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/
30 * Lustre is a trademark of Sun Microsystems, Inc.
32 * lustre/osd-zfs/osd_oi.c
33 * OI functions to map fid to dnode
35 * Author: Alex Zhuravlev <bzzz@whamcloud.com>
36 * Author: Mike Pershin <tappro@whamcloud.com>
37 * Author: Di Wang <di.wang@intel.com>
40 #define DEBUG_SUBSYSTEM S_OSD
42 #include <libcfs/libcfs.h>
43 #include <obd_support.h>
44 #include <lustre_net.h>
46 #include <obd_class.h>
47 #include <lustre_disk.h>
48 #include <lustre_fid.h>
50 #include "osd_internal.h"
52 #include <sys/dnode.h>
57 #include <sys/spa_impl.h>
58 #include <sys/zfs_znode.h>
59 #include <sys/dmu_tx.h>
60 #include <sys/dmu_objset.h>
61 #include <sys/dsl_prop.h>
62 #include <sys/sa_impl.h>
64 #include <lustre_scrub.h>
66 #define OSD_OI_FID_NR (1UL << 7)
67 unsigned int osd_oi_count = OSD_OI_FID_NR;
71 * zfs osd maintains names for known fids in the name hierarchy
72 * so that one can mount filesystem with regular ZFS stack and
80 static const struct named_oid oids[] = {
81 { .oid = LAST_RECV_OID, .name = LAST_RCVD },
82 { .oid = OFD_LAST_GROUP_OID, .name = "LAST_GROUP" },
83 { .oid = LLOG_CATALOGS_OID, .name = "CATALOGS" },
84 { .oid = MGS_CONFIGS_OID, /*MOUNT_CONFIGS_DIR*/ },
85 { .oid = FID_SEQ_SRV_OID, .name = "seq_srv" },
86 { .oid = FID_SEQ_CTL_OID, .name = "seq_ctl" },
87 { .oid = FLD_INDEX_OID, .name = "fld" },
88 { .oid = MDD_LOV_OBJ_OID, .name = LOV_OBJID },
89 { .oid = OFD_HEALTH_CHECK_OID, .name = HEALTH_CHECK },
90 { .oid = REPLY_DATA_OID, .name = REPLY_DATA },
91 { .oid = MDD_LOV_OBJ_OSEQ, .name = LOV_OBJSEQ },
92 { .oid = BATCHID_COMMITTED_OID, .name = "BATCHID" },
96 static inline bool fid_is_objseq(const struct lu_fid *fid)
98 return fid->f_seq == FID_SEQ_LOCAL_FILE &&
99 fid->f_oid == MDD_LOV_OBJ_OSEQ;
102 static inline bool fid_is_batchid(const struct lu_fid *fid)
104 return fid->f_seq == FID_SEQ_LOCAL_FILE &&
105 fid->f_oid == BATCHID_COMMITTED_OID;
108 static char *oid2name(const unsigned long oid)
112 while (oids[i].oid) {
113 if (oids[i].oid == oid)
121 * Lookup an existing OI by the given name.
