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, 2013, 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>
46 #define DEBUG_SUBSYSTEM S_OSD
48 #include <lustre_ver.h>
49 #include <libcfs/libcfs.h>
50 #include <obd_support.h>
51 #include <lustre_net.h>
53 #include <obd_class.h>
54 #include <lustre_disk.h>
55 #include <lustre_fid.h>
57 #include "osd_internal.h"
59 #include <sys/dnode.h>
64 #include <sys/spa_impl.h>
65 #include <sys/zfs_znode.h>
66 #include <sys/dmu_tx.h>
67 #include <sys/dmu_objset.h>
68 #include <sys/dsl_prop.h>
69 #include <sys/sa_impl.h>
72 static char *oi_tag = "osd_mount, oi";
74 #define OSD_OI_FID_NR (1UL << 7)
75 #define OSD_OI_FID_NR_MAX (1UL << OSD_OI_FID_OID_BITS_MAX)
76 unsigned int osd_oi_count = OSD_OI_FID_NR;
80 * zfs osd maintains names for known fids in the name hierarchy
81 * so that one can mount filesystem with regular ZFS stack and
89 static const struct named_oid oids[] = {
90 { LAST_RECV_OID, LAST_RCVD },
91 { OFD_LAST_GROUP_OID, "LAST_GROUP" },
92 { LLOG_CATALOGS_OID, "CATALOGS" },
93 { MGS_CONFIGS_OID, NULL /*MOUNT_CONFIGS_DIR*/ },
94 { FID_SEQ_SRV_OID, "seq_srv" },
95 { FID_SEQ_CTL_OID, "seq_ctl" },
96 { FLD_INDEX_OID, "fld" },
97 { MDD_LOV_OBJ_OID, LOV_OBJID },
98 { OFD_HEALTH_CHECK_OID, HEALTH_CHECK },
99 { ACCT_USER_OID, "acct_usr_inode" },
100 { ACCT_GROUP_OID, "acct_grp_inode" },
104 static char *oid2name(const unsigned long oid)
108 while (oids[i].oid) {
109 if (oids[i].oid == oid)
117 * Lookup an existing OI by the given name.
120 osd_oi_lookup(const struct lu_env *env, struct osd_device *o,
121 uint64_t parent, const char *name, struct osd_oi *oi)
123 struct zpl_direntry *zde = &osd_oti_get(env)->oti_zde.lzd_reg;
126 rc = -zap_lookup(o->od_objset.os, parent, name, 8, 1, (void *)zde);
130 rc = strlcpy(oi->oi_name, name, sizeof(oi->oi_name));
131 if (rc >= sizeof(oi->oi_name))
135 oi->oi_zapid = zde->zde_dnode;
141 * Create a new OI with the given name.
144 osd_oi_create(const struct lu_env *env, struct osd_device *o,
145 uint64_t parent, const char *name, uint64_t *child)
147 struct zpl_direntry *zde = &osd_oti_get(env)->oti_zde.lzd_reg;
148 struct lu_attr *la = &osd_oti_get(env)->oti_la;
153 /* verify it doesn't already exist */
154 rc = -zap_lookup(o->od_objset.os, parent, name, 8, 1, (void *)zde);
158 /* create fid-to-dnode index */
159 tx = dmu_tx_create(o->od_objset.os);
163 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, 1, NULL);
164 dmu_tx_hold_bonus(tx, parent);
165 dmu_tx_hold_zap(tx, parent, TRUE, name);
166 LASSERT(tx->tx_objset->os_sa);
167 dmu_tx_hold_sa_create(tx, ZFS_SA_BASE_ATTR_SIZE);
169 rc = -dmu_tx_assign(tx, TXG_WAIT);
175 la->la_valid = LA_MODE | LA_UID | LA_GID;
176 la->la_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
177 la->la_uid = la->la_gid = 0;
178 __osd_zap_create(env, &o->od_objset, &db, tx, la, parent, oi_tag, 0);
180 zde->zde_dnode = db->db_object;
182 zde->zde_type = IFTODT(S_IFDIR);
184 rc = -zap_add(o->od_objset.os, parent, name, 8, 1, (void *)zde, tx);
188 *child = db->db_object;
189 sa_buf_rele(db, oi_tag);
195 osd_oi_find_or_create(const struct lu_env *env, struct osd_device *o,
196 uint64_t parent, const char *name, uint64_t *child)
201 rc = osd_oi_lookup(env, o, parent, name, &oi);
203 *child = oi.oi_zapid;
204 else if (rc == -ENOENT)
205 rc = osd_oi_create(env, o, parent, name, child);
211 * Lookup the target index/flags of the fid, so it will know where
212 * the object is located (tgt index) and it is MDT or OST object.
