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, 2016, 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>
65 #define OSD_OI_FID_NR (1UL << 7)
66 #define OSD_OI_FID_NR_MAX (1UL << OSD_OI_FID_OID_BITS_MAX)
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 },
94 static char *oid2name(const unsigned long oid)
99 if (oids[i].oid == oid)
107 * Lookup an existing OI by the given name.
110 osd_oi_lookup(const struct lu_env *env, struct osd_device *o,
111 uint64_t parent, const char *name, struct osd_oi *oi)
113 struct zpl_direntry *zde = &osd_oti_get(env)->oti_zde.lzd_reg;
116 rc = -zap_lookup(o->od_os, parent, name, 8, 1, (void *)zde);
120 rc = strlcpy(oi->oi_name, name, sizeof(oi->oi_name));
121 if (rc >= sizeof(oi->oi_name))
124 oi->oi_zapid = zde->zde_dnode;
130 * Create a new OI with the given name.
133 osd_oi_create(const struct lu_env *env, struct osd_device *o,
134 uint64_t parent, const char *name, uint64_t *child)
136 struct zpl_direntry *zde = &osd_oti_get(env)->oti_zde.lzd_reg;
137 struct lu_attr *la = &osd_oti_get(env)->oti_la;
138 sa_handle_t *sa_hdl = NULL;
143 /* verify it doesn't already exist */
144 rc = -zap_lookup(o->od_os, parent, name, 8, 1, (void *)zde);
148 if (o->od_dt_dev.dd_rdonly)
151 /* create fid-to-dnode index */
152 tx = dmu_tx_create(o->od_os);
156 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, 1, NULL);
157 dmu_tx_hold_bonus(tx, parent);
158 dmu_tx_hold_zap(tx, parent, TRUE, name);
159 dmu_tx_hold_sa_create(tx, ZFS_SA_BASE_ATTR_SIZE);
161 rc = -dmu_tx_assign(tx, TXG_WAIT);
167 oid = osd_zap_create_flags(o->od_os, 0, ZAP_FLAG_HASH64,
168 DMU_OT_DIRECTORY_CONTENTS,
169 14, /* == ZFS fzap_default_block_shift */
173 rc = -sa_handle_get(o->od_os, oid, NULL, SA_HDL_PRIVATE, &sa_hdl);
176 la->la_valid = LA_MODE | LA_UID | LA_GID;
177 la->la_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
178 la->la_uid = la->la_gid = 0;
179 rc = __osd_attr_init(env, o, sa_hdl, tx, la, parent, NULL);
180 sa_handle_destroy(sa_hdl);
184 zde->zde_dnode = oid;
186 zde->zde_type = IFTODT(S_IFDIR);
188 rc = -zap_add(o->od_os, parent, name, 8, 1, (void *)zde, tx);
192 dmu_object_free(o->od_os, oid, tx);
202 osd_oi_find_or_create(const struct lu_env *env, struct osd_device *o,
203 uint64_t parent, const char *name, uint64_t *child)
208 rc = osd_oi_lookup(env, o, parent, name, &oi);
210 *child = oi.oi_zapid;
211 else if (rc == -ENOENT)
212 rc = osd_oi_create(env, o, parent, name, child);
218 * Lookup the target index/flags of the fid, so it will know where
219 * the object is located (tgt index) and it is MDT or OST object.
