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, 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 { MDD_CAPA_KEYS_OID, NULL /*CAPA_KEYS*/ },
100 { FLD_INDEX_OID, "fld" },
101 { MDD_LOV_OBJ_OID, LOV_OBJID },
102 { OFD_HEALTH_CHECK_OID, HEALTH_CHECK },
103 { ACCT_USER_OID, "acct_usr_inode" },
104 { ACCT_GROUP_OID, "acct_grp_inode" },
105 { MDD_ORPHAN_OID, NULL },
109 static char *oid2name(const unsigned long oid)
113 while (oids[i].oid) {
114 if (oids[i].oid == oid)
122 * Lookup an existing OI by the given name.
125 osd_oi_lookup(const struct lu_env *env, struct osd_device *o,
126 uint64_t parent, const char *name, struct osd_oi *oi)
128 struct zpl_direntry *zde = &osd_oti_get(env)->oti_zde.lzd_reg;
131 rc = -zap_lookup(o->od_objset.os, parent, name, 8, 1, (void *)zde);
135 strncpy(oi->oi_name, name, OSD_OI_NAME_SIZE - 1);
136 oi->oi_zapid = zde->zde_dnode;
142 * Create a new OI with the given name.
145 osd_oi_create(const struct lu_env *env, struct osd_device *o,
146 uint64_t parent, const char *name, uint64_t *child)
148 struct zpl_direntry *zde = &osd_oti_get(env)->oti_zde.lzd_reg;
149 struct lu_attr *la = &osd_oti_get(env)->oti_la;
154 /* verify it doesn't already exist */
155 rc = -zap_lookup(o->od_objset.os, parent, name, 8, 1, (void *)zde);
159 /* create fid-to-dnode index */
160 tx = dmu_tx_create(o->od_objset.os);
164 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, 1, NULL);
165 dmu_tx_hold_bonus(tx, parent);
166 dmu_tx_hold_zap(tx, parent, TRUE, name);
167 LASSERT(tx->tx_objset->os_sa);
168 dmu_tx_hold_sa_create(tx, ZFS_SA_BASE_ATTR_SIZE);
170 rc = -dmu_tx_assign(tx, TXG_WAIT);
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 __osd_zap_create(env, &o->od_objset, &db, tx, la, parent, oi_tag, 0);
181 zde->zde_dnode = db->db_object;
183 zde->zde_type = IFTODT(S_IFDIR);
185 rc = -zap_add(o->od_objset.os, parent, name, 8, 1, (void *)zde, tx);
189 *child = db->db_object;
190 sa_buf_rele(db, oi_tag);
196 osd_oi_find_or_create(const struct lu_env *env, struct osd_device *o,
197 uint64_t parent, const char *name, uint64_t *child)
202 rc = osd_oi_lookup(env, o, parent, name, &oi);
204 *child = oi.oi_zapid;
205 else if (rc == -ENOENT)
206 rc = osd_oi_create(env, o, parent, name, child);
212 * Lookup the target index/flags of the fid, so it will know where
213 * the object is located (tgt index) and it is MDT or OST object.
