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, 2014, 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" },
100 static char *oid2name(const unsigned long oid)
104 while (oids[i].oid) {
105 if (oids[i].oid == oid)
113 * Lookup an existing OI by the given name.
116 osd_oi_lookup(const struct lu_env *env, struct osd_device *o,
117 uint64_t parent, const char *name, struct osd_oi *oi)
119 struct zpl_direntry *zde = &osd_oti_get(env)->oti_zde.lzd_reg;
122 rc = -zap_lookup(o->od_os, parent, name, 8, 1, (void *)zde);
126 rc = strlcpy(oi->oi_name, name, sizeof(oi->oi_name));
127 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;
149 /* verify it doesn't already exist */
150 rc = -zap_lookup(o->od_os, parent, name, 8, 1, (void *)zde);
154 /* create fid-to-dnode index */
155 tx = dmu_tx_create(o->od_os);
159 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, 1, NULL);
160 dmu_tx_hold_bonus(tx, parent);
161 dmu_tx_hold_zap(tx, parent, TRUE, name);
162 LASSERT(tx->tx_objset->os_sa);
163 dmu_tx_hold_sa_create(tx, ZFS_SA_BASE_ATTR_SIZE);
165 rc = -dmu_tx_assign(tx, TXG_WAIT);
171 la->la_valid = LA_MODE | LA_UID | LA_GID;
172 la->la_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
173 la->la_uid = la->la_gid = 0;
174 __osd_zap_create(env, o, &db, tx, la, parent, 0);
176 zde->zde_dnode = db->db_object;
178 zde->zde_type = IFTODT(S_IFDIR);
180 rc = -zap_add(o->od_os, parent, name, 8, 1, (void *)zde, tx);
184 *child = db->db_object;
185 sa_buf_rele(db, osd_obj_tag);
191 osd_oi_find_or_create(const struct lu_env *env, struct osd_device *o,
192 uint64_t parent, const char *name, uint64_t *child)
197 rc = osd_oi_lookup(env, o, parent, name, &oi);
199 *child = oi.oi_zapid;
200 else if (rc == -ENOENT)
201 rc = osd_oi_create(env, o, parent, name, child);
207 * Lookup the target index/flags of the fid, so it will know where
208 * the object is located (tgt index) and it is MDT or OST object.
210 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
211 u64 seq, struct lu_seq_range *range)
213 struct seq_server_site *ss = osd_seq_site(osd);
215 if (fid_seq_is_idif(seq)) {
216 fld_range_set_ost(range);
217 range->lsr_index = idif_ost_idx(seq);
221 if (!fid_seq_in_fldb(seq)) {
222 fld_range_set_mdt(range);
224 /* FIXME: If ss is NULL, it suppose not get lsr_index
226 range->lsr_index = ss->ss_node_id;
231 fld_range_set_any(range);
232 /* OSD will only do local fld lookup */
233 return fld_local_lookup(env, ss->ss_server_fld, seq, range);
236 int fid_is_on_ost(const struct lu_env *env, struct osd_device *osd,
237 const struct lu_fid *fid)
239 struct lu_seq_range *range = &osd_oti_get(env)->oti_seq_range;
243 if (fid_is_idif(fid))
246 if (unlikely(fid_is_local_file(fid) || fid_is_llog(fid)) ||
247 fid_is_name_llog(fid) || fid_is_quota(fid))
250 rc = osd_fld_lookup(env, osd, fid_seq(fid), range);
253 CERROR("%s: "DFID" lookup failed: rc = %d\n",
254 osd_name(osd), PFID(fid), rc);
258 if (fld_range_is_ost(range))
264 static struct osd_seq *osd_seq_find_locked(struct osd_seq_list *seq_list,
267 struct osd_seq *osd_seq;
269 list_for_each_entry(osd_seq, &seq_list->osl_seq_list, os_seq_list) {
270 if (osd_seq->os_seq == seq)
276 static struct osd_seq *osd_seq_find(struct osd_seq_list *seq_list, u64 seq)
