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 <lustre_ver.h>
43 #include <libcfs/libcfs.h>
44 #include <obd_support.h>
45 #include <lustre_net.h>
47 #include <obd_class.h>
48 #include <lustre_disk.h>
49 #include <lustre_fid.h>
51 #include "osd_internal.h"
53 #include <sys/dnode.h>
58 #include <sys/spa_impl.h>
59 #include <sys/zfs_znode.h>
60 #include <sys/dmu_tx.h>
61 #include <sys/dmu_objset.h>
62 #include <sys/dsl_prop.h>
63 #include <sys/sa_impl.h>
66 #define OSD_OI_FID_NR (1UL << 7)
67 #define OSD_OI_FID_NR_MAX (1UL << OSD_OI_FID_OID_BITS_MAX)
68 unsigned int osd_oi_count = OSD_OI_FID_NR;
72 * zfs osd maintains names for known fids in the name hierarchy
73 * so that one can mount filesystem with regular ZFS stack and
81 static const struct named_oid oids[] = {
82 { .oid = LAST_RECV_OID, .name = LAST_RCVD },
83 { .oid = OFD_LAST_GROUP_OID, .name = "LAST_GROUP" },
84 { .oid = LLOG_CATALOGS_OID, .name = "CATALOGS" },
85 { .oid = MGS_CONFIGS_OID, /*MOUNT_CONFIGS_DIR*/ },
86 { .oid = FID_SEQ_SRV_OID, .name = "seq_srv" },
87 { .oid = FID_SEQ_CTL_OID, .name = "seq_ctl" },
88 { .oid = FLD_INDEX_OID, .name = "fld" },
89 { .oid = MDD_LOV_OBJ_OID, .name = LOV_OBJID },
90 { .oid = OFD_HEALTH_CHECK_OID, .name = HEALTH_CHECK },
91 { .oid = ACCT_USER_OID, .name = "acct_usr_inode" },
92 { .oid = ACCT_GROUP_OID, .name = "acct_grp_inode" },
93 { .oid = REPLY_DATA_OID, .name = REPLY_DATA },
97 static char *oid2name(const unsigned long oid)
101 while (oids[i].oid) {
102 if (oids[i].oid == oid)
110 * Lookup an existing OI by the given name.
113 osd_oi_lookup(const struct lu_env *env, struct osd_device *o,
114 uint64_t parent, const char *name, struct osd_oi *oi)
116 struct zpl_direntry *zde = &osd_oti_get(env)->oti_zde.lzd_reg;
119 rc = -zap_lookup(o->od_os, parent, name, 8, 1, (void *)zde);
123 rc = strlcpy(oi->oi_name, name, sizeof(oi->oi_name));
124 if (rc >= sizeof(oi->oi_name))
127 oi->oi_zapid = zde->zde_dnode;
133 * Create a new OI with the given name.
