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.
30 * Copyright (c) 2012, 2014, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/osd-zfs/osd_index.c
38 * Author: Alex Zhuravlev <bzzz@whamcloud.com>
39 * Author: Mike Pershin <tappro@whamcloud.com>
42 #define DEBUG_SUBSYSTEM S_OSD
44 #include <lustre_ver.h>
45 #include <libcfs/libcfs.h>
46 #include <obd_support.h>
47 #include <lustre_net.h>
49 #include <obd_class.h>
50 #include <lustre_disk.h>
51 #include <lustre_fid.h>
53 #include "osd_internal.h"
55 #include <sys/dnode.h>
60 #include <sys/spa_impl.h>
61 #include <sys/zfs_znode.h>
62 #include <sys/dmu_tx.h>
63 #include <sys/dmu_objset.h>
64 #include <sys/dsl_prop.h>
65 #include <sys/sa_impl.h>
68 static inline int osd_object_is_zap(dmu_buf_t *db)
70 dmu_buf_impl_t *dbi = (dmu_buf_impl_t *) db;
76 rc = (dn->dn_type == DMU_OT_DIRECTORY_CONTENTS ||
77 dn->dn_type == DMU_OT_USERGROUP_USED);
83 /* We don't actually have direct access to the zap_hashbits() function
84 * so just pretend like we do for now. If this ever breaks we can look at
86 #define zap_hashbits(zc) 48
89 * | cd (16 bits) | hash (48 bits) |
90 * we need it in other form:
91 * |0| hash (48 bit) | cd (15 bit) |
92 * to be a full 64-bit ordered hash so that Lustre readdir can use it to merge
93 * the readdir hashes from multiple directory stripes uniformly on the client.
94 * Another point is sign bit, the hash range should be in [0, 2^63-1] because
95 * loff_t (for llseek) needs to be a positive value. This means the "cd" field
96 * should only be the low 15 bits.
98 uint64_t osd_zap_cursor_serialize(zap_cursor_t *zc)
100 uint64_t zfs_hash = zap_cursor_serialize(zc) & (~0ULL >> 1);
102 return (zfs_hash >> zap_hashbits(zc)) |
103 (zfs_hash << (63 - zap_hashbits(zc)));
106 void osd_zap_cursor_init_serialized(zap_cursor_t *zc, struct objset *os,
107 uint64_t id, uint64_t dirhash)
109 uint64_t zfs_hash = ((dirhash << zap_hashbits(zc)) & (~0ULL >> 1)) |
110 (dirhash >> (63 - zap_hashbits(zc)));
112 zap_cursor_init_serialized(zc, os, id, zfs_hash);
115 int osd_zap_cursor_init(zap_cursor_t **zc, struct objset *os,
116 uint64_t id, uint64_t dirhash)
121 if (unlikely(t == NULL))
124 osd_zap_cursor_init_serialized(t, os, id, dirhash);
130 void osd_zap_cursor_fini(zap_cursor_t *zc)
136 static inline void osd_obj_cursor_init_serialized(zap_cursor_t *zc,
137 struct osd_object *o,
140 struct osd_device *d = osd_obj2dev(o);
141 osd_zap_cursor_init_serialized(zc, d->od_os,
142 o->oo_db->db_object, dirhash);
145 static inline int osd_obj_cursor_init(zap_cursor_t **zc, struct osd_object *o,
148 struct osd_device *d = osd_obj2dev(o);
149 return osd_zap_cursor_init(zc, d->od_os, o->oo_db->db_object, dirhash);
152 static struct dt_it *osd_index_it_init(const struct lu_env *env,
153 struct dt_object *dt,
155 struct lustre_capa *capa)
157 struct osd_thread_info *info = osd_oti_get(env);
158 struct osd_zap_it *it;
159 struct osd_object *obj = osd_dt_obj(dt);
160 struct lu_object *lo = &dt->do_lu;
164 /* XXX: check capa ? */
166 LASSERT(lu_object_exists(lo));
168 LASSERT(osd_object_is_zap(obj->oo_db));
171 if (info->oti_it_inline) {
174 RETURN(ERR_PTR(-ENOMEM));
176 it = &info->oti_it_zap;
177 info->oti_it_inline = 1;
180 rc = osd_obj_cursor_init(&it->ozi_zc, obj, 0);
182 if (it != &info->oti_it_zap)
185 info->oti_it_inline = 0;
195 RETURN((struct dt_it *)it);
198 static void osd_index_it_fini(const struct lu_env *env, struct dt_it *di)
200 struct osd_thread_info *info = osd_oti_get(env);
201 struct osd_zap_it *it = (struct osd_zap_it *)di;
202 struct osd_object *obj;
206 LASSERT(it->ozi_obj);
210 osd_zap_cursor_fini(it->ozi_zc);
211 lu_object_put(env, &obj->oo_dt.do_lu);
212 if (it != &info->oti_it_zap)
215 info->oti_it_inline = 0;
221 static void osd_index_it_put(const struct lu_env *env, struct dt_it *di)
223 /* PBS: do nothing : ref are incremented at retrive and decreamented
227 static inline void osd_it_append_attrs(struct lu_dirent *ent, __u32 attr,
230 const unsigned align = sizeof(struct luda_type) - 1;
231 struct luda_type *lt;
233 /* check if file type is required */
234 if (attr & LUDA_TYPE) {
235 len = (len + align) & ~align;
237 lt = (void *)ent->lde_name + len;
238 lt->lt_type = cpu_to_le16(DTTOIF(type));
239 ent->lde_attrs |= LUDA_TYPE;
242 ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
246 * as we don't know FID, we can't use LU object, so this function
247 * partially duplicate __osd_xattr_get() which is built around
248 * LU-object and uses it to cache data like regular EA dnode, etc
250 static int osd_find_parent_by_dnode(const struct lu_env *env,