278 struct osd_seq *osd_seq;
280 read_lock(&seq_list->osl_seq_list_lock);
281 osd_seq = osd_seq_find_locked(seq_list, seq);
282 read_unlock(&seq_list->osl_seq_list_lock);
287 static struct osd_seq *osd_find_or_add_seq(const struct lu_env *env,
288 struct osd_device *osd, u64 seq)
290 struct osd_seq_list *seq_list = &osd->od_seq_list;
291 struct osd_seq *osd_seq;
292 char *key = osd_oti_get(env)->oti_buf;
293 char *seq_name = osd_oti_get(env)->oti_str;
300 osd_seq = osd_seq_find(seq_list, seq);
304 down(&seq_list->osl_seq_init_sem);
305 /* Check again, in case some one else already add it
307 osd_seq = osd_seq_find(seq_list, seq);
311 OBD_ALLOC_PTR(osd_seq);
313 GOTO(out, rc = -ENOMEM);
315 INIT_LIST_HEAD(&osd_seq->os_seq_list);
316 osd_seq->os_seq = seq;
318 /* Init subdir count to be 32, but each seq can have
319 * different subdir count */
320 osd_seq->os_subdir_count = OSD_OST_MAP_SIZE;
321 OBD_ALLOC(osd_seq->os_compat_dirs,
322 sizeof(uint64_t) * osd_seq->os_subdir_count);
323 if (osd_seq->os_compat_dirs == NULL)
324 GOTO(out, rc = -ENOMEM);
326 oi.oi_zapid = osd->od_O_id;
327 sprintf(seq_name, (fid_seq_is_rsvd(seq) ||
328 fid_seq_is_mdt0(seq)) ? LPU64 : LPX64i,
329 fid_seq_is_idif(seq) ? 0 : seq);
331 rc = osd_oi_find_or_create(env, osd, oi.oi_zapid, seq_name, &odb);
333 CERROR("%s: Can not create %s : rc = %d\n",
334 osd_name(osd), seq_name, rc);
338 for (i = 0; i < OSD_OST_MAP_SIZE; i++) {
339 sprintf(key, "d%d", i);
340 rc = osd_oi_find_or_create(env, osd, odb, key, &sdb);
343 osd_seq->os_compat_dirs[i] = sdb;
346 write_lock(&seq_list->osl_seq_list_lock);
347 list_add(&osd_seq->os_seq_list, &seq_list->osl_seq_list);
348 write_unlock(&seq_list->osl_seq_list_lock);
350 up(&seq_list->osl_seq_init_sem);
352 if (osd_seq != NULL && osd_seq->os_compat_dirs != NULL)
353 OBD_FREE(osd_seq->os_compat_dirs,
354 sizeof(uint64_t) * osd_seq->os_subdir_count);
356 OBD_FREE_PTR(osd_seq);
357 osd_seq = ERR_PTR(rc);
363 * objects w/o a natural reference (unlike a file on a MDS)
364 * are put under a special hierarchy /O/<seq>/d0..dXX
365 * this function returns a directory specific fid belongs to
368 osd_get_idx_for_ost_obj(const struct lu_env *env, struct osd_device *osd,
369 const struct lu_fid *fid, char *buf)
371 struct osd_seq *osd_seq;
376 osd_seq = osd_find_or_add_seq(env, osd, fid_seq(fid));
377 if (IS_ERR(osd_seq)) {
378 CERROR("%s: Can not find seq group "DFID"\n", osd_name(osd),
380 return PTR_ERR(osd_seq);
383 if (fid_is_last_id(fid)) {
386 rc = fid_to_ostid(fid, &osd_oti_get(env)->oti_ostid);
387 LASSERT(rc == 0); /* we should not get here with IGIF */
388 id = ostid_id(&osd_oti_get(env)->oti_ostid);
391 b = id % OSD_OST_MAP_SIZE;
392 LASSERT(osd_seq->os_compat_dirs[b]);
394 sprintf(buf, LPU64, id);
396 return osd_seq->os_compat_dirs[b];
399 /* XXX: f_ver is not counted, but may differ too */
400 static void osd_fid2str(char *buf, const struct lu_fid *fid)
402 sprintf(buf, DFID_NOBRACE, PFID(fid));
406 * Determine the zap object id which is being used as the OI for the
407 * given fid. The lowest N bits in the sequence ID are used as the
408 * index key. On failure 0 is returned which zfs treats internally
409 * as an invalid object id.