124 osd_oi_lookup(const struct lu_env *env, struct osd_device *o,
125 uint64_t parent, const char *name, struct osd_oi *oi)
127 struct zpl_direntry *zde = &osd_oti_get(env)->oti_zde.lzd_reg;
130 rc = -zap_lookup(o->od_os, parent, name, 8, 1, (void *)zde);
134 rc = strlcpy(oi->oi_name, name, sizeof(oi->oi_name));
135 if (rc >= sizeof(oi->oi_name))
138 oi->oi_zapid = zde->zde_dnode;
143 static int osd_obj_create(const struct lu_env *env, struct osd_device *o,
144 uint64_t parent, const char *name, uint64_t *child,
145 const struct lu_fid *fid, bool isdir)
147 struct osd_thread_info *info = osd_oti_get(env);
148 struct zpl_direntry *zde = &info->oti_zde.lzd_reg;
149 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
150 struct lu_attr *la = &info->oti_la;
151 sa_handle_t *sa_hdl = NULL;
152 nvlist_t *nvbuf = NULL;
155 __u32 compat = LMAC_NOT_IN_OI;
159 if (o->od_dt_dev.dd_rdonly)
162 memset(la, 0, sizeof(*la));
163 la->la_valid = LA_MODE | LA_UID | LA_GID;
164 la->la_mode = S_IRUGO | S_IWUSR | (isdir ? S_IXUGO | S_IFDIR : S_IFREG);
167 rc = -nvlist_alloc(&nvbuf, NV_UNIQUE_NAME, KM_SLEEP);
172 compat |= LMAC_FID_ON_OST;
173 lustre_lma_init(lma, fid, compat, 0);
174 lustre_lma_swab(lma);
175 rc = -nvlist_add_byte_array(nvbuf, XATTR_NAME_LMA,
176 (uchar_t *)lma, sizeof(*lma));
181 /* create fid-to-dnode index */
182 tx = dmu_tx_create(o->od_os);
184 GOTO(out, rc = -ENOMEM);
186 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
187 dmu_tx_hold_bonus(tx, parent);
188 dmu_tx_hold_zap(tx, parent, TRUE, name);
189 dmu_tx_hold_sa_create(tx, ZFS_SA_BASE_ATTR_SIZE);
190 rc = -dmu_tx_assign(tx, TXG_WAIT);
197 oid = osd_zap_create_flags(o->od_os, 0, ZAP_FLAG_HASH64,
198 DMU_OT_DIRECTORY_CONTENTS,
199 14, DN_MAX_INDBLKSHIFT, 0, tx);
201 oid = osd_dmu_object_alloc(o->od_os, DMU_OTN_UINT8_METADATA,
203 rc = -sa_handle_get(o->od_os, oid, NULL, SA_HDL_PRIVATE, &sa_hdl);
207 rc = __osd_attr_init(env, o, NULL, sa_hdl, tx, la, parent, nvbuf);
208 sa_handle_destroy(sa_hdl);
212 zde->zde_dnode = oid;
214 zde->zde_type = S_DT(isdir ? S_IFDIR : S_IFREG);
215 rc = -zap_add(o->od_os, parent, name, 8, 1, (void *)zde, tx);
221 dmu_object_free(o->od_os, oid, tx);
231 static int osd_oi_destroy(const struct lu_env *env, struct osd_device *o,
242 if (o->od_dt_dev.dd_rdonly)
245 rc = osd_oi_lookup(env, o, o->od_rootid, name, &oi);
253 rc = __osd_obj2dnode(o->od_os, o->od_rootid, &rootdn);
257 tx = dmu_tx_create(o->od_os);
258 dmu_tx_mark_netfree(tx);
259 dmu_tx_hold_free(tx, oid, 0, DMU_OBJECT_END);
260 osd_tx_hold_zap(tx, oid, rootdn, FALSE, NULL);
261 rc = -dmu_tx_assign(tx, TXG_WAIT);
267 rc = -dmu_object_free(o->od_os, oid, tx);
269 CERROR("%s: failed to free %s %llu: rc = %d\n",
270 o->od_svname, name, oid, rc);
274 rc = osd_zap_remove(o, o->od_rootid, rootdn, name, tx);
276 CERROR("%s: zap_remove %s failed: rc = %d\n",
277 o->od_svname, name, rc);
285 osd_dnode_rele(rootdn);
291 osd_oi_find_or_create(const struct lu_env *env, struct osd_device *o,
292 uint64_t parent, const char *name, uint64_t *child)
297 rc = osd_oi_lookup(env, o, parent, name, &oi);
299 *child = oi.oi_zapid;
300 else if (rc == -ENOENT)
301 rc = osd_obj_create(env, o, parent, name, child, NULL, true);
306 int osd_obj_find_or_create(const struct lu_env *env, struct osd_device *o,
307 uint64_t parent, const char *name, uint64_t *child,
308 const struct lu_fid *fid, bool isdir)
313 rc = osd_oi_lookup(env, o, parent, name, &oi);
315 *child = oi.oi_zapid;
316 else if (rc == -ENOENT)
317 rc = osd_obj_create(env, o, parent, name, child, fid, isdir);
323 * Lookup the target index/flags of the fid, so it will know where
324 * the object is located (tgt index) and it is MDT or OST object.