254 struct osd_device *osd = osd_obj2dev(osd_dt_obj(o));
255 struct lustre_mdt_attrs *lma;
258 nvlist_t *nvbuf = NULL;
264 /* first of all, get parent dnode from own attributes */
265 LASSERT(osd_dt_obj(o)->oo_db);
266 rc = -sa_handle_get(osd->od_os, osd_dt_obj(o)->oo_db->db_object,
267 NULL, SA_HDL_PRIVATE, &sa_hdl);
271 dnode = ZFS_NO_OBJECT;
272 rc = -sa_lookup(sa_hdl, SA_ZPL_PARENT(osd), &dnode, 8);
273 sa_handle_destroy(sa_hdl);
277 /* now get EA buffer */
278 rc = __osd_xattr_load(osd, dnode, &nvbuf);
282 /* XXX: if we get that far.. should we cache the result? */
284 /* try to find LMA attribute */
285 LASSERT(nvbuf != NULL);
286 rc = -nvlist_lookup_byte_array(nvbuf, XATTR_NAME_LMA, &value, &size);
287 if (rc == 0 && size >= sizeof(*lma)) {
288 lma = (struct lustre_mdt_attrs *)value;
289 lustre_lma_swab(lma);
290 *fid = lma->lma_self_fid;
295 /* no LMA attribute in SA, let's try regular EA */
297 /* first of all, get parent dnode storing regular EA */
298 rc = -sa_handle_get(osd->od_os, dnode, NULL, SA_HDL_PRIVATE, &sa_hdl);
302 dnode = ZFS_NO_OBJECT;
303 rc = -sa_lookup(sa_hdl, SA_ZPL_XATTR(osd), &dnode, 8);
304 sa_handle_destroy(sa_hdl);
308 CLASSERT(sizeof(*lma) <= sizeof(osd_oti_get(env)->oti_buf));
309 buf.lb_buf = osd_oti_get(env)->oti_buf;
310 buf.lb_len = sizeof(osd_oti_get(env)->oti_buf);
312 /* now try to find LMA */
313 rc = __osd_xattr_get_large(env, osd, dnode, &buf,
314 XATTR_NAME_LMA, &size);
315 if (rc == 0 && size >= sizeof(*lma)) {
317 lustre_lma_swab(lma);
318 *fid = lma->lma_self_fid;
323 GOTO(out, rc = -EIO);
332 static int osd_find_parent_fid(const struct lu_env *env, struct dt_object *o,
335 struct link_ea_header *leh;
336 struct link_ea_entry *lee;
341 buf.lb_buf = osd_oti_get(env)->oti_buf;
342 buf.lb_len = sizeof(osd_oti_get(env)->oti_buf);
344 rc = osd_xattr_get(env, o, &buf, XATTR_NAME_LINK, BYPASS_CAPA);
346 rc = osd_xattr_get(env, o, &LU_BUF_NULL,
347 XATTR_NAME_LINK, BYPASS_CAPA);
351 OBD_ALLOC(buf.lb_buf, rc);
352 if (buf.lb_buf == NULL)
355 rc = osd_xattr_get(env, o, &buf, XATTR_NAME_LINK, BYPASS_CAPA);
359 if (rc < sizeof(*leh) + sizeof(*lee))
360 GOTO(out, rc = -EINVAL);
363 if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) {
364 leh->leh_magic = LINK_EA_MAGIC;
365 leh->leh_reccount = __swab32(leh->leh_reccount);
366 leh->leh_len = __swab64(leh->leh_len);
368 if (leh->leh_magic != LINK_EA_MAGIC)
369 GOTO(out, rc = -EINVAL);
370 if (leh->leh_reccount == 0)
371 GOTO(out, rc = -ENODATA);
373 lee = (struct link_ea_entry *)(leh + 1);
374 fid_be_to_cpu(fid, (const struct lu_fid *)&lee->lee_parent_fid);
378 if (buf.lb_buf != osd_oti_get(env)->oti_buf)
379 OBD_FREE(buf.lb_buf, buf.lb_len);
382 /* this block can be enabled for additional verification
383 * it's trying to match FID from LinkEA vs. FID from LMA */
387 rc2 = osd_find_parent_by_dnode(env, o, &fid2);
389 if (lu_fid_eq(fid, &fid2) == 0)
390 CERROR("wrong parent: "DFID" != "DFID"\n",
391 PFID(fid), PFID(&fid2));
395 /* no LinkEA is found, let's try to find the fid in parent's LMA */
396 if (unlikely(rc != 0))
397 rc = osd_find_parent_by_dnode(env, o, fid);
402 static int osd_dir_lookup(const struct lu_env *env, struct dt_object *dt,
403 struct dt_rec *rec, const struct dt_key *key,
404 struct lustre_capa *capa)
406 struct osd_thread_info *oti = osd_oti_get(env);
407 struct osd_object *obj = osd_dt_obj(dt);
408 struct osd_device *osd = osd_obj2dev(obj);
409 char *name = (char *)key;
413 LASSERT(osd_object_is_zap(obj->oo_db));
415 if (name[0] == '.') {
417 const struct lu_fid *f = lu_object_fid(&dt->do_lu);
418 memcpy(rec, f, sizeof(*f));
420 } else if (name[1] == '.' && name[2] == 0) {
421 rc = osd_find_parent_fid(env, dt, (struct lu_fid *)rec);
422 RETURN(rc == 0 ? 1 : rc);
426 rc = -zap_lookup(osd->od_os, obj->oo_db->db_object,
427 (char *)key, 8, sizeof(oti->oti_zde) / 8,
428 (void *)&oti->oti_zde);
429 memcpy(rec, &oti->oti_zde.lzd_fid, sizeof(struct lu_fid));
431 RETURN(rc == 0 ? 1 : rc);
434 static int osd_declare_dir_insert(const struct lu_env *env,
435 struct dt_object *dt,
436 const struct dt_rec *rec,
437 const struct dt_key *key,
440 struct osd_object *obj = osd_dt_obj(dt);
441 struct osd_thandle *oh;
446 oh = container_of0(th, struct osd_thandle, ot_super);
448 /* This is for inserting dot/dotdot for new created dir. */
449 if (obj->oo_db == NULL)
450 object = DMU_NEW_OBJECT;
452 object = obj->oo_db->db_object;
454 dmu_tx_hold_bonus(oh->ot_tx, object);
455 dmu_tx_hold_zap(oh->ot_tx, object, TRUE, (char *)key);
461 * Find the osd object for given fid.
