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_root(fid)) {
222 range->lsr_flags = LU_SEQ_RANGE_MDT;
223 range->lsr_index = 0;
227 if (fid_is_idif(fid)) {
228 range->lsr_flags = LU_SEQ_RANGE_OST;
229 range->lsr_index = fid_idif_ost_idx(fid);
233 if (!fid_is_norm(fid)) {
234 range->lsr_flags = LU_SEQ_RANGE_MDT;
235 /* If ss is NULL, it suppose not get lsr_index at all */
237 range->lsr_index = ss->ss_node_id;
242 range->lsr_flags = -1;
243 rc = fld_server_lookup(env, ss->ss_server_fld, fid_seq(fid), range);
245 CERROR("%s can not find "DFID": rc = %d\n",
246 osd_name(osd), PFID(fid), rc);
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 rc = osd_fld_lookup(env, osd, fid, range);
262 CERROR("%s: Can not lookup fld for "DFID"\n",
263 osd2lu_dev(osd)->ld_obd->obd_name, PFID(fid));
267 CDEBUG(D_INFO, "fid "DFID" range "DRANGE"\n", PFID(fid),
270 if (range->lsr_flags == LU_SEQ_RANGE_OST)
276 static struct osd_seq *osd_seq_find_locked(struct osd_seq_list *seq_list,
279 struct osd_seq *osd_seq;
281 cfs_list_for_each_entry(osd_seq, &seq_list->osl_seq_list, os_seq_list) {
282 if (osd_seq->os_seq == seq)
288 static struct osd_seq *osd_seq_find(struct osd_seq_list *seq_list,
291 struct osd_seq *osd_seq;
293 read_lock(&seq_list->osl_seq_list_lock);
294 osd_seq = osd_seq_find_locked(seq_list, seq);
295 read_unlock(&seq_list->osl_seq_list_lock);
300 static struct osd_seq *osd_find_or_add_seq(const struct lu_env *env,
301 struct osd_device *osd, obd_seq seq)
303 struct osd_seq_list *seq_list = &osd->od_seq_list;
304 struct osd_seq *osd_seq;
305 char *key = osd_oti_get(env)->oti_buf;
306 char *seq_name = osd_oti_get(env)->oti_str;
313 osd_seq = osd_seq_find(seq_list, seq);
317 down(&seq_list->osl_seq_init_sem);
318 /* Check again, in case some one else already add it
320 osd_seq = osd_seq_find(seq_list, seq);
324 OBD_ALLOC_PTR(osd_seq);
326 GOTO(out, rc = -ENOMEM);
328 CFS_INIT_LIST_HEAD(&osd_seq->os_seq_list);
329 osd_seq->os_seq = seq;
331 /* Init subdir count to be 32, but each seq can have
332 * different subdir count */
333 osd_seq->os_subdir_count = OSD_OST_MAP_SIZE;
334 OBD_ALLOC(osd_seq->os_compat_dirs,
335 sizeof(uint64_t) * osd_seq->os_subdir_count);
336 if (osd_seq->os_compat_dirs == NULL)
337 GOTO(out, rc = -ENOMEM);
339 rc = osd_oi_lookup(env, osd, osd->od_root, "O", &oi);
341 CERROR("%s: Can not find O: rc = %d\n", osd_name(osd), rc);
345 sprintf(seq_name, (fid_seq_is_rsvd(seq) ||
346 fid_seq_is_mdt0(seq)) ? LPU64 : LPX64i,
347 fid_seq_is_idif(seq) ? 0 : seq);
349 rc = osd_oi_find_or_create(env, osd, oi.oi_zapid, seq_name, &odb);
351 CERROR("%s: Can not create %s : rc = %d\n",
352 osd_name(osd), seq_name, rc);
356 for (i = 0; i < OSD_OST_MAP_SIZE; i++) {
357 sprintf(key, "d%d", i);
358 rc = osd_oi_find_or_create(env, osd, odb, key, &sdb);
360 GOTO(out, osd_seq = ERR_PTR(rc));
361 osd_seq->os_compat_dirs[i] = sdb;
364 write_lock(&seq_list->osl_seq_list_lock);
365 cfs_list_add(&osd_seq->os_seq_list, &seq_list->osl_seq_list);
366 write_unlock(&seq_list->osl_seq_list_lock);
368 up(&seq_list->osl_seq_init_sem);
370 if (osd_seq != NULL && osd_seq->os_compat_dirs != NULL)
371 OBD_FREE(osd_seq->os_compat_dirs,
372 sizeof(uint64_t) * osd_seq->os_subdir_count);