136 osd_oi_create(const struct lu_env *env, struct osd_device *o,
137 uint64_t parent, const char *name, uint64_t *child)
139 struct zpl_direntry *zde = &osd_oti_get(env)->oti_zde.lzd_reg;
140 struct lu_attr *la = &osd_oti_get(env)->oti_la;
141 sa_handle_t *sa_hdl = NULL;
146 /* verify it doesn't already exist */
147 rc = -zap_lookup(o->od_os, parent, name, 8, 1, (void *)zde);
151 if (o->od_dt_dev.dd_rdonly)
154 /* create fid-to-dnode index */
155 tx = dmu_tx_create(o->od_os);
159 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, 1, NULL);
160 dmu_tx_hold_bonus(tx, parent);
161 dmu_tx_hold_zap(tx, parent, TRUE, name);
162 dmu_tx_hold_sa_create(tx, ZFS_SA_BASE_ATTR_SIZE);
164 rc = -dmu_tx_assign(tx, TXG_WAIT);
170 oid = zap_create_flags(o->od_os, 0, ZAP_FLAG_HASH64,
171 DMU_OT_DIRECTORY_CONTENTS,
172 14, /* == ZFS fzap_default_block_shift */
173 DN_MAX_INDBLKSHIFT, /* indirect block shift */
174 DMU_OT_SA, DN_MAX_BONUSLEN, tx);
176 rc = -sa_handle_get(o->od_os, oid, NULL, SA_HDL_PRIVATE, &sa_hdl);
179 la->la_valid = LA_MODE | LA_UID | LA_GID;
180 la->la_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
181 la->la_uid = la->la_gid = 0;
182 rc = __osd_attr_init(env, o, sa_hdl, tx, la, parent);
183 sa_handle_destroy(sa_hdl);
187 zde->zde_dnode = oid;
189 zde->zde_type = IFTODT(S_IFDIR);
191 rc = -zap_add(o->od_os, parent, name, 8, 1, (void *)zde, tx);
195 dmu_object_free(o->od_os, oid, tx);
205 osd_oi_find_or_create(const struct lu_env *env, struct osd_device *o,
206 uint64_t parent, const char *name, uint64_t *child)
211 rc = osd_oi_lookup(env, o, parent, name, &oi);
213 *child = oi.oi_zapid;
214 else if (rc == -ENOENT)
215 rc = osd_oi_create(env, o, parent, name, child);
221 * Lookup the target index/flags of the fid, so it will know where
222 * the object is located (tgt index) and it is MDT or OST object.
224 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
225 u64 seq, struct lu_seq_range *range)
227 struct seq_server_site *ss = osd_seq_site(osd);
229 if (fid_seq_is_idif(seq)) {
230 fld_range_set_ost(range);
231 range->lsr_index = idif_ost_idx(seq);
235 if (!fid_seq_in_fldb(seq)) {
236 fld_range_set_mdt(range);
238 /* FIXME: If ss is NULL, it suppose not get lsr_index
240 range->lsr_index = ss->ss_node_id;
245 fld_range_set_any(range);
246 /* OSD will only do local fld lookup */
247 return fld_local_lookup(env, ss->ss_server_fld, seq, range);
250 int fid_is_on_ost(const struct lu_env *env, struct osd_device *osd,
251 const struct lu_fid *fid)
253 struct lu_seq_range *range = &osd_oti_get(env)->oti_seq_range;
257 if (fid_is_idif(fid))
260 if (unlikely(fid_is_local_file(fid) || fid_is_llog(fid)) ||
261 fid_is_name_llog(fid) || fid_is_quota(fid))
264 rc = osd_fld_lookup(env, osd, fid_seq(fid), range);
267 CERROR("%s: "DFID" lookup failed: rc = %d\n",
268 osd_name(osd), PFID(fid), rc);
272 if (fld_range_is_ost(range))
278 static struct osd_seq *osd_seq_find_locked(struct osd_seq_list *seq_list,
281 struct osd_seq *osd_seq;
283 list_for_each_entry(osd_seq, &seq_list->osl_seq_list, os_seq_list) {
284 if (osd_seq->os_seq == seq)
290 static struct osd_seq *osd_seq_find(struct osd_seq_list *seq_list, u64 seq)
292 struct osd_seq *osd_seq;