463 * \param fid need to find the osd object having this fid
465 * \retval osd_object on success
466 * \retval -ve on error
468 struct osd_object *osd_object_find(const struct lu_env *env,
469 struct dt_object *dt,
470 const struct lu_fid *fid)
472 struct lu_device *ludev = dt->do_lu.lo_dev;
473 struct osd_object *child = NULL;
474 struct lu_object *luch;
475 struct lu_object *lo;
478 * at this point topdev might not exist yet
479 * (i.e. MGS is preparing profiles). so we can
480 * not rely on topdev and instead lookup with
481 * our device passed as topdev. this can't work
482 * if the object isn't cached yet (as osd doesn't
483 * allocate lu_header). IOW, the object must be
484 * in the cache, otherwise lu_object_alloc() crashes
487 luch = lu_object_find_at(env, ludev, fid, NULL);
491 if (lu_object_exists(luch)) {
492 lo = lu_object_locate(luch->lo_header, ludev->ld_type);
496 LU_OBJECT_DEBUG(D_ERROR, env, luch,
497 "%s: object can't be located "DFID"\n",
498 osd_dev(ludev)->od_svname, PFID(fid));
501 lu_object_put(env, luch);
502 CERROR("%s: Unable to get osd_object "DFID"\n",
503 osd_dev(ludev)->od_svname, PFID(fid));
504 child = ERR_PTR(-ENOENT);
507 LU_OBJECT_DEBUG(D_ERROR, env, luch,
508 "%s: lu_object does not exists "DFID"\n",
509 osd_dev(ludev)->od_svname, PFID(fid));
510 lu_object_put(env, luch);
511 child = ERR_PTR(-ENOENT);
518 * Put the osd object once done with it.
520 * \param obj osd object that needs to be put
522 static inline void osd_object_put(const struct lu_env *env,
523 struct osd_object *obj)
525 lu_object_put(env, &obj->oo_dt.do_lu);
528 static int osd_seq_exists(const struct lu_env *env, struct osd_device *osd,
531 struct lu_seq_range *range = &osd_oti_get(env)->oti_seq_range;
532 struct seq_server_site *ss = osd_seq_site(osd);
537 LASSERT(ss->ss_server_fld != NULL);
539 rc = osd_fld_lookup(env, osd, seq, range);
541 CERROR("%s: Can not lookup fld for "LPX64"\n",
546 RETURN(ss->ss_node_id == range->lsr_index);
549 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
550 const struct lu_fid *fid)
552 struct seq_server_site *ss = osd_seq_site(osd);
555 /* FID seqs not in FLDB, must be local seq */
556 if (unlikely(!fid_seq_in_fldb(fid_seq(fid))))
559 /* If FLD is not being initialized yet, it only happens during the
560 * initialization, likely during mgs initialization, and we assume
561 * this is local FID. */
562 if (ss == NULL || ss->ss_server_fld == NULL)
565 /* Only check the local FLDB here */
566 if (osd_seq_exists(env, osd, fid_seq(fid)))
573 * Inserts (key, value) pair in \a directory object.
575 * \param dt osd index object
576 * \param key key for index
577 * \param rec record reference
578 * \param th transaction handler
579 * \param capa capability descriptor
580 * \param ignore_quota update should not affect quota
583 * \retval -ve failure
585 static int osd_dir_insert(const struct lu_env *env, struct dt_object *dt,
586 const struct dt_rec *rec, const struct dt_key *key,
587 struct thandle *th, struct lustre_capa *capa,
590 struct osd_thread_info *oti = osd_oti_get(env);
591 struct osd_object *parent = osd_dt_obj(dt);
592 struct osd_device *osd = osd_obj2dev(parent);
593 struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec;
594 const struct lu_fid *fid = rec1->rec_fid;
595 struct osd_thandle *oh;
596 struct osd_object *child = NULL;
598 char *name = (char *)key;
602 LASSERT(parent->oo_db);
603 LASSERT(osd_object_is_zap(parent->oo_db));
605 LASSERT(dt_object_exists(dt));
606 LASSERT(osd_invariant(parent));
609 oh = container_of0(th, struct osd_thandle, ot_super);
611 rc = osd_remote_fid(env, osd, fid);
613 CERROR("%s: Can not find object "DFID": rc = %d\n",
614 osd->od_svname, PFID(fid), rc);
618 if (unlikely(rc == 1)) {
619 /* Insert remote entry */
620 memset(&oti->oti_zde.lzd_reg, 0, sizeof(oti->oti_zde.lzd_reg));
621 oti->oti_zde.lzd_reg.zde_type = IFTODT(rec1->rec_type & S_IFMT);
624 * To simulate old Orion setups with ./.. stored in the
627 /* Insert local entry */
628 child = osd_object_find(env, dt, fid);
630 RETURN(PTR_ERR(child));
632 LASSERT(child->oo_db);
633 if (name[0] == '.') {
635 /* do not store ".", instead generate it
636 * during iteration */
638 } else if (name[1] == '.' && name[2] == 0) {
639 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT)) {
640 struct lu_fid tfid = *fid;
642 osd_object_put(env, child);
644 child = osd_object_find(env, dt, &tfid);
646 RETURN(PTR_ERR(child));
648 LASSERT(child->oo_db);
651 /* update parent dnode in the child.
