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.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2009, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2012, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
31 * lustre/osd-zfs/osd_oi.c
32 * OI functions to map fid to dnode
34 * Author: Alex Zhuravlev <bzzz@whamcloud.com>
35 * Author: Mike Pershin <tappro@whamcloud.com>
36 * Author: Di Wang <di.wang@intel.com>
39 #define DEBUG_SUBSYSTEM S_OSD
41 #include <libcfs/libcfs.h>
42 #include <obd_support.h>
43 #include <lustre_net.h>
45 #include <obd_class.h>
46 #include <lustre_disk.h>
47 #include <lustre_fid.h>
49 #include "osd_internal.h"
51 #include <sys/dnode.h>
56 #include <sys/spa_impl.h>
57 #include <sys/zfs_znode.h>
58 #include <sys/dmu_tx.h>
59 #include <sys/dmu_objset.h>
60 #include <sys/dsl_prop.h>
61 #include <sys/sa_impl.h>
63 #include <lustre_scrub.h>
65 #define OSD_OI_FID_NR (1UL << 7)
66 unsigned int osd_oi_count = OSD_OI_FID_NR;
70 * zfs osd maintains names for known fids in the name hierarchy
71 * so that one can mount filesystem with regular ZFS stack and
79 static const struct named_oid oids[] = {
80 { .oid = LAST_RECV_OID, .name = LAST_RCVD },
81 { .oid = OFD_LAST_GROUP_OID, .name = "LAST_GROUP" },
82 { .oid = LLOG_CATALOGS_OID, .name = "CATALOGS" },
83 { .oid = MGS_CONFIGS_OID, /*MOUNT_CONFIGS_DIR*/ },
84 { .oid = FID_SEQ_SRV_OID, .name = "seq_srv" },
85 { .oid = FID_SEQ_CTL_OID, .name = "seq_ctl" },
86 { .oid = FLD_INDEX_OID, .name = "fld" },
87 { .oid = MDD_LOV_OBJ_OID, .name = LOV_OBJID },
88 { .oid = OFD_HEALTH_CHECK_OID, .name = HEALTH_CHECK },
89 { .oid = REPLY_DATA_OID, .name = REPLY_DATA },
90 { .oid = MDD_LOV_OBJ_OSEQ, .name = LOV_OBJSEQ },
91 { .oid = BATCHID_COMMITTED_OID, .name = "BATCHID" },
95 static inline bool fid_is_objseq(const struct lu_fid *fid)
97 return fid->f_seq == FID_SEQ_LOCAL_FILE &&
98 fid->f_oid == MDD_LOV_OBJ_OSEQ;
101 static inline bool fid_is_batchid(const struct lu_fid *fid)
103 return fid->f_seq == FID_SEQ_LOCAL_FILE &&
104 fid->f_oid == BATCHID_COMMITTED_OID;
107 static char *oid2name(const unsigned long oid)
111 while (oids[i].oid) {
112 if (oids[i].oid == oid)
120 * Lookup an existing OI by the given name.