294 read_lock(&seq_list->osl_seq_list_lock);
295 osd_seq = osd_seq_find_locked(seq_list, seq);
296 read_unlock(&seq_list->osl_seq_list_lock);
301 static struct osd_seq *osd_find_or_add_seq(const struct lu_env *env,
302 struct osd_device *osd, u64 seq)
304 struct osd_seq_list *seq_list = &osd->od_seq_list;
305 struct osd_seq *osd_seq;
306 char *key = osd_oti_get(env)->oti_buf;
307 char *seq_name = osd_oti_get(env)->oti_str;
314 osd_seq = osd_seq_find(seq_list, seq);
318 down(&seq_list->osl_seq_init_sem);
319 /* Check again, in case some one else already add it
321 osd_seq = osd_seq_find(seq_list, seq);
325 OBD_ALLOC_PTR(osd_seq);
327 GOTO(out, rc = -ENOMEM);
329 INIT_LIST_HEAD(&osd_seq->os_seq_list);
330 osd_seq->os_seq = seq;
332 /* Init subdir count to be 32, but each seq can have
333 * different subdir count */
334 osd_seq->os_subdir_count = OSD_OST_MAP_SIZE;
335 OBD_ALLOC(osd_seq->os_compat_dirs,
336 sizeof(uint64_t) * osd_seq->os_subdir_count);
337 if (osd_seq->os_compat_dirs == NULL)
338 GOTO(out, rc = -ENOMEM);
340 oi.oi_zapid = osd->od_O_id;
341 sprintf(seq_name, (fid_seq_is_rsvd(seq) ||
342 fid_seq_is_mdt0(seq)) ? "%llu" : "%llx",
343 fid_seq_is_idif(seq) ? 0 : seq);
345 rc = osd_oi_find_or_create(env, osd, oi.oi_zapid, seq_name, &odb);
347 CERROR("%s: Can not create %s : rc = %d\n",
348 osd_name(osd), seq_name, rc);
352 for (i = 0; i < OSD_OST_MAP_SIZE; i++) {
353 sprintf(key, "d%d", i);
354 rc = osd_oi_find_or_create(env, osd, odb, key, &sdb);
357 osd_seq->os_compat_dirs[i] = sdb;
360 write_lock(&seq_list->osl_seq_list_lock);
361 list_add(&osd_seq->os_seq_list, &seq_list->osl_seq_list);
362 write_unlock(&seq_list->osl_seq_list_lock);
364 up(&seq_list->osl_seq_init_sem);
366 if (osd_seq != NULL && osd_seq->os_compat_dirs != NULL)
367 OBD_FREE(osd_seq->os_compat_dirs,
368 sizeof(uint64_t) * osd_seq->os_subdir_count);
370 OBD_FREE_PTR(osd_seq);
371 osd_seq = ERR_PTR(rc);
377 * objects w/o a natural reference (unlike a file on a MDS)
378 * are put under a special hierarchy /O/<seq>/d0..dXX
379 * this function returns a directory specific fid belongs to
382 osd_get_idx_for_ost_obj(const struct lu_env *env, struct osd_device *osd,
383 const struct lu_fid *fid, char *buf, int bufsize)
385 struct osd_seq *osd_seq;
390 osd_seq = osd_find_or_add_seq(env, osd, fid_seq(fid));
391 if (IS_ERR(osd_seq)) {
392 CERROR("%s: Can not find seq group "DFID"\n", osd_name(osd),
394 return PTR_ERR(osd_seq);
397 if (fid_is_last_id(fid)) {
400 rc = fid_to_ostid(fid, &osd_oti_get(env)->oti_ostid);
401 LASSERT(rc == 0); /* we should not get here with IGIF */
402 id = ostid_id(&osd_oti_get(env)->oti_ostid);
405 b = id % OSD_OST_MAP_SIZE;
406 LASSERT(osd_seq->os_compat_dirs[b]);
409 snprintf(buf, bufsize, "%llu", id);
411 return osd_seq->os_compat_dirs[b];
414 /* XXX: f_ver is not counted, but may differ too */
415 static void osd_fid2str(char *buf, const struct lu_fid *fid)
417 sprintf(buf, DFID_NOBRACE, PFID(fid));
421 * Determine the zap object id which is being used as the OI for the
422 * given fid. The lowest N bits in the sequence ID are used as the
423 * index key. On failure 0 is returned which zfs treats internally
424 * as an invalid object id.