326 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
327 u64 seq, struct lu_seq_range *range)
329 struct seq_server_site *ss = osd_seq_site(osd);
331 if (fid_seq_is_idif(seq)) {
332 fld_range_set_ost(range);
333 range->lsr_index = idif_ost_idx(seq);
337 if (!fid_seq_in_fldb(seq)) {
338 fld_range_set_mdt(range);
340 /* FIXME: If ss is NULL, it suppose not get lsr_index
342 range->lsr_index = ss->ss_node_id;
346 /* The seq_server_site may be NOT ready during initial OI scrub */
347 if (unlikely(!ss || !ss->ss_server_fld ||
348 !ss->ss_server_fld->lsf_cache))
351 fld_range_set_any(range);
352 /* OSD will only do local fld lookup */
353 return fld_local_lookup(env, ss->ss_server_fld, seq, range);
356 int fid_is_on_ost(const struct lu_env *env, struct osd_device *osd,
357 const struct lu_fid *fid)
359 struct lu_seq_range *range = &osd_oti_get(env)->oti_seq_range;
363 if (fid_is_idif(fid))
366 if (unlikely(fid_is_local_file(fid) || fid_is_llog(fid)) ||
367 fid_is_name_llog(fid) || fid_is_quota(fid) ||
371 rc = osd_fld_lookup(env, osd, fid_seq(fid), range);
373 /* During upgrade, OST FLDB might not be loaded because
374 * OST FLDB is not created until 2.6, so if some DNE
375 * filesystem upgrade from 2.5 to 2.7/2.8, they will
376 * not be able to find the sequence from local FLDB
377 * cache see fld_index_init(). */
378 if (rc == -ENOENT && osd->od_is_ost)
382 CERROR("%s: "DFID" lookup failed: rc = %d\n",
383 osd_name(osd), PFID(fid), rc);
387 if (fld_range_is_ost(range))
393 static struct osd_seq *osd_seq_find_locked(struct osd_seq_list *seq_list,
396 struct osd_seq *osd_seq;
398 list_for_each_entry(osd_seq, &seq_list->osl_seq_list, os_seq_list) {
399 if (osd_seq->os_seq == seq)
405 static struct osd_seq *osd_seq_find(struct osd_seq_list *seq_list, u64 seq)
407 struct osd_seq *osd_seq;
409 read_lock(&seq_list->osl_seq_list_lock);
410 osd_seq = osd_seq_find_locked(seq_list, seq);
411 read_unlock(&seq_list->osl_seq_list_lock);
416 static struct osd_seq *osd_find_or_add_seq(const struct lu_env *env,
417 struct osd_device *osd, u64 seq)
419 struct osd_seq_list *seq_list = &osd->od_seq_list;
420 struct osd_seq *osd_seq;
421 char *key = osd_oti_get(env)->oti_buf;
422 char *seq_name = osd_oti_get(env)->oti_str;
429 osd_seq = osd_seq_find(seq_list, seq);
433 down(&seq_list->osl_seq_init_sem);
434 /* Check again, in case some one else already add it
436 osd_seq = osd_seq_find(seq_list, seq);
440 OBD_ALLOC_PTR(osd_seq);
442 GOTO(out, rc = -ENOMEM);
444 INIT_LIST_HEAD(&osd_seq->os_seq_list);
445 osd_seq->os_seq = seq;
447 /* Init subdir count to be 32, but each seq can have
448 * different subdir count */
449 osd_seq->os_subdir_count = OSD_OST_MAP_SIZE;
450 OBD_ALLOC_PTR_ARRAY(osd_seq->os_compat_dirs, osd_seq->os_subdir_count);
451 if (osd_seq->os_compat_dirs == NULL)