221 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
222 u64 seq, struct lu_seq_range *range)
224 struct seq_server_site *ss = osd_seq_site(osd);
226 if (fid_seq_is_idif(seq)) {
227 fld_range_set_ost(range);
228 range->lsr_index = idif_ost_idx(seq);
232 if (!fid_seq_in_fldb(seq)) {
233 fld_range_set_mdt(range);
235 /* FIXME: If ss is NULL, it suppose not get lsr_index
237 range->lsr_index = ss->ss_node_id;
242 fld_range_set_any(range);
243 /* OSD will only do local fld lookup */
244 return fld_local_lookup(env, ss->ss_server_fld, seq, range);
247 int fid_is_on_ost(const struct lu_env *env, struct osd_device *osd,
248 const struct lu_fid *fid)
250 struct lu_seq_range *range = &osd_oti_get(env)->oti_seq_range;
254 if (fid_is_idif(fid))
257 if (unlikely(fid_is_local_file(fid) || fid_is_llog(fid)) ||
258 fid_is_name_llog(fid) || fid_is_quota(fid))
261 rc = osd_fld_lookup(env, osd, fid_seq(fid), range);
263 /* During upgrade, OST FLDB might not be loaded because
264 * OST FLDB is not created until 2.6, so if some DNE
265 * filesystem upgrade from 2.5 to 2.7/2.8, they will
266 * not be able to find the sequence from local FLDB
267 * cache see fld_index_init(). */
268 if (rc == -ENOENT && osd->od_is_ost)
272 CERROR("%s: "DFID" lookup failed: rc = %d\n",
273 osd_name(osd), PFID(fid), rc);
277 if (fld_range_is_ost(range))
283 static struct osd_seq *osd_seq_find_locked(struct osd_seq_list *seq_list,
286 struct osd_seq *osd_seq;
288 list_for_each_entry(osd_seq, &seq_list->osl_seq_list, os_seq_list) {
289 if (osd_seq->os_seq == seq)
295 static struct osd_seq *osd_seq_find(struct osd_seq_list *seq_list, u64 seq)
297 struct osd_seq *osd_seq;
299 read_lock(&seq_list->osl_seq_list_lock);
300 osd_seq = osd_seq_find_locked(seq_list, seq);
301 read_unlock(&seq_list->osl_seq_list_lock);
306 static struct osd_seq *osd_find_or_add_seq(const struct lu_env *env,
307 struct osd_device *osd, u64 seq)
309 struct osd_seq_list *seq_list = &osd->od_seq_list;
310 struct osd_seq *osd_seq;
311 char *key = osd_oti_get(env)->oti_buf;
312 char *seq_name = osd_oti_get(env)->oti_str;
319 osd_seq = osd_seq_find(seq_list, seq);
323 down(&seq_list->osl_seq_init_sem);
324 /* Check again, in case some one else already add it
326 osd_seq = osd_seq_find(seq_list, seq);
330 OBD_ALLOC_PTR(osd_seq);
332 GOTO(out, rc = -ENOMEM);
334 INIT_LIST_HEAD(&osd_seq->os_seq_list);
335 osd_seq->os_seq = seq;
337 /* Init subdir count to be 32, but each seq can have
338 * different subdir count */
339 osd_seq->os_subdir_count = OSD_OST_MAP_SIZE;
340 OBD_ALLOC(osd_seq->os_compat_dirs,
341 sizeof(uint64_t) * osd_seq->os_subdir_count);
342 if (osd_seq->os_compat_dirs == NULL)
343 GOTO(out, rc = -ENOMEM);
345 oi.oi_zapid = osd->od_O_id;
346 sprintf(seq_name, (fid_seq_is_rsvd(seq) ||
347 fid_seq_is_mdt0(seq)) ? "%llu" : "%llx",
348 fid_seq_is_idif(seq) ? 0 : seq);
350 rc = osd_oi_find_or_create(env, osd, oi.oi_zapid, seq_name, &odb);
352 CERROR("%s: Can not create %s : rc = %d\n",
353 osd_name(osd), seq_name, rc);
357 for (i = 0; i < OSD_OST_MAP_SIZE; i++) {
358 sprintf(key, "d%d", i);
359 rc = osd_oi_find_or_create(env, osd, odb, key, &sdb);
362 osd_seq->os_compat_dirs[i] = sdb;
365 write_lock(&seq_list->osl_seq_list_lock);
366 list_add(&osd_seq->os_seq_list, &seq_list->osl_seq_list);
367 write_unlock(&seq_list->osl_seq_list_lock);
369 up(&seq_list->osl_seq_init_sem);
371 if (osd_seq != NULL && osd_seq->os_compat_dirs != NULL)
