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.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2009, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
31 * Copyright (c) 2012, 2013, Intel Corporation.
32 * Use is subject to license terms.
35 * This file is part of Lustre, http://www.lustre.org/
36 * Lustre is a trademark of Sun Microsystems, Inc.
38 * lustre/osd-zfs/osd_oi.c
39 * OI functions to map fid to dnode
41 * Author: Alex Zhuravlev <bzzz@whamcloud.com>
42 * Author: Mike Pershin <tappro@whamcloud.com>
43 * Author: Di Wang <di.wang@intel.com>
47 # define EXPORT_SYMTAB
49 #define DEBUG_SUBSYSTEM S_OSD
51 #include <lustre_ver.h>
52 #include <libcfs/libcfs.h>
53 #include <obd_support.h>
54 #include <lustre_net.h>
56 #include <obd_class.h>
57 #include <lustre_disk.h>
58 #include <lustre_fid.h>
60 #include "osd_internal.h"
62 #include <sys/dnode.h>
67 #include <sys/spa_impl.h>
68 #include <sys/zfs_znode.h>
69 #include <sys/dmu_tx.h>
70 #include <sys/dmu_objset.h>
71 #include <sys/dsl_prop.h>
72 #include <sys/sa_impl.h>
75 static char *oi_tag = "osd_mount, oi";
77 #define OSD_OI_FID_NR (1UL << 7)
78 #define OSD_OI_FID_NR_MAX (1UL << OSD_OI_FID_OID_BITS_MAX)
79 unsigned int osd_oi_count = OSD_OI_FID_NR;
83 * zfs osd maintains names for known fids in the name hierarchy
84 * so that one can mount filesystem with regular ZFS stack and
92 static const struct named_oid oids[] = {
93 { LAST_RECV_OID, LAST_RCVD },
94 { OFD_LAST_GROUP_OID, "LAST_GROUP" },
95 { LLOG_CATALOGS_OID, "CATALOGS" },
96 { MGS_CONFIGS_OID, NULL /*MOUNT_CONFIGS_DIR*/ },
97 { FID_SEQ_SRV_OID, "seq_srv" },
98 { FID_SEQ_CTL_OID, "seq_ctl" },
99 { FLD_INDEX_OID, "fld" },
100 { MDD_LOV_OBJ_OID, LOV_OBJID },
101 { OFD_HEALTH_CHECK_OID, HEALTH_CHECK },
102 { ACCT_USER_OID, "acct_usr_inode" },
103 { ACCT_GROUP_OID, "acct_grp_inode" },
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_objset.os, parent, name, 8, 1, (void *)zde);
133 strncpy(oi->oi_name, name, OSD_OI_NAME_SIZE - 1);
134 oi->oi_zapid = zde->zde_dnode;
140 * Create a new OI with the given name.
143 osd_oi_create(const struct lu_env *env, struct osd_device *o,
144 uint64_t parent, const char *name, uint64_t *child)
146 struct zpl_direntry *zde = &osd_oti_get(env)->oti_zde.lzd_reg;
147 struct lu_attr *la = &osd_oti_get(env)->oti_la;
152 /* verify it doesn't already exist */
153 rc = -zap_lookup(o->od_objset.os, parent, name, 8, 1, (void *)zde);
157 /* create fid-to-dnode index */
158 tx = dmu_tx_create(o->od_objset.os);
162 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, 1, NULL);
163 dmu_tx_hold_bonus(tx, parent);
164 dmu_tx_hold_zap(tx, parent, TRUE, name);
165 LASSERT(tx->tx_objset->os_sa);
166 dmu_tx_hold_sa_create(tx, ZFS_SA_BASE_ATTR_SIZE);
168 rc = -dmu_tx_assign(tx, TXG_WAIT);
174 la->la_valid = LA_MODE | LA_UID | LA_GID;
175 la->la_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
176 la->la_uid = la->la_gid = 0;
177 __osd_zap_create(env, &o->od_objset, &db, tx, la, parent, oi_tag, 0);
179 zde->zde_dnode = db->db_object;
181 zde->zde_type = IFTODT(S_IFDIR);
183 rc = -zap_add(o->od_objset.os, parent, name, 8, 1, (void *)zde, tx);
187 *child = db->db_object;
188 sa_buf_rele(db, oi_tag);
194 osd_oi_find_or_create(const struct lu_env *env, struct osd_device *o,
195 uint64_t parent, const char *name, uint64_t *child)
200 rc = osd_oi_lookup(env, o, parent, name, &oi);
202 *child = oi.oi_zapid;
203 else if (rc == -ENOENT)
204 rc = osd_oi_create(env, o, parent, name, child);
210 * Lookup the target index/flags of the fid, so it will know where
211 * the object is located (tgt index) and it is MDT or OST object.