452 GOTO(out, rc = -ENOMEM);
454 oi.oi_zapid = osd->od_O_id;
455 sprintf(seq_name, (fid_seq_is_rsvd(seq) ||
456 fid_seq_is_mdt0(seq)) ? "%llu" : "%llx",
457 fid_seq_is_idif(seq) ? 0 : seq);
459 rc = osd_oi_find_or_create(env, osd, oi.oi_zapid, seq_name, &odb);
461 CERROR("%s: Can not create %s : rc = %d\n",
462 osd_name(osd), seq_name, rc);
466 osd_seq->os_oid = odb;
467 for (i = 0; i < OSD_OST_MAP_SIZE; i++) {
468 sprintf(key, "d%d", i);
469 rc = osd_oi_find_or_create(env, osd, odb, key, &sdb);
472 osd_seq->os_compat_dirs[i] = sdb;
475 write_lock(&seq_list->osl_seq_list_lock);
476 list_add(&osd_seq->os_seq_list, &seq_list->osl_seq_list);
477 write_unlock(&seq_list->osl_seq_list_lock);
479 up(&seq_list->osl_seq_init_sem);
481 if (osd_seq != NULL && osd_seq->os_compat_dirs != NULL)
482 OBD_FREE_PTR_ARRAY(osd_seq->os_compat_dirs,
483 osd_seq->os_subdir_count);
485 OBD_FREE_PTR(osd_seq);
486 osd_seq = ERR_PTR(rc);
492 osd_get_idx_for_ost_obj_compat(const struct lu_env *env, struct osd_device *osd,
493 const struct lu_fid *fid, char *buf, int bufsize)
495 struct osd_seq *osd_seq;
500 osd_seq = osd_find_or_add_seq(env, osd, fid_seq(fid));
501 if (IS_ERR(osd_seq)) {
502 CERROR("%s: Can not find seq group "DFID"\n", osd_name(osd),
504 return PTR_ERR(osd_seq);
507 if (fid_is_last_id(fid)) {
510 rc = fid_to_ostid(fid, &osd_oti_get(env)->oti_ostid);
511 LASSERT(rc == 0); /* we should not get here with IGIF */
512 id = ostid_id(&osd_oti_get(env)->oti_ostid);
515 b = id % OSD_OST_MAP_SIZE;
516 LASSERT(osd_seq->os_compat_dirs[b]);
519 snprintf(buf, bufsize, "%llu", id);
521 return osd_seq->os_compat_dirs[b];
525 * objects w/o a natural reference (unlike a file on a MDS)
526 * are put under a special hierarchy /O/<seq>/d0..dXX
527 * this function returns a directory specific fid belongs to
530 osd_get_idx_for_ost_obj(const struct lu_env *env, struct osd_device *osd,
531 const struct lu_fid *fid, char *buf, int bufsize)
533 struct osd_seq *osd_seq;
538 osd_seq = osd_find_or_add_seq(env, osd, fid_seq(fid));
539 if (IS_ERR(osd_seq)) {
540 CERROR("%s: Can not find seq group "DFID"\n", osd_name(osd),
542 return PTR_ERR(osd_seq);
545 if (fid_is_last_id(fid)) {
547 snprintf(buf, bufsize, "LAST_ID");
549 return osd_seq->os_oid;
552 rc = fid_to_ostid(fid, &osd_oti_get(env)->oti_ostid);
553 LASSERT(rc == 0); /* we should not get here with IGIF */
555 id = ostid_id(&osd_oti_get(env)->oti_ostid);
556 b = id % OSD_OST_MAP_SIZE;
557 LASSERT(osd_seq->os_compat_dirs[b]);
560 snprintf(buf, bufsize, "%llu", id);
562 return osd_seq->os_compat_dirs[b];
566 * Determine the zap object id which is being used as the OI for the
567 * given fid. The lowest N bits in the sequence ID are used as the
568 * index key. On failure 0 is returned which zfs treats internally
569 * as an invalid object id.
