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 = IFTODT(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);
232 osd_oi_find_or_create(const struct lu_env *env, struct osd_device *o,
233 uint64_t parent, const char *name, uint64_t *child)
238 rc = osd_oi_lookup(env, o, parent, name, &oi);
240 *child = oi.oi_zapid;
241 else if (rc == -ENOENT)
242 rc = osd_obj_create(env, o, parent, name, child, NULL, true);
247 int osd_obj_find_or_create(const struct lu_env *env, struct osd_device *o,
248 uint64_t parent, const char *name, uint64_t *child,
249 const struct lu_fid *fid, bool isdir)
254 rc = osd_oi_lookup(env, o, parent, name, &oi);
256 *child = oi.oi_zapid;
257 else if (rc == -ENOENT)
258 rc = osd_obj_create(env, o, parent, name, child, fid, isdir);
264 * Lookup the target index/flags of the fid, so it will know where
265 * the object is located (tgt index) and it is MDT or OST object.
267 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
268 u64 seq, struct lu_seq_range *range)
270 struct seq_server_site *ss = osd_seq_site(osd);
272 if (fid_seq_is_idif(seq)) {
273 fld_range_set_ost(range);
274 range->lsr_index = idif_ost_idx(seq);
278 if (!fid_seq_in_fldb(seq)) {
279 fld_range_set_mdt(range);
281 /* FIXME: If ss is NULL, it suppose not get lsr_index
283 range->lsr_index = ss->ss_node_id;
287 /* The seq_server_site may be NOT ready during initial OI scrub */
288 if (unlikely(!ss || !ss->ss_server_fld ||
289 !ss->ss_server_fld->lsf_cache))
292 fld_range_set_any(range);
293 /* OSD will only do local fld lookup */
294 return fld_local_lookup(env, ss->ss_server_fld, seq, range);
297 int fid_is_on_ost(const struct lu_env *env, struct osd_device *osd,
298 const struct lu_fid *fid)
300 struct lu_seq_range *range = &osd_oti_get(env)->oti_seq_range;
304 if (fid_is_idif(fid))
307 if (unlikely(fid_is_local_file(fid) || fid_is_llog(fid)) ||
308 fid_is_name_llog(fid) || fid_is_quota(fid) ||
312 rc = osd_fld_lookup(env, osd, fid_seq(fid), range);
314 /* During upgrade, OST FLDB might not be loaded because
315 * OST FLDB is not created until 2.6, so if some DNE
316 * filesystem upgrade from 2.5 to 2.7/2.8, they will
317 * not be able to find the sequence from local FLDB
318 * cache see fld_index_init(). */
319 if (rc == -ENOENT && osd->od_is_ost)
323 CERROR("%s: "DFID" lookup failed: rc = %d\n",
324 osd_name(osd), PFID(fid), rc);
328 if (fld_range_is_ost(range))
334 static struct osd_seq *osd_seq_find_locked(struct osd_seq_list *seq_list,
337 struct osd_seq *osd_seq;
339 list_for_each_entry(osd_seq, &seq_list->osl_seq_list, os_seq_list) {
340 if (osd_seq->os_seq == seq)
346 static struct osd_seq *osd_seq_find(struct osd_seq_list *seq_list, u64 seq)
348 struct osd_seq *osd_seq;
350 read_lock(&seq_list->osl_seq_list_lock);
351 osd_seq = osd_seq_find_locked(seq_list, seq);
352 read_unlock(&seq_list->osl_seq_list_lock);
357 static struct osd_seq *osd_find_or_add_seq(const struct lu_env *env,
358 struct osd_device *osd, u64 seq)
360 struct osd_seq_list *seq_list = &osd->od_seq_list;
361 struct osd_seq *osd_seq;
362 char *key = osd_oti_get(env)->oti_buf;
363 char *seq_name = osd_oti_get(env)->oti_str;
370 osd_seq = osd_seq_find(seq_list, seq);
374 down(&seq_list->osl_seq_init_sem);
375 /* Check again, in case some one else already add it