123 osd_oi_lookup(const struct lu_env *env, struct osd_device *o,
124 uint64_t parent, const char *name, struct osd_oi *oi)
126 struct zpl_direntry *zde = &osd_oti_get(env)->oti_zde.lzd_reg;
129 rc = -zap_lookup(o->od_os, parent, name, 8, 1, (void *)zde);
133 rc = strlcpy(oi->oi_name, name, sizeof(oi->oi_name));
134 if (rc >= sizeof(oi->oi_name))
137 oi->oi_zapid = zde->zde_dnode;
142 static int osd_obj_create(const struct lu_env *env, struct osd_device *o,
143 uint64_t parent, const char *name, uint64_t *child,
144 const struct lu_fid *fid, bool isdir)
146 struct osd_thread_info *info = osd_oti_get(env);
147 struct zpl_direntry *zde = &info->oti_zde.lzd_reg;
148 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
149 struct lu_attr *la = &info->oti_la;
150 sa_handle_t *sa_hdl = NULL;
151 nvlist_t *nvbuf = NULL;
154 __u32 compat = LMAC_NOT_IN_OI;
158 if (o->od_dt_dev.dd_rdonly)
161 memset(la, 0, sizeof(*la));
162 la->la_valid = LA_MODE | LA_UID | LA_GID;
163 la->la_mode = S_IRUGO | S_IWUSR | (isdir ? S_IXUGO | S_IFDIR : S_IFREG);
166 rc = -nvlist_alloc(&nvbuf, NV_UNIQUE_NAME, KM_SLEEP);
171 compat |= LMAC_FID_ON_OST;
172 lustre_lma_init(lma, fid, compat, 0);
173 lustre_lma_swab(lma);
174 rc = -nvlist_add_byte_array(nvbuf, XATTR_NAME_LMA,
175 (uchar_t *)lma, sizeof(*lma));
180 /* create fid-to-dnode index */
181 tx = dmu_tx_create(o->od_os);
183 GOTO(out, rc = -ENOMEM);
185 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
186 dmu_tx_hold_bonus(tx, parent);
187 dmu_tx_hold_zap(tx, parent, TRUE, name);
188 dmu_tx_hold_sa_create(tx, ZFS_SA_BASE_ATTR_SIZE);
189 rc = -dmu_tx_assign(tx, TXG_WAIT);
196 oid = osd_zap_create_flags(o->od_os, 0, ZAP_FLAG_HASH64,
197 DMU_OT_DIRECTORY_CONTENTS,
198 14, DN_MAX_INDBLKSHIFT, 0, tx);
200 oid = osd_dmu_object_alloc(o->od_os, DMU_OTN_UINT8_METADATA,
202 rc = -sa_handle_get(o->od_os, oid, NULL, SA_HDL_PRIVATE, &sa_hdl);
206 rc = __osd_attr_init(env, o, NULL, sa_hdl, tx, la, parent, nvbuf);
207 sa_handle_destroy(sa_hdl);
211 zde->zde_dnode = oid;
213 zde->zde_type = S_DT(isdir ? S_IFDIR : S_IFREG);
214 rc = -zap_add(o->od_os, parent, name, 8, 1, (void *)zde, tx);
220 dmu_object_free(o->od_os, oid, tx);
230 static int osd_oi_destroy(const struct lu_env *env, struct osd_device *o,
241 if (o->od_dt_dev.dd_rdonly)
244 rc = osd_oi_lookup(env, o, o->od_rootid, name, &oi);
252 rc = __osd_obj2dnode(o->od_os, o->od_rootid, &rootdn);
256 tx = dmu_tx_create(o->od_os);
257 dmu_tx_mark_netfree(tx);
258 dmu_tx_hold_free(tx, oid, 0, DMU_OBJECT_END);
259 osd_tx_hold_zap(tx, oid, rootdn, FALSE, NULL);
260 rc = -dmu_tx_assign(tx, TXG_WAIT);
266 rc = -dmu_object_free(o->od_os, oid, tx);
268 CERROR("%s: failed to free %s %llu: rc = %d\n",
269 o->od_svname, name, oid, rc);
273 rc = osd_zap_remove(o, o->od_rootid, rootdn, name, tx);
275 CERROR("%s: zap_remove %s failed: rc = %d\n",
276 o->od_svname, name, rc);
284 osd_dnode_rele(rootdn);
290 osd_oi_find_or_create(const struct lu_env *env, struct osd_device *o,
291 uint64_t parent, const char *name, uint64_t *child)
296 rc = osd_oi_lookup(env, o, parent, name, &oi);
298 *child = oi.oi_zapid;
299 else if (rc == -ENOENT)
300 rc = osd_obj_create(env, o, parent, name, child, NULL, true);
305 int osd_obj_find_or_create(const struct lu_env *env, struct osd_device *o,
306 uint64_t parent, const char *name, uint64_t *child,
307 const struct lu_fid *fid, bool isdir)
312 rc = osd_oi_lookup(env, o, parent, name, &oi);
314 *child = oi.oi_zapid;
315 else if (rc == -ENOENT)
316 rc = osd_obj_create(env, o, parent, name, child, fid, isdir);
322 * Lookup the target index/flags of the fid, so it will know where
323 * the object is located (tgt index) and it is MDT or OST object.