372 OBD_FREE(osd_seq->os_compat_dirs,
373 sizeof(uint64_t) * osd_seq->os_subdir_count);
375 OBD_FREE_PTR(osd_seq);
376 osd_seq = ERR_PTR(rc);
382 * objects w/o a natural reference (unlike a file on a MDS)
383 * are put under a special hierarchy /O/<seq>/d0..dXX
384 * this function returns a directory specific fid belongs to
387 osd_get_idx_for_ost_obj(const struct lu_env *env, struct osd_device *osd,
388 const struct lu_fid *fid, char *buf, int bufsize)
390 struct osd_seq *osd_seq;
395 osd_seq = osd_find_or_add_seq(env, osd, fid_seq(fid));
396 if (IS_ERR(osd_seq)) {
397 CERROR("%s: Can not find seq group "DFID"\n", osd_name(osd),
399 return PTR_ERR(osd_seq);
402 if (fid_is_last_id(fid)) {
405 rc = fid_to_ostid(fid, &osd_oti_get(env)->oti_ostid);
406 LASSERT(rc == 0); /* we should not get here with IGIF */
407 id = ostid_id(&osd_oti_get(env)->oti_ostid);
410 b = id % OSD_OST_MAP_SIZE;
411 LASSERT(osd_seq->os_compat_dirs[b]);
414 snprintf(buf, bufsize, "%llu", id);
416 return osd_seq->os_compat_dirs[b];
419 /* XXX: f_ver is not counted, but may differ too */
420 static void osd_fid2str(char *buf, const struct lu_fid *fid)
422 sprintf(buf, DFID_NOBRACE, PFID(fid));
426 * Determine the zap object id which is being used as the OI for the
427 * given fid. The lowest N bits in the sequence ID are used as the
428 * index key. On failure 0 is returned which zfs treats internally
429 * as an invalid object id.
432 osd_get_idx_for_fid(struct osd_device *osd, const struct lu_fid *fid,
433 char *buf, dnode_t **zdn)
437 LASSERT(osd->od_oi_table != NULL);
438 oi = osd->od_oi_table[fid_seq(fid) & (osd->od_oi_count - 1)];
440 osd_fid2str(buf, fid);
447 uint64_t osd_get_name_n_idx(const struct lu_env *env, struct osd_device *osd,
448 const struct lu_fid *fid, char *buf, int bufsize,
454 LASSERT(!fid_is_acct(fid));
459 if (fid_is_on_ost(env, osd, fid) == 1 || fid_seq(fid) == FID_SEQ_ECHO) {
460 zapid = osd_get_idx_for_ost_obj(env, osd, fid, buf, bufsize);
461 } else if (unlikely(fid_seq(fid) == FID_SEQ_LOCAL_FILE)) {
462 /* special objects with fixed known fids get their name */
463 char *name = oid2name(fid_oid(fid));
466 zapid = osd->od_root;
468 strncpy(buf, name, bufsize);
470 zapid = osd_get_idx_for_fid(osd, fid, buf, NULL);
473 zapid = osd_get_idx_for_fid(osd, fid, buf, zdn);
479 static inline int fid_is_fs_root(const struct lu_fid *fid)
481 /* Map root inode to special local object FID */
482 return fid_seq(fid) == FID_SEQ_LOCAL_FILE &&
483 fid_oid(fid) == OSD_FS_ROOT_OID;
486 int osd_fid_lookup(const struct lu_env *env, struct osd_device *dev,
487 const struct lu_fid *fid, uint64_t *oid)
489 struct osd_thread_info *info = osd_oti_get(env);
490 char *buf = info->oti_buf;
496 if (OBD_FAIL_CHECK(OBD_FAIL_SRV_ENOENT))
499 LASSERT(!fid_is_acct(fid));
501 if (unlikely(fid_is_fs_root(fid))) {
504 zapid = osd_get_name_n_idx(env, dev, fid, buf,
505 sizeof(info->oti_buf), &zdn);
506 rc = osd_zap_lookup(dev, zapid, zdn, buf,
507 8, 1, &info->oti_zde);
510 *oid = info->oti_zde.lzd_reg.zde_dnode;
514 osd_dmu_prefetch(dev->od_os, *oid, 0, 0, 0,
515 ZIO_PRIORITY_ASYNC_READ);
521 * Close an entry in a specific slot.
524 osd_oi_remove_table(const struct lu_env *env, struct osd_device *o, int key)
528 LASSERT(key < o->od_oi_count);
530 oi = o->od_oi_table[key];
533 osd_dnode_rele(oi->oi_dn);
535 o->od_oi_table[key] = NULL;
540 * Allocate and open a new entry in the specified unused slot.