215 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
216 const struct lu_fid *fid, struct lu_seq_range *range)
218 struct seq_server_site *ss = osd_seq_site(osd);
221 if (fid_is_idif(fid)) {
222 fld_range_set_ost(range);
223 range->lsr_index = fid_idif_ost_idx(fid);
227 if (!fid_seq_in_fldb(fid_seq(fid))) {
228 fld_range_set_mdt(range);
230 /* FIXME: If ss is NULL, it suppose not get lsr_index
232 range->lsr_index = ss->ss_node_id;
237 fld_range_set_any(range);
238 rc = fld_server_lookup(env, ss->ss_server_fld, fid_seq(fid), range);
240 CERROR("%s: cannot find FLD range for "DFID": rc = %d\n",
241 osd_name(osd), PFID(fid), rc);
245 int fid_is_on_ost(const struct lu_env *env, struct osd_device *osd,
246 const struct lu_fid *fid)
248 struct lu_seq_range *range = &osd_oti_get(env)->oti_seq_range;
252 if (fid_is_idif(fid))
255 rc = osd_fld_lookup(env, osd, fid, range);
257 CERROR("%s: Can not lookup fld for "DFID"\n",
258 osd_name(osd), PFID(fid));
262 CDEBUG(D_INFO, "fid "DFID" range "DRANGE"\n", PFID(fid),
265 if (fld_range_is_ost(range))
271 static struct osd_seq *osd_seq_find_locked(struct osd_seq_list *seq_list,
274 struct osd_seq *osd_seq;
276 cfs_list_for_each_entry(osd_seq, &seq_list->osl_seq_list, os_seq_list) {
277 if (osd_seq->os_seq == seq)
283 static struct osd_seq *osd_seq_find(struct osd_seq_list *seq_list,
286 struct osd_seq *osd_seq;
288 read_lock(&seq_list->osl_seq_list_lock);
289 osd_seq = osd_seq_find_locked(seq_list, seq);
290 read_unlock(&seq_list->osl_seq_list_lock);
295 static struct osd_seq *osd_find_or_add_seq(const struct lu_env *env,
296 struct osd_device *osd, obd_seq seq)
298 struct osd_seq_list *seq_list = &osd->od_seq_list;
299 struct osd_seq *osd_seq;
300 char *key = osd_oti_get(env)->oti_buf;
301 char *seq_name = osd_oti_get(env)->oti_str;
308 osd_seq = osd_seq_find(seq_list, seq);
312 down(&seq_list->osl_seq_init_sem);
313 /* Check again, in case some one else already add it
315 osd_seq = osd_seq_find(seq_list, seq);
319 OBD_ALLOC_PTR(osd_seq);
321 GOTO(out, rc = -ENOMEM);
323 CFS_INIT_LIST_HEAD(&osd_seq->os_seq_list);
324 osd_seq->os_seq = seq;
326 /* Init subdir count to be 32, but each seq can have
327 * different subdir count */
328 osd_seq->os_subdir_count = OSD_OST_MAP_SIZE;
329 OBD_ALLOC(osd_seq->os_compat_dirs,
330 sizeof(uint64_t) * osd_seq->os_subdir_count);
331 if (osd_seq->os_compat_dirs == NULL)
332 GOTO(out, rc = -ENOMEM);
334 rc = osd_oi_lookup(env, osd, osd->od_root, "O", &oi);
336 CERROR("%s: Can not find O: rc = %d\n", osd_name(osd), rc);
340 sprintf(seq_name, (fid_seq_is_rsvd(seq) ||
341 fid_seq_is_mdt0(seq)) ? LPU64 : LPX64i,
342 fid_seq_is_idif(seq) ? 0 : seq);
344 rc = osd_oi_find_or_create(env, osd, oi.oi_zapid, seq_name, &odb);
346 CERROR("%s: Can not create %s : rc = %d\n",
347 osd_name(osd), seq_name, rc);
351 for (i = 0; i < OSD_OST_MAP_SIZE; i++) {
352 sprintf(key, "d%d", i);
353 rc = osd_oi_find_or_create(env, osd, odb, key, &sdb);
355 GOTO(out, osd_seq = ERR_PTR(rc));
356 osd_seq->os_compat_dirs[i] = sdb;
359 write_lock(&seq_list->osl_seq_list_lock);
360 cfs_list_add(&osd_seq->os_seq_list, &seq_list->osl_seq_list);
361 write_unlock(&seq_list->osl_seq_list_lock);
363 up(&seq_list->osl_seq_init_sem);
365 if (osd_seq != NULL && osd_seq->os_compat_dirs != NULL)
366 OBD_FREE(osd_seq->os_compat_dirs,
367 sizeof(uint64_t) * osd_seq->os_subdir_count);
369 OBD_FREE_PTR(osd_seq);
370 osd_seq = ERR_PTR(rc);
376 * objects w/o a natural reference (unlike a file on a MDS)
377 * are put under a special hierarchy /O/<seq>/d0..dXX
378 * this function returns a directory specific fid belongs to
381 osd_get_idx_for_ost_obj(const struct lu_env *env, struct osd_device *osd,
382 const struct lu_fid *fid, char *buf)
384 struct osd_seq *osd_seq;
388 osd_seq = osd_find_or_add_seq(env, osd, fid_seq(fid));
389 if (IS_ERR(osd_seq)) {
390 CERROR("%s: Can not find seq group "DFID"\n", osd_name(osd),
392 return PTR_ERR(osd_seq);
395 rc = fid_to_ostid(fid, &osd_oti_get(env)->oti_ostid);
396 LASSERT(rc == 0); /* we should not get here with IGIF */
397 b = ostid_id(&osd_oti_get(env)->oti_ostid) % OSD_OST_MAP_SIZE;
398 LASSERT(osd_seq->os_compat_dirs[b]);
400 sprintf(buf, LPU64, ostid_id(&osd_oti_get(env)->oti_ostid));
402 return osd_seq->os_compat_dirs[b];
405 /* XXX: f_ver is not counted, but may differ too */
406 static void osd_fid2str(char *buf, const struct lu_fid *fid)
408 sprintf(buf, DFID_NOBRACE, PFID(fid));
412 * Determine the zap object id which is being used as the OI for the
413 * given fid. The lowest N bits in the sequence ID are used as the
414 * index key. On failure 0 is returned which zfs treats internally
415 * as an invalid object id.