572 osd_get_idx_for_fid(struct osd_device *osd, const struct lu_fid *fid,
573 char *buf, dnode_t **zdn, int bufsize)
577 oi = osd_fid2oi(osd, fid);
579 osd_fid2str(buf, fid, bufsize);
587 osd_get_name_n_idx_compat(const struct lu_env *env, struct osd_device *osd,
588 const struct lu_fid *fid, char *buf, int bufsize,
594 LASSERT(!fid_is_acct(fid));
599 if (fid_is_echo(fid) || fid_is_on_ost(env, osd, fid)) {
600 zapid = osd_get_idx_for_ost_obj_compat(env, osd, fid,
602 } else if (unlikely(fid_seq(fid) == FID_SEQ_LOCAL_FILE)) {
603 /* special objects with fixed known fids get their name */
604 char *name = oid2name(fid_oid(fid));
607 zapid = osd->od_root;
609 strncpy(buf, name, bufsize);
611 zapid = osd_get_idx_for_fid(osd, fid, buf, NULL,
615 zapid = osd_get_idx_for_fid(osd, fid, buf, zdn, bufsize);
621 uint64_t osd_get_name_n_idx(const struct lu_env *env, struct osd_device *osd,
622 const struct lu_fid *fid, char *buf, int bufsize,
628 LASSERT(!fid_is_acct(fid));
633 if (fid_is_echo(fid) || fid_is_last_id(fid) ||
634 fid_is_on_ost(env, osd, fid)) {
635 zapid = osd_get_idx_for_ost_obj(env, osd, fid, buf, bufsize);
636 } else if (unlikely(fid_seq(fid) == FID_SEQ_LOCAL_FILE)) {
637 /* special objects with fixed known fids get their name */
638 char *name = oid2name(fid_oid(fid));
641 zapid = osd->od_root;
643 strncpy(buf, name, bufsize);
645 zapid = osd_get_idx_for_fid(osd, fid, buf, NULL,
649 zapid = osd_get_idx_for_fid(osd, fid, buf, zdn, bufsize);
655 static inline int fid_is_fs_root(const struct lu_fid *fid)
657 /* Map root inode to special local object FID */
658 return fid_seq(fid) == FID_SEQ_LOCAL_FILE &&
659 fid_oid(fid) == OSD_FS_ROOT_OID;
662 int osd_fid_lookup(const struct lu_env *env, struct osd_device *dev,
663 const struct lu_fid *fid, uint64_t *oid)
665 struct osd_thread_info *info = osd_oti_get(env);
666 char *buf = info->oti_buf;
672 if (OBD_FAIL_CHECK(OBD_FAIL_SRV_ENOENT))
675 LASSERT(!fid_is_acct(fid));
677 if (unlikely(fid_is_fs_root(fid))) {
680 zapid = osd_get_name_n_idx(env, dev, fid, buf,
681 sizeof(info->oti_buf), &zdn);
682 rc = osd_zap_lookup(dev, zapid, zdn, buf,
683 8, 1, &info->oti_zde);
685 if (unlikely(fid_is_last_id(fid))) {
686 zapid = osd_get_name_n_idx_compat(env, dev, fid,
687 buf, sizeof(info->oti_buf), &zdn);
688 rc = osd_zap_lookup(dev, zapid, zdn, buf,
689 8, 1, &info->oti_zde);
690 } else if (fid_is_objseq(fid) || fid_is_batchid(fid)) {
691 zapid = osd_get_idx_for_fid(dev, fid,
692 buf, NULL, sizeof(info->oti_buf));
693 rc = osd_zap_lookup(dev, zapid, zdn, buf,
694 8, 1, &info->oti_zde);
700 *oid = info->oti_zde.lzd_reg.zde_dnode;
704 osd_dmu_prefetch(dev->od_os, *oid, 0, 0, 0,
705 ZIO_PRIORITY_ASYNC_READ);
711 * Close an entry in a specific slot.
714 osd_oi_remove_table(const struct lu_env *env, struct osd_device *o, int key)
718 LASSERT(key < o->od_oi_count);
720 oi = o->od_oi_table[key];
723 osd_dnode_rele(oi->oi_dn);
725 o->od_oi_table[key] = NULL;
730 * Allocate and open a new entry in the specified unused slot.
733 osd_oi_add_table(const struct lu_env *env, struct osd_device *o,
739 LASSERT(key < o->od_oi_count);
740 LASSERT(o->od_oi_table[key] == NULL);
746 rc = osd_oi_lookup(env, o, o->od_root, name, oi);
752 o->od_oi_table[key] = oi;
753 __osd_obj2dnode(o->od_os, oi->oi_zapid, &oi->oi_dn);
759 * Depopulate the OI table.