325 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
326 u64 seq, struct lu_seq_range *range)
328 struct seq_server_site *ss = osd_seq_site(osd);
330 if (fid_seq_is_idif(seq)) {
331 fld_range_set_ost(range);
332 range->lsr_index = idif_ost_idx(seq);
336 if (!fid_seq_in_fldb(seq)) {
337 fld_range_set_mdt(range);
339 /* FIXME: If ss is NULL, it suppose not get lsr_index
341 range->lsr_index = ss->ss_node_id;
345 /* The seq_server_site may be NOT ready during initial OI scrub */
346 if (unlikely(!ss || !ss->ss_server_fld ||
347 !ss->ss_server_fld->lsf_cache))
350 fld_range_set_any(range);
351 /* OSD will only do local fld lookup */
352 return fld_local_lookup(env, ss->ss_server_fld, seq, range);
355 int fid_is_on_ost(const struct lu_env *env, struct osd_device *osd,
356 const struct lu_fid *fid)
358 struct lu_seq_range *range = &osd_oti_get(env)->oti_seq_range;
362 if (fid_is_idif(fid))
365 if (unlikely(fid_is_local_file(fid) || fid_is_llog(fid)) ||
366 fid_is_name_llog(fid) || fid_is_quota(fid) ||
370 rc = osd_fld_lookup(env, osd, fid_seq(fid), range);
372 /* During upgrade, OST FLDB might not be loaded because
373 * OST FLDB is not created until 2.6, so if some DNE
374 * filesystem upgrade from 2.5 to 2.7/2.8, they will
375 * not be able to find the sequence from local FLDB
376 * cache see fld_index_init(). */
377 if (rc == -ENOENT && osd->od_is_ost)
381 CERROR("%s: "DFID" lookup failed: rc = %d\n",
382 osd_name(osd), PFID(fid), rc);
386 if (fld_range_is_ost(range))
392 static struct osd_seq *osd_seq_find_locked(struct osd_seq_list *seq_list,
395 struct osd_seq *osd_seq;
397 list_for_each_entry(osd_seq, &seq_list->osl_seq_list, os_seq_list) {
398 if (osd_seq->os_seq == seq)
404 static struct osd_seq *osd_seq_find(struct osd_seq_list *seq_list, u64 seq)
406 struct osd_seq *osd_seq;
408 read_lock(&seq_list->osl_seq_list_lock);
409 osd_seq = osd_seq_find_locked(seq_list, seq);
410 read_unlock(&seq_list->osl_seq_list_lock);
415 static struct osd_seq *osd_find_or_add_seq(const struct lu_env *env,
416 struct osd_device *osd, u64 seq)
418 struct osd_seq_list *seq_list = &osd->od_seq_list;
419 struct osd_seq *osd_seq;
420 char *key = osd_oti_get(env)->oti_buf;
421 char *seq_name = osd_oti_get(env)->oti_str;
428 osd_seq = osd_seq_find(seq_list, seq);
432 down(&seq_list->osl_seq_init_sem);
433 /* Check again, in case some one else already add it
435 osd_seq = osd_seq_find(seq_list, seq);
439 OBD_ALLOC_PTR(osd_seq);
441 GOTO(out, rc = -ENOMEM);
443 INIT_LIST_HEAD(&osd_seq->os_seq_list);
444 osd_seq->os_seq = seq;
446 /* Init subdir count to be 32, but each seq can have
447 * different subdir count */
448 osd_seq->os_subdir_count = OSD_OST_MAP_SIZE;
449 OBD_ALLOC_PTR_ARRAY(osd_seq->os_compat_dirs, osd_seq->os_subdir_count);
450 if (osd_seq->os_compat_dirs == NULL)