374 OBD_FREE_PTR(osd_seq);
375 osd_seq = ERR_PTR(rc);
381 * objects w/o a natural reference (unlike a file on a MDS)
382 * are put under a special hierarchy /O/<seq>/d0..dXX
383 * this function returns a directory specific fid belongs to
386 osd_get_idx_for_ost_obj(const struct lu_env *env, struct osd_device *osd,
387 const struct lu_fid *fid, char *buf)
389 struct osd_seq *osd_seq;
393 osd_seq = osd_find_or_add_seq(env, osd, fid_seq(fid));
394 if (IS_ERR(osd_seq)) {
395 CERROR("%s: Can not find seq group "DFID"\n", osd_name(osd),
397 return PTR_ERR(osd_seq);
400 rc = fid_to_ostid(fid, &osd_oti_get(env)->oti_ostid);
401 LASSERT(rc == 0); /* we should not get here with IGIF */
402 b = ostid_id(&osd_oti_get(env)->oti_ostid) % OSD_OST_MAP_SIZE;
403 LASSERT(osd_seq->os_compat_dirs[b]);
405 sprintf(buf, LPU64, ostid_id(&osd_oti_get(env)->oti_ostid));
407 return osd_seq->os_compat_dirs[b];
410 /* XXX: f_ver is not counted, but may differ too */
411 static void osd_fid2str(char *buf, const struct lu_fid *fid)
413 sprintf(buf, DFID_NOBRACE, PFID(fid));
417 * Determine the zap object id which is being used as the OI for the
418 * given fid. The lowest N bits in the sequence ID are used as the
419 * index key. On failure 0 is returned which zfs treats internally
420 * as an invalid object id.
423 osd_get_idx_for_fid(struct osd_device *osd, const struct lu_fid *fid,
428 LASSERT(osd->od_oi_table != NULL);
429 oi = osd->od_oi_table[fid_seq(fid) & (osd->od_oi_count - 1)];
430 osd_fid2str(buf, fid);
435 uint64_t osd_get_name_n_idx(const struct lu_env *env, struct osd_device *osd,
436 const struct lu_fid *fid, char *buf)
443 if (fid_is_on_ost(env, osd, fid) == 1 || fid_seq(fid) == FID_SEQ_ECHO) {
444 zapid = osd_get_idx_for_ost_obj(env, osd, fid, buf);
445 } else if (unlikely(fid_seq(fid) == FID_SEQ_LOCAL_FILE)) {
446 /* special objects with fixed known fids get their name */
447 char *name = oid2name(fid_oid(fid));
450 zapid = osd->od_root;
452 if (fid_is_acct(fid))
453 zapid = MASTER_NODE_OBJ;
455 zapid = osd_get_idx_for_fid(osd, fid, buf);
458 zapid = osd_get_idx_for_fid(osd, fid, buf);
464 static inline int fid_is_fs_root(const struct lu_fid *fid)
466 /* Map root inode to special local object FID */
467 return fid_seq(fid) == FID_SEQ_LOCAL_FILE &&
468 fid_oid(fid) == OSD_FS_ROOT_OID;
471 int osd_fid_lookup(const struct lu_env *env, struct osd_device *dev,
472 const struct lu_fid *fid, uint64_t *oid)
474 struct osd_thread_info *info = osd_oti_get(env);
475 char *buf = info->oti_buf;
480 if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOENT))
483 if (unlikely(fid_is_acct(fid))) {
484 if (fid_oid(fid) == ACCT_USER_OID)
485 *oid = dev->od_iusr_oid;
487 *oid = dev->od_igrp_oid;
488 } else if (unlikely(fid_is_fs_root(fid))) {
491 zapid = osd_get_name_n_idx(env, dev, fid, buf);
493 rc = -zap_lookup(dev->od_objset.os, zapid, buf,
494 8, 1, &info->oti_zde);
497 *oid = info->oti_zde.lzd_reg.zde_dnode;
504 * Close an entry in a specific slot.
507 osd_oi_remove_table(const struct lu_env *env, struct osd_device *o, int key)
511 LASSERT(key < o->od_oi_count);
513 oi = o->od_oi_table[key];
516 o->od_oi_table[key] = NULL;
521 * Allocate and open a new entry in the specified unused slot.
