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 { OFD_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 { MDT_LAST_RECV_OID, LAST_RCVD },
103 { OFD_HEALTH_CHECK_OID, HEALTH_CHECK },
104 { ACCT_USER_OID, "acct_usr_inode" },
105 { ACCT_GROUP_OID, "acct_grp_inode" },
106 { MDD_ROOT_INDEX_OID, NULL },
107 { MDD_ORPHAN_OID, NULL },
111 static char *oid2name(const unsigned long oid)
115 while (oids[i].oid) {
116 if (oids[i].oid == oid)
124 * Lookup an existing OI by the given name.
127 osd_oi_lookup(const struct lu_env *env, struct osd_device *o,
128 uint64_t parent, const char *name, struct osd_oi *oi)
130 struct zpl_direntry *zde = &osd_oti_get(env)->oti_zde.lzd_reg;
133 rc = -zap_lookup(o->od_objset.os, parent, name, 8, 1, (void *)zde);
137 strncpy(oi->oi_name, name, OSD_OI_NAME_SIZE - 1);
138 oi->oi_zapid = zde->zde_dnode;
144 * Create a new OI with the given name.
147 osd_oi_create(const struct lu_env *env, struct osd_device *o,
148 uint64_t parent, const char *name, uint64_t *child)
150 struct zpl_direntry *zde = &osd_oti_get(env)->oti_zde.lzd_reg;
151 struct lu_attr *la = &osd_oti_get(env)->oti_la;
156 /* verify it doesn't already exist */
157 rc = -zap_lookup(o->od_objset.os, parent, name, 8, 1, (void *)zde);
161 /* create fid-to-dnode index */
162 tx = dmu_tx_create(o->od_objset.os);
166 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, 1, NULL);
167 dmu_tx_hold_bonus(tx, parent);
168 dmu_tx_hold_zap(tx, parent, TRUE, name);
169 LASSERT(tx->tx_objset->os_sa);
170 dmu_tx_hold_sa_create(tx, ZFS_SA_BASE_ATTR_SIZE);
172 rc = -dmu_tx_assign(tx, TXG_WAIT);
178 la->la_valid = LA_MODE | LA_UID | LA_GID;
179 la->la_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
180 la->la_uid = la->la_gid = 0;
181 __osd_zap_create(env, &o->od_objset, &db, tx, la, oi_tag, 0);
183 zde->zde_dnode = db->db_object;
185 zde->zde_type = IFTODT(S_IFDIR);
187 rc = -zap_add(o->od_objset.os, parent, name, 8, 1, (void *)zde, tx);
191 *child = db->db_object;
192 sa_buf_rele(db, oi_tag);
198 osd_oi_find_or_create(const struct lu_env *env, struct osd_device *o,
199 uint64_t parent, const char *name, uint64_t *child)
204 rc = osd_oi_lookup(env, o, parent, name, &oi);
206 *child = oi.oi_zapid;
207 else if (rc == -ENOENT)
208 rc = osd_oi_create(env, o, parent, name, child);
214 * Lookup the target index/flags of the fid, so it will know where
215 * the object is located (tgt index) and it is MDT or OST object.