377 osd_seq = osd_seq_find(seq_list, seq);
381 OBD_ALLOC_PTR(osd_seq);
383 GOTO(out, rc = -ENOMEM);
385 INIT_LIST_HEAD(&osd_seq->os_seq_list);
386 osd_seq->os_seq = seq;
388 /* Init subdir count to be 32, but each seq can have
389 * different subdir count */
390 osd_seq->os_subdir_count = OSD_OST_MAP_SIZE;
391 OBD_ALLOC(osd_seq->os_compat_dirs,
392 sizeof(uint64_t) * osd_seq->os_subdir_count);
393 if (osd_seq->os_compat_dirs == NULL)
394 GOTO(out, rc = -ENOMEM);
396 oi.oi_zapid = osd->od_O_id;
397 sprintf(seq_name, (fid_seq_is_rsvd(seq) ||
398 fid_seq_is_mdt0(seq)) ? "%llu" : "%llx",
399 fid_seq_is_idif(seq) ? 0 : seq);
401 rc = osd_oi_find_or_create(env, osd, oi.oi_zapid, seq_name, &odb);
403 CERROR("%s: Can not create %s : rc = %d\n",
404 osd_name(osd), seq_name, rc);
408 osd_seq->os_oid = odb;
409 for (i = 0; i < OSD_OST_MAP_SIZE; i++) {
410 sprintf(key, "d%d", i);
411 rc = osd_oi_find_or_create(env, osd, odb, key, &sdb);
414 osd_seq->os_compat_dirs[i] = sdb;
417 write_lock(&seq_list->osl_seq_list_lock);
418 list_add(&osd_seq->os_seq_list, &seq_list->osl_seq_list);
419 write_unlock(&seq_list->osl_seq_list_lock);
421 up(&seq_list->osl_seq_init_sem);
423 if (osd_seq != NULL && osd_seq->os_compat_dirs != NULL)
424 OBD_FREE(osd_seq->os_compat_dirs,
425 sizeof(uint64_t) * osd_seq->os_subdir_count);
427 OBD_FREE_PTR(osd_seq);
428 osd_seq = ERR_PTR(rc);
434 osd_get_idx_for_ost_obj_compat(const struct lu_env *env, struct osd_device *osd,
435 const struct lu_fid *fid, char *buf, int bufsize)
437 struct osd_seq *osd_seq;
442 osd_seq = osd_find_or_add_seq(env, osd, fid_seq(fid));
443 if (IS_ERR(osd_seq)) {
444 CERROR("%s: Can not find seq group "DFID"\n", osd_name(osd),
446 return PTR_ERR(osd_seq);
449 if (fid_is_last_id(fid)) {
452 rc = fid_to_ostid(fid, &osd_oti_get(env)->oti_ostid);
453 LASSERT(rc == 0); /* we should not get here with IGIF */
454 id = ostid_id(&osd_oti_get(env)->oti_ostid);
457 b = id % OSD_OST_MAP_SIZE;
458 LASSERT(osd_seq->os_compat_dirs[b]);
461 snprintf(buf, bufsize, "%llu", id);
463 return osd_seq->os_compat_dirs[b];
467 * objects w/o a natural reference (unlike a file on a MDS)
468 * are put under a special hierarchy /O/<seq>/d0..dXX
469 * this function returns a directory specific fid belongs to
472 osd_get_idx_for_ost_obj(const struct lu_env *env, struct osd_device *osd,
473 const struct lu_fid *fid, char *buf, int bufsize)
475 struct osd_seq *osd_seq;
480 osd_seq = osd_find_or_add_seq(env, osd, fid_seq(fid));
481 if (IS_ERR(osd_seq)) {
482 CERROR("%s: Can not find seq group "DFID"\n", osd_name(osd),
484 return PTR_ERR(osd_seq);
487 if (fid_is_last_id(fid)) {
489 snprintf(buf, bufsize, "LAST_ID");
491 return osd_seq->os_oid;
494 rc = fid_to_ostid(fid, &osd_oti_get(env)->oti_ostid);
495 LASSERT(rc == 0); /* we should not get here with IGIF */
497 id = ostid_id(&osd_oti_get(env)->oti_ostid);
498 b = id % OSD_OST_MAP_SIZE;
499 LASSERT(osd_seq->os_compat_dirs[b]);
502 snprintf(buf, bufsize, "%llu", id);
504 return osd_seq->os_compat_dirs[b];
508 * Determine the zap object id which is being used as the OI for the
509 * given fid. The lowest N bits in the sequence ID are used as the
510 * index key. On failure 0 is returned which zfs treats internally
511 * as an invalid object id.