524 osd_oi_add_table(const struct lu_env *env, struct osd_device *o,
530 LASSERT(key < o->od_oi_count);
531 LASSERT(o->od_oi_table[key] == NULL);
537 rc = osd_oi_lookup(env, o, o->od_root, name, oi);
543 o->od_oi_table[key] = oi;
549 * Depopulate the OI table.
552 osd_oi_close_table(const struct lu_env *env, struct osd_device *o)
556 for (i = 0; i < o->od_oi_count; i++)
557 osd_oi_remove_table(env, o, i);
561 * Populate the OI table based.
564 osd_oi_open_table(const struct lu_env *env, struct osd_device *o, int count)
570 for (i = 0; i < count; i++) {
571 sprintf(name, "%s.%d", DMU_OSD_OI_NAME_BASE, i);
572 rc = osd_oi_add_table(env, o, name, i);
574 osd_oi_close_table(env, o);
583 * Determine if the type and number of OIs used by this file system.
586 osd_oi_probe(const struct lu_env *env, struct osd_device *o, int *count)
588 uint64_t root_oid = o->od_root;
595 * Check for multiple OIs and determine the count. There is no
596 * gap handling, if an OI is missing the wrong size can be returned.
597 * The only safeguard is that we know the number of OIs must be a
598 * power of two and this is checked for basic sanity.
600 for (*count = 0; *count < OSD_OI_FID_NR_MAX; (*count)++) {
601 sprintf(name, "%s.%d", DMU_OSD_OI_NAME_BASE, *count);
602 rc = osd_oi_lookup(env, o, root_oid, name, &oi);
610 if ((*count & (*count - 1)) != 0)
620 * No OIs exist, this must be a new filesystem.
627 static void osd_ost_seq_init(const struct lu_env *env, struct osd_device *osd)
629 struct osd_seq_list *osl = &osd->od_seq_list;
631 CFS_INIT_LIST_HEAD(&osl->osl_seq_list);
632 rwlock_init(&osl->osl_seq_list_lock);
633 sema_init(&osl->osl_seq_init_sem, 1);
636 static void osd_ost_seq_fini(const struct lu_env *env, struct osd_device *osd)
638 struct osd_seq_list *osl = &osd->od_seq_list;
639 struct osd_seq *osd_seq, *tmp;
641 write_lock(&osl->osl_seq_list_lock);
642 cfs_list_for_each_entry_safe(osd_seq, tmp, &osl->osl_seq_list,
644 cfs_list_del(&osd_seq->os_seq_list);
645 OBD_FREE(osd_seq->os_compat_dirs,
646 sizeof(uint64_t) * osd_seq->os_subdir_count);
647 OBD_FREE(osd_seq, sizeof(*osd_seq));
649 write_unlock(&osl->osl_seq_list_lock);
655 * Create /O subdirectory to map legacy OST objects for compatibility.
658 osd_oi_init_compat(const struct lu_env *env, struct osd_device *o)
664 rc = osd_oi_find_or_create(env, o, o->od_root, "O", &sdb);
668 osd_ost_seq_init(env, o);
669 /* Create on-disk indexes to maintain per-UID/GID inode usage.
670 * Those new indexes are created in the top-level ZAP outside the
671 * namespace in order not to confuse ZPL which might interpret those
672 * indexes as directories and assume the values are object IDs */
673 rc = osd_oi_find_or_create(env, o, MASTER_NODE_OBJ,
674 oid2name(ACCT_USER_OID), &odb);
677 o->od_iusr_oid = odb;
679 rc = osd_oi_find_or_create(env, o, MASTER_NODE_OBJ,
680 oid2name(ACCT_GROUP_OID), &odb);
683 o->od_igrp_oid = odb;
688 static char *root2convert = "ROOT";
690 * due to DNE requirements we have to change sequence of /ROOT object
691 * so that it doesn't belong to the local sequence FID_SEQ_LOCAL_FILE
692 * but a normal sequence living on MDS#0
693 * this is the sole purpose of this function.
695 * This is only needed for pre-production 2.4 ZFS filesystems, and
696 * can be removed in the future.
