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.
30 * Copyright (c) 2012, 2015, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/osd-zfs/osd_oi.c
37 * OI functions to map fid to dnode
39 * Author: Alex Zhuravlev <bzzz@whamcloud.com>
40 * Author: Mike Pershin <tappro@whamcloud.com>
41 * Author: Di Wang <di.wang@intel.com>
44 #define DEBUG_SUBSYSTEM S_OSD
46 #include <lustre_ver.h>
47 #include <libcfs/libcfs.h>
48 #include <obd_support.h>
49 #include <lustre_net.h>
51 #include <obd_class.h>
52 #include <lustre_disk.h>
53 #include <lustre_fid.h>
55 #include "osd_internal.h"
57 #include <sys/dnode.h>
62 #include <sys/spa_impl.h>
63 #include <sys/zfs_znode.h>
64 #include <sys/dmu_tx.h>
65 #include <sys/dmu_objset.h>
66 #include <sys/dsl_prop.h>
67 #include <sys/sa_impl.h>
70 #define OSD_OI_FID_NR (1UL << 7)
71 #define OSD_OI_FID_NR_MAX (1UL << OSD_OI_FID_OID_BITS_MAX)
72 unsigned int osd_oi_count = OSD_OI_FID_NR;
76 * zfs osd maintains names for known fids in the name hierarchy
77 * so that one can mount filesystem with regular ZFS stack and
85 static const struct named_oid oids[] = {
86 { LAST_RECV_OID, LAST_RCVD },
87 { OFD_LAST_GROUP_OID, "LAST_GROUP" },
88 { LLOG_CATALOGS_OID, "CATALOGS" },
89 { MGS_CONFIGS_OID, NULL /*MOUNT_CONFIGS_DIR*/ },
90 { FID_SEQ_SRV_OID, "seq_srv" },
91 { FID_SEQ_CTL_OID, "seq_ctl" },
92 { FLD_INDEX_OID, "fld" },
93 { MDD_LOV_OBJ_OID, LOV_OBJID },
94 { OFD_HEALTH_CHECK_OID, HEALTH_CHECK },
95 { ACCT_USER_OID, "acct_usr_inode" },
96 { ACCT_GROUP_OID, "acct_grp_inode" },
97 { REPLY_DATA_OID, REPLY_DATA },
101 static char *oid2name(const unsigned long oid)
105 while (oids[i].oid) {
106 if (oids[i].oid == oid)
114 * Lookup an existing OI by the given name.
117 osd_oi_lookup(const struct lu_env *env, struct osd_device *o,
118 uint64_t parent, const char *name, struct osd_oi *oi)
120 struct zpl_direntry *zde = &osd_oti_get(env)->oti_zde.lzd_reg;
123 rc = -zap_lookup(o->od_os, parent, name, 8, 1, (void *)zde);
127 rc = strlcpy(oi->oi_name, name, sizeof(oi->oi_name));
128 if (rc >= sizeof(oi->oi_name))
131 oi->oi_zapid = zde->zde_dnode;
137 * Create a new OI with the given name.