412 osd_get_idx_for_fid(struct osd_device *osd, const struct lu_fid *fid,
417 LASSERT(osd->od_oi_table != NULL);
418 oi = osd->od_oi_table[fid_seq(fid) & (osd->od_oi_count - 1)];
419 osd_fid2str(buf, fid);
424 uint64_t osd_get_name_n_idx(const struct lu_env *env, struct osd_device *osd,
425 const struct lu_fid *fid, char *buf)
432 if (fid_is_on_ost(env, osd, fid) == 1 || fid_seq(fid) == FID_SEQ_ECHO) {
433 zapid = osd_get_idx_for_ost_obj(env, osd, fid, buf);
434 } else if (unlikely(fid_seq(fid) == FID_SEQ_LOCAL_FILE)) {
435 /* special objects with fixed known fids get their name */
436 char *name = oid2name(fid_oid(fid));
439 zapid = osd->od_root;
441 if (fid_is_acct(fid))
442 zapid = MASTER_NODE_OBJ;
444 zapid = osd_get_idx_for_fid(osd, fid, buf);
447 zapid = osd_get_idx_for_fid(osd, fid, buf);
453 static inline int fid_is_fs_root(const struct lu_fid *fid)
455 /* Map root inode to special local object FID */
456 return fid_seq(fid) == FID_SEQ_LOCAL_FILE &&
457 fid_oid(fid) == OSD_FS_ROOT_OID;
460 int osd_fid_lookup(const struct lu_env *env, struct osd_device *dev,
461 const struct lu_fid *fid, uint64_t *oid)
463 struct osd_thread_info *info = osd_oti_get(env);
464 char *buf = info->oti_buf;
469 if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOENT))
472 if (unlikely(fid_is_acct(fid))) {
473 if (fid_oid(fid) == ACCT_USER_OID)
474 *oid = dev->od_iusr_oid;
476 *oid = dev->od_igrp_oid;
477 } else if (unlikely(fid_is_fs_root(fid))) {
480 zapid = osd_get_name_n_idx(env, dev, fid, buf);
482 rc = -zap_lookup(dev->od_os, zapid, buf,
483 8, 1, &info->oti_zde);
486 *oid = info->oti_zde.lzd_reg.zde_dnode;
490 dmu_prefetch(dev->od_os, *oid, 0, 0);
496 * Close an entry in a specific slot.
499 osd_oi_remove_table(const struct lu_env *env, struct osd_device *o, int key)
503 LASSERT(key < o->od_oi_count);
505 oi = o->od_oi_table[key];
508 o->od_oi_table[key] = NULL;
513 * Allocate and open a new entry in the specified unused slot.
516 osd_oi_add_table(const struct lu_env *env, struct osd_device *o,
522 LASSERT(key < o->od_oi_count);
523 LASSERT(o->od_oi_table[key] == NULL);
529 rc = osd_oi_lookup(env, o, o->od_root, name, oi);
535 o->od_oi_table[key] = oi;
541 * Depopulate the OI table.
544 osd_oi_close_table(const struct lu_env *env, struct osd_device *o)
548 for (i = 0; i < o->od_oi_count; i++)
549 osd_oi_remove_table(env, o, i);
553 * Populate the OI table based.
556 osd_oi_open_table(const struct lu_env *env, struct osd_device *o, int count)
562 for (i = 0; i < count; i++) {
563 sprintf(name, "%s.%d", DMU_OSD_OI_NAME_BASE, i);
564 rc = osd_oi_add_table(env, o, name, i);
566 osd_oi_close_table(env, o);
575 * Determine if the type and number of OIs used by this file system.