213 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
214 const struct lu_fid *fid, struct lu_seq_range *range)
216 struct seq_server_site *ss = osd_seq_site(osd);
219 if (fid_is_idif(fid)) {
220 fld_range_set_ost(range);
221 range->lsr_index = fid_idif_ost_idx(fid);
225 if (!fid_seq_in_fldb(fid_seq(fid))) {
226 fld_range_set_mdt(range);
228 /* FIXME: If ss is NULL, it suppose not get lsr_index
230 range->lsr_index = ss->ss_node_id;
235 fld_range_set_any(range);
236 rc = fld_server_lookup(env, ss->ss_server_fld, fid_seq(fid), range);
238 CERROR("%s: cannot find FLD range for "DFID": rc = %d\n",
239 osd_name(osd), PFID(fid), rc);
243 int fid_is_on_ost(const struct lu_env *env, struct osd_device *osd,
244 const struct lu_fid *fid)
246 struct lu_seq_range *range = &osd_oti_get(env)->oti_seq_range;
250 if (fid_is_idif(fid))
253 rc = osd_fld_lookup(env, osd, fid, range);
255 CERROR("%s: Can not lookup fld for "DFID"\n",
256 osd_name(osd), PFID(fid));
260 CDEBUG(D_INFO, "fid "DFID" range "DRANGE"\n", PFID(fid),
263 if (fld_range_is_ost(range))
269 static struct osd_seq *osd_seq_find_locked(struct osd_seq_list *seq_list,
272 struct osd_seq *osd_seq;
274 cfs_list_for_each_entry(osd_seq, &seq_list->osl_seq_list, os_seq_list) {
275 if (osd_seq->os_seq == seq)
281 static struct osd_seq *osd_seq_find(struct osd_seq_list *seq_list,
284 struct osd_seq *osd_seq;
286 read_lock(&seq_list->osl_seq_list_lock);
287 osd_seq = osd_seq_find_locked(seq_list, seq);
288 read_unlock(&seq_list->osl_seq_list_lock);
293 static struct osd_seq *osd_find_or_add_seq(const struct lu_env *env,
294 struct osd_device *osd, obd_seq seq)
296 struct osd_seq_list *seq_list = &osd->od_seq_list;
297 struct osd_seq *osd_seq;
298 char *key = osd_oti_get(env)->oti_buf;
299 char *seq_name = osd_oti_get(env)->oti_str;
306 osd_seq = osd_seq_find(seq_list, seq);
310 down(&seq_list->osl_seq_init_sem);
311 /* Check again, in case some one else already add it
313 osd_seq = osd_seq_find(seq_list, seq);
317 OBD_ALLOC_PTR(osd_seq);
319 GOTO(out, rc = -ENOMEM);
321 CFS_INIT_LIST_HEAD(&osd_seq->os_seq_list);
322 osd_seq->os_seq = seq;
324 /* Init subdir count to be 32, but each seq can have
325 * different subdir count */
326 osd_seq->os_subdir_count = OSD_OST_MAP_SIZE;
327 OBD_ALLOC(osd_seq->os_compat_dirs,
328 sizeof(uint64_t) * osd_seq->os_subdir_count);
329 if (osd_seq->os_compat_dirs == NULL)
330 GOTO(out, rc = -ENOMEM);
332 rc = osd_oi_lookup(env, osd, osd->od_root, "O", &oi);
334 CERROR("%s: Can not find O: rc = %d\n", osd_name(osd), rc);
338 sprintf(seq_name, (fid_seq_is_rsvd(seq) ||
339 fid_seq_is_mdt0(seq)) ? LPU64 : LPX64i,
340 fid_seq_is_idif(seq) ? 0 : seq);
342 rc = osd_oi_find_or_create(env, osd, oi.oi_zapid, seq_name, &odb);
344 CERROR("%s: Can not create %s : rc = %d\n",
345 osd_name(osd), seq_name, rc);
349 for (i = 0; i < OSD_OST_MAP_SIZE; i++) {
350 sprintf(key, "d%d", i);
351 rc = osd_oi_find_or_create(env, osd, odb, key, &sdb);
353 GOTO(out, osd_seq = ERR_PTR(rc));
354 osd_seq->os_compat_dirs[i] = sdb;
357 write_lock(&seq_list->osl_seq_list_lock);
358 cfs_list_add(&osd_seq->os_seq_list, &seq_list->osl_seq_list);
359 write_unlock(&seq_list->osl_seq_list_lock);
