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.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2009, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2012, 2016, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
32 * lustre/osd-zfs/osd_oi.c
33 * OI functions to map fid to dnode
35 * Author: Alex Zhuravlev <bzzz@whamcloud.com>
36 * Author: Mike Pershin <tappro@whamcloud.com>
37 * Author: Di Wang <di.wang@intel.com>
40 #define DEBUG_SUBSYSTEM S_OSD
42 #include <lustre_ver.h>
43 #include <libcfs/libcfs.h>
44 #include <obd_support.h>
45 #include <lustre_net.h>
47 #include <obd_class.h>
48 #include <lustre_disk.h>
49 #include <lustre_fid.h>
51 #include "osd_internal.h"
53 #include <sys/dnode.h>
58 #include <sys/spa_impl.h>
59 #include <sys/zfs_znode.h>
60 #include <sys/dmu_tx.h>
61 #include <sys/dmu_objset.h>
62 #include <sys/dsl_prop.h>
63 #include <sys/sa_impl.h>
66 #define OSD_OI_FID_NR (1UL << 7)
67 #define OSD_OI_FID_NR_MAX (1UL << OSD_OI_FID_OID_BITS_MAX)
68 unsigned int osd_oi_count = OSD_OI_FID_NR;
72 * zfs osd maintains names for known fids in the name hierarchy
73 * so that one can mount filesystem with regular ZFS stack and
81 static const struct named_oid oids[] = {
82 { .oid = LAST_RECV_OID, .name = LAST_RCVD },
83 { .oid = OFD_LAST_GROUP_OID, .name = "LAST_GROUP" },
84 { .oid = LLOG_CATALOGS_OID, .name = "CATALOGS" },
85 { .oid = MGS_CONFIGS_OID, /*MOUNT_CONFIGS_DIR*/ },
86 { .oid = FID_SEQ_SRV_OID, .name = "seq_srv" },
87 { .oid = FID_SEQ_CTL_OID, .name = "seq_ctl" },
88 { .oid = FLD_INDEX_OID, .name = "fld" },
89 { .oid = MDD_LOV_OBJ_OID, .name = LOV_OBJID },
90 { .oid = OFD_HEALTH_CHECK_OID, .name = HEALTH_CHECK },
91 { .oid = ACCT_USER_OID, .name = "acct_usr_inode" },
92 { .oid = ACCT_GROUP_OID, .name = "acct_grp_inode" },
93 { .oid = REPLY_DATA_OID, .name = REPLY_DATA },
97 static char *oid2name(const unsigned long oid)
101 while (oids[i].oid) {
102 if (oids[i].oid == oid)
110 * Lookup an existing OI by the given name.
113 osd_oi_lookup(const struct lu_env *env, struct osd_device *o,
114 uint64_t parent, const char *name, struct osd_oi *oi)
116 struct zpl_direntry *zde = &osd_oti_get(env)->oti_zde.lzd_reg;
119 rc = -zap_lookup(o->od_os, parent, name, 8, 1, (void *)zde);
123 rc = strlcpy(oi->oi_name, name, sizeof(oi->oi_name));
124 if (rc >= sizeof(oi->oi_name))
127 oi->oi_zapid = zde->zde_dnode;
133 * Create a new OI with the given name.