543 osd_oi_add_table(const struct lu_env *env, struct osd_device *o,
549 LASSERT(key < o->od_oi_count);
550 LASSERT(o->od_oi_table[key] == NULL);
556 rc = osd_oi_lookup(env, o, o->od_root, name, oi);
562 o->od_oi_table[key] = oi;
563 __osd_obj2dnode(o->od_os, oi->oi_zapid, &oi->oi_dn);
569 * Depopulate the OI table.
572 osd_oi_close_table(const struct lu_env *env, struct osd_device *o)
576 for (i = 0; i < o->od_oi_count; i++)
577 osd_oi_remove_table(env, o, i);
581 * Populate the OI table based.
584 osd_oi_open_table(const struct lu_env *env, struct osd_device *o, int count)
590 for (i = 0; i < count; i++) {
591 sprintf(name, "%s.%d", DMU_OSD_OI_NAME_BASE, i);
592 rc = osd_oi_add_table(env, o, name, i);
594 osd_oi_close_table(env, o);
603 * Determine if the type and number of OIs used by this file system.
606 osd_oi_probe(const struct lu_env *env, struct osd_device *o, int *count)
608 uint64_t root_oid = o->od_root;
615 * Check for multiple OIs and determine the count. There is no
616 * gap handling, if an OI is missing the wrong size can be returned.
617 * The only safeguard is that we know the number of OIs must be a
618 * power of two and this is checked for basic sanity.
620 for (*count = 0; *count < OSD_OI_FID_NR_MAX; (*count)++) {
621 sprintf(name, "%s.%d", DMU_OSD_OI_NAME_BASE, *count);
622 rc = osd_oi_lookup(env, o, root_oid, name, &oi);
630 if ((*count & (*count - 1)) != 0)
640 * No OIs exist, this must be a new filesystem.
647 static void osd_ost_seq_fini(const struct lu_env *env, struct osd_device *osd)
649 struct osd_seq_list *osl = &osd->od_seq_list;
650 struct osd_seq *osd_seq, *tmp;
652 write_lock(&osl->osl_seq_list_lock);
653 list_for_each_entry_safe(osd_seq, tmp, &osl->osl_seq_list,
655 list_del(&osd_seq->os_seq_list);
656 OBD_FREE(osd_seq->os_compat_dirs,
657 sizeof(uint64_t) * osd_seq->os_subdir_count);
658 OBD_FREE(osd_seq, sizeof(*osd_seq));
660 write_unlock(&osl->osl_seq_list_lock);
666 * Create /O subdirectory to map legacy OST objects for compatibility.
669 osd_oi_init_compat(const struct lu_env *env, struct osd_device *o)
675 rc = osd_oi_find_or_create(env, o, o->od_root, "O", &sdb);
683 * Initialize the OIs by either opening or creating them as needed.
685 int osd_oi_init(const struct lu_env *env, struct osd_device *o)
687 char *key = osd_oti_get(env)->oti_buf;
688 int i, rc, count = 0;
691 rc = osd_oi_probe(env, o, &count);
698 count = osd_oi_count;
701 for (i = 0; i < count; i++) {
702 sprintf(key, "%s.%d", DMU_OSD_OI_NAME_BASE, i);
703 rc = osd_oi_find_or_create(env, o, odb, key, &sdb);
709 rc = osd_oi_init_compat(env, o);
713 LASSERT((count & (count - 1)) == 0);
714 o->od_oi_count = count;
715 OBD_ALLOC(o->od_oi_table, sizeof(struct osd_oi *) * count);
716 if (o->od_oi_table == NULL)
719 rc = osd_oi_open_table(env, o, count);
721 OBD_FREE(o->od_oi_table, sizeof(struct osd_oi *) * count);
722 o->od_oi_table = NULL;
728 void osd_oi_fini(const struct lu_env *env, struct osd_device *o)
732 osd_ost_seq_fini(env, o);
734 if (o->od_oi_table != NULL) {
735 (void) osd_oi_close_table(env, o);
736 OBD_FREE(o->od_oi_table,
737 sizeof(struct osd_oi *) * o->od_oi_count);
738 o->od_oi_table = NULL;
745 int osd_options_init(void)
747 /* osd_oi_count - Default number of OIs, 128 works well for ZFS */
748 if (osd_oi_count == 0 || osd_oi_count > OSD_OI_FID_NR_MAX)
749 osd_oi_count = OSD_OI_FID_NR;
751 if ((osd_oi_count & (osd_oi_count - 1)) != 0) {
752 LCONSOLE_WARN("Round up osd_oi_count %d to power2 %d\n",
753 osd_oi_count, size_roundup_power2(osd_oi_count));
754 osd_oi_count = size_roundup_power2(osd_oi_count);
761 * the following set of functions are used to maintain per-thread
762 * cache of FID->ino mapping. this mechanism is used to avoid
763 * expensive LU/OI lookups.