451 GOTO(out, rc = -ENOMEM);
453 oi.oi_zapid = osd->od_O_id;
454 sprintf(seq_name, (fid_seq_is_rsvd(seq) ||
455 fid_seq_is_mdt0(seq)) ? "%llu" : "%llx",
456 fid_seq_is_idif(seq) ? 0 : seq);
458 rc = osd_oi_find_or_create(env, osd, oi.oi_zapid, seq_name, &odb);
460 CERROR("%s: Can not create %s : rc = %d\n",
461 osd_name(osd), seq_name, rc);
465 osd_seq->os_oid = odb;
466 for (i = 0; i < OSD_OST_MAP_SIZE; i++) {
467 sprintf(key, "d%d", i);
468 rc = osd_oi_find_or_create(env, osd, odb, key, &sdb);
471 osd_seq->os_compat_dirs[i] = sdb;
474 write_lock(&seq_list->osl_seq_list_lock);
475 list_add(&osd_seq->os_seq_list, &seq_list->osl_seq_list);
476 write_unlock(&seq_list->osl_seq_list_lock);
478 up(&seq_list->osl_seq_init_sem);
480 if (osd_seq != NULL && osd_seq->os_compat_dirs != NULL)
481 OBD_FREE_PTR_ARRAY(osd_seq->os_compat_dirs,
482 osd_seq->os_subdir_count);
484 OBD_FREE_PTR(osd_seq);
485 osd_seq = ERR_PTR(rc);
491 osd_get_idx_for_ost_obj_compat(const struct lu_env *env, struct osd_device *osd,
492 const struct lu_fid *fid, char *buf, int bufsize)
494 struct osd_seq *osd_seq;
499 osd_seq = osd_find_or_add_seq(env, osd, fid_seq(fid));
500 if (IS_ERR(osd_seq)) {
501 CERROR("%s: Can not find seq group "DFID"\n", osd_name(osd),
503 return PTR_ERR(osd_seq);
506 if (fid_is_last_id(fid)) {
509 rc = fid_to_ostid(fid, &osd_oti_get(env)->oti_ostid);
510 LASSERT(rc == 0); /* we should not get here with IGIF */
511 id = ostid_id(&osd_oti_get(env)->oti_ostid);
514 b = id % OSD_OST_MAP_SIZE;
515 LASSERT(osd_seq->os_compat_dirs[b]);
518 snprintf(buf, bufsize, "%llu", id);
520 return osd_seq->os_compat_dirs[b];
524 * objects w/o a natural reference (unlike a file on a MDS)
525 * are put under a special hierarchy /O/<seq>/d0..dXX
526 * this function returns a directory specific fid belongs to
529 osd_get_idx_for_ost_obj(const struct lu_env *env, struct osd_device *osd,
530 const struct lu_fid *fid, char *buf, int bufsize)
532 struct osd_seq *osd_seq;
537 osd_seq = osd_find_or_add_seq(env, osd, fid_seq(fid));
538 if (IS_ERR(osd_seq)) {
539 CERROR("%s: Can not find seq group "DFID"\n", osd_name(osd),
541 return PTR_ERR(osd_seq);
544 if (fid_is_last_id(fid)) {
546 snprintf(buf, bufsize, "LAST_ID");
548 return osd_seq->os_oid;
551 rc = fid_to_ostid(fid, &osd_oti_get(env)->oti_ostid);
552 LASSERT(rc == 0); /* we should not get here with IGIF */
554 id = ostid_id(&osd_oti_get(env)->oti_ostid);
555 b = id % OSD_OST_MAP_SIZE;
556 LASSERT(osd_seq->os_compat_dirs[b]);
559 snprintf(buf, bufsize, "%llu", id);
561 return osd_seq->os_compat_dirs[b];
565 * Determine the zap object id which is being used as the OI for the
566 * given fid. The lowest N bits in the sequence ID are used as the
567 * index key. On failure 0 is returned which zfs treats internally
568 * as an invalid object id.