136 osd_oi_create(const struct lu_env *env, struct osd_device *o,
137 uint64_t parent, const char *name, uint64_t *child)
139 struct zpl_direntry *zde = &osd_oti_get(env)->oti_zde.lzd_reg;
140 struct lu_attr *la = &osd_oti_get(env)->oti_la;
141 sa_handle_t *sa_hdl = NULL;
146 /* verify it doesn't already exist */
147 rc = -zap_lookup(o->od_os, parent, name, 8, 1, (void *)zde);
151 /* create fid-to-dnode index */
152 tx = dmu_tx_create(o->od_os);
156 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, 1, NULL);
157 dmu_tx_hold_bonus(tx, parent);
158 dmu_tx_hold_zap(tx, parent, TRUE, name);
159 LASSERT(tx->tx_objset->os_sa);
160 dmu_tx_hold_sa_create(tx, ZFS_SA_BASE_ATTR_SIZE);
162 rc = -dmu_tx_assign(tx, TXG_WAIT);
168 oid = zap_create_flags(o->od_os, 0, ZAP_FLAG_HASH64,
169 DMU_OT_DIRECTORY_CONTENTS,
170 14, /* == ZFS fzap_default_block_shift */
171 DN_MAX_INDBLKSHIFT, /* indirect block shift */
172 DMU_OT_SA, DN_MAX_BONUSLEN, tx);
174 rc = -sa_handle_get(o->od_os, oid, NULL, SA_HDL_PRIVATE, &sa_hdl);
177 la->la_valid = LA_MODE | LA_UID | LA_GID;
178 la->la_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
179 la->la_uid = la->la_gid = 0;
180 rc = __osd_attr_init(env, o, sa_hdl, tx, la, parent);
181 sa_handle_destroy(sa_hdl);
185 zde->zde_dnode = oid;
187 zde->zde_type = IFTODT(S_IFDIR);
189 rc = -zap_add(o->od_os, parent, name, 8, 1, (void *)zde, tx);
193 dmu_object_free(o->od_os, oid, tx);
203 osd_oi_find_or_create(const struct lu_env *env, struct osd_device *o,
204 uint64_t parent, const char *name, uint64_t *child)
209 rc = osd_oi_lookup(env, o, parent, name, &oi);
211 *child = oi.oi_zapid;
212 else if (rc == -ENOENT)
213 rc = osd_oi_create(env, o, parent, name, child);
219 * Lookup the target index/flags of the fid, so it will know where
220 * the object is located (tgt index) and it is MDT or OST object.
222 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
223 u64 seq, struct lu_seq_range *range)
225 struct seq_server_site *ss = osd_seq_site(osd);
227 if (fid_seq_is_idif(seq)) {
228 fld_range_set_ost(range);
229 range->lsr_index = idif_ost_idx(seq);
233 if (!fid_seq_in_fldb(seq)) {
234 fld_range_set_mdt(range);
236 /* FIXME: If ss is NULL, it suppose not get lsr_index
238 range->lsr_index = ss->ss_node_id;
243 fld_range_set_any(range);
244 /* OSD will only do local fld lookup */
245 return fld_local_lookup(env, ss->ss_server_fld, seq, range);
248 int fid_is_on_ost(const struct lu_env *env, struct osd_device *osd,
249 const struct lu_fid *fid)
251 struct lu_seq_range *range = &osd_oti_get(env)->oti_seq_range;
255 if (fid_is_idif(fid))
258 if (unlikely(fid_is_local_file(fid) || fid_is_llog(fid)) ||
259 fid_is_name_llog(fid) || fid_is_quota(fid))
262 rc = osd_fld_lookup(env, osd, fid_seq(fid), range);
265 CERROR("%s: "DFID" lookup failed: rc = %d\n",
266 osd_name(osd), PFID(fid), rc);
270 if (fld_range_is_ost(range))
276 static struct osd_seq *osd_seq_find_locked(struct osd_seq_list *seq_list,
279 struct osd_seq *osd_seq;
281 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, u64 seq)
