env *env,
1566 const struct dt_it *di, __u64 hash)
1568 struct osd_metadnode_it *it = (struct osd_metadnode_it *)di;
1571 it->mit_prefetched = 0;
1572 it->mit_prefetched_dnode = 0;
1574 return osd_zfs_otable_it_next(env, (struct dt_it *)di);
1577 static int osd_zfs_otable_it_key_rec(const struct lu_env *env,
1578 const struct dt_it *di, void *key_rec)
1583 const struct dt_index_operations osd_zfs_otable_ops = {
1585 .init = osd_zfs_otable_it_init,
1586 .fini = osd_zfs_otable_it_fini,
1587 .get = osd_zfs_otable_it_get,
1588 .put = osd_zfs_otable_it_put,
1589 .next = osd_zfs_otable_it_next,
1590 .key = osd_zfs_otable_it_key,
1591 .key_size = osd_zfs_otable_it_key_size,
1592 .rec = osd_zfs_otable_it_rec,
1593 .store = osd_zfs_otable_it_store,
1594 .load = osd_zfs_otable_it_load,
1595 .key_rec = osd_zfs_otable_it_key_rec,
1599 int osd_index_try(const struct lu_env *env, struct dt_object *dt,
1600 const struct dt_index_features *feat)
1602 struct osd_object *obj = osd_dt_obj(dt);
1605 LASSERT(dt_object_exists(dt));
1608 * XXX: implement support for fixed-size keys sorted with natural
1609 * numerical way (not using internal hash value)
1611 if (feat->dif_flags & DT_IND_RANGE)
1614 if (unlikely(feat == &dt_otable_features)) {
1615 dt->do_index_ops = &osd_zfs_otable_ops;
1619 LASSERT(obj->oo_db != NULL);
1620 if (likely(feat == &dt_directory_features)) {
1621 if (udmu_object_is_zap(obj->oo_db))
1622 dt->do_index_ops = &osd_dir_ops;
1625 } else if (unlikely(feat == &dt_acct_features)) {
1626 LASSERT(fid_is_acct(lu_object_fid(&dt->do_lu)));
1627 dt->do_index_ops = &osd_acct_index_ops;
1628 } else if (udmu_object_is_zap(obj->oo_db) &&
1629 dt->do_index_ops == NULL) {
1630 /* For index file, we don't support variable key & record sizes
1631 * and the key has to be unique */
1632 if ((feat->dif_flags & ~DT_IND_UPDATE) != 0)
1635 if (feat->dif_keysize_max > ZAP_MAXNAMELEN)
1637 if (feat->dif_keysize_max != feat->dif_keysize_min)
1640 /* As for the record size, it should be a multiple of 8 bytes
1641 * and smaller than the maximum value length supported by ZAP.
1643 if (feat->dif_recsize_max > ZAP_MAXVALUELEN)
1645 if (feat->dif_recsize_max != feat->dif_recsize_min)
1648 obj->oo_keysize = feat->dif_keysize_max;
1649 obj->oo_recsize = feat->dif_recsize_max;
1650 obj->oo_recusize = 1;
1652 /* ZFS prefers to work with array of 64bits */
1653 if ((obj->oo_recsize & 7) == 0) {
1654 obj->oo_recsize >>= 3;
1655 obj->oo_recusize = 8;
1657 dt->do_index_ops = &osd_index_ops;