571 osd_get_idx_for_fid(struct osd_device *osd, const struct lu_fid *fid,
572 char *buf, dnode_t **zdn, int bufsize)
576 oi = osd_fid2oi(osd, fid);
578 osd_fid2str(buf, fid, bufsize);
586 osd_get_name_n_idx_compat(const struct lu_env *env, struct osd_device *osd,
587 const struct lu_fid *fid, char *buf, int bufsize,
593 LASSERT(!fid_is_acct(fid));
598 if (fid_is_echo(fid) || fid_is_on_ost(env, osd, fid)) {
599 zapid = osd_get_idx_for_ost_obj_compat(env, osd, fid,
601 } else if (unlikely(fid_seq(fid) == FID_SEQ_LOCAL_FILE)) {
602 /* special objects with fixed known fids get their name */
603 char *name = oid2name(fid_oid(fid));
606 zapid = osd->od_root;
608 strncpy(buf, name, bufsize);
610 zapid = osd_get_idx_for_fid(osd, fid, buf, NULL,
614 zapid = osd_get_idx_for_fid(osd, fid, buf, zdn, bufsize);
620 uint64_t osd_get_name_n_idx(const struct lu_env *env, struct osd_device *osd,
621 const struct lu_fid *fid, char *buf, int bufsize,
627 LASSERT(!fid_is_acct(fid));
632 if (fid_is_echo(fid) || fid_is_last_id(fid) ||
633 fid_is_on_ost(env, osd, fid)) {
634 zapid = osd_get_idx_for_ost_obj(env, osd, fid, buf, bufsize);
635 } else if (unlikely(fid_seq(fid) == FID_SEQ_LOCAL_FILE)) {
636 /* special objects with fixed known fids get their name */
637 char *name = oid2name(fid_oid(fid));
640 zapid = osd->od_root;
642 strncpy(buf, name, bufsize);
644 zapid = osd_get_idx_for_fid(osd, fid, buf, NULL,
648 zapid = osd_get_idx_for_fid(osd, fid, buf, zdn, bufsize);
654 static inline int fid_is_fs_root(const struct lu_fid *fid)
656 /* Map root inode to special local object FID */
657 return fid_seq(fid) == FID_SEQ_LOCAL_FILE &&
658 fid_oid(fid) == OSD_FS_ROOT_OID;
661 int osd_fid_lookup(const struct lu_env *env, struct osd_device *dev,
662 const struct lu_fid *fid, uint64_t *oid)
664 struct osd_thread_info *info = osd_oti_get(env);
665 char *buf = info->oti_buf;
671 if (CFS_FAIL_CHECK(OBD_FAIL_SRV_ENOENT))
674 LASSERT(!fid_is_acct(fid));
676 if (unlikely(fid_is_fs_root(fid))) {
679 zapid = osd_get_name_n_idx(env, dev, fid, buf,
680 sizeof(info->oti_buf), &zdn);
681 rc = osd_zap_lookup(dev, zapid, zdn, buf,
682 8, 1, &info->oti_zde);
684 if (unlikely(fid_is_last_id(fid))) {
685 zapid = osd_get_name_n_idx_compat(env, dev, fid,
686 buf, sizeof(info->oti_buf), &zdn);
687 rc = osd_zap_lookup(dev, zapid, zdn, buf,
688 8, 1, &info->oti_zde);
689 } else if (fid_is_objseq(fid) || fid_is_batchid(fid)) {
690 zapid = osd_get_idx_for_fid(dev, fid,
691 buf, NULL, sizeof(info->oti_buf));
692 rc = osd_zap_lookup(dev, zapid, zdn, buf,
693 8, 1, &info->oti_zde);
699 *oid = info->oti_zde.lzd_reg.zde_dnode;
703 osd_dmu_prefetch(dev->od_os, *oid, 0, 0, 0,
704 ZIO_PRIORITY_ASYNC_READ);
710 * Close an entry in a specific slot.
713 osd_oi_remove_table(const struct lu_env *env, struct osd_device *o, int key)
717 LASSERT(key < o->od_oi_count);
719 oi = o->od_oi_table[key];
722 osd_dnode_rele(oi->oi_dn);
724 o->od_oi_table[key] = NULL;
729 * Allocate and open a new entry in the specified unused slot.
732 osd_oi_add_table(const struct lu_env *env, struct osd_device *o,
738 LASSERT(key < o->od_oi_count);
739 LASSERT(o->od_oi_table[key] == NULL);
745 rc = osd_oi_lookup(env, o, o->od_root, name, oi);
751 o->od_oi_table[key] = oi;
752 __osd_obj2dnode(o->od_os, oi->oi_zapid, &oi->oi_dn);
758 * Depopulate the OI table.