290 struct osd_seq *osd_seq;
292 read_lock(&seq_list->osl_seq_list_lock);
293 osd_seq = osd_seq_find_locked(seq_list, seq);
294 read_unlock(&seq_list->osl_seq_list_lock);
299 static struct osd_seq *osd_find_or_add_seq(const struct lu_env *env,
300 struct osd_device *osd, u64 seq)
302 struct osd_seq_list *seq_list = &osd->od_seq_list;
303 struct osd_seq *osd_seq;
304 char *key = osd_oti_get(env)->oti_buf;
305 char *seq_name = osd_oti_get(env)->oti_str;
312 osd_seq = osd_seq_find(seq_list, seq);
316 down(&seq_list->osl_seq_init_sem);
317 /* Check again, in case some one else already add it
319 osd_seq = osd_seq_find(seq_list, seq);
323 OBD_ALLOC_PTR(osd_seq);
325 GOTO(out, rc = -ENOMEM);
327 INIT_LIST_HEAD(&osd_seq->os_seq_list);
328 osd_seq->os_seq = seq;
330 /* Init subdir count to be 32, but each seq can have
331 * different subdir count */
332 osd_seq->os_subdir_count = OSD_OST_MAP_SIZE;
333 OBD_ALLOC(osd_seq->os_compat_dirs,
334 sizeof(uint64_t) * osd_seq->os_subdir_count);
335 if (osd_seq->os_compat_dirs == NULL)
336 GOTO(out, rc = -ENOMEM);
338 oi.oi_zapid = osd->od_O_id;
339 sprintf(seq_name, (fid_seq_is_rsvd(seq) ||
340 fid_seq_is_mdt0(seq)) ? "%llu" : "%llx",
341 fid_seq_is_idif(seq) ? 0 : seq);
343 rc = osd_oi_find_or_create(env, osd, oi.oi_zapid, seq_name, &odb);
345 CERROR("%s: Can not create %s : rc = %d\n",
346 osd_name(osd), seq_name, rc);
350 for (i = 0; i < OSD_OST_MAP_SIZE; i++) {
351 sprintf(key, "d%d", i);
352 rc = osd_oi_find_or_create(env, osd, odb, key, &sdb);
355 osd_seq->os_compat_dirs[i] = sdb;
358 write_lock(&seq_list->osl_seq_list_lock);
359 list_add(&osd_seq->os_seq_list, &seq_list->osl_seq_list);
360 write_unlock(&seq_list->osl_seq_list_lock);
362 up(&seq_list->osl_seq_init_sem);
364 if (osd_seq != NULL && osd_seq->os_compat_dirs != NULL)
365 OBD_FREE(osd_seq->os_compat_dirs,
366 sizeof(uint64_t) * osd_seq->os_subdir_count);
368 OBD_FREE_PTR(osd_seq);
369 osd_seq = ERR_PTR(rc);
375 * objects w/o a natural reference (unlike a file on a MDS)
376 * are put under a special hierarchy /O/<seq>/d0..dXX
377 * this function returns a directory specific fid belongs to
380 osd_get_idx_for_ost_obj(const struct lu_env *env, struct osd_device *osd,
381 const struct lu_fid *fid, char *buf, int bufsize)
383 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 if (fid_is_last_id(fid)) {
398 rc = fid_to_ostid(fid, &osd_oti_get(env)->oti_ostid);
399 LASSERT(rc == 0); /* we should not get here with IGIF */
400 id = ostid_id(&osd_oti_get(env)->oti_ostid);
403 b = id % OSD_OST_MAP_SIZE;
404 LASSERT(osd_seq->os_compat_dirs[b]);
407 snprintf(buf, bufsize, "%llu", id);
409 return osd_seq->os_compat_dirs[b];
412 /* XXX: f_ver is not counted, but may differ too */
413 static void osd_fid2str(char *buf, const struct lu_fid *fid)
415 sprintf(buf, DFID_NOBRACE, PFID(fid));
419 * Determine the zap object id which is being used as the OI for the
420 * given fid. The lowest N bits in the sequence ID are used as the
421 * index key. On failure 0 is returned which zfs treats internally
422 * as an invalid object id.