361 up(&seq_list->osl_seq_init_sem);
363 if (osd_seq != NULL && osd_seq->os_compat_dirs != NULL)
364 OBD_FREE(osd_seq->os_compat_dirs,
365 sizeof(uint64_t) * osd_seq->os_subdir_count);
367 OBD_FREE_PTR(osd_seq);
368 osd_seq = ERR_PTR(rc);
374 * objects w/o a natural reference (unlike a file on a MDS)
375 * are put under a special hierarchy /O/<seq>/d0..dXX
376 * this function returns a directory specific fid belongs to
379 osd_get_idx_for_ost_obj(const struct lu_env *env, struct osd_device *osd,
380 const struct lu_fid *fid, char *buf)
382 struct osd_seq *osd_seq;
386 osd_seq = osd_find_or_add_seq(env, osd, fid_seq(fid));
387 if (IS_ERR(osd_seq)) {
388 CERROR("%s: Can not find seq group "DFID"\n", osd_name(osd),
390 return PTR_ERR(osd_seq);
393 rc = fid_to_ostid(fid, &osd_oti_get(env)->oti_ostid);
394 LASSERT(rc == 0); /* we should not get here with IGIF */
395 b = ostid_id(&osd_oti_get(env)->oti_ostid) % OSD_OST_MAP_SIZE;
396 LASSERT(osd_seq->os_compat_dirs[b]);
398 sprintf(buf, LPU64, ostid_id(&osd_oti_get(env)->oti_ostid));
400 return osd_seq->os_compat_dirs[b];
403 /* XXX: f_ver is not counted, but may differ too */
404 static void osd_fid2str(char *buf, const struct lu_fid *fid)
406 sprintf(buf, DFID_NOBRACE, PFID(fid));
410 * Determine the zap object id which is being used as the OI for the
411 * given fid. The lowest N bits in the sequence ID are used as the
412 * index key. On failure 0 is returned which zfs treats internally
413 * as an invalid object id.
416 osd_get_idx_for_fid(struct osd_device *osd, const struct lu_fid *fid,
421 LASSERT(osd->od_oi_table != NULL);
422 oi = osd->od_oi_table[fid_seq(fid) & (osd->od_oi_count - 1)];
423 osd_fid2str(buf, fid);
428 uint64_t osd_get_name_n_idx(const struct lu_env *env, struct osd_device *osd,
429 const struct lu_fid *fid, char *buf)
436 if (fid_is_on_ost(env, osd, fid) == 1 || fid_seq(fid) == FID_SEQ_ECHO) {
437 zapid = osd_get_idx_for_ost_obj(env, osd, fid, buf);
438 } else if (unlikely(fid_seq(fid) == FID_SEQ_LOCAL_FILE)) {
439 /* special objects with fixed known fids get their name */
440 char *name = oid2name(fid_oid(fid));
443 zapid = osd->od_root;
445 if (fid_is_acct(fid))
446 zapid = MASTER_NODE_OBJ;
448 zapid = osd_get_idx_for_fid(osd, fid, buf);
451 zapid = osd_get_idx_for_fid(osd, fid, buf);
457 static inline int fid_is_fs_root(const struct lu_fid *fid)
459 /* Map root inode to special local object FID */
460 return fid_seq(fid) == FID_SEQ_LOCAL_FILE &&
461 fid_oid(fid) == OSD_FS_ROOT_OID;
464 int osd_fid_lookup(const struct lu_env *env, struct osd_device *dev,
465 const struct lu_fid *fid, uint64_t *oid)
467 struct osd_thread_info *info = osd_oti_get(env);
468 char *buf = info->oti_buf;
473 if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOENT))
476 if (unlikely(fid_is_acct(fid))) {
477 if (fid_oid(fid) == ACCT_USER_OID)
478 *oid = dev->od_iusr_oid;
480 *oid = dev->od_igrp_oid;
481 } else if (unlikely(fid_is_fs_root(fid))) {
484 zapid = osd_get_name_n_idx(env, dev, fid, buf);
486 rc = -zap_lookup(dev->od_objset.os, zapid, buf,
487 8, 1, &info->oti_zde);
490 *oid = info->oti_zde.lzd_reg.zde_dnode;
497 * Close an entry in a specific slot.