762 osd_oi_close_table(const struct lu_env *env, struct osd_device *o)
766 for (i = 0; i < o->od_oi_count; i++)
767 osd_oi_remove_table(env, o, i);
771 * Populate the OI table based.
774 osd_oi_open_table(const struct lu_env *env, struct osd_device *o, int count)
780 for (i = 0; i < count; i++) {
781 sprintf(name, "%s.%d", DMU_OSD_OI_NAME_BASE, i);
782 rc = osd_oi_add_table(env, o, name, i);
784 osd_oi_close_table(env, o);
793 * Determine if the type and number of OIs used by this file system.
795 static int osd_oi_probe(const struct lu_env *env, struct osd_device *o)
797 struct lustre_scrub *scrub = &o->od_scrub;
798 struct scrub_file *sf = &scrub->os_file;
801 int max = sf->sf_oi_count > 0 ? sf->sf_oi_count : OSD_OI_FID_NR_MAX;
807 * Check for multiple OIs and determine the count. There is no
808 * gap handling, if an OI is missing the wrong size can be returned.
809 * The only safeguard is that we know the number of OIs must be a
810 * power of two and this is checked for basic sanity.
812 for (count = 0; count < max; count++) {
813 snprintf(name, sizeof(name) - 1, "%s.%d",
814 DMU_OSD_OI_NAME_BASE, count);
815 rc = osd_oi_lookup(env, o, o->od_root, name, &oi);
820 if (sf->sf_oi_count == 0)
823 zfs_set_bit(count, sf->sf_oi_bitmap);
834 static void osd_ost_seq_fini(const struct lu_env *env, struct osd_device *osd)
836 struct osd_seq_list *osl = &osd->od_seq_list;
837 struct osd_seq *osd_seq, *tmp;
839 write_lock(&osl->osl_seq_list_lock);
840 list_for_each_entry_safe(osd_seq, tmp, &osl->osl_seq_list,
842 list_del(&osd_seq->os_seq_list);
843 OBD_FREE_PTR_ARRAY(osd_seq->os_compat_dirs,
844 osd_seq->os_subdir_count);
845 OBD_FREE(osd_seq, sizeof(*osd_seq));
847 write_unlock(&osl->osl_seq_list_lock);
851 * Create /O subdirectory to map legacy OST objects for compatibility.
854 osd_oi_init_compat(const struct lu_env *env, struct osd_device *o)
860 rc = osd_oi_find_or_create(env, o, o->od_root, "O", &sdb);
868 osd_oi_init_index_backup(const struct lu_env *env, struct osd_device *o)
870 struct lu_fid *fid = &osd_oti_get(env)->oti_fid;
874 lu_local_obj_fid(fid, INDEX_BACKUP_OID);
875 rc = osd_obj_find_or_create(env, o, o->od_root, INDEX_BACKUP_DIR,
876 &o->od_index_backup_id, fid, true);
882 osd_oi_init_remote_parent(const struct lu_env *env, struct osd_device *o)
889 o->od_remote_parent_dir = ZFS_NO_OBJECT;
891 /* Remote parent only used for cross-MDT objects,
892 * it is usless for single MDT case or under read
893 * only mode. So ignore the failure. */
894 rc = osd_oi_find_or_create(env, o, o->od_root,
895 REMOTE_PARENT_DIR, &sdb);
897 o->od_remote_parent_dir = sdb;
899 o->od_remote_parent_dir = ZFS_NO_OBJECT;
904 * Initialize the OIs by either opening or creating them as needed.