427 osd_get_idx_for_fid(struct osd_device *osd, const struct lu_fid *fid,
432 LASSERT(osd->od_oi_table != NULL);
433 oi = osd->od_oi_table[fid_seq(fid) & (osd->od_oi_count - 1)];
435 osd_fid2str(buf, fid);
440 uint64_t osd_get_name_n_idx(const struct lu_env *env, struct osd_device *osd,
441 const struct lu_fid *fid, char *buf, int bufsize)
447 if (fid_is_on_ost(env, osd, fid) == 1 || fid_seq(fid) == FID_SEQ_ECHO) {
448 zapid = osd_get_idx_for_ost_obj(env, osd, fid, buf, bufsize);
449 } else if (unlikely(fid_seq(fid) == FID_SEQ_LOCAL_FILE)) {
450 /* special objects with fixed known fids get their name */
451 char *name = oid2name(fid_oid(fid));
454 zapid = osd->od_root;
456 strncpy(buf, name, bufsize);
457 if (fid_is_acct(fid))
458 zapid = MASTER_NODE_OBJ;
460 zapid = osd_get_idx_for_fid(osd, fid, buf);
463 zapid = osd_get_idx_for_fid(osd, fid, buf);
469 static inline int fid_is_fs_root(const struct lu_fid *fid)
471 /* Map root inode to special local object FID */
472 return fid_seq(fid) == FID_SEQ_LOCAL_FILE &&
473 fid_oid(fid) == OSD_FS_ROOT_OID;
476 int osd_fid_lookup(const struct lu_env *env, struct osd_device *dev,
477 const struct lu_fid *fid, uint64_t *oid)
479 struct osd_thread_info *info = osd_oti_get(env);
480 char *buf = info->oti_buf;
485 if (OBD_FAIL_CHECK(OBD_FAIL_SRV_ENOENT))
488 if (unlikely(fid_is_acct(fid))) {
489 if (fid_oid(fid) == ACCT_USER_OID)
490 *oid = dev->od_iusr_oid;
492 *oid = dev->od_igrp_oid;
493 } else if (unlikely(fid_is_fs_root(fid))) {
496 zapid = osd_get_name_n_idx(env, dev, fid, buf,
497 sizeof(info->oti_buf));
498 rc = -zap_lookup(dev->od_os, zapid, buf,
499 8, 1, &info->oti_zde);
502 *oid = info->oti_zde.lzd_reg.zde_dnode;
506 osd_dmu_prefetch(dev->od_os, *oid, 0, 0, 0,
507 ZIO_PRIORITY_ASYNC_READ);
513 * Close an entry in a specific slot.
516 osd_oi_remove_table(const struct lu_env *env, struct osd_device *o, int key)
520 LASSERT(key < o->od_oi_count);
522 oi = o->od_oi_table[key];
525 osd_dnode_rele(oi->oi_dn);
527 o->od_oi_table[key] = NULL;
532 * Allocate and open a new entry in the specified unused slot.
535 osd_oi_add_table(const struct lu_env *env, struct osd_device *o,
541 LASSERT(key < o->od_oi_count);
542 LASSERT(o->od_oi_table[key] == NULL);
548 rc = osd_oi_lookup(env, o, o->od_root, name, oi);
554 o->od_oi_table[key] = oi;
555 __osd_obj2dnode(env, o->od_os, oi->oi_zapid, &oi->oi_dn);
561 * Depopulate the OI table.
564 osd_oi_close_table(const struct lu_env *env, struct osd_device *o)
568 for (i = 0; i < o->od_oi_count; i++)
569 osd_oi_remove_table(env, o, i);
573 * Populate the OI table based.
576 osd_oi_open_table(const struct lu_env *env, struct osd_device *o, int count)
582 for (i = 0; i < count; i++) {
583 sprintf(name, "%s.%d", DMU_OSD_OI_NAME_BASE, i);
584 rc = osd_oi_add_table(env, o, name, i);
586 osd_oi_close_table(env, o);
595 * Determine if the type and number of OIs used by this file system.