698 int osd_convert_root_to_new_seq(const struct lu_env *env,
699 struct osd_device *o)
701 struct luz_direntry *lze = &osd_oti_get(env)->oti_zde;
702 char *buf = osd_oti_get(env)->oti_str;
703 struct lu_fid newfid;
710 if (strstr(o->od_svname, "MDT") == NULL)
714 rc = -zap_lookup(o->od_objset.os, o->od_root, root2convert, 8,
715 sizeof(*lze) / 8, (void *)lze);
716 /* doesn't exist or let actual user to handle the error */
720 CDEBUG(D_OTHER, "%s: /ROOT -> "DFID" -> "LPU64"\n", o->od_svname,
721 PFID(&lze->lzd_fid), (long long int) lze->lzd_reg.zde_dnode);
723 /* already right one? */
724 if (fid_seq(&lze->lzd_fid) == FID_SEQ_ROOT)
727 tx = dmu_tx_create(o->od_objset.os);
731 dmu_tx_hold_bonus(tx, o->od_root);
733 /* declare delete/insert of the name */
734 dmu_tx_hold_zap(tx, o->od_root, TRUE, root2convert);
735 dmu_tx_hold_zap(tx, o->od_root, FALSE, root2convert);
737 /* declare that we'll remove object from fid-dnode mapping */
738 zapid = osd_get_name_n_idx(env, o, &lze->lzd_fid, buf);
739 dmu_tx_hold_bonus(tx, zapid);
740 dmu_tx_hold_zap(tx, zapid, FALSE, buf);
742 /* declare that we'll add object to fid-dnode mapping */
743 newfid.f_seq = FID_SEQ_ROOT;
746 zapid = osd_get_name_n_idx(env, o, &newfid, buf);
747 dmu_tx_hold_bonus(tx, zapid);
748 dmu_tx_hold_zap(tx, zapid, TRUE, buf);
750 rc = -dmu_tx_assign(tx, TXG_WAIT);
754 rc = -zap_remove(o->od_objset.os, o->od_root, root2convert, tx);
759 zapid = osd_get_name_n_idx(env, o, &lze->lzd_fid, buf);
760 rc = -zap_remove(o->od_objset.os, zapid, buf, tx);
764 lze->lzd_fid = newfid;
765 rc = -zap_add(o->od_objset.os, o->od_root, root2convert,
766 8, sizeof(*lze) / 8, (void *)lze, tx);
770 /* add to OI with the new fid */
771 zapid = osd_get_name_n_idx(env, o, &newfid, buf);
772 rc = -zap_add(o->od_objset.os, zapid, buf, 8, 1, &lze->lzd_reg, tx);
777 /* LMA will be updated in mdd_compat_fixes */
785 CERROR("%s: can't convert to new fid: rc = %d\n", o->od_svname, rc);
790 * Initialize the OIs by either opening or creating them as needed.
792 int osd_oi_init(const struct lu_env *env, struct osd_device *o)
794 char *key = osd_oti_get(env)->oti_buf;
795 int i, rc, count = 0;
798 rc = osd_oi_probe(env, o, &count);
805 count = osd_oi_count;
808 for (i = 0; i < count; i++) {
809 sprintf(key, "%s.%d", DMU_OSD_OI_NAME_BASE, i);
810 rc = osd_oi_find_or_create(env, o, odb, key, &sdb);
816 rc = osd_oi_init_compat(env, o);
820 LASSERT((count & (count - 1)) == 0);
821 o->od_oi_count = count;
822 OBD_ALLOC(o->od_oi_table, sizeof(struct osd_oi *) * count);
823 if (o->od_oi_table == NULL)
826 rc = osd_oi_open_table(env, o, count);
828 OBD_FREE(o->od_oi_table, sizeof(struct osd_oi *) * count);
829 o->od_oi_table = NULL;
835 void osd_oi_fini(const struct lu_env *env, struct osd_device *o)
839 osd_ost_seq_fini(env, o);
841 if (o->od_oi_table != NULL) {
842 (void) osd_oi_close_table(env, o);
843 OBD_FREE(o->od_oi_table,
844 sizeof(struct osd_oi *) * o->od_oi_count);
845 o->od_oi_table = NULL;
852 int osd_options_init(void)
854 /* osd_oi_count - Default number of OIs, 128 works well for ZFS */
855 if (osd_oi_count == 0 || osd_oi_count > OSD_OI_FID_NR_MAX)
856 osd_oi_count = OSD_OI_FID_NR;
858 if ((osd_oi_count & (osd_oi_count - 1)) != 0) {
859 LCONSOLE_WARN("Round up osd_oi_count %d to power2 %d\n",
860 osd_oi_count, size_roundup_power2(osd_oi_count));
861 osd_oi_count = size_roundup_power2(osd_oi_count);