906 int osd_oi_init(const struct lu_env *env, struct osd_device *o, bool reset)
908 struct lustre_scrub *scrub = &o->od_scrub;
909 struct scrub_file *sf = &scrub->os_file;
910 char *key = osd_oti_get(env)->oti_buf;
915 LASSERTF((sf->sf_oi_count & (sf->sf_oi_count - 1)) == 0,
916 "Invalid OI count in scrub file %d\n", sf->sf_oi_count);
918 rc = osd_oi_init_index_backup(env, o);
922 osd_oi_init_remote_parent(env, o);
924 rc = osd_oi_init_compat(env, o);
928 count = osd_oi_probe(env, o);
930 GOTO(out, rc = count);
933 if (count == sf->sf_oi_count && !reset)
936 if (sf->sf_oi_count == 0) {
937 if (likely((count & (count - 1)) == 0)) {
938 sf->sf_oi_count = count;
939 rc = scrub_file_store(env, scrub);
946 LCONSOLE_ERROR("%s: invalid oi count %d. You can "
947 "remove all OIs, then remount it\n",
949 GOTO(out, rc = -EDOM);
952 scrub_file_reset(scrub, o->od_uuid, SF_RECREATED);
953 count = sf->sf_oi_count;
955 if (sf->sf_oi_count > 0) {
956 count = sf->sf_oi_count;
957 memset(sf->sf_oi_bitmap, 0, SCRUB_OI_BITMAP_SIZE);
958 for (i = 0; i < count; i++)
959 zfs_set_bit(i, sf->sf_oi_bitmap);
960 scrub_file_reset(scrub, o->od_uuid, SF_RECREATED);
962 count = sf->sf_oi_count = osd_oi_count;
966 rc = scrub_file_store(env, scrub);
971 LCONSOLE_WARN("%s: reset Object Index mappings\n",
974 for (i = 0; i < count; i++) {
975 LASSERT(sizeof(osd_oti_get(env)->oti_buf) >= 32);
977 snprintf(key, sizeof(osd_oti_get(env)->oti_buf) - 1,
978 "%s.%d", DMU_OSD_OI_NAME_BASE, i);
980 rc = osd_oi_destroy(env, o, key);
984 rc = osd_oi_find_or_create(env, o, o->od_root, key, &sdb);
990 LASSERT((count & (count - 1)) == 0);
991 o->od_oi_count = count;
992 OBD_ALLOC_PTR_ARRAY(o->od_oi_table, count);
993 if (o->od_oi_table == NULL)
994 GOTO(out, rc = -ENOMEM);
996 rc = osd_oi_open_table(env, o, count);
1002 osd_ost_seq_fini(env, o);
1004 if (o->od_oi_table) {
1005 OBD_FREE_PTR_ARRAY(o->od_oi_table, count);
1006 o->od_oi_table = NULL;
1013 void osd_oi_fini(const struct lu_env *env, struct osd_device *o)
1017 osd_ost_seq_fini(env, o);
1019 if (o->od_oi_table != NULL) {
1020 (void) osd_oi_close_table(env, o);
1021 OBD_FREE_PTR_ARRAY(o->od_oi_table, o->od_oi_count);
1022 o->od_oi_table = NULL;
1029 int osd_options_init(void)
1031 /* osd_oi_count - Default number of OIs, 128 works well for ZFS */
1032 if (osd_oi_count == 0 || osd_oi_count > OSD_OI_FID_NR_MAX)
1033 osd_oi_count = OSD_OI_FID_NR;
1035 if ((osd_oi_count & (osd_oi_count - 1)) != 0) {
1036 LCONSOLE_WARN("Round up osd_oi_count %d to power2 %d\n",
1037 osd_oi_count, size_roundup_power2(osd_oi_count));
1038 osd_oi_count = size_roundup_power2(osd_oi_count);
1045 * the following set of functions are used to maintain per-thread
1046 * cache of FID->ino mapping. this mechanism is used to avoid
1047 * expensive LU/OI lookups.