598 osd_oi_probe(const struct lu_env *env, struct osd_device *o, int *count)
600 uint64_t root_oid = o->od_root;
607 * Check for multiple OIs and determine the count. There is no
608 * gap handling, if an OI is missing the wrong size can be returned.
609 * The only safeguard is that we know the number of OIs must be a
610 * power of two and this is checked for basic sanity.
612 for (*count = 0; *count < OSD_OI_FID_NR_MAX; (*count)++) {
613 sprintf(name, "%s.%d", DMU_OSD_OI_NAME_BASE, *count);
614 rc = osd_oi_lookup(env, o, root_oid, name, &oi);
622 if ((*count & (*count - 1)) != 0)
632 * No OIs exist, this must be a new filesystem.
639 static void osd_ost_seq_fini(const struct lu_env *env, struct osd_device *osd)
641 struct osd_seq_list *osl = &osd->od_seq_list;
642 struct osd_seq *osd_seq, *tmp;
644 write_lock(&osl->osl_seq_list_lock);
645 list_for_each_entry_safe(osd_seq, tmp, &osl->osl_seq_list,
647 list_del(&osd_seq->os_seq_list);
648 OBD_FREE(osd_seq->os_compat_dirs,
649 sizeof(uint64_t) * osd_seq->os_subdir_count);
650 OBD_FREE(osd_seq, sizeof(*osd_seq));
652 write_unlock(&osl->osl_seq_list_lock);
658 * Create /O subdirectory to map legacy OST objects for compatibility.
661 osd_oi_init_compat(const struct lu_env *env, struct osd_device *o)
667 rc = osd_oi_find_or_create(env, o, o->od_root, "O", &sdb);
673 /* Create on-disk indexes to maintain per-UID/GID inode usage.
674 * Those new indexes are created in the top-level ZAP outside the
675 * namespace in order not to confuse ZPL which might interpret those
676 * indexes as directories and assume the values are object IDs */
677 rc = osd_oi_find_or_create(env, o, MASTER_NODE_OBJ,
678 oid2name(ACCT_USER_OID), &odb);
681 o->od_iusr_oid = odb;
683 rc = osd_oi_find_or_create(env, o, MASTER_NODE_OBJ,
684 oid2name(ACCT_GROUP_OID), &odb);
687 o->od_igrp_oid = odb;
693 * Initialize the OIs by either opening or creating them as needed.
695 int osd_oi_init(const struct lu_env *env, struct osd_device *o)
697 char *key = osd_oti_get(env)->oti_buf;
698 int i, rc, count = 0;
701 rc = osd_oi_probe(env, o, &count);
708 count = osd_oi_count;
711 for (i = 0; i < count; i++) {
712 sprintf(key, "%s.%d", DMU_OSD_OI_NAME_BASE, i);
713 rc = osd_oi_find_or_create(env, o, odb, key, &sdb);
719 rc = osd_oi_init_compat(env, o);
723 LASSERT((count & (count - 1)) == 0);
724 o->od_oi_count = count;
725 OBD_ALLOC(o->od_oi_table, sizeof(struct osd_oi *) * count);
726 if (o->od_oi_table == NULL)
729 rc = osd_oi_open_table(env, o, count);
731 OBD_FREE(o->od_oi_table, sizeof(struct osd_oi *) * count);
732 o->od_oi_table = NULL;
738 void osd_oi_fini(const struct lu_env *env, struct osd_device *o)
742 osd_ost_seq_fini(env, o);
744 if (o->od_oi_table != NULL) {
745 (void) osd_oi_close_table(env, o);
746 OBD_FREE(o->od_oi_table,
747 sizeof(struct osd_oi *) * o->od_oi_count);
748 o->od_oi_table = NULL;
755 int osd_options_init(void)
757 /* osd_oi_count - Default number of OIs, 128 works well for ZFS */
758 if (osd_oi_count == 0 || osd_oi_count > OSD_OI_FID_NR_MAX)
759 osd_oi_count = OSD_OI_FID_NR;
761 if ((osd_oi_count & (osd_oi_count - 1)) != 0) {
762 LCONSOLE_WARN("Round up osd_oi_count %d to power2 %d\n",
763 osd_oi_count, size_roundup_power2(osd_oi_count));
764 osd_oi_count = size_roundup_power2(osd_oi_count);
771 * the following set of functions are used to maintain per-thread
772 * cache of FID->ino mapping. this mechanism is used to avoid
773 * expensive LU/OI lookups.