514 osd_get_idx_for_fid(struct osd_device *osd, const struct lu_fid *fid,
515 char *buf, dnode_t **zdn, int bufsize)
519 oi = osd_fid2oi(osd, fid);
521 osd_fid2str(buf, fid, bufsize);
529 osd_get_name_n_idx_compat(const struct lu_env *env, struct osd_device *osd,
530 const struct lu_fid *fid, char *buf, int bufsize,
536 LASSERT(!fid_is_acct(fid));
541 if (fid_is_echo(fid) || fid_is_on_ost(env, osd, fid)) {
542 zapid = osd_get_idx_for_ost_obj_compat(env, osd, fid,
544 } else if (unlikely(fid_seq(fid) == FID_SEQ_LOCAL_FILE)) {
545 /* special objects with fixed known fids get their name */
546 char *name = oid2name(fid_oid(fid));
549 zapid = osd->od_root;
551 strncpy(buf, name, bufsize);
553 zapid = osd_get_idx_for_fid(osd, fid, buf, NULL,
557 zapid = osd_get_idx_for_fid(osd, fid, buf, zdn, bufsize);
563 uint64_t osd_get_name_n_idx(const struct lu_env *env, struct osd_device *osd,
564 const struct lu_fid *fid, char *buf, int bufsize,
570 LASSERT(!fid_is_acct(fid));
575 if (fid_is_echo(fid) || fid_is_last_id(fid) ||
576 fid_is_on_ost(env, osd, fid)) {
577 zapid = osd_get_idx_for_ost_obj(env, osd, fid, buf, bufsize);
578 } else if (unlikely(fid_seq(fid) == FID_SEQ_LOCAL_FILE)) {
579 /* special objects with fixed known fids get their name */
580 char *name = oid2name(fid_oid(fid));
583 zapid = osd->od_root;
585 strncpy(buf, name, bufsize);
587 zapid = osd_get_idx_for_fid(osd, fid, buf, NULL,
591 zapid = osd_get_idx_for_fid(osd, fid, buf, zdn, bufsize);
597 static inline int fid_is_fs_root(const struct lu_fid *fid)
599 /* Map root inode to special local object FID */
600 return fid_seq(fid) == FID_SEQ_LOCAL_FILE &&
601 fid_oid(fid) == OSD_FS_ROOT_OID;
604 int osd_fid_lookup(const struct lu_env *env, struct osd_device *dev,
605 const struct lu_fid *fid, uint64_t *oid)
607 struct osd_thread_info *info = osd_oti_get(env);
608 char *buf = info->oti_buf;
614 if (OBD_FAIL_CHECK(OBD_FAIL_SRV_ENOENT))
617 LASSERT(!fid_is_acct(fid));
619 if (unlikely(fid_is_fs_root(fid))) {
622 zapid = osd_get_name_n_idx(env, dev, fid, buf,
623 sizeof(info->oti_buf), &zdn);
624 rc = osd_zap_lookup(dev, zapid, zdn, buf,
625 8, 1, &info->oti_zde);
627 if (unlikely(fid_is_last_id(fid))) {
628 zapid = osd_get_name_n_idx_compat(env, dev, fid,
629 buf, sizeof(info->oti_buf), &zdn);
630 rc = osd_zap_lookup(dev, zapid, zdn, buf,
631 8, 1, &info->oti_zde);
632 } else if (fid_is_objseq(fid) || fid_is_batchid(fid)) {
633 zapid = osd_get_idx_for_fid(dev, fid,
634 buf, NULL, sizeof(info->oti_buf));
635 rc = osd_zap_lookup(dev, zapid, zdn, buf,
636 8, 1, &info->oti_zde);
642 *oid = info->oti_zde.lzd_reg.zde_dnode;
646 osd_dmu_prefetch(dev->od_os, *oid, 0, 0, 0,
647 ZIO_PRIORITY_ASYNC_READ);
653 * Close an entry in a specific slot.