217 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
218 const struct lu_fid *fid, struct lu_seq_range *range)
220 struct seq_server_site *ss = osd_seq_site(osd);
223 if (fid_is_idif(fid)) {
224 range->lsr_flags = LU_SEQ_RANGE_OST;
225 range->lsr_index = fid_idif_ost_idx(fid);
229 if (!fid_is_norm(fid)) {
230 range->lsr_flags = LU_SEQ_RANGE_MDT;
232 range->lsr_index = ss->ss_node_id;
237 range->lsr_flags = -1;
238 rc = fld_server_lookup(env, ss->ss_server_fld, fid_seq(fid), range);
240 CERROR("%s can not find "DFID": rc = %d\n",
241 osd2lu_dev(osd)->ld_obd->obd_name, PFID(fid), rc);
246 int fid_is_on_ost(const struct lu_env *env, struct osd_device *osd,
247 const struct lu_fid *fid)
249 struct lu_seq_range *range = &osd_oti_get(env)->oti_seq_range;
253 if (fid_is_idif(fid))
256 rc = osd_fld_lookup(env, osd, fid, range);
258 CERROR("%s: Can not lookup fld for "DFID"\n",
259 osd2lu_dev(osd)->ld_obd->obd_name, PFID(fid));
263 CDEBUG(D_INFO, "fid "DFID" range "DRANGE"\n", PFID(fid),
266 if (range->lsr_flags == LU_SEQ_RANGE_OST)
272 static struct osd_seq *osd_seq_find_locked(struct osd_seq_list *seq_list,
275 struct osd_seq *osd_seq;
277 cfs_list_for_each_entry(osd_seq, &seq_list->osl_seq_list, os_seq_list) {
278 if (osd_seq->os_seq == seq)
284 static struct osd_seq *osd_seq_find(struct osd_seq_list *seq_list,
287 struct osd_seq *osd_seq;
289 read_lock(&seq_list->osl_seq_list_lock);
290 osd_seq = osd_seq_find_locked(seq_list, seq);
291 read_unlock(&seq_list->osl_seq_list_lock);
296 static struct osd_seq *osd_find_or_add_seq(const struct lu_env *env,
297 struct osd_device *osd, obd_seq seq)
299 struct osd_seq_list *seq_list = &osd->od_seq_list;
300 struct osd_seq *osd_seq;
301 char *key = osd_oti_get(env)->oti_buf;
302 char *seq_name = osd_oti_get(env)->oti_str;
309 osd_seq = osd_seq_find(seq_list, seq);
313 down(&seq_list->osl_seq_init_sem);
314 /* Check again, in case some one else already add it
316 osd_seq = osd_seq_find(seq_list, seq);
320 OBD_ALLOC_PTR(osd_seq);
322 GOTO(out, rc = -ENOMEM);
324 CFS_INIT_LIST_HEAD(&osd_seq->os_seq_list);
325 osd_seq->os_seq = seq;
327 /* Init subdir count to be 32, but each seq can have
328 * different subdir count */
329 osd_seq->os_subdir_count = OSD_OST_MAP_SIZE;
330 OBD_ALLOC(osd_seq->os_compat_dirs,
331 sizeof(uint64_t) * osd_seq->os_subdir_count);
332 if (osd_seq->os_compat_dirs == NULL)
333 GOTO(out, rc = -ENOMEM);
335 rc = osd_oi_lookup(env, osd, osd->od_root, "O", &oi);
337 CERROR("%s: Can not find O: rc = %d\n", osd_name(osd), rc);
341 sprintf(seq_name, (fid_seq_is_rsvd(seq) ||
342 fid_seq_is_mdt0(seq)) ? LPU64 : LPX64i,
343 fid_seq_is_idif(seq) ? 0 : seq);
345 rc = osd_oi_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);
353 osd->od_ost_compat_grp0 = odb;
355 for (i = 0; i < OSD_OST_MAP_SIZE; i++) {
356 sprintf(key, "d%d", i);
357 rc = osd_oi_find_or_create(env, osd, odb, key, &sdb);
359 GOTO(out, osd_seq = ERR_PTR(rc));
360 osd_seq->os_compat_dirs[i] = sdb;
363 write_lock(&seq_list->osl_seq_list_lock);
364 cfs_list_add(&osd_seq->os_seq_list, &seq_list->osl_seq_list);
365 write_unlock(&seq_list->osl_seq_list_lock);
367 up(&seq_list->osl_seq_init_sem);
369 if (osd_seq != NULL && osd_seq->os_compat_dirs != NULL)
370 OBD_FREE(osd_seq->os_compat_dirs,
371 sizeof(uint64_t) * osd_seq->os_subdir_count);