500 osd_oi_remove_table(const struct lu_env *env, struct osd_device *o, int key)
504 LASSERT(key < o->od_oi_count);
506 oi = o->od_oi_table[key];
509 o->od_oi_table[key] = NULL;
514 * Allocate and open a new entry in the specified unused slot.
517 osd_oi_add_table(const struct lu_env *env, struct osd_device *o,
523 LASSERT(key < o->od_oi_count);
524 LASSERT(o->od_oi_table[key] == NULL);
530 rc = osd_oi_lookup(env, o, o->od_root, name, oi);
536 o->od_oi_table[key] = oi;
542 * Depopulate the OI table.
545 osd_oi_close_table(const struct lu_env *env, struct osd_device *o)
549 for (i = 0; i < o->od_oi_count; i++)
550 osd_oi_remove_table(env, o, i);
554 * Populate the OI table based.
557 osd_oi_open_table(const struct lu_env *env, struct osd_device *o, int count)
563 for (i = 0; i < count; i++) {
564 sprintf(name, "%s.%d", DMU_OSD_OI_NAME_BASE, i);
565 rc = osd_oi_add_table(env, o, name, i);
567 osd_oi_close_table(env, o);
576 * Determine if the type and number of OIs used by this file system.
579 osd_oi_probe(const struct lu_env *env, struct osd_device *o, int *count)
581 uint64_t root_oid = o->od_root;
588 * Check for multiple OIs and determine the count. There is no
589 * gap handling, if an OI is missing the wrong size can be returned.
590 * The only safeguard is that we know the number of OIs must be a
591 * power of two and this is checked for basic sanity.
593 for (*count = 0; *count < OSD_OI_FID_NR_MAX; (*count)++) {
594 sprintf(name, "%s.%d", DMU_OSD_OI_NAME_BASE, *count);
595 rc = osd_oi_lookup(env, o, root_oid, name, &oi);
603 if ((*count & (*count - 1)) != 0)
613 * No OIs exist, this must be a new filesystem.
620 static void osd_ost_seq_init(const struct lu_env *env, struct osd_device *osd)
622 struct osd_seq_list *osl = &osd->od_seq_list;
624 CFS_INIT_LIST_HEAD(&osl->osl_seq_list);
625 rwlock_init(&osl->osl_seq_list_lock);
626 sema_init(&osl->osl_seq_init_sem, 1);
629 static void osd_ost_seq_fini(const struct lu_env *env, struct osd_device *osd)
631 struct osd_seq_list *osl = &osd->od_seq_list;
632 struct osd_seq *osd_seq, *tmp;
634 write_lock(&osl->osl_seq_list_lock);
635 cfs_list_for_each_entry_safe(osd_seq, tmp, &osl->osl_seq_list,
637 cfs_list_del(&osd_seq->os_seq_list);
638 OBD_FREE(osd_seq->os_compat_dirs,
639 sizeof(uint64_t) * osd_seq->os_subdir_count);
640 OBD_FREE(osd_seq, sizeof(*osd_seq));
642 write_unlock(&osl->osl_seq_list_lock);
648 * Create /O subdirectory to map legacy OST objects for compatibility.
651 osd_oi_init_compat(const struct lu_env *env, struct osd_device *o)
657 rc = osd_oi_find_or_create(env, o, o->od_root, "O", &sdb);
661 osd_ost_seq_init(env, o);
662 /* Create on-disk indexes to maintain per-UID/GID inode usage.
663 * Those new indexes are created in the top-level ZAP outside the
664 * namespace in order not to confuse ZPL which might interpret those
665 * indexes as directories and assume the values are object IDs */
666 rc = osd_oi_find_or_create(env, o, MASTER_NODE_OBJ,
667 oid2name(ACCT_USER_OID), &odb);
670 o->od_iusr_oid = odb;
672 rc = osd_oi_find_or_create(env, o, MASTER_NODE_OBJ,
673 oid2name(ACCT_GROUP_OID), &odb);
676 o->od_igrp_oid = odb;
681 static char *root2convert = "ROOT";
683 * due to DNE requirements we have to change sequence of /ROOT object
684 * so that it doesn't belong to the local sequence FID_SEQ_LOCAL_FILE
685 * but a normal sequence living on MDS#0
686 * this is the sole purpose of this function.
688 * This is only needed for pre-production 2.4 ZFS filesystems, and
689 * can be removed in the future.