656 osd_oi_remove_table(const struct lu_env *env, struct osd_device *o, int key)
660 LASSERT(key < o->od_oi_count);
662 oi = o->od_oi_table[key];
665 osd_dnode_rele(oi->oi_dn);
667 o->od_oi_table[key] = NULL;
672 * Allocate and open a new entry in the specified unused slot.
675 osd_oi_add_table(const struct lu_env *env, struct osd_device *o,
681 LASSERT(key < o->od_oi_count);
682 LASSERT(o->od_oi_table[key] == NULL);
688 rc = osd_oi_lookup(env, o, o->od_root, name, oi);
694 o->od_oi_table[key] = oi;
695 __osd_obj2dnode(o->od_os, oi->oi_zapid, &oi->oi_dn);
701 * Depopulate the OI table.
704 osd_oi_close_table(const struct lu_env *env, struct osd_device *o)
708 for (i = 0; i < o->od_oi_count; i++)
709 osd_oi_remove_table(env, o, i);
713 * Populate the OI table based.
716 osd_oi_open_table(const struct lu_env *env, struct osd_device *o, int count)
722 for (i = 0; i < count; i++) {
723 sprintf(name, "%s.%d", DMU_OSD_OI_NAME_BASE, i);
724 rc = osd_oi_add_table(env, o, name, i);
726 osd_oi_close_table(env, o);
735 * Determine if the type and number of OIs used by this file system.
737 static int osd_oi_probe(const struct lu_env *env, struct osd_device *o)
739 struct lustre_scrub *scrub = &o->od_scrub;
740 struct scrub_file *sf = &scrub->os_file;
743 int max = sf->sf_oi_count > 0 ? sf->sf_oi_count : OSD_OI_FID_NR_MAX;
749 * Check for multiple OIs and determine the count. There is no
750 * gap handling, if an OI is missing the wrong size can be returned.
751 * The only safeguard is that we know the number of OIs must be a
752 * power of two and this is checked for basic sanity.
754 for (count = 0; count < max; count++) {
755 snprintf(name, sizeof(name) - 1, "%s.%d",
756 DMU_OSD_OI_NAME_BASE, count);
757 rc = osd_oi_lookup(env, o, o->od_root, name, &oi);
762 if (sf->sf_oi_count == 0)
765 zfs_set_bit(count, sf->sf_oi_bitmap);
776 static void osd_ost_seq_fini(const struct lu_env *env, struct osd_device *osd)
778 struct osd_seq_list *osl = &osd->od_seq_list;
779 struct osd_seq *osd_seq, *tmp;
781 write_lock(&osl->osl_seq_list_lock);
782 list_for_each_entry_safe(osd_seq, tmp, &osl->osl_seq_list,
784 list_del(&osd_seq->os_seq_list);
785 OBD_FREE(osd_seq->os_compat_dirs,
786 sizeof(uint64_t) * osd_seq->os_subdir_count);
787 OBD_FREE(osd_seq, sizeof(*osd_seq));
789 write_unlock(&osl->osl_seq_list_lock);
795 * Create /O subdirectory to map legacy OST objects for compatibility.
798 osd_oi_init_compat(const struct lu_env *env, struct osd_device *o)
804 rc = osd_oi_find_or_create(env, o, o->od_root, "O", &sdb);
812 osd_oi_init_index_backup(const struct lu_env *env, struct osd_device *o)
814 struct lu_fid *fid = &osd_oti_get(env)->oti_fid;
818 lu_local_obj_fid(fid, INDEX_BACKUP_OID);
819 rc = osd_obj_find_or_create(env, o, o->od_root, INDEX_BACKUP_DIR,
820 &o->od_index_backup_id, fid, true);
826 osd_oi_init_remote_parent(const struct lu_env *env, struct osd_device *o)
833 o->od_remote_parent_dir = ZFS_NO_OBJECT;
835 /* Remote parent only used for cross-MDT objects,
836 * it is usless for single MDT case or under read
837 * only mode. So ignore the failure. */
838 rc = osd_oi_find_or_create(env, o, o->od_root,
839 REMOTE_PARENT_DIR, &sdb);
841 o->od_remote_parent_dir = sdb;
843 o->od_remote_parent_dir = ZFS_NO_OBJECT;
848 * Initialize the OIs by either opening or creating them as needed.