765 struct osd_idmap_cache *osd_idc_find(const struct lu_env *env,
766 struct osd_device *osd,
767 const struct lu_fid *fid)
769 struct osd_thread_info *oti = osd_oti_get(env);
770 struct osd_idmap_cache *idc = oti->oti_ins_cache;
773 for (i = 0; i < oti->oti_ins_cache_used; i++) {
774 if (!lu_fid_eq(&idc[i].oic_fid, fid))
776 if (idc[i].oic_dev != osd)
785 struct osd_idmap_cache *osd_idc_add(const struct lu_env *env,
786 struct osd_device *osd,
787 const struct lu_fid *fid)
789 struct osd_thread_info *oti = osd_oti_get(env);
790 struct osd_idmap_cache *idc;
793 if (unlikely(oti->oti_ins_cache_used >= oti->oti_ins_cache_size)) {
794 i = oti->oti_ins_cache_size * 2;
797 i = OSD_INS_CACHE_SIZE;
798 OBD_ALLOC(idc, sizeof(*idc) * i);
800 return ERR_PTR(-ENOMEM);
801 if (oti->oti_ins_cache != NULL) {
802 memcpy(idc, oti->oti_ins_cache,
803 oti->oti_ins_cache_used * sizeof(*idc));
804 OBD_FREE(oti->oti_ins_cache,
805 oti->oti_ins_cache_used * sizeof(*idc));
807 oti->oti_ins_cache = idc;
808 oti->oti_ins_cache_size = i;
811 idc = &oti->oti_ins_cache[oti->oti_ins_cache_used++];
821 * Lookup mapping for the given fid in the cache
823 * Initialize a new one if not found. the initialization checks whether
824 * the object is local or remote. for the local objects, OI is used to
825 * learn dnode#. the function is used when the caller has no information
826 * about the object, e.g. at dt_insert().
828 struct osd_idmap_cache *osd_idc_find_or_init(const struct lu_env *env,
829 struct osd_device *osd,
830 const struct lu_fid *fid)
832 struct osd_idmap_cache *idc;
835 LASSERT(!fid_is_acct(fid));
837 idc = osd_idc_find(env, osd, fid);
841 /* new mapping is needed */
842 idc = osd_idc_add(env, osd, fid);
847 rc = osd_remote_fid(env, osd, fid);
848 if (unlikely(rc < 0))
852 /* the object is local, lookup in OI */
855 rc = osd_fid_lookup(env, osd, fid, &dnode);
856 if (unlikely(rc < 0)) {
857 CERROR("%s: can't lookup: rc = %d\n",
861 LASSERT(dnode < (1ULL << DN_MAX_OBJECT_SHIFT));
862 idc->oic_dnode = dnode;
864 /* the object is remote */
872 * lookup mapping for given FID and fill it from the given object.
873 * the object is local by definition.
875 int osd_idc_find_and_init(const struct lu_env *env, struct osd_device *osd,
876 struct osd_object *obj)
878 const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
879 struct osd_idmap_cache *idc;
881 idc = osd_idc_find(env, osd, fid);
883 if (obj->oo_dn == NULL)
885 idc->oic_dnode = obj->oo_dn->dn_object;
889 /* new mapping is needed */
890 idc = osd_idc_add(env, osd, fid);
895 idc->oic_dnode = obj->oo_dn->dn_object;