140 osd_oi_create(const struct lu_env *env, struct osd_device *o,
141 uint64_t parent, const char *name, uint64_t *child)
143 struct zpl_direntry *zde = &osd_oti_get(env)->oti_zde.lzd_reg;
144 struct lu_attr *la = &osd_oti_get(env)->oti_la;
145 sa_handle_t *sa_hdl = NULL;
150 /* verify it doesn't already exist */
151 rc = -zap_lookup(o->od_os, parent, name, 8, 1, (void *)zde);
155 /* create fid-to-dnode index */
156 tx = dmu_tx_create(o->od_os);
160 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, 1, NULL);
161 dmu_tx_hold_bonus(tx, parent);
162 dmu_tx_hold_zap(tx, parent, TRUE, name);
163 LASSERT(tx->tx_objset->os_sa);
164 dmu_tx_hold_sa_create(tx, ZFS_SA_BASE_ATTR_SIZE);
166 rc = -dmu_tx_assign(tx, TXG_WAIT);
172 oid = zap_create_flags(o->od_os, 0, ZAP_FLAG_HASH64,
173 DMU_OT_DIRECTORY_CONTENTS,
174 14, /* == ZFS fzap_default_block_shift */
175 DN_MAX_INDBLKSHIFT, /* indirect block shift */
176 DMU_OT_SA, DN_MAX_BONUSLEN, tx);
178 rc = -sa_handle_get(o->od_os, oid, NULL, SA_HDL_PRIVATE, &sa_hdl);
181 la->la_valid = LA_MODE | LA_UID | LA_GID;
182 la->la_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
183 la->la_uid = la->la_gid = 0;
184 rc = __osd_attr_init(env, o, sa_hdl, tx, la, parent);
185 sa_handle_destroy(sa_hdl);
189 zde->zde_dnode = oid;
191 zde->zde_type = IFTODT(S_IFDIR);
193 rc = -zap_add(o->od_os, parent, name, 8, 1, (void *)zde, tx);
197 dmu_object_free(o->od_os, oid, tx);
207 osd_oi_find_or_create(const struct lu_env *env, struct osd_device *o,
208 uint64_t parent, const char *name, uint64_t *child)
213 rc = osd_oi_lookup(env, o, parent, name, &oi);
215 *child = oi.oi_zapid;
216 else if (rc == -ENOENT)
217 rc = osd_oi_create(env, o, parent, name, child);
223 * Lookup the target index/flags of the fid, so it will know where
224 * the object is located (tgt index) and it is MDT or OST object.
226 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
227 u64 seq, struct lu_seq_range *range)
229 struct seq_server_site *ss = osd_seq_site(osd);
231 if (fid_seq_is_idif(seq)) {
232 fld_range_set_ost(range);
233 range->lsr_index = idif_ost_idx(seq);
237 if (!fid_seq_in_fldb(seq)) {
238 fld_range_set_mdt(range);
240 /* FIXME: If ss is NULL, it suppose not get lsr_index
242 range->lsr_index = ss->ss_node_id;
247 fld_range_set_any(range);
248 /* OSD will only do local fld lookup */
249 return fld_local_lookup(env, ss->ss_server_fld, seq, range);
252 int fid_is_on_ost(const struct lu_env *env, struct osd_device *osd,
253 const struct lu_fid *fid)
255 struct lu_seq_range *range = &osd_oti_get(env)->oti_seq_range;
259 if (fid_is_idif(fid))
262 if (unlikely(fid_is_local_file(fid) || fid_is_llog(fid)) ||
263 fid_is_name_llog(fid) || fid_is_quota(fid))
266 rc = osd_fld_lookup(env, osd, fid_seq(fid), range);
269 CERROR("%s: "DFID" lookup failed: rc = %d\n",
270 osd_name(osd), PFID(fid), rc);
274 if (fld_range_is_ost(range))
280 static struct osd_seq *osd_seq_find_locked(struct osd_seq_list *seq_list,
283 struct osd_seq *osd_seq;
285 list_for_each_entry(osd_seq, &seq_list->osl_seq_list, os_seq_list) {
286 if (osd_seq->os_seq == seq)