850 int osd_oi_init(const struct lu_env *env, struct osd_device *o)
852 struct lustre_scrub *scrub = &o->od_scrub;
853 struct scrub_file *sf = &scrub->os_file;
854 char *key = osd_oti_get(env)->oti_buf;
859 LASSERTF((sf->sf_oi_count & (sf->sf_oi_count - 1)) == 0,
860 "Invalid OI count in scrub file %d\n", sf->sf_oi_count);
862 rc = osd_oi_init_index_backup(env, o);
866 osd_oi_init_remote_parent(env, o);
868 rc = osd_oi_init_compat(env, o);
872 count = osd_oi_probe(env, o);
874 GOTO(out, rc = count);
877 if (count == sf->sf_oi_count)
880 if (sf->sf_oi_count == 0) {
881 if (likely((count & (count - 1)) == 0)) {
882 sf->sf_oi_count = count;
883 rc = scrub_file_store(env, scrub);
890 LCONSOLE_ERROR("%s: invalid oi count %d. You can "
891 "remove all OIs, then remount it\n",
893 GOTO(out, rc = -EDOM);
896 scrub_file_reset(scrub, o->od_uuid, SF_RECREATED);
897 count = sf->sf_oi_count;
899 if (sf->sf_oi_count > 0) {
900 count = sf->sf_oi_count;
901 memset(sf->sf_oi_bitmap, 0, SCRUB_OI_BITMAP_SIZE);
902 for (i = 0; i < count; i++)
903 zfs_set_bit(i, sf->sf_oi_bitmap);
904 scrub_file_reset(scrub, o->od_uuid, SF_RECREATED);
906 count = sf->sf_oi_count = osd_oi_count;
910 rc = scrub_file_store(env, scrub);
914 for (i = 0; i < count; i++) {
915 LASSERT(sizeof(osd_oti_get(env)->oti_buf) >= 32);
917 snprintf(key, sizeof(osd_oti_get(env)->oti_buf) - 1,
918 "%s.%d", DMU_OSD_OI_NAME_BASE, i);
919 rc = osd_oi_find_or_create(env, o, o->od_root, key, &sdb);
925 LASSERT((count & (count - 1)) == 0);
926 o->od_oi_count = count;
927 OBD_ALLOC(o->od_oi_table, sizeof(struct osd_oi *) * count);
928 if (o->od_oi_table == NULL)
929 GOTO(out, rc = -ENOMEM);
931 rc = osd_oi_open_table(env, o, count);
937 osd_ost_seq_fini(env, o);
939 if (o->od_oi_table) {
940 OBD_FREE(o->od_oi_table,
941 sizeof(struct osd_oi *) * count);
942 o->od_oi_table = NULL;
949 void osd_oi_fini(const struct lu_env *env, struct osd_device *o)
953 osd_ost_seq_fini(env, o);
955 if (o->od_oi_table != NULL) {
956 (void) osd_oi_close_table(env, o);
957 OBD_FREE(o->od_oi_table,
958 sizeof(struct osd_oi *) * o->od_oi_count);
959 o->od_oi_table = NULL;
966 int osd_options_init(void)
968 /* osd_oi_count - Default number of OIs, 128 works well for ZFS */
969 if (osd_oi_count == 0 || osd_oi_count > OSD_OI_FID_NR_MAX)
970 osd_oi_count = OSD_OI_FID_NR;
972 if ((osd_oi_count & (osd_oi_count - 1)) != 0) {
973 LCONSOLE_WARN("Round up osd_oi_count %d to power2 %d\n",
974 osd_oi_count, size_roundup_power2(osd_oi_count));
975 osd_oi_count = size_roundup_power2(osd_oi_count);
982 * the following set of functions are used to maintain per-thread
983 * cache of FID->ino mapping. this mechanism is used to avoid
984 * expensive LU/OI lookups.