578 osd_oi_probe(const struct lu_env *env, struct osd_device *o, int *count)
580 uint64_t root_oid = o->od_root;
587 * Check for multiple OIs and determine the count. There is no
588 * gap handling, if an OI is missing the wrong size can be returned.
589 * The only safeguard is that we know the number of OIs must be a
590 * power of two and this is checked for basic sanity.
592 for (*count = 0; *count < OSD_OI_FID_NR_MAX; (*count)++) {
593 sprintf(name, "%s.%d", DMU_OSD_OI_NAME_BASE, *count);
594 rc = osd_oi_lookup(env, o, root_oid, name, &oi);
602 if ((*count & (*count - 1)) != 0)
612 * No OIs exist, this must be a new filesystem.
619 static void osd_ost_seq_init(const struct lu_env *env, struct osd_device *osd)
621 struct osd_seq_list *osl = &osd->od_seq_list;
623 INIT_LIST_HEAD(&osl->osl_seq_list);
624 rwlock_init(&osl->osl_seq_list_lock);
625 sema_init(&osl->osl_seq_init_sem, 1);
628 static void osd_ost_seq_fini(const struct lu_env *env, struct osd_device *osd)
630 struct osd_seq_list *osl = &osd->od_seq_list;
631 struct osd_seq *osd_seq, *tmp;
633 write_lock(&osl->osl_seq_list_lock);
634 list_for_each_entry_safe(osd_seq, tmp, &osl->osl_seq_list,
636 list_del(&osd_seq->os_seq_list);
637 OBD_FREE(osd_seq->os_compat_dirs,
638 sizeof(uint64_t) * osd_seq->os_subdir_count);
639 OBD_FREE(osd_seq, sizeof(*osd_seq));
641 write_unlock(&osl->osl_seq_list_lock);
647 * Create /O subdirectory to map legacy OST objects for compatibility.
650 osd_oi_init_compat(const struct lu_env *env, struct osd_device *o)
656 rc = osd_oi_find_or_create(env, o, o->od_root, "O", &sdb);
662 osd_ost_seq_init(env, o);
663 /* Create on-disk indexes to maintain per-UID/GID inode usage.
664 * Those new indexes are created in the top-level ZAP outside the
665 * namespace in order not to confuse ZPL which might interpret those
666 * indexes as directories and assume the values are object IDs */
667 rc = osd_oi_find_or_create(env, o, MASTER_NODE_OBJ,
668 oid2name(ACCT_USER_OID), &odb);
671 o->od_iusr_oid = odb;
673 rc = osd_oi_find_or_create(env, o, MASTER_NODE_OBJ,
674 oid2name(ACCT_GROUP_OID), &odb);
677 o->od_igrp_oid = odb;
682 static char *root2convert = "ROOT";
684 * due to DNE requirements we have to change sequence of /ROOT object
685 * so that it doesn't belong to the local sequence FID_SEQ_LOCAL_FILE
686 * but a normal sequence living on MDS#0
687 * this is the sole purpose of this function.
689 * This is only needed for pre-production 2.4 ZFS filesystems, and
690 * can be removed in the future.