652 * later it will be used to generate ".." */
653 rc = osd_object_sa_update(parent,
655 &child->oo_db->db_object,
661 CLASSERT(sizeof(oti->oti_zde.lzd_reg) == 8);
662 CLASSERT(sizeof(oti->oti_zde) % 8 == 0);
663 attr = child->oo_dt.do_lu.lo_header ->loh_attr;
664 oti->oti_zde.lzd_reg.zde_type = IFTODT(attr & S_IFMT);
665 oti->oti_zde.lzd_reg.zde_dnode = child->oo_db->db_object;
668 oti->oti_zde.lzd_fid = *fid;
669 /* Insert (key,oid) into ZAP */
670 rc = -zap_add(osd->od_os, parent->oo_db->db_object,
671 (char *)key, 8, sizeof(oti->oti_zde) / 8,
672 (void *)&oti->oti_zde, oh->ot_tx);
673 if (unlikely(rc == -EEXIST &&
674 name[0] == '.' && name[1] == '.' && name[2] == 0))
675 /* Update (key,oid) in ZAP */
676 rc = -zap_update(osd->od_os, parent->oo_db->db_object,
677 (char *)key, 8, sizeof(oti->oti_zde) / 8,
678 (void *)&oti->oti_zde, oh->ot_tx);
682 osd_object_put(env, child);
687 static int osd_declare_dir_delete(const struct lu_env *env,
688 struct dt_object *dt,
689 const struct dt_key *key,
692 struct osd_object *obj = osd_dt_obj(dt);
693 struct osd_thandle *oh;
696 LASSERT(dt_object_exists(dt));
697 LASSERT(osd_invariant(obj));
700 oh = container_of0(th, struct osd_thandle, ot_super);
703 LASSERT(osd_object_is_zap(obj->oo_db));
705 dmu_tx_hold_zap(oh->ot_tx, obj->oo_db->db_object, TRUE, (char *)key);
710 static int osd_dir_delete(const struct lu_env *env, struct dt_object *dt,
711 const struct dt_key *key, struct thandle *th,
712 struct lustre_capa *capa)
714 struct osd_object *obj = osd_dt_obj(dt);
715 struct osd_device *osd = osd_obj2dev(obj);
716 struct osd_thandle *oh;
717 dmu_buf_t *zap_db = obj->oo_db;
718 char *name = (char *)key;
723 LASSERT(osd_object_is_zap(obj->oo_db));
726 oh = container_of0(th, struct osd_thandle, ot_super);
729 * In Orion . and .. were stored in the directory (not generated upon
730 * request as now). we preserve them for backward compatibility
732 if (name[0] == '.') {
735 } else if (name[1] == '.' && name[2] == 0) {
740 /* Remove key from the ZAP */
741 rc = -zap_remove(osd->od_os, zap_db->db_object,
742 (char *) key, oh->ot_tx);
744 if (unlikely(rc && rc != -ENOENT))
745 CERROR("%s: zap_remove failed: rc = %d\n", osd->od_svname, rc);
750 static struct dt_it *osd_dir_it_init(const struct lu_env *env,
751 struct dt_object *dt,
753 struct lustre_capa *capa)
755 struct osd_zap_it *it;
757 it = (struct osd_zap_it *)osd_index_it_init(env, dt, unused, capa);
761 RETURN((struct dt_it *)it);
765 * Move Iterator to record specified by \a key
767 * \param di osd iterator
768 * \param key key for index
770 * \retval +ve di points to record with least key not larger than key
771 * \retval 0 di points to exact matched key
772 * \retval -ve failure
774 static int osd_dir_it_get(const struct lu_env *env,
775 struct dt_it *di, const struct dt_key *key)
777 struct osd_zap_it *it = (struct osd_zap_it *)di;
778 struct osd_object *obj = it->ozi_obj;
779 char *name = (char *)key;
786 /* reset the cursor */
787 zap_cursor_fini(it->ozi_zc);
788 osd_obj_cursor_init_serialized(it->ozi_zc, obj, 0);
790 /* XXX: implementation of the API is broken at the moment */
791 LASSERT(((const char *)key)[0] == 0);
798 if (name[0] == '.') {
802 } else if (name[1] == '.' && name[2] == 0) {
808 /* neither . nor .. - some real record */
816 static void osd_dir_it_put(const struct lu_env *env, struct dt_it *di)
818 /* PBS: do nothing : ref are incremented at retrive and decreamented
823 * in Orion . and .. were stored in the directory, while ZPL
824 * and current osd-zfs generate them up on request. so, we
825 * need to ignore previously stored . and ..
827 static int osd_index_retrieve_skip_dots(struct osd_zap_it *it,
833 rc = -zap_cursor_retrieve(it->ozi_zc, za);
836 if (unlikely(rc == 0 && za->za_name[0] == '.')) {
837 if (za->za_name[1] == 0) {
839 } else if (za->za_name[1] == '.' &&
840 za->za_name[2] == 0) {
844 zap_cursor_advance(it->ozi_zc);
846 } while (unlikely(rc == 0 && isdot));
852 * to load a directory entry at a time and stored it in
853 * iterator's in-memory data structure.
855 * \param di, struct osd_it_ea, iterator's in memory structure
857 * \retval +ve, iterator reached to end
858 * \retval 0, iterator not reached to end
859 * \retval -ve, on error
861 static int osd_dir_it_next(const struct lu_env *env, 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;
869 /* temp. storage should be enough for any key supported by ZFS */
870 CLASSERT(sizeof(za->za_name) <= sizeof(it->ozi_name));
873 * the first ->next() moves the cursor to .
874 * the second ->next() moves the cursor to ..
875 * then we get to the real records and have to verify any exist
877 if (it->ozi_pos <= 2) {
883 zap_cursor_advance(it->ozi_zc);
887 * According to current API we need to return error if its last entry.