292 static struct osd_seq *osd_seq_find(struct osd_seq_list *seq_list, u64 seq)
294 struct osd_seq *osd_seq;
296 read_lock(&seq_list->osl_seq_list_lock);
297 osd_seq = osd_seq_find_locked(seq_list, seq);
298 read_unlock(&seq_list->osl_seq_list_lock);
303 static struct osd_seq *osd_find_or_add_seq(const struct lu_env *env,
304 struct osd_device *osd, u64 seq)
306 struct osd_seq_list *seq_list = &osd->od_seq_list;
307 struct osd_seq *osd_seq;
308 char *key = osd_oti_get(env)->oti_buf;
309 char *seq_name = osd_oti_get(env)->oti_str;
316 osd_seq = osd_seq_find(seq_list, seq);
320 down(&seq_list->osl_seq_init_sem);
321 /* Check again, in case some one else already add it
323 osd_seq = osd_seq_find(seq_list, seq);
327 OBD_ALLOC_PTR(osd_seq);
329 GOTO(out, rc = -ENOMEM);
331 INIT_LIST_HEAD(&osd_seq->os_seq_list);
332 osd_seq->os_seq = seq;
334 /* Init subdir count to be 32, but each seq can have
335 * different subdir count */
336 osd_seq->os_subdir_count = OSD_OST_MAP_SIZE;
337 OBD_ALLOC(osd_seq->os_compat_dirs,
338 sizeof(uint64_t) * osd_seq->os_subdir_count);
339 if (osd_seq->os_compat_dirs == NULL)
340 GOTO(out, rc = -ENOMEM);
342 oi.oi_zapid = osd->od_O_id;
343 sprintf(seq_name, (fid_seq_is_rsvd(seq) ||
344 fid_seq_is_mdt0(seq)) ? LPU64 : LPX64i,
345 fid_seq_is_idif(seq) ? 0 : seq);
347 rc = osd_oi_find_or_create(env, osd, oi.oi_zapid, seq_name, &odb);
349 CERROR("%s: Can not create %s : rc = %d\n",
350 osd_name(osd), seq_name, rc);
354 for (i = 0; i < OSD_OST_MAP_SIZE; i++) {
355 sprintf(key, "d%d", i);
356 rc = osd_oi_find_or_create(env, osd, odb, key, &sdb);
359 osd_seq->os_compat_dirs[i] = sdb;
362 write_lock(&seq_list->osl_seq_list_lock);
363 list_add(&osd_seq->os_seq_list, &seq_list->osl_seq_list);
364 write_unlock(&seq_list->osl_seq_list_lock);
366 up(&seq_list->osl_seq_init_sem);
368 if (osd_seq != NULL && osd_seq->os_compat_dirs != NULL)
369 OBD_FREE(osd_seq->os_compat_dirs,
370 sizeof(uint64_t) * osd_seq->os_subdir_count);
372 OBD_FREE_PTR(osd_seq);
373 osd_seq = ERR_PTR(rc);
379 * objects w/o a natural reference (unlike a file on a MDS)
380 * are put under a special hierarchy /O/<seq>/d0..dXX
381 * this function returns a directory specific fid belongs to
384 osd_get_idx_for_ost_obj(const struct lu_env *env, struct osd_device *osd,
385 const struct lu_fid *fid, char *buf, int bufsize)
387 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);
399 if (fid_is_last_id(fid)) {
402 rc = fid_to_ostid(fid, &osd_oti_get(env)->oti_ostid);
403 LASSERT(rc == 0); /* we should not get here with IGIF */
404 id = ostid_id(&osd_oti_get(env)->oti_ostid);
407 b = id % OSD_OST_MAP_SIZE;
408 LASSERT(osd_seq->os_compat_dirs[b]);
411 snprintf(buf, bufsize, LPU64, id);
413 return osd_seq->os_compat_dirs[b];
416 /* XXX: f_ver is not counted, but may differ too */
417 static void osd_fid2str(char *buf, const struct lu_fid *fid)
419 sprintf(buf, DFID_NOBRACE, PFID(fid));
423 * Determine the zap object id which is being used as the OI for the
424 * given fid. The lowest N bits in the sequence ID are used as the
425 * index key. On failure 0 is returned which zfs treats internally
426 * as an invalid object id.