986 struct osd_idmap_cache *osd_idc_find(const struct lu_env *env,
987 struct osd_device *osd,
988 const struct lu_fid *fid)
990 struct osd_thread_info *oti = osd_oti_get(env);
991 struct osd_idmap_cache *idc = oti->oti_ins_cache;
994 for (i = 0; i < oti->oti_ins_cache_used; i++) {
995 if (!lu_fid_eq(&idc[i].oic_fid, fid))
997 if (idc[i].oic_dev != osd)
1006 struct osd_idmap_cache *osd_idc_add(const struct lu_env *env,
1007 struct osd_device *osd,
1008 const struct lu_fid *fid)
1010 struct osd_thread_info *oti = osd_oti_get(env);
1011 struct osd_idmap_cache *idc;
1014 if (unlikely(oti->oti_ins_cache_used >= oti->oti_ins_cache_size)) {
1015 i = oti->oti_ins_cache_size * 2;
1018 i = OSD_INS_CACHE_SIZE;
1019 OBD_ALLOC(idc, sizeof(*idc) * i);
1021 return ERR_PTR(-ENOMEM);
1022 if (oti->oti_ins_cache != NULL) {
1023 memcpy(idc, oti->oti_ins_cache,
1024 oti->oti_ins_cache_used * sizeof(*idc));
1025 OBD_FREE(oti->oti_ins_cache,
1026 oti->oti_ins_cache_used * sizeof(*idc));
1028 oti->oti_ins_cache = idc;
1029 oti->oti_ins_cache_size = i;
1032 idc = &oti->oti_ins_cache[oti->oti_ins_cache_used++];
1033 idc->oic_fid = *fid;
1036 idc->oic_remote = 0;
1042 * Lookup mapping for the given fid in the cache
1044 * Initialize a new one if not found. the initialization checks whether
1045 * the object is local or remote. for the local objects, OI is used to
1046 * learn dnode#. the function is used when the caller has no information
1047 * about the object, e.g. at dt_insert().
1049 struct osd_idmap_cache *osd_idc_find_or_init(const struct lu_env *env,
1050 struct osd_device *osd,
1051 const struct lu_fid *fid)
1053 struct osd_idmap_cache *idc;
1056 LASSERT(!fid_is_acct(fid));
1058 idc = osd_idc_find(env, osd, fid);
1062 CDEBUG(D_INODE, "%s: FID "DFID" not in the id map cache\n",
1063 osd->od_svname, PFID(fid));
1065 /* new mapping is needed */
1066 idc = osd_idc_add(env, osd, fid);
1068 CERROR("%s: FID "DFID" add id map cache failed: %ld\n",
1069 osd->od_svname, PFID(fid), PTR_ERR(idc));
1074 rc = osd_remote_fid(env, osd, fid);
1075 if (unlikely(rc < 0))
1079 /* the object is local, lookup in OI */
1082 rc = osd_fid_lookup(env, osd, fid, &dnode);
1083 if (unlikely(rc < 0)) {
1084 CERROR("%s: can't lookup: rc = %d\n",
1085 osd->od_svname, rc);
1088 LASSERT(dnode < (1ULL << DN_MAX_OBJECT_SHIFT));
1089 idc->oic_dnode = dnode;
1091 /* the object is remote */
1092 idc->oic_remote = 1;
1099 * lookup mapping for given FID and fill it from the given object.
1100 * the object is local by definition.
1102 int osd_idc_find_and_init(const struct lu_env *env, struct osd_device *osd,
1103 struct osd_object *obj)
1105 const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
1106 struct osd_idmap_cache *idc;
1108 idc = osd_idc_find(env, osd, fid);
1110 if (obj->oo_dn == NULL)
1112 idc->oic_dnode = obj->oo_dn->dn_object;
1116 CDEBUG(D_INODE, "%s: FID "DFID" not in the id map cache\n",
1117 osd->od_svname, PFID(fid));
1119 /* new mapping is needed */
1120 idc = osd_idc_add(env, osd, fid);
1122 CERROR("%s: FID "DFID" add id map cache failed: %ld\n",
1123 osd->od_svname, PFID(fid), PTR_ERR(idc));
1124 return PTR_ERR(idc);
1128 idc->oic_dnode = obj->oo_dn->dn_object;
1133 int osd_idc_find_and_init_with_oid(const struct lu_env *env,
1134 struct osd_device *osd,
1135 const struct lu_fid *fid,
1138 struct osd_idmap_cache *idc;
1140 idc = osd_idc_find(env, osd, fid);
1142 idc = osd_idc_add(env, osd, fid);
1144 return PTR_ERR(idc);
1147 idc->oic_dnode = oid;
1148 idc->oic_remote = 0;