692 int osd_convert_root_to_new_seq(const struct lu_env *env, struct osd_device *o)
694 struct luz_direntry *lze = &osd_oti_get(env)->oti_zde;
695 char *buf = osd_oti_get(env)->oti_str;
696 struct lu_fid newfid;
703 if (strstr(o->od_svname, "MDT") == NULL)
707 rc = -zap_lookup(o->od_os, o->od_root, root2convert, 8,
708 sizeof(*lze) / 8, (void *)lze);
709 /* doesn't exist or let actual user to handle the error */
713 CDEBUG(D_OTHER, "%s: /ROOT -> "DFID" -> "LPU64"\n", o->od_svname,
714 PFID(&lze->lzd_fid), (long long int) lze->lzd_reg.zde_dnode);
716 /* already right one? */
717 if (fid_seq(&lze->lzd_fid) == FID_SEQ_ROOT)
720 tx = dmu_tx_create(o->od_os);
724 dmu_tx_hold_bonus(tx, o->od_root);
726 /* declare delete/insert of the name */
727 dmu_tx_hold_zap(tx, o->od_root, TRUE, root2convert);
728 dmu_tx_hold_zap(tx, o->od_root, FALSE, root2convert);
730 /* declare that we'll remove object from fid-dnode mapping */
731 zapid = osd_get_name_n_idx(env, o, &lze->lzd_fid, buf);
732 dmu_tx_hold_bonus(tx, zapid);
733 dmu_tx_hold_zap(tx, zapid, FALSE, buf);
735 /* declare that we'll add object to fid-dnode mapping */
736 newfid.f_seq = FID_SEQ_ROOT;
737 newfid.f_oid = FID_OID_ROOT;
739 zapid = osd_get_name_n_idx(env, o, &newfid, buf);
740 dmu_tx_hold_bonus(tx, zapid);
741 dmu_tx_hold_zap(tx, zapid, TRUE, buf);
743 rc = -dmu_tx_assign(tx, TXG_WAIT);
747 rc = -zap_remove(o->od_os, o->od_root, root2convert, tx);
752 zapid = osd_get_name_n_idx(env, o, &lze->lzd_fid, buf);
753 rc = -zap_remove(o->od_os, zapid, buf, tx);
757 lze->lzd_fid = newfid;
758 rc = -zap_add(o->od_os, o->od_root, root2convert,
759 8, sizeof(*lze) / 8, (void *)lze, tx);
763 /* add to OI with the new fid */
764 zapid = osd_get_name_n_idx(env, o, &newfid, buf);
765 rc = -zap_add(o->od_os, zapid, buf, 8, 1, &lze->lzd_reg, tx);
770 /* LMA will be updated in mdd_compat_fixes */
778 CERROR("%s: can't convert to new fid: rc = %d\n", o->od_svname, rc);
783 * Initialize the OIs by either opening or creating them as needed.
785 int osd_oi_init(const struct lu_env *env, struct osd_device *o)
787 char *key = osd_oti_get(env)->oti_buf;
788 int i, rc, count = 0;
791 rc = osd_oi_probe(env, o, &count);
798 count = osd_oi_count;
801 for (i = 0; i < count; i++) {
802 sprintf(key, "%s.%d", DMU_OSD_OI_NAME_BASE, i);
803 rc = osd_oi_find_or_create(env, o, odb, key, &sdb);
809 rc = osd_oi_init_compat(env, o);
813 LASSERT((count & (count - 1)) == 0);
814 o->od_oi_count = count;
815 OBD_ALLOC(o->od_oi_table, sizeof(struct osd_oi *) * count);
816 if (o->od_oi_table == NULL)
819 rc = osd_oi_open_table(env, o, count);
821 OBD_FREE(o->od_oi_table, sizeof(struct osd_oi *) * count);
822 o->od_oi_table = NULL;
828 void osd_oi_fini(const struct lu_env *env, struct osd_device *o)
832 osd_ost_seq_fini(env, o);
834 if (o->od_oi_table != NULL) {
835 (void) osd_oi_close_table(env, o);
836 OBD_FREE(o->od_oi_table,
837 sizeof(struct osd_oi *) * o->od_oi_count);
838 o->od_oi_table = NULL;
845 int osd_options_init(void)
847 /* osd_oi_count - Default number of OIs, 128 works well for ZFS */
848 if (osd_oi_count == 0 || osd_oi_count > OSD_OI_FID_NR_MAX)
849 osd_oi_count = OSD_OI_FID_NR;
851 if ((osd_oi_count & (osd_oi_count - 1)) != 0) {
852 LCONSOLE_WARN("Round up osd_oi_count %d to power2 %d\n",
853 osd_oi_count, size_roundup_power2(osd_oi_count));
854 osd_oi_count = size_roundup_power2(osd_oi_count);