373 OBD_FREE_PTR(osd_seq);
374 osd_seq = ERR_PTR(rc);
380 * objects w/o a natural reference (unlike a file on a MDS)
381 * are put under a special hierarchy /O/<seq>/d0..dXX
382 * this function returns a directory specific fid belongs to
385 osd_get_idx_for_ost_obj(const struct lu_env *env, struct osd_device *osd,
386 const struct lu_fid *fid, char *buf)
388 struct osd_seq *osd_seq;
392 osd_seq = osd_find_or_add_seq(env, osd, fid_seq(fid));
393 if (IS_ERR(osd_seq)) {
394 CERROR("%s: Can not find seq group "DFID"\n", osd_name(osd),
396 return PTR_ERR(osd_seq);
398 rc = fid_ostid_pack(fid, &osd_oti_get(env)->oti_ostid);
399 LASSERT(rc == 0); /* we should not get here with IGIF */
400 b = osd_oti_get(env)->oti_ostid.oi_id % OSD_OST_MAP_SIZE;
401 LASSERT(osd_seq->os_compat_dirs[b]);
403 sprintf(buf, LPU64, osd_oti_get(env)->oti_ostid.oi_id);
405 return osd_seq->os_compat_dirs[b];
408 /* XXX: f_ver is not counted, but may differ too */
409 static void osd_fid2str(char *buf, const struct lu_fid *fid)
411 sprintf(buf, DFID_NOBRACE, PFID(fid));
415 * Determine the zap object id which is being used as the OI for the
416 * given fid. The lowest N bits in the sequence ID are used as the
417 * index key. On failure 0 is returned which zfs treats internally
418 * as an invalid object id.
421 osd_get_idx_for_fid(struct osd_device *osd, const struct lu_fid *fid,
426 LASSERT(osd->od_oi_table != NULL);
427 oi = osd->od_oi_table[fid_seq(fid) & (osd->od_oi_count - 1)];
428 osd_fid2str(buf, fid);
433 uint64_t osd_get_name_n_idx(const struct lu_env *env, struct osd_device *osd,
434 const struct lu_fid *fid, char *buf)
441 if (fid_is_on_ost(env, osd, fid) == 1) {
442 zapid = osd_get_idx_for_ost_obj(env, osd, fid, buf);
443 } else if (fid_is_last_id(fid)) {
444 zapid = osd->od_ost_compat_grp0;
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;
689 * Initialize the OIs by either opening or creating them as needed.
691 int osd_oi_init(const struct lu_env *env, struct osd_device *o)
693 char *key = osd_oti_get(env)->oti_buf;
694 int i, rc, count = 0;
697 rc = osd_oi_probe(env, o, &count);
704 count = osd_oi_count;
707 for (i = 0; i < count; i++) {
708 sprintf(key, "%s.%d", DMU_OSD_OI_NAME_BASE, i);
709 rc = osd_oi_find_or_create(env, o, odb, key, &sdb);
715 rc = osd_oi_init_compat(env, o);
719 LASSERT((count & (count - 1)) == 0);
720 o->od_oi_count = count;
721 OBD_ALLOC(o->od_oi_table, sizeof(struct osd_oi *) * count);
722 if (o->od_oi_table == NULL)
725 rc = osd_oi_open_table(env, o, count);
727 OBD_FREE(o->od_oi_table, sizeof(struct osd_oi *) * count);
728 o->od_oi_table = NULL;
734 void osd_oi_fini(const struct lu_env *env, struct osd_device *o)
738 osd_ost_seq_fini(env, o);
740 if (o->od_oi_table != NULL) {
741 (void) osd_oi_close_table(env, o);
742 OBD_FREE(o->od_oi_table,
743 sizeof(struct osd_oi *) * o->od_oi_count);
744 o->od_oi_table = NULL;
751 int osd_options_init(void)
753 /* osd_oi_count - Default number of OIs, 128 works well for ZFS */
754 if (osd_oi_count == 0 || osd_oi_count > OSD_OI_FID_NR_MAX)
755 osd_oi_count = OSD_OI_FID_NR;
757 if ((osd_oi_count & (osd_oi_count - 1)) != 0) {
758 LCONSOLE_WARN("Round up osd_oi_count %d to power2 %d\n",
759 osd_oi_count, size_roundup_power2(osd_oi_count));
760 osd_oi_count = size_roundup_power2(osd_oi_count);