888 * zap_cursor_advance() does not return any value. So we need to call
889 * retrieve to check if there is any record. We should make
890 * changes to Iterator API to not return status for this API
892 rc = osd_index_retrieve_skip_dots(it, za);
894 if (rc == -ENOENT) /* end of dir */
900 static struct dt_key *osd_dir_it_key(const struct lu_env *env,
901 const struct dt_it *di)
903 struct osd_zap_it *it = (struct osd_zap_it *)di;
904 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
908 if (it->ozi_pos <= 1) {
910 RETURN((struct dt_key *)".");
911 } else if (it->ozi_pos == 2) {
912 RETURN((struct dt_key *)"..");
915 if ((rc = -zap_cursor_retrieve(it->ozi_zc, za)))
918 strcpy(it->ozi_name, za->za_name);
920 RETURN((struct dt_key *)it->ozi_name);
923 static int osd_dir_it_key_size(const struct lu_env *env, const struct dt_it *di)
925 struct osd_zap_it *it = (struct osd_zap_it *)di;
926 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
930 if (it->ozi_pos <= 1) {
933 } else if (it->ozi_pos == 2) {
937 if ((rc = -zap_cursor_retrieve(it->ozi_zc, za)) == 0)
938 rc = strlen(za->za_name);
943 static int osd_dir_it_rec(const struct lu_env *env, const struct dt_it *di,
944 struct dt_rec *dtrec, __u32 attr)
946 struct osd_zap_it *it = (struct osd_zap_it *)di;
947 struct lu_dirent *lde = (struct lu_dirent *)dtrec;
948 struct luz_direntry *zde = &osd_oti_get(env)->oti_zde;
949 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
953 if (it->ozi_pos <= 1) {
954 lde->lde_hash = cpu_to_le64(1);
955 strcpy(lde->lde_name, ".");
956 lde->lde_namelen = cpu_to_le16(1);
957 lde->lde_fid = *lu_object_fid(&it->ozi_obj->oo_dt.do_lu);
958 lde->lde_attrs = LUDA_FID;
959 /* append lustre attributes */
960 osd_it_append_attrs(lde, attr, 1, IFTODT(S_IFDIR));
961 lde->lde_reclen = cpu_to_le16(lu_dirent_calc_size(1, attr));
965 } else if (it->ozi_pos == 2) {
966 lde->lde_hash = cpu_to_le64(2);
967 strcpy(lde->lde_name, "..");
968 lde->lde_namelen = cpu_to_le16(2);
969 lde->lde_attrs = LUDA_FID;
970 /* append lustre attributes */
971 osd_it_append_attrs(lde, attr, 2, IFTODT(S_IFDIR));
972 lde->lde_reclen = cpu_to_le16(lu_dirent_calc_size(2, attr));
973 rc = osd_find_parent_fid(env, &it->ozi_obj->oo_dt, &lde->lde_fid);
975 /* ENOENT happens at the root of filesystem so ignore it */
983 rc = -zap_cursor_retrieve(it->ozi_zc, za);
984 if (unlikely(rc != 0))
987 lde->lde_hash = cpu_to_le64(osd_zap_cursor_serialize(it->ozi_zc));
988 namelen = strlen(za->za_name);
989 if (namelen > NAME_MAX)
990 GOTO(out, rc = -EOVERFLOW);
991 strcpy(lde->lde_name, za->za_name);
992 lde->lde_namelen = cpu_to_le16(namelen);
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);
999 GOTO(out, rc = -EIO);
1002 rc = -zap_lookup(it->ozi_zc->zc_objset, it->ozi_zc->zc_zapobj,
1003 za->za_name, za->za_integer_length, 3, zde);
1007 lde->lde_fid = zde->lzd_fid;
1008 lde->lde_attrs = LUDA_FID;
1010 /* append lustre attributes */
1011 osd_it_append_attrs(lde, attr, namelen, zde->lzd_reg.zde_type);
1013 lde->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
1019 static int osd_dir_it_rec_size(const struct lu_env *env, const struct dt_it *di,
1022 struct osd_zap_it *it = (struct osd_zap_it *)di;
1023 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
1028 if (it->ozi_pos <= 1)
1030 else if (it->ozi_pos == 2)
1034 rc = lu_dirent_calc_size(namelen, attr);
1038 rc = -zap_cursor_retrieve(it->ozi_zc, za);
1039 if (unlikely(rc != 0))
1042 if (za->za_integer_length != 8 || za->za_num_integers < 3) {
1043 CERROR("%s: unsupported direntry format: %d %d\n",
1044 osd_obj2dev(it->ozi_obj)->od_svname,
1045 za->za_integer_length, (int)za->za_num_integers);
1049 namelen = strlen(za->za_name);
1050 if (namelen > NAME_MAX)
1053 rc = lu_dirent_calc_size(namelen, attr);
1058 static __u64 osd_dir_it_store(const struct lu_env *env, const struct dt_it *di)
1060 struct osd_zap_it *it = (struct osd_zap_it *)di;
1064 if (it->ozi_pos <= 2)
1067 pos = osd_zap_cursor_serialize(it->ozi_zc);
1074 * rc == 0 -> end of directory.
1075 * rc > 0 -> ok, proceed.
1076 * rc < 0 -> error. ( EOVERFLOW can be masked.)
1078 static int osd_dir_it_load(const struct lu_env *env,
1079 const struct dt_it *di, __u64 hash)
1081 struct osd_zap_it *it = (struct osd_zap_it *)di;
1082 struct osd_object *obj = it->ozi_obj;
1083 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
1087 /* reset the cursor */
1088 zap_cursor_fini(it->ozi_zc);
1089 osd_obj_cursor_init_serialized(it->ozi_zc, obj, hash);
1096 /* to return whether the end has been reached */
1097 rc = osd_index_retrieve_skip_dots(it, za);
1100 else if (rc == -ENOENT)
1107 struct dt_index_operations osd_dir_ops = {
1108 .dio_lookup = osd_dir_lookup,
1109 .dio_declare_insert = osd_declare_dir_insert,
1110 .dio_insert = osd_dir_insert,
1111 .dio_declare_delete = osd_declare_dir_delete,
1112 .dio_delete = osd_dir_delete,
1114 .init = osd_dir_it_init,
1115 .fini = osd_index_it_fini,
1116 .get = osd_dir_it_get,
1117 .put = osd_dir_it_put,
1118 .next = osd_dir_it_next,
1119 .key = osd_dir_it_key,
1120 .key_size = osd_dir_it_key_size,
1121 .rec = osd_dir_it_rec,
1122 .rec_size = osd_dir_it_rec_size,
1123 .store = osd_dir_it_store,
1124 .load = osd_dir_it_load
1129 * Primitives for index files using binary keys.