429 osd_get_idx_for_fid(struct osd_device *osd, const struct lu_fid *fid,
434 LASSERT(osd->od_oi_table != NULL);
435 oi = osd->od_oi_table[fid_seq(fid) & (osd->od_oi_count - 1)];
437 osd_fid2str(buf, fid);
442 uint64_t osd_get_name_n_idx(const struct lu_env *env, struct osd_device *osd,
443 const struct lu_fid *fid, char *buf, int bufsize)
449 if (fid_is_on_ost(env, osd, fid) == 1 || fid_seq(fid) == FID_SEQ_ECHO) {
450 zapid = osd_get_idx_for_ost_obj(env, osd, fid, buf, bufsize);
451 } else if (unlikely(fid_seq(fid) == FID_SEQ_LOCAL_FILE)) {
452 /* special objects with fixed known fids get their name */
453 char *name = oid2name(fid_oid(fid));
456 zapid = osd->od_root;
458 strncpy(buf, name, bufsize);
459 if (fid_is_acct(fid))
460 zapid = MASTER_NODE_OBJ;
462 zapid = osd_get_idx_for_fid(osd, fid, buf);
465 zapid = osd_get_idx_for_fid(osd, fid, buf);
471 static inline int fid_is_fs_root(const struct lu_fid *fid)
473 /* Map root inode to special local object FID */
474 return fid_seq(fid) == FID_SEQ_LOCAL_FILE &&
475 fid_oid(fid) == OSD_FS_ROOT_OID;
478 int osd_fid_lookup(const struct lu_env *env, struct osd_device *dev,
479 const struct lu_fid *fid, uint64_t *oid)
481 struct osd_thread_info *info = osd_oti_get(env);
482 char *buf = info->oti_buf;
487 if (OBD_FAIL_CHECK(OBD_FAIL_SRV_ENOENT))
490 if (unlikely(fid_is_acct(fid))) {
491 if (fid_oid(fid) == ACCT_USER_OID)
492 *oid = dev->od_iusr_oid;
494 *oid = dev->od_igrp_oid;
495 } else if (unlikely(fid_is_fs_root(fid))) {
498 zapid = osd_get_name_n_idx(env, dev, fid, buf,
499 sizeof(info->oti_buf));
500 rc = -zap_lookup(dev->od_os, zapid, buf,
501 8, 1, &info->oti_zde);
504 *oid = info->oti_zde.lzd_reg.zde_dnode;
508 osd_dmu_prefetch(dev->od_os, *oid, 0, 0, 0,
509 ZIO_PRIORITY_ASYNC_READ);
515 * Close an entry in a specific slot.
518 osd_oi_remove_table(const struct lu_env *env, struct osd_device *o, int key)
522 LASSERT(key < o->od_oi_count);
524 oi = o->od_oi_table[key];
527 sa_buf_rele(oi->oi_db, osd_obj_tag);
529 o->od_oi_table[key] = NULL;
534 * Allocate and open a new entry in the specified unused slot.
537 osd_oi_add_table(const struct lu_env *env, struct osd_device *o,
543 LASSERT(key < o->od_oi_count);
544 LASSERT(o->od_oi_table[key] == NULL);
550 rc = osd_oi_lookup(env, o, o->od_root, name, oi);
556 o->od_oi_table[key] = oi;
557 __osd_obj2dbuf(env, o->od_os, oi->oi_zapid, &oi->oi_db);
563 * Depopulate the OI table.
566 osd_oi_close_table(const struct lu_env *env, struct osd_device *o)
570 for (i = 0; i < o->od_oi_count; i++)
571 osd_oi_remove_table(env, o, i);
575 * Populate the OI table based.
578 osd_oi_open_table(const struct lu_env *env, struct osd_device *o, int count)
584 for (i = 0; i < count; i++) {
585 sprintf(name, "%s.%d", DMU_OSD_OI_NAME_BASE, i);
586 rc = osd_oi_add_table(env, o, name, i);
588 osd_oi_close_table(env, o);
597 * Determine if the type and number of OIs used by this file system.
600 osd_oi_probe(const struct lu_env *env, struct osd_device *o, int *count)
602 uint64_t root_oid = o->od_root;
609 * Check for multiple OIs and determine the count. There is no
610 * gap handling, if an OI is missing the wrong size can be returned.
611 * The only safeguard is that we know the number of OIs must be a
612 * power of two and this is checked for basic sanity.