1049 struct osd_idmap_cache *osd_idc_find(const struct lu_env *env,
1050 struct osd_device *osd,
1051 const struct lu_fid *fid)
1053 struct osd_thread_info *oti = osd_oti_get(env);
1054 struct osd_idmap_cache *idc = oti->oti_ins_cache;
1057 for (i = 0; i < oti->oti_ins_cache_used; i++) {
1058 if (!lu_fid_eq(&idc[i].oic_fid, fid))
1060 if (idc[i].oic_dev != osd)
1069 struct osd_idmap_cache *osd_idc_add(const struct lu_env *env,
1070 struct osd_device *osd,
1071 const struct lu_fid *fid)
1073 struct osd_thread_info *oti = osd_oti_get(env);
1074 struct osd_idmap_cache *idc;
1077 if (unlikely(oti->oti_ins_cache_used >= oti->oti_ins_cache_size)) {
1078 i = oti->oti_ins_cache_size * 2;
1080 i = OSD_INS_CACHE_SIZE;
1081 OBD_ALLOC_PTR_ARRAY_LARGE(idc, i);
1083 return ERR_PTR(-ENOMEM);
1084 if (oti->oti_ins_cache != NULL) {
1085 memcpy(idc, oti->oti_ins_cache,
1086 oti->oti_ins_cache_used * sizeof(*idc));
1087 OBD_FREE_PTR_ARRAY_LARGE(oti->oti_ins_cache,
1088 oti->oti_ins_cache_used);
1090 oti->oti_ins_cache = idc;
1091 oti->oti_ins_cache_size = i;
1094 idc = &oti->oti_ins_cache[oti->oti_ins_cache_used++];
1095 idc->oic_fid = *fid;
1098 idc->oic_remote = 0;
1104 * Lookup mapping for the given fid in the cache
1106 * Initialize a new one if not found. the initialization checks whether
1107 * the object is local or remote. for the local objects, OI is used to
1108 * learn dnode#. the function is used when the caller has no information
1109 * about the object, e.g. at dt_insert().
1111 struct osd_idmap_cache *osd_idc_find_or_init(const struct lu_env *env,
1112 struct osd_device *osd,
1113 const struct lu_fid *fid)
1115 struct osd_idmap_cache *idc;
1118 LASSERT(!fid_is_acct(fid));
1120 idc = osd_idc_find(env, osd, fid);
1124 CDEBUG(D_INODE, "%s: FID "DFID" not in the id map cache\n",
1125 osd->od_svname, PFID(fid));
1127 /* new mapping is needed */
1128 idc = osd_idc_add(env, osd, fid);
1130 CERROR("%s: FID "DFID" add id map cache failed: %ld\n",
1131 osd->od_svname, PFID(fid), PTR_ERR(idc));
1136 rc = osd_remote_fid(env, osd, fid);
1137 if (unlikely(rc < 0))
1141 /* the object is local, lookup in OI */
1144 rc = osd_fid_lookup(env, osd, fid, &dnode);
1145 if (unlikely(rc < 0)) {
1146 CERROR("%s: can't lookup: rc = %d\n",
1147 osd->od_svname, rc);
1150 LASSERT(dnode < (1ULL << DN_MAX_OBJECT_SHIFT));
1151 idc->oic_dnode = dnode;
1153 /* the object is remote */
1154 idc->oic_remote = 1;
1161 * lookup mapping for given FID and fill it from the given object.
1162 * the object is local by definition.
1164 int osd_idc_find_and_init(const struct lu_env *env, struct osd_device *osd,
1165 struct osd_object *obj)
1167 const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
1168 struct osd_idmap_cache *idc;
1170 idc = osd_idc_find(env, osd, fid);
1172 if (obj->oo_dn == NULL)
1174 idc->oic_dnode = obj->oo_dn->dn_object;
1178 CDEBUG(D_INODE, "%s: FID "DFID" not in the id map cache\n",
1179 osd->od_svname, PFID(fid));
1181 /* new mapping is needed */
1182 idc = osd_idc_add(env, osd, fid);
1184 CERROR("%s: FID "DFID" add id map cache failed: %ld\n",
1185 osd->od_svname, PFID(fid), PTR_ERR(idc));
1186 return PTR_ERR(idc);
1190 idc->oic_dnode = obj->oo_dn->dn_object;
1195 int osd_idc_find_and_init_with_oid(const struct lu_env *env,
1196 struct osd_device *osd,
1197 const struct lu_fid *fid,
1200 struct osd_idmap_cache *idc;
1202 idc = osd_idc_find(env, osd, fid);
1204 idc = osd_idc_add(env, osd, fid);
1206 return PTR_ERR(idc);
1209 idc->oic_dnode = oid;
1210 idc->oic_remote = 0;