1132 /* key integer_size is 8 */
1133 static int osd_prepare_key_uint64(struct osd_object *o, __u64 *dst,
1134 const struct dt_key *src)
1141 /* align keysize to 64bit */
1142 size = (o->oo_keysize + sizeof(__u64) - 1) / sizeof(__u64);
1143 size *= sizeof(__u64);
1145 LASSERT(size <= MAXNAMELEN);
1147 if (unlikely(size > o->oo_keysize))
1148 memset(dst + o->oo_keysize, 0, size - o->oo_keysize);
1149 memcpy(dst, (const char *)src, o->oo_keysize);
1151 return (size/sizeof(__u64));
1154 static int osd_index_lookup(const struct lu_env *env, struct dt_object *dt,
1155 struct dt_rec *rec, const struct dt_key *key,
1156 struct lustre_capa *capa)
1158 struct osd_object *obj = osd_dt_obj(dt);
1159 struct osd_device *osd = osd_obj2dev(obj);
1160 __u64 *k = osd_oti_get(env)->oti_key64;
1164 rc = osd_prepare_key_uint64(obj, k, key);
1166 rc = -zap_lookup_uint64(osd->od_os, obj->oo_db->db_object,
1167 k, rc, obj->oo_recusize, obj->oo_recsize,
1169 RETURN(rc == 0 ? 1 : rc);
1172 static int osd_declare_index_insert(const struct lu_env *env,
1173 struct dt_object *dt,
1174 const struct dt_rec *rec,
1175 const struct dt_key *key,
1178 struct osd_object *obj = osd_dt_obj(dt);
1179 struct osd_thandle *oh;
1182 LASSERT(th != NULL);
1183 oh = container_of0(th, struct osd_thandle, ot_super);
1185 LASSERT(obj->oo_db);
1187 dmu_tx_hold_bonus(oh->ot_tx, obj->oo_db->db_object);
1189 /* It is not clear what API should be used for binary keys, so we pass
1190 * a null name which has the side effect of over-reserving space,
1191 * accounting for the worst case. See zap_count_write() */
1192 dmu_tx_hold_zap(oh->ot_tx, obj->oo_db->db_object, TRUE, NULL);
1197 static int osd_index_insert(const struct lu_env *env, struct dt_object *dt,
1198 const struct dt_rec *rec, const struct dt_key *key,
1199 struct thandle *th, struct lustre_capa *capa,
1202 struct osd_object *obj = osd_dt_obj(dt);
1203 struct osd_device *osd = osd_obj2dev(obj);
1204 struct osd_thandle *oh;
1205 __u64 *k = osd_oti_get(env)->oti_key64;
1209 LASSERT(obj->oo_db);
1210 LASSERT(dt_object_exists(dt));
1211 LASSERT(osd_invariant(obj));
1212 LASSERT(th != NULL);
1214 oh = container_of0(th, struct osd_thandle, ot_super);
1216 rc = osd_prepare_key_uint64(obj, k, key);
1218 /* Insert (key,oid) into ZAP */
1219 rc = -zap_add_uint64(osd->od_os, obj->oo_db->db_object,
1220 k, rc, obj->oo_recusize, obj->oo_recsize,
1221 (void *)rec, oh->ot_tx);
1225 static int osd_declare_index_delete(const struct lu_env *env,
1226 struct dt_object *dt,
1227 const struct dt_key *key,
1230 struct osd_object *obj = osd_dt_obj(dt);
1231 struct osd_thandle *oh;
1234 LASSERT(dt_object_exists(dt));
1235 LASSERT(osd_invariant(obj));
1236 LASSERT(th != NULL);
1237 LASSERT(obj->oo_db);
1239 oh = container_of0(th, struct osd_thandle, ot_super);
1240 dmu_tx_hold_zap(oh->ot_tx, obj->oo_db->db_object, TRUE, NULL);
1245 static int osd_index_delete(const struct lu_env *env, struct dt_object *dt,
1246 const struct dt_key *key, struct thandle *th,
1247 struct lustre_capa *capa)
1249 struct osd_object *obj = osd_dt_obj(dt);
1250 struct osd_device *osd = osd_obj2dev(obj);
1251 struct osd_thandle *oh;
1252 __u64 *k = osd_oti_get(env)->oti_key64;
1256 LASSERT(obj->oo_db);
1257 LASSERT(th != NULL);
1258 oh = container_of0(th, struct osd_thandle, ot_super);
1260 rc = osd_prepare_key_uint64(obj, k, key);
1262 /* Remove binary key from the ZAP */
1263 rc = -zap_remove_uint64(osd->od_os, obj->oo_db->db_object,
1268 static int osd_index_it_get(const struct lu_env *env, struct dt_it *di,
1269 const struct dt_key *key)
1271 struct osd_zap_it *it = (struct osd_zap_it *)di;
1272 struct osd_object *obj = it->ozi_obj;
1273 struct osd_device *osd = osd_obj2dev(obj);
1277 LASSERT(it->ozi_zc);
1280 * XXX: we need a binary version of zap_cursor_move_to_key()
1281 * to implement this API */
1282 if (*((const __u64 *)key) != 0)
1283 CERROR("NOT IMPLEMETED YET (move to "LPX64")\n",
1286 zap_cursor_fini(it->ozi_zc);
1287 zap_cursor_init(it->ozi_zc, osd->od_os, obj->oo_db->db_object);
1293 static int osd_index_it_next(const struct lu_env *env, struct dt_it *di)
1295 struct osd_zap_it *it = (struct osd_zap_it *)di;
1296 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
1300 if (it->ozi_reset == 0)
1301 zap_cursor_advance(it->ozi_zc);
1305 * According to current API we need to return error if it's last entry.