614 for (*count = 0; *count < OSD_OI_FID_NR_MAX; (*count)++) {
615 sprintf(name, "%s.%d", DMU_OSD_OI_NAME_BASE, *count);
616 rc = osd_oi_lookup(env, o, root_oid, name, &oi);
624 if ((*count & (*count - 1)) != 0)
634 * No OIs exist, this must be a new filesystem.
641 static void osd_ost_seq_fini(const struct lu_env *env, struct osd_device *osd)
643 struct osd_seq_list *osl = &osd->od_seq_list;
644 struct osd_seq *osd_seq, *tmp;
646 write_lock(&osl->osl_seq_list_lock);
647 list_for_each_entry_safe(osd_seq, tmp, &osl->osl_seq_list,
649 list_del(&osd_seq->os_seq_list);
650 OBD_FREE(osd_seq->os_compat_dirs,
651 sizeof(uint64_t) * osd_seq->os_subdir_count);
652 OBD_FREE(osd_seq, sizeof(*osd_seq));
654 write_unlock(&osl->osl_seq_list_lock);
660 * Create /O subdirectory to map legacy OST objects for compatibility.
663 osd_oi_init_compat(const struct lu_env *env, struct osd_device *o)
669 rc = osd_oi_find_or_create(env, o, o->od_root, "O", &sdb);
675 /* Create on-disk indexes to maintain per-UID/GID inode usage.
676 * Those new indexes are created in the top-level ZAP outside the
677 * namespace in order not to confuse ZPL which might interpret those
678 * indexes as directories and assume the values are object IDs */
679 rc = osd_oi_find_or_create(env, o, MASTER_NODE_OBJ,
680 oid2name(ACCT_USER_OID), &odb);
683 o->od_iusr_oid = odb;
685 rc = osd_oi_find_or_create(env, o, MASTER_NODE_OBJ,
686 oid2name(ACCT_GROUP_OID), &odb);
689 o->od_igrp_oid = odb;
695 * Initialize the OIs by either opening or creating them as needed.
697 int osd_oi_init(const struct lu_env *env, struct osd_device *o)
699 char *key = osd_oti_get(env)->oti_buf;
700 int i, rc, count = 0;
703 rc = osd_oi_probe(env, o, &count);
710 count = osd_oi_count;
713 for (i = 0; i < count; i++) {
714 sprintf(key, "%s.%d", DMU_OSD_OI_NAME_BASE, i);
715 rc = osd_oi_find_or_create(env, o, odb, key, &sdb);
721 rc = osd_oi_init_compat(env, o);
725 LASSERT((count & (count - 1)) == 0);
726 o->od_oi_count = count;
727 OBD_ALLOC(o->od_oi_table, sizeof(struct osd_oi *) * count);
728 if (o->od_oi_table == NULL)
731 rc = osd_oi_open_table(env, o, count);
733 OBD_FREE(o->od_oi_table, sizeof(struct osd_oi *) * count);
734 o->od_oi_table = NULL;
740 void osd_oi_fini(const struct lu_env *env, struct osd_device *o)
744 osd_ost_seq_fini(env, o);
746 if (o->od_oi_table != NULL) {
747 (void) osd_oi_close_table(env, o);
748 OBD_FREE(o->od_oi_table,
749 sizeof(struct osd_oi *) * o->od_oi_count);
750 o->od_oi_table = NULL;
757 int osd_options_init(void)
759 /* osd_oi_count - Default number of OIs, 128 works well for ZFS */
760 if (osd_oi_count == 0 || osd_oi_count > OSD_OI_FID_NR_MAX)
761 osd_oi_count = OSD_OI_FID_NR;
763 if ((osd_oi_count & (osd_oi_count - 1)) != 0) {
764 LCONSOLE_WARN("Round up osd_oi_count %d to power2 %d\n",
765 osd_oi_count, size_roundup_power2(osd_oi_count));
766 osd_oi_count = size_roundup_power2(osd_oi_count);