775 struct osd_idmap_cache *osd_idc_find(const struct lu_env *env,
776 struct osd_device *osd,
777 const struct lu_fid *fid)
779 struct osd_thread_info *oti = osd_oti_get(env);
780 struct osd_idmap_cache *idc = oti->oti_ins_cache;
783 for (i = 0; i < oti->oti_ins_cache_used; i++) {
784 if (!lu_fid_eq(&idc[i].oic_fid, fid))
786 if (idc[i].oic_dev != osd)
795 struct osd_idmap_cache *osd_idc_add(const struct lu_env *env,
796 struct osd_device *osd,
797 const struct lu_fid *fid)
799 struct osd_thread_info *oti = osd_oti_get(env);
800 struct osd_idmap_cache *idc;
803 if (unlikely(oti->oti_ins_cache_used >= oti->oti_ins_cache_size)) {
804 i = oti->oti_ins_cache_size * 2;
807 i = OSD_INS_CACHE_SIZE;
808 OBD_ALLOC(idc, sizeof(*idc) * i);
810 return ERR_PTR(-ENOMEM);
811 if (oti->oti_ins_cache != NULL) {
812 memcpy(idc, oti->oti_ins_cache,
813 oti->oti_ins_cache_used * sizeof(*idc));
814 OBD_FREE(oti->oti_ins_cache,
815 oti->oti_ins_cache_used * sizeof(*idc));
817 oti->oti_ins_cache = idc;
818 oti->oti_ins_cache_size = i;
821 idc = &oti->oti_ins_cache[oti->oti_ins_cache_used++];
831 * Lookup mapping for the given fid in the cache
833 * Initialize a new one if not found. the initialization checks whether
834 * the object is local or remote. for the local objects, OI is used to
835 * learn dnode#. the function is used when the caller has no information
836 * about the object, e.g. at dt_insert().
838 struct osd_idmap_cache *osd_idc_find_or_init(const struct lu_env *env,
839 struct osd_device *osd,
840 const struct lu_fid *fid)
842 struct osd_idmap_cache *idc;
845 idc = osd_idc_find(env, osd, fid);
849 /* new mapping is needed */
850 idc = osd_idc_add(env, osd, fid);
855 rc = osd_remote_fid(env, osd, fid);
856 if (unlikely(rc < 0))
860 /* the object is local, lookup in OI */
863 rc = osd_fid_lookup(env, osd, fid, &dnode);
864 if (unlikely(rc < 0)) {
865 CERROR("%s: can't lookup: rc = %d\n",
869 LASSERT(dnode < (1ULL << DN_MAX_OBJECT_SHIFT));
870 idc->oic_dnode = dnode;
872 /* the object is remote */
880 * lookup mapping for given FID and fill it from the given object.
881 * the object is local by definition.
883 int osd_idc_find_and_init(const struct lu_env *env, struct osd_device *osd,
884 struct osd_object *obj)
886 const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
887 struct osd_idmap_cache *idc;
889 idc = osd_idc_find(env, osd, fid);
891 if (obj->oo_dn == NULL)
893 idc->oic_dnode = obj->oo_dn->dn_object;
897 /* new mapping is needed */
898 idc = osd_idc_add(env, osd, fid);
903 idc->oic_dnode = obj->oo_dn->dn_object;