1306 * zap_cursor_advance() does not return any value. So we need to call
1307 * retrieve to check if there is any record. We should make
1308 * changes to Iterator API to not return status for this API
1310 rc = -zap_cursor_retrieve(it->ozi_zc, za);
1317 static struct dt_key *osd_index_it_key(const struct lu_env *env,
1318 const struct dt_it *di)
1320 struct osd_zap_it *it = (struct osd_zap_it *)di;
1321 struct osd_object *obj = it->ozi_obj;
1322 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
1327 rc = -zap_cursor_retrieve(it->ozi_zc, za);
1329 RETURN(ERR_PTR(rc));
1331 /* the binary key is stored in the name */
1332 memcpy(&it->ozi_key, za->za_name, obj->oo_keysize);
1334 RETURN((struct dt_key *)&it->ozi_key);
1337 static int osd_index_it_key_size(const struct lu_env *env,
1338 const struct dt_it *di)
1340 struct osd_zap_it *it = (struct osd_zap_it *)di;
1341 struct osd_object *obj = it->ozi_obj;
1342 RETURN(obj->oo_keysize);
1345 static int osd_index_it_rec(const struct lu_env *env, const struct dt_it *di,
1346 struct dt_rec *rec, __u32 attr)
1348 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
1349 struct osd_zap_it *it = (struct osd_zap_it *)di;
1350 struct osd_object *obj = it->ozi_obj;
1351 struct osd_device *osd = osd_obj2dev(obj);
1352 __u64 *k = osd_oti_get(env)->oti_key64;
1357 rc = -zap_cursor_retrieve(it->ozi_zc, za);
1361 rc = osd_prepare_key_uint64(obj, k, (const struct dt_key *)za->za_name);
1363 rc = -zap_lookup_uint64(osd->od_os, obj->oo_db->db_object,
1364 k, rc, obj->oo_recusize, obj->oo_recsize,
1369 static __u64 osd_index_it_store(const struct lu_env *env,
1370 const struct dt_it *di)
1372 struct osd_zap_it *it = (struct osd_zap_it *)di;
1375 RETURN((__u64)zap_cursor_serialize(it->ozi_zc));
1378 static int osd_index_it_load(const struct lu_env *env, const struct dt_it *di,
1381 struct osd_zap_it *it = (struct osd_zap_it *)di;
1382 struct osd_object *obj = it->ozi_obj;
1383 struct osd_device *osd = osd_obj2dev(obj);
1384 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
1388 /* reset the cursor */
1389 zap_cursor_fini(it->ozi_zc);
1390 zap_cursor_init_serialized(it->ozi_zc, osd->od_os,
1391 obj->oo_db->db_object, hash);
1394 rc = -zap_cursor_retrieve(it->ozi_zc, za);
1397 else if (rc == -ENOENT)
1403 static struct dt_index_operations osd_index_ops = {
1404 .dio_lookup = osd_index_lookup,
1405 .dio_declare_insert = osd_declare_index_insert,
1406 .dio_insert = osd_index_insert,
1407 .dio_declare_delete = osd_declare_index_delete,
1408 .dio_delete = osd_index_delete,
1410 .init = osd_index_it_init,
1411 .fini = osd_index_it_fini,
1412 .get = osd_index_it_get,
1413 .put = osd_index_it_put,
1414 .next = osd_index_it_next,
1415 .key = osd_index_it_key,
1416 .key_size = osd_index_it_key_size,
1417 .rec = osd_index_it_rec,
1418 .store = osd_index_it_store,
1419 .load = osd_index_it_load
1423 struct osd_metadnode_it {
1424 struct osd_device *mit_dev;
1426 struct lu_fid mit_fid;
1428 __u64 mit_prefetched_dnode;
1431 static struct dt_it *osd_zfs_otable_it_init(const struct lu_env *env,
1432 struct dt_object *dt, __u32 attr,
1433 struct lustre_capa *capa)
1435 struct osd_device *dev = osd_dev(dt->do_lu.lo_dev);
1436 struct osd_metadnode_it *it;
1440 if (unlikely(it == NULL))
1441 RETURN(ERR_PTR(-ENOMEM));
1445 /* XXX: dmu_object_next() does NOT find dnodes allocated
1446 * in the current non-committed txg, so we force txg
1447 * commit to find all existing dnodes ... */
1448 txg_wait_synced(dmu_objset_pool(dev->od_os), 0ULL);
1450 RETURN((struct dt_it *)it);
1453 static void osd_zfs_otable_it_fini(const struct lu_env *env, struct dt_it *di)
1455 struct osd_metadnode_it *it = (struct osd_metadnode_it *)di;
1460 static int osd_zfs_otable_it_get(const struct lu_env *env,
1461 struct dt_it *di, const struct dt_key *key)
1466 static void osd_zfs_otable_it_put(const struct lu_env *env, struct dt_it *di)
1470 #define OTABLE_PREFETCH 256
1472 static void osd_zfs_otable_prefetch(const struct lu_env *env,
1473 struct osd_metadnode_it *it)
1475 struct osd_device *dev = it->mit_dev;
1478 /* can go negative on the very first access to the iterator
1479 * or if some non-Lustre objects were found */
1480 if (unlikely(it->mit_prefetched < 0))
1481 it->mit_prefetched = 0;
1483 if (it->mit_prefetched >= (OTABLE_PREFETCH >> 1))
1486 if (it->mit_prefetched_dnode == 0)
1487 it->mit_prefetched_dnode = it->mit_pos;
1489 while (it->mit_prefetched < OTABLE_PREFETCH) {
1490 rc = -dmu_object_next(dev->od_os, &it->mit_prefetched_dnode,
1492 if (unlikely(rc != 0))
1495 /* dmu_prefetch() was exported in 0.6.2, if you use with
1496 * an older release, just comment it out - this is an
1498 dmu_prefetch(dev->od_os, it->mit_prefetched_dnode, 0, 0);
1500 it->mit_prefetched++;
1504 static int osd_zfs_otable_it_next(const struct lu_env *env, struct dt_it *di)
1506 struct osd_metadnode_it *it = (struct osd_metadnode_it *)di;
1507 struct lustre_mdt_attrs *lma;
1508 struct osd_device *dev = it->mit_dev;
1509 nvlist_t *nvbuf = NULL;
1514 memset(&it->mit_fid, 0, sizeof(it->mit_fid));
1516 dnode = it->mit_pos;
1518 rc = -dmu_object_next(dev->od_os, &it->mit_pos, B_FALSE, 0);
1519 if (unlikely(rc != 0))
1521 it->mit_prefetched--;
1523 /* LMA is required for this to be a Lustre object.