418 osd_get_idx_for_fid(struct osd_device *osd, const struct lu_fid *fid,
423 LASSERT(osd->od_oi_table != NULL);
424 oi = osd->od_oi_table[fid_seq(fid) & (osd->od_oi_count - 1)];
425 osd_fid2str(buf, fid);
430 uint64_t osd_get_name_n_idx(const struct lu_env *env, struct osd_device *osd,
431 const struct lu_fid *fid, char *buf)
438 if (fid_is_on_ost(env, osd, fid) == 1 || fid_seq(fid) == FID_SEQ_ECHO) {
439 zapid = osd_get_idx_for_ost_obj(env, osd, fid, buf);
440 } else if (unlikely(fid_seq(fid) == FID_SEQ_LOCAL_FILE)) {
441 /* special objects with fixed known fids get their name */
442 char *name = oid2name(fid_oid(fid));
445 zapid = osd->od_root;
447 if (fid_is_acct(fid))
448 zapid = MASTER_NODE_OBJ;
450 zapid = osd_get_idx_for_fid(osd, fid, buf);
453 zapid = osd_get_idx_for_fid(osd, fid, buf);
459 static inline int fid_is_fs_root(const struct lu_fid *fid)
461 /* Map root inode to special local object FID */
462 return fid_seq(fid) == FID_SEQ_LOCAL_FILE &&
463 fid_oid(fid) == OSD_FS_ROOT_OID;
466 int osd_fid_lookup(const struct lu_env *env, struct osd_device *dev,
467 const struct lu_fid *fid, uint64_t *oid)
469 struct osd_thread_info *info = osd_oti_get(env);
470 char *buf = info->oti_buf;
475 if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOENT))
478 if (unlikely(fid_is_acct(fid))) {
479 if (fid_oid(fid) == ACCT_USER_OID)
480 *oid = dev->od_iusr_oid;
482 *oid = dev->od_igrp_oid;
483 } else if (unlikely(fid_is_fs_root(fid))) {
486 zapid = osd_get_name_n_idx(env, dev, fid, buf);
488 rc = -zap_lookup(dev->od_objset.os, zapid, buf,
489 8, 1, &info->oti_zde);
492 *oid = info->oti_zde.lzd_reg.zde_dnode;
499 * Close an entry in a specific slot.
502 osd_oi_remove_table(const struct lu_env *env, struct osd_device *o, int key)
506 LASSERT(key < o->od_oi_count);
508 oi = o->od_oi_table[key];
511 o->od_oi_table[key] = NULL;
516 * Allocate and open a new entry in the specified unused slot.
519 osd_oi_add_table(const struct lu_env *env, struct osd_device *o,
525 LASSERT(key < o->od_oi_count);
526 LASSERT(o->od_oi_table[key] == NULL);
532 rc = osd_oi_lookup(env, o, o->od_root, name, oi);
538 o->od_oi_table[key] = oi;
544 * Depopulate the OI table.
547 osd_oi_close_table(const struct lu_env *env, struct osd_device *o)
551 for (i = 0; i < o->od_oi_count; i++)
552 osd_oi_remove_table(env, o, i);
556 * Populate the OI table based.