425 osd_get_idx_for_fid(struct osd_device *osd, const struct lu_fid *fid,
430 LASSERT(osd->od_oi_table != NULL);
431 oi = osd->od_oi_table[fid_seq(fid) & (osd->od_oi_count - 1)];
433 osd_fid2str(buf, fid);
438 uint64_t osd_get_name_n_idx(const struct lu_env *env, struct osd_device *osd,
439 const struct lu_fid *fid, char *buf, int bufsize)
445 if (fid_is_on_ost(env, osd, fid) == 1 || fid_seq(fid) == FID_SEQ_ECHO) {
446 zapid = osd_get_idx_for_ost_obj(env, osd, fid, buf, bufsize);
447 } else if (unlikely(fid_seq(fid) == FID_SEQ_LOCAL_FILE)) {
448 /* special objects with fixed known fids get their name */
449 char *name = oid2name(fid_oid(fid));
452 zapid = osd->od_root;
454 strncpy(buf, name, bufsize);
455 if (fid_is_acct(fid))
456 zapid = MASTER_NODE_OBJ;
458 zapid = osd_get_idx_for_fid(osd, fid, buf);
461 zapid = osd_get_idx_for_fid(osd, fid, buf);
467 static inline int fid_is_fs_root(const struct lu_fid *fid)
469 /* Map root inode to special local object FID */
470 return fid_seq(fid) == FID_SEQ_LOCAL_FILE &&
471 fid_oid(fid) == OSD_FS_ROOT_OID;
474 int osd_fid_lookup(const struct lu_env *env, struct osd_device *dev,
475 const struct lu_fid *fid, uint64_t *oid)
477 struct osd_thread_info *info = osd_oti_get(env);
478 char *buf = info->oti_buf;
483 if (OBD_FAIL_CHECK(OBD_FAIL_SRV_ENOENT))
486 if (unlikely(fid_is_acct(fid))) {
487 if (fid_oid(fid) == ACCT_USER_OID)
488 *oid = dev->od_iusr_oid;
490 *oid = dev->od_igrp_oid;
491 } else if (unlikely(fid_is_fs_root(fid))) {
494 zapid = osd_get_name_n_idx(env, dev, fid, buf,
495 sizeof(info->oti_buf));
496 rc = -zap_lookup(dev->od_os, zapid, buf,
497 8, 1, &info->oti_zde);
500 *oid = info->oti_zde.lzd_reg.zde_dnode;
504 osd_dmu_prefetch(dev->od_os, *oid, 0, 0, 0,
505 ZIO_PRIORITY_ASYNC_READ);
511 * Close an entry in a specific slot.
514 osd_oi_remove_table(const struct lu_env *env, struct osd_device *o, int key)
518 LASSERT(key < o->od_oi_count);
520 oi = o->od_oi_table[key];
523 sa_buf_rele(oi->oi_db, osd_obj_tag);
525 o->od_oi_table[key] = NULL;
530 * Allocate and open a new entry in the specified unused slot.
533 osd_oi_add_table(const struct lu_env *env, struct osd_device *o,
539 LASSERT(key < o->od_oi_count);
540 LASSERT(o->od_oi_table[key] == NULL);
546 rc = osd_oi_lookup(env, o, o->od_root, name, oi);
552 o->od_oi_table[key] = oi;
553 __osd_obj2dbuf(env, o->od_os, oi->oi_zapid, &oi->oi_db);
559 * Depopulate the OI table.
562 osd_oi_close_table(const struct lu_env *env, struct osd_device *o)
566 for (i = 0; i < o->od_oi_count; i++)
567 osd_oi_remove_table(env, o, i);
571 * Populate the OI table based.
574 osd_oi_open_table(const struct lu_env *env, struct osd_device *o, int count)
580 for (i = 0; i < count; i++) {
581 sprintf(name, "%s.%d", DMU_OSD_OI_NAME_BASE, i);
582 rc = osd_oi_add_table(env, o, name, i);
584 osd_oi_close_table(env, o);
593 * Determine if the type and number of OIs used by this file system.
596 osd_oi_probe(const struct lu_env *env, struct osd_device *o, int *count)
598 uint64_t root_oid = o->od_root;
605 * Check for multiple OIs and determine the count. There is no
606 * gap handling, if an OI is missing the wrong size can be returned.
607 * The only safeguard is that we know the number of OIs must be a
608 * power of two and this is checked for basic sanity.