691 int osd_convert_root_to_new_seq(const struct lu_env *env,
692 struct osd_device *o)
694 struct luz_direntry *lze = &osd_oti_get(env)->oti_zde;
695 char *buf = osd_oti_get(env)->oti_str;
696 struct lu_fid newfid;
703 if (strstr(o->od_svname, "MDT") == NULL)
707 rc = -zap_lookup(o->od_objset.os, o->od_root, root2convert, 8,
708 sizeof(*lze) / 8, (void *)lze);
709 /* doesn't exist or let actual user to handle the error */
713 CDEBUG(D_OTHER, "%s: /ROOT -> "DFID" -> "LPU64"\n", o->od_svname,
714 PFID(&lze->lzd_fid), (long long int) lze->lzd_reg.zde_dnode);
716 /* already right one? */
717 if (fid_seq(&lze->lzd_fid) == FID_SEQ_ROOT)
720 tx = dmu_tx_create(o->od_objset.os);
724 dmu_tx_hold_bonus(tx, o->od_root);
726 /* declare delete/insert of the name */
727 dmu_tx_hold_zap(tx, o->od_root, TRUE, root2convert);
728 dmu_tx_hold_zap(tx, o->od_root, FALSE, root2convert);
730 /* declare that we'll remove object from fid-dnode mapping */
731 zapid = osd_get_name_n_idx(env, o, &lze->lzd_fid, buf);
732 dmu_tx_hold_bonus(tx, zapid);
733 dmu_tx_hold_zap(tx, zapid, FALSE, buf);
735 /* declare that we'll add object to fid-dnode mapping */
736 newfid.f_seq = FID_SEQ_ROOT;
739 zapid = osd_get_name_n_idx(env, o, &newfid, buf);
740 dmu_tx_hold_bonus(tx, zapid);
741 dmu_tx_hold_zap(tx, zapid, TRUE, buf);
743 rc = -dmu_tx_assign(tx, TXG_WAIT);
747 rc = -zap_remove(o->od_objset.os, o->od_root, root2convert, tx);
752 zapid = osd_get_name_n_idx(env, o, &lze->lzd_fid, buf);
753 rc = -zap_remove(o->od_objset.os, zapid, buf, tx);
757 lze->lzd_fid = newfid;
758 rc = -zap_add(o->od_objset.os, o->od_root, root2convert,
759 8, sizeof(*lze) / 8, (void *)lze, tx);
763 /* add to OI with the new fid */
764 zapid = osd_get_name_n_idx(env, o, &newfid, buf);
765 rc = -zap_add(o->od_objset.os, zapid, buf, 8, 1, &lze->lzd_reg, tx);
770 /* LMA will be updated in mdd_compat_fixes */
778 CERROR("%s: can't convert to new fid: rc = %d\n", o->od_svname, rc);
783 * Initialize the OIs by either opening or creating them as needed.
785 int osd_oi_init(const struct lu_env *env, struct osd_device *o)
787 char *key = osd_oti_get(env)->oti_buf;
788 int i, rc, count = 0;
791 rc = osd_oi_probe(env, o, &count);
798 count = osd_oi_count;
801 for (i = 0; i < count; i++) {
802 sprintf(key, "%s.%d", DMU_OSD_OI_NAME_BASE, i);
803 rc = osd_oi_find_or_create(env, o, odb, key, &sdb);
809 rc = osd_oi_init_compat(env, o);
813 LASSERT((count & (count - 1)) == 0);
814 o->od_oi_count = count;
815 OBD_ALLOC(o->od_oi_table, sizeof(struct osd_oi *) * count);
816 if (o->od_oi_table == NULL)
819 rc = osd_oi_open_table(env, o, count);
821 OBD_FREE(o->od_oi_table, sizeof(struct osd_oi *) * count);
822 o->od_oi_table = NULL;
828 void osd_oi_fini(const struct lu_env *env, struct osd_device *o)
832 osd_ost_seq_fini(env, o);
834 if (o->od_oi_table != NULL) {
835 (void) osd_oi_close_table(env, o);
836 OBD_FREE(o->od_oi_table,
837 sizeof(struct osd_oi *) * o->od_oi_count);
838 o->od_oi_table = NULL;
845 int osd_options_init(void)
847 /* osd_oi_count - Default number of OIs, 128 works well for ZFS */
848 if (osd_oi_count == 0 || osd_oi_count > OSD_OI_FID_NR_MAX)
849 osd_oi_count = OSD_OI_FID_NR;
851 if ((osd_oi_count & (osd_oi_count - 1)) != 0) {
852 LCONSOLE_WARN("Round up osd_oi_count %d to power2 %d\n",
853 osd_oi_count, size_roundup_power2(osd_oi_count));
854 osd_oi_count = size_roundup_power2(osd_oi_count);