559 osd_oi_open_table(const struct lu_env *env, struct osd_device *o, int count)
565 for (i = 0; i < count; i++) {
566 sprintf(name, "%s.%d", DMU_OSD_OI_NAME_BASE, i);
567 rc = osd_oi_add_table(env, o, name, i);
569 osd_oi_close_table(env, o);
578 * Determine if the type and number of OIs used by this file system.
581 osd_oi_probe(const struct lu_env *env, struct osd_device *o, int *count)
583 uint64_t root_oid = o->od_root;
590 * Check for multiple OIs and determine the count. There is no
591 * gap handling, if an OI is missing the wrong size can be returned.
592 * The only safeguard is that we know the number of OIs must be a
593 * power of two and this is checked for basic sanity.
595 for (*count = 0; *count < OSD_OI_FID_NR_MAX; (*count)++) {
596 sprintf(name, "%s.%d", DMU_OSD_OI_NAME_BASE, *count);
597 rc = osd_oi_lookup(env, o, root_oid, name, &oi);
605 if ((*count & (*count - 1)) != 0)
615 * No OIs exist, this must be a new filesystem.
622 static void osd_ost_seq_init(const struct lu_env *env, struct osd_device *osd)
624 struct osd_seq_list *osl = &osd->od_seq_list;
626 CFS_INIT_LIST_HEAD(&osl->osl_seq_list);
627 rwlock_init(&osl->osl_seq_list_lock);
628 sema_init(&osl->osl_seq_init_sem, 1);
631 static void osd_ost_seq_fini(const struct lu_env *env, struct osd_device *osd)
633 struct osd_seq_list *osl = &osd->od_seq_list;
634 struct osd_seq *osd_seq, *tmp;
636 write_lock(&osl->osl_seq_list_lock);
637 cfs_list_for_each_entry_safe(osd_seq, tmp, &osl->osl_seq_list,
639 cfs_list_del(&osd_seq->os_seq_list);
640 OBD_FREE(osd_seq->os_compat_dirs,
641 sizeof(uint64_t) * osd_seq->os_subdir_count);
642 OBD_FREE(osd_seq, sizeof(*osd_seq));
644 write_unlock(&osl->osl_seq_list_lock);
650 * Create /O subdirectory to map legacy OST objects for compatibility.
653 osd_oi_init_compat(const struct lu_env *env, struct osd_device *o)
659 rc = osd_oi_find_or_create(env, o, o->od_root, "O", &sdb);
663 osd_ost_seq_init(env, o);
664 /* Create on-disk indexes to maintain per-UID/GID inode usage.
665 * Those new indexes are created in the top-level ZAP outside the
666 * namespace in order not to confuse ZPL which might interpret those
667 * indexes as directories and assume the values are object IDs */
668 rc = osd_oi_find_or_create(env, o, MASTER_NODE_OBJ,
669 oid2name(ACCT_USER_OID), &odb);
672 o->od_iusr_oid = odb;
674 rc = osd_oi_find_or_create(env, o, MASTER_NODE_OBJ,
675 oid2name(ACCT_GROUP_OID), &odb);
678 o->od_igrp_oid = odb;
683 static char *root2convert = "ROOT";
685 * due to DNE requirements we have to change sequence of /ROOT object
686 * so that it doesn't belong to the local sequence FID_SEQ_LOCAL_FILE
687 * but a normal sequence living on MDS#0
688 * this is the sole purpose of this function.
690 * This is only needed for pre-production 2.4 ZFS filesystems, and
691 * can be removed in the future.