761 osd_oi_close_table(const struct lu_env *env, struct osd_device *o)
765 for (i = 0; i < o->od_oi_count; i++)
766 osd_oi_remove_table(env, o, i);
770 * Populate the OI table based.
773 osd_oi_open_table(const struct lu_env *env, struct osd_device *o, int count)
779 for (i = 0; i < count; i++) {
780 sprintf(name, "%s.%d", DMU_OSD_OI_NAME_BASE, i);
781 rc = osd_oi_add_table(env, o, name, i);
783 osd_oi_close_table(env, o);
792 * Determine if the type and number of OIs used by this file system.
794 static int osd_oi_probe(const struct lu_env *env, struct osd_device *o)
796 struct lustre_scrub *scrub = &o->od_scrub;
797 struct scrub_file *sf = &scrub->os_file;
800 int max = sf->sf_oi_count > 0 ? sf->sf_oi_count : OSD_OI_FID_NR_MAX;
806 * Check for multiple OIs and determine the count. There is no
807 * gap handling, if an OI is missing the wrong size can be returned.
808 * The only safeguard is that we know the number of OIs must be a
809 * power of two and this is checked for basic sanity.
811 for (count = 0; count < max; count++) {
812 snprintf(name, sizeof(name) - 1, "%s.%d",
813 DMU_OSD_OI_NAME_BASE, count);
814 rc = osd_oi_lookup(env, o, o->od_root, name, &oi);
819 if (sf->sf_oi_count == 0)
822 zfs_set_bit(count, sf->sf_oi_bitmap);
833 static void osd_ost_seq_fini(const struct lu_env *env, struct osd_device *osd)
835 struct osd_seq_list *osl = &osd->od_seq_list;
836 struct osd_seq *osd_seq, *tmp;
838 write_lock(&osl->osl_seq_list_lock);
839 list_for_each_entry_safe(osd_seq, tmp, &osl->osl_seq_list,
841 list_del(&osd_seq->os_seq_list);
842 OBD_FREE_PTR_ARRAY(osd_seq->os_compat_dirs,
843 osd_seq->os_subdir_count);
844 OBD_FREE(osd_seq, sizeof(*osd_seq));
846 write_unlock(&osl->osl_seq_list_lock);
850 * Create /O subdirectory to map legacy OST objects for compatibility.
853 osd_oi_init_compat(const struct lu_env *env, struct osd_device *o)
859 rc = osd_oi_find_or_create(env, o, o->od_root, "O", &sdb);
867 osd_oi_init_index_backup(const struct lu_env *env, struct osd_device *o)
869 struct lu_fid *fid = &osd_oti_get(env)->oti_fid;
873 lu_local_obj_fid(fid, INDEX_BACKUP_OID);
874 rc = osd_obj_find_or_create(env, o, o->od_root, INDEX_BACKUP_DIR,
875 &o->od_index_backup_id, fid, true);
881 osd_oi_init_remote_parent(const struct lu_env *env, struct osd_device *o)
888 o->od_remote_parent_dir = ZFS_NO_OBJECT;
890 /* Remote parent only used for cross-MDT objects,
891 * it is usless for single MDT case or under read
892 * only mode. So ignore the failure. */
893 rc = osd_oi_find_or_create(env, o, o->od_root,
894 REMOTE_PARENT_DIR, &sdb);
896 o->od_remote_parent_dir = sdb;
898 o->od_remote_parent_dir = ZFS_NO_OBJECT;
903 * Initialize the OIs by either opening or creating them as needed.