610 for (*count = 0; *count < OSD_OI_FID_NR_MAX; (*count)++) {
611 sprintf(name, "%s.%d", DMU_OSD_OI_NAME_BASE, *count);
612 rc = osd_oi_lookup(env, o, root_oid, name, &oi);
620 if ((*count & (*count - 1)) != 0)
630 * No OIs exist, this must be a new filesystem.
637 static void osd_ost_seq_fini(const struct lu_env *env, struct osd_device *osd)
639 struct osd_seq_list *osl = &osd->od_seq_list;
640 struct osd_seq *osd_seq, *tmp;
642 write_lock(&osl->osl_seq_list_lock);
643 list_for_each_entry_safe(osd_seq, tmp, &osl->osl_seq_list,
645 list_del(&osd_seq->os_seq_list);
646 OBD_FREE(osd_seq->os_compat_dirs,
647 sizeof(uint64_t) * osd_seq->os_subdir_count);
648 OBD_FREE(osd_seq, sizeof(*osd_seq));
650 write_unlock(&osl->osl_seq_list_lock);
656 * Create /O subdirectory to map legacy OST objects for compatibility.
659 osd_oi_init_compat(const struct lu_env *env, struct osd_device *o)
665 rc = osd_oi_find_or_create(env, o, o->od_root, "O", &sdb);
671 /* Create on-disk indexes to maintain per-UID/GID inode usage.
672 * Those new indexes are created in the top-level ZAP outside the
673 * namespace in order not to confuse ZPL which might interpret those
674 * indexes as directories and assume the values are object IDs */
675 rc = osd_oi_find_or_create(env, o, MASTER_NODE_OBJ,
676 oid2name(ACCT_USER_OID), &odb);
679 o->od_iusr_oid = odb;
681 rc = osd_oi_find_or_create(env, o, MASTER_NODE_OBJ,
682 oid2name(ACCT_GROUP_OID), &odb);
685 o->od_igrp_oid = odb;
691 * Initialize the OIs by either opening or creating them as needed.
693 int osd_oi_init(const struct lu_env *env, struct osd_device *o)
695 char *key = osd_oti_get(env)->oti_buf;
696 int i, rc, count = 0;
699 rc = osd_oi_probe(env, o, &count);
706 count = osd_oi_count;
709 for (i = 0; i < count; i++) {
710 sprintf(key, "%s.%d", DMU_OSD_OI_NAME_BASE, i);
711 rc = osd_oi_find_or_create(env, o, odb, key, &sdb);
717 rc = osd_oi_init_compat(env, o);
721 LASSERT((count & (count - 1)) == 0);
722 o->od_oi_count = count;
723 OBD_ALLOC(o->od_oi_table, sizeof(struct osd_oi *) * count);
724 if (o->od_oi_table == NULL)
727 rc = osd_oi_open_table(env, o, count);
729 OBD_FREE(o->od_oi_table, sizeof(struct osd_oi *) * count);
730 o->od_oi_table = NULL;
736 void osd_oi_fini(const struct lu_env *env, struct osd_device *o)
740 osd_ost_seq_fini(env, o);
742 if (o->od_oi_table != NULL) {
743 (void) osd_oi_close_table(env, o);
744 OBD_FREE(o->od_oi_table,
745 sizeof(struct osd_oi *) * o->od_oi_count);
746 o->od_oi_table = NULL;
753 int osd_options_init(void)
755 /* osd_oi_count - Default number of OIs, 128 works well for ZFS */
756 if (osd_oi_count == 0 || osd_oi_count > OSD_OI_FID_NR_MAX)
757 osd_oi_count = OSD_OI_FID_NR;
759 if ((osd_oi_count & (osd_oi_count - 1)) != 0) {
760 LCONSOLE_WARN("Round up osd_oi_count %d to power2 %d\n",
761 osd_oi_count, size_roundup_power2(osd_oi_count));
762 osd_oi_count = size_roundup_power2(osd_oi_count);