693 int osd_convert_root_to_new_seq(const struct lu_env *env,
694 struct osd_device *o)
696 struct luz_direntry *lze = &osd_oti_get(env)->oti_zde;
697 char *buf = osd_oti_get(env)->oti_str;
698 struct lu_fid newfid;
705 if (strstr(o->od_svname, "MDT") == NULL)
709 rc = -zap_lookup(o->od_objset.os, o->od_root, root2convert, 8,
710 sizeof(*lze) / 8, (void *)lze);
711 /* doesn't exist or let actual user to handle the error */
715 CDEBUG(D_OTHER, "%s: /ROOT -> "DFID" -> "LPU64"\n", o->od_svname,
716 PFID(&lze->lzd_fid), (long long int) lze->lzd_reg.zde_dnode);
718 /* already right one? */
719 if (fid_seq(&lze->lzd_fid) == FID_SEQ_ROOT)
722 tx = dmu_tx_create(o->od_objset.os);
726 dmu_tx_hold_bonus(tx, o->od_root);
728 /* declare delete/insert of the name */
729 dmu_tx_hold_zap(tx, o->od_root, TRUE, root2convert);
730 dmu_tx_hold_zap(tx, o->od_root, FALSE, root2convert);
732 /* declare that we'll remove object from fid-dnode mapping */
733 zapid = osd_get_name_n_idx(env, o, &lze->lzd_fid, buf);
734 dmu_tx_hold_bonus(tx, zapid);
735 dmu_tx_hold_zap(tx, zapid, FALSE, buf);
737 /* declare that we'll add object to fid-dnode mapping */
738 newfid.f_seq = FID_SEQ_ROOT;
741 zapid = osd_get_name_n_idx(env, o, &newfid, buf);
742 dmu_tx_hold_bonus(tx, zapid);
743 dmu_tx_hold_zap(tx, zapid, TRUE, buf);
745 rc = -dmu_tx_assign(tx, TXG_WAIT);
749 rc = -zap_remove(o->od_objset.os, o->od_root, root2convert, tx);
754 zapid = osd_get_name_n_idx(env, o, &lze->lzd_fid, buf);
755 rc = -zap_remove(o->od_objset.os, zapid, buf, tx);
759 lze->lzd_fid = newfid;
760 rc = -zap_add(o->od_objset.os, o->od_root, root2convert,
761 8, sizeof(*lze) / 8, (void *)lze, tx);
765 /* add to OI with the new fid */
766 zapid = osd_get_name_n_idx(env, o, &newfid, buf);
767 rc = -zap_add(o->od_objset.os, zapid, buf, 8, 1, &lze->lzd_reg, tx);
772 /* LMA will be updated in mdd_compat_fixes */
780 CERROR("%s: can't convert to new fid: rc = %d\n", o->od_svname, rc);
785 * Initialize the OIs by either opening or creating them as needed.
787 int osd_oi_init(const struct lu_env *env, struct osd_device *o)
789 char *key = osd_oti_get(env)->oti_buf;
790 int i, rc, count = 0;
793 rc = osd_oi_probe(env, o, &count);
800 count = osd_oi_count;
803 for (i = 0; i < count; i++) {
804 sprintf(key, "%s.%d", DMU_OSD_OI_NAME_BASE, i);
805 rc = osd_oi_find_or_create(env, o, odb, key, &sdb);
811 rc = osd_oi_init_compat(env, o);
815 LASSERT((count & (count - 1)) == 0);
816 o->od_oi_count = count;
817 OBD_ALLOC(o->od_oi_table, sizeof(struct osd_oi *) * count);
818 if (o->od_oi_table == NULL)
821 rc = osd_oi_open_table(env, o, count);
823 OBD_FREE(o->od_oi_table, sizeof(struct osd_oi *) * count);
824 o->od_oi_table = NULL;
830 void osd_oi_fini(const struct lu_env *env, struct osd_device *o)
834 osd_ost_seq_fini(env, o);
836 if (o->od_oi_table != NULL) {
837 (void) osd_oi_close_table(env, o);
838 OBD_FREE(o->od_oi_table,
839 sizeof(struct osd_oi *) * o->od_oi_count);
840 o->od_oi_table = NULL;
847 int osd_options_init(void)
849 /* osd_oi_count - Default number of OIs, 128 works well for ZFS */
850 if (osd_oi_count == 0 || osd_oi_count > OSD_OI_FID_NR_MAX)
851 osd_oi_count = OSD_OI_FID_NR;
853 if ((osd_oi_count & (osd_oi_count - 1)) != 0) {
854 LCONSOLE_WARN("Round up osd_oi_count %d to power2 %d\n",
855 osd_oi_count, size_roundup_power2(osd_oi_count));
856 osd_oi_count = size_roundup_power2(osd_oi_count);