905 int osd_oi_init(const struct lu_env *env, struct osd_device *o, bool reset)
907 struct lustre_scrub *scrub = &o->od_scrub;
908 struct scrub_file *sf = &scrub->os_file;
909 char *key = osd_oti_get(env)->oti_buf;
914 LASSERTF((sf->sf_oi_count & (sf->sf_oi_count - 1)) == 0,
915 "Invalid OI count in scrub file %d\n", sf->sf_oi_count);
917 rc = osd_oi_init_index_backup(env, o);
921 osd_oi_init_remote_parent(env, o);
923 rc = osd_oi_init_compat(env, o);
927 count = osd_oi_probe(env, o);
929 GOTO(out, rc = count);
932 if (count == sf->sf_oi_count && !reset)
935 if (sf->sf_oi_count == 0) {
936 if (likely((count & (count - 1)) == 0)) {
937 sf->sf_oi_count = count;
938 rc = scrub_file_store(env, scrub);
945 LCONSOLE_ERROR("%s: invalid oi count %d. You can "
946 "remove all OIs, then remount it\n",
948 GOTO(out, rc = -EDOM);
951 scrub_file_reset(scrub, o->od_uuid, SF_RECREATED);
952 count = sf->sf_oi_count;
954 if (sf->sf_oi_count > 0) {
955 count = sf->sf_oi_count;
956 memset(sf->sf_oi_bitmap, 0, SCRUB_OI_BITMAP_SIZE);
957 for (i = 0; i < count; i++)
958 zfs_set_bit(i, sf->sf_oi_bitmap);
959 scrub_file_reset(scrub, o->od_uuid, SF_RECREATED);
961 count = sf->sf_oi_count = osd_oi_count;
965 rc = scrub_file_store(env, scrub);
970 LCONSOLE_WARN("%s: reset Object Index mappings\n",
973 for (i = 0; i < count; i++) {
974 LASSERT(sizeof(osd_oti_get(env)->oti_buf) >= 32);
976 snprintf(key, sizeof(osd_oti_get(env)->oti_buf) - 1,
977 "%s.%d", DMU_OSD_OI_NAME_BASE, i);
979 rc = osd_oi_destroy(env, o, key);
983 rc = osd_oi_find_or_create(env, o, o->od_root, key, &sdb);
989 LASSERT((count & (count - 1)) == 0);
990 o->od_oi_count = count;
991 OBD_ALLOC_PTR_ARRAY(o->od_oi_table, count);
992 if (o->od_oi_table == NULL)
993 GOTO(out, rc = -ENOMEM);
995 rc = osd_oi_open_table(env, o, count);
1001 osd_ost_seq_fini(env, o);
1003 if (o->od_oi_table) {
1004 OBD_FREE_PTR_ARRAY(o->od_oi_table, count);
1005 o->od_oi_table = NULL;
1012 void osd_oi_fini(const struct lu_env *env, struct osd_device *o)
1016 osd_ost_seq_fini(env, o);
1018 if (o->od_oi_table != NULL) {
1019 (void) osd_oi_close_table(env, o);
1020 OBD_FREE_PTR_ARRAY(o->od_oi_table, o->od_oi_count);
1021 o->od_oi_table = NULL;
1028 int osd_options_init(void)
1030 /* osd_oi_count - Default number of OIs, 128 works well for ZFS */
1031 if (osd_oi_count == 0 || osd_oi_count > OSD_OI_FID_NR_MAX)
1032 osd_oi_count = OSD_OI_FID_NR;
1034 if ((osd_oi_count & (osd_oi_count - 1)) != 0) {
1035 LCONSOLE_WARN("Round up osd_oi_count %d to power2 %d\n",
1036 osd_oi_count, size_roundup_power2(osd_oi_count));
1037 osd_oi_count = size_roundup_power2(osd_oi_count);
1044 * the following set of functions are used to maintain per-thread
1045 * cache of FID->ino mapping. this mechanism is used to avoid
1046 * expensive LU/OI lookups.