1524 * If there is no xattr skip it. */
1525 rc = __osd_xattr_load(dev, it->mit_pos, &nvbuf);
1526 if (unlikely(rc != 0))
1529 LASSERT(nvbuf != NULL);
1530 rc = -nvlist_lookup_byte_array(nvbuf, XATTR_NAME_LMA, &v, &s);
1531 if (likely(rc == 0)) {
1533 lma = (struct lustre_mdt_attrs *)v;
1534 lustre_lma_swab(lma);
1535 it->mit_fid = lma->lma_self_fid;
1539 /* not a Lustre object, try next one */
1546 /* we aren't prefetching in the above loop because the number of
1547 * non-Lustre objects is very small and we will be repeating very
1548 * rare. in case we want to use this to iterate over non-Lustre
1549 * objects (i.e. when we convert regular ZFS in Lustre) it makes
1550 * sense to initiate prefetching in the loop */
1552 /* 0 - there are more items, +1 - the end */
1553 if (likely(rc == 0))
1554 osd_zfs_otable_prefetch(env, it);
1556 CDEBUG(D_OTHER, "advance: %llu -> %llu "DFID": %d\n", dnode,
1557 it->mit_pos, PFID(&it->mit_fid), rc);
1563 static struct dt_key *osd_zfs_otable_it_key(const struct lu_env *env,
1564 const struct dt_it *di)
1569 static int osd_zfs_otable_it_key_size(const struct lu_env *env,
1570 const struct dt_it *di)
1572 return sizeof(__u64);
1575 static int osd_zfs_otable_it_rec(const struct lu_env *env,
1576 const struct dt_it *di,
1577 struct dt_rec *rec, __u32 attr)
1579 struct osd_metadnode_it *it = (struct osd_metadnode_it *)di;
1580 struct lu_fid *fid = (struct lu_fid *)rec;
1589 static __u64 osd_zfs_otable_it_store(const struct lu_env *env,
1590 const struct dt_it *di)
1592 struct osd_metadnode_it *it = (struct osd_metadnode_it *)di;
1597 static int osd_zfs_otable_it_load(const struct lu_env *env,
1598 const struct dt_it *di, __u64 hash)
1600 struct osd_metadnode_it *it = (struct osd_metadnode_it *)di;
1603 it->mit_prefetched = 0;
1604 it->mit_prefetched_dnode = 0;
1606 return osd_zfs_otable_it_next(env, (struct dt_it *)di);
1609 static int osd_zfs_otable_it_key_rec(const struct lu_env *env,
1610 const struct dt_it *di, void *key_rec)
1615 const struct dt_index_operations osd_zfs_otable_ops = {
1617 .init = osd_zfs_otable_it_init,
1618 .fini = osd_zfs_otable_it_fini,
1619 .get = osd_zfs_otable_it_get,
1620 .put = osd_zfs_otable_it_put,
1621 .next = osd_zfs_otable_it_next,
1622 .key = osd_zfs_otable_it_key,
1623 .key_size = osd_zfs_otable_it_key_size,
1624 .rec = osd_zfs_otable_it_rec,
1625 .store = osd_zfs_otable_it_store,
1626 .load = osd_zfs_otable_it_load,
1627 .key_rec = osd_zfs_otable_it_key_rec,
1631 int osd_index_try(const struct lu_env *env, struct dt_object *dt,
1632 const struct dt_index_features *feat)
1634 struct osd_object *obj = osd_dt_obj(dt);
1637 LASSERT(dt_object_exists(dt));
1640 * XXX: implement support for fixed-size keys sorted with natural
1641 * numerical way (not using internal hash value)
1643 if (feat->dif_flags & DT_IND_RANGE)
1646 if (unlikely(feat == &dt_otable_features)) {
1647 dt->do_index_ops = &osd_zfs_otable_ops;
1651 LASSERT(obj->oo_db != NULL);
1652 if (likely(feat == &dt_directory_features)) {
1653 if (osd_object_is_zap(obj->oo_db))
1654 dt->do_index_ops = &osd_dir_ops;
1657 } else if (unlikely(feat == &dt_acct_features)) {
1658 LASSERT(fid_is_acct(lu_object_fid(&dt->do_lu)));
1659 dt->do_index_ops = &osd_acct_index_ops;
1660 } else if (osd_object_is_zap(obj->oo_db) &&
1661 dt->do_index_ops == NULL) {
1662 /* For index file, we don't support variable key & record sizes
1663 * and the key has to be unique */
1664 if ((feat->dif_flags & ~DT_IND_UPDATE) != 0)
1667 if (feat->dif_keysize_max > ZAP_MAXNAMELEN)
1669 if (feat->dif_keysize_max != feat->dif_keysize_min)
1672 /* As for the record size, it should be a multiple of 8 bytes
1673 * and smaller than the maximum value length supported by ZAP.
1675 if (feat->dif_recsize_max > ZAP_MAXVALUELEN)
1677 if (feat->dif_recsize_max != feat->dif_recsize_min)
1680 obj->oo_keysize = feat->dif_keysize_max;
1681 obj->oo_recsize = feat->dif_recsize_max;
1682 obj->oo_recusize = 1;
1684 /* ZFS prefers to work with array of 64bits */
1685 if ((obj->oo_recsize & 7) == 0) {
1686 obj->oo_recsize >>= 3;
1687 obj->oo_recusize = 8;
1689 dt->do_index_ops = &osd_index_ops;