1048 struct osd_idmap_cache *osd_idc_find(const struct lu_env *env,
1049 struct osd_device *osd,
1050 const struct lu_fid *fid)
1052 struct osd_thread_info *oti = osd_oti_get(env);
1053 struct osd_idmap_cache *idc = oti->oti_ins_cache;
1056 for (i = 0; i < oti->oti_ins_cache_used; i++) {
1057 if (!lu_fid_eq(&idc[i].oic_fid, fid))
1059 if (idc[i].oic_dev != osd)
1068 struct osd_idmap_cache *osd_idc_add(const struct lu_env *env,
1069 struct osd_device *osd,
1070 const struct lu_fid *fid)
1072 struct osd_thread_info *oti = osd_oti_get(env);
1073 struct osd_idmap_cache *idc;
1076 if (unlikely(oti->oti_ins_cache_used >= oti->oti_ins_cache_size)) {
1077 i = oti->oti_ins_cache_size * 2;
1079 i = OSD_INS_CACHE_SIZE;
1080 OBD_ALLOC_PTR_ARRAY_LARGE(idc, i);
1082 return ERR_PTR(-ENOMEM);
1083 if (oti->oti_ins_cache != NULL) {
1084 memcpy(idc, oti->oti_ins_cache,
1085 oti->oti_ins_cache_used * sizeof(*idc));
1086 OBD_FREE_PTR_ARRAY_LARGE(oti->oti_ins_cache,
1087 oti->oti_ins_cache_used);
1089 oti->oti_ins_cache = idc;
1090 oti->oti_ins_cache_size = i;
1093 idc = &oti->oti_ins_cache[oti->oti_ins_cache_used++];
1094 idc->oic_fid = *fid;
1097 idc->oic_remote = 0;
1103 * Lookup mapping for the given fid in the cache
1105 * Initialize a new one if not found. the initialization checks whether
1106 * the object is local or remote. for the local objects, OI is used to
1107 * learn dnode#. the function is used when the caller has no information
1108 * about the object, e.g. at dt_insert().
1110 struct osd_idmap_cache *osd_idc_find_or_init(const struct lu_env *env,
1111 struct osd_device *osd,
1112 const struct lu_fid *fid)
1114 struct osd_idmap_cache *idc;
1117 LASSERT(!fid_is_acct(fid));
1119 idc = osd_idc_find(env, osd, fid);
1123 CDEBUG(D_INODE, "%s: FID "DFID" not in the id map cache\n",
1124 osd->od_svname, PFID(fid));
1126 /* new mapping is needed */
1127 idc = osd_idc_add(env, osd, fid);
1129 CERROR("%s: FID "DFID" add id map cache failed: %ld\n",
1130 osd->od_svname, PFID(fid), PTR_ERR(idc));
1135 rc = osd_remote_fid(env, osd, fid);
1136 if (unlikely(rc < 0))
1140 /* the object is local, lookup in OI */
1143 rc = osd_fid_lookup(env, osd, fid, &dnode);
1144 if (unlikely(rc < 0)) {
1145 CERROR("%s: can't lookup: rc = %d\n",
1146 osd->od_svname, rc);
1149 LASSERT(dnode < (1ULL << DN_MAX_OBJECT_SHIFT));
1150 idc->oic_dnode = dnode;
1152 /* the object is remote */
1153 idc->oic_remote = 1;
1160 * lookup mapping for given FID and fill it from the given object.
1161 * the object is local by definition.
1163 int osd_idc_find_and_init(const struct lu_env *env, struct osd_device *osd,
1164 struct osd_object *obj)
1166 const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
1167 struct osd_idmap_cache *idc;
1169 idc = osd_idc_find(env, osd, fid);
1171 if (obj->oo_dn == NULL)
1173 idc->oic_dnode = obj->oo_dn->dn_object;
1177 CDEBUG(D_INODE, "%s: FID "DFID" not in the id map cache\n",
1178 osd->od_svname, PFID(fid));
1180 /* new mapping is needed */
1181 idc = osd_idc_add(env, osd, fid);
1183 CERROR("%s: FID "DFID" add id map cache failed: %ld\n",
1184 osd->od_svname, PFID(fid), PTR_ERR(idc));
1185 return PTR_ERR(idc);
1189 idc->oic_dnode = obj->oo_dn->dn_object;
1194 int osd_idc_find_and_init_with_oid(const struct lu_env *env,
1195 struct osd_device *osd,
1196 const struct lu_fid *fid,
1199 struct osd_idmap_cache *idc;
1201 idc = osd_idc_find(env, osd, fid);
1203 idc = osd_idc_add(env, osd, fid);
1205 return PTR_ERR(idc);
1208 idc->oic_dnode = oid;
1209 idc->oic_remote = 0;