214 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
215 obd_seq seq, struct lu_seq_range *range)
217 struct seq_server_site *ss = osd_seq_site(osd);
219 if (fid_seq_is_idif(seq)) {
220 fld_range_set_ost(range);
221 range->lsr_index = idif_ost_idx(seq);
225 if (!fid_seq_in_fldb(seq)) {
226 fld_range_set_mdt(range);
228 /* FIXME: If ss is NULL, it suppose not get lsr_index
230 range->lsr_index = ss->ss_node_id;
235 fld_range_set_any(range);
236 /* OSD will only do local fld lookup */
237 return fld_local_lookup(env, ss->ss_server_fld, seq, range);
240 int fid_is_on_ost(const struct lu_env *env, struct osd_device *osd,
241 const struct lu_fid *fid)
243 struct lu_seq_range *range = &osd_oti_get(env)->oti_seq_range;
247 if (fid_is_idif(fid))
250 if (unlikely(fid_is_local_file(fid) || fid_is_llog(fid)) ||
251 fid_is_name_llog(fid) || fid_is_quota(fid))
254 rc = osd_fld_lookup(env, osd, fid_seq(fid), range);
257 CERROR("%s: "DFID" lookup failed: rc = %d\n",
258 osd_name(osd), PFID(fid), rc);
262 if (fld_range_is_ost(range))
268 static struct osd_seq *osd_seq_find_locked(struct osd_seq_list *seq_list,
271 struct osd_seq *osd_seq;
273 list_for_each_entry(osd_seq, &seq_list->osl_seq_list, os_seq_list) {
274 if (osd_seq->os_seq == seq)
280 static struct osd_seq *osd_seq_find(struct osd_seq_list *seq_list,
283 struct osd_seq *osd_seq;
285 read_lock(&seq_list->osl_seq_list_lock);
286 osd_seq = osd_seq_find_locked(seq_list, seq);
287 read_unlock(&seq_list->osl_seq_list_lock);
292 static struct osd_seq *osd_find_or_add_seq(const struct lu_env *env,
293 struct osd_device *osd, obd_seq seq)
295 struct osd_seq_list *seq_list = &osd->od_seq_list;
296 struct osd_seq *osd_seq;
297 char *key = osd_oti_get(env)->oti_buf;
298 char *seq_name = osd_oti_get(env)->oti_str;
305 osd_seq = osd_seq_find(seq_list, seq);
309 down(&seq_list->osl_seq_init_sem);
310 /* Check again, in case some one else already add it
312 osd_seq = osd_seq_find(seq_list, seq);
316 OBD_ALLOC_PTR(osd_seq);
318 GOTO(out, rc = -ENOMEM);
320 INIT_LIST_HEAD(&osd_seq->os_seq_list);
321 osd_seq->os_seq = seq;
323 /* Init subdir count to be 32, but each seq can have
324 * different subdir count */
325 osd_seq->os_subdir_count = OSD_OST_MAP_SIZE;
326 OBD_ALLOC(osd_seq->os_compat_dirs,
327 sizeof(uint64_t) * osd_seq->os_subdir_count);
328 if (osd_seq->os_compat_dirs == NULL)
329 GOTO(out, rc = -ENOMEM);
331 oi.oi_zapid = osd->od_O_id;
332 sprintf(seq_name, (fid_seq_is_rsvd(seq) ||
333 fid_seq_is_mdt0(seq)) ? LPU64 : LPX64i,
334 fid_seq_is_idif(seq) ? 0 : seq);
336 rc = osd_oi_find_or_create(env, osd, oi.oi_zapid, seq_name, &odb);
338 CERROR("%s: Can not create %s : rc = %d\n",
339 osd_name(osd), seq_name, rc);
343 for (i = 0; i < OSD_OST_MAP_SIZE; i++) {
344 sprintf(key, "d%d", i);
345 rc = osd_oi_find_or_create(env, osd, odb, key, &sdb);
348 osd_seq->os_compat_dirs[i] = sdb;
351 write_lock(&seq_list->osl_seq_list_lock);
352 list_add(&osd_seq->os_seq_list, &seq_list->osl_seq_list);
353 write_unlock(&seq_list->osl_seq_list_lock);
355 up(&seq_list->osl_seq_init_sem);
357 if (osd_seq != NULL && osd_seq->os_compat_dirs != NULL)
358 OBD_FREE(osd_seq->os_compat_dirs,
359 sizeof(uint64_t) * osd_seq->os_subdir_count);
361 OBD_FREE_PTR(osd_seq);
362 osd_seq = ERR_PTR(rc);
368 * objects w/o a natural reference (unlike a file on a MDS)
369 * are put under a special hierarchy /O/<seq>/d0..dXX
370 * this function returns a directory specific fid belongs to
373 osd_get_idx_for_ost_obj(const struct lu_env *env, struct osd_device *osd,
374 const struct lu_fid *fid, char *buf)
376 struct osd_seq *osd_seq;
381 osd_seq = osd_find_or_add_seq(env, osd, fid_seq(fid));
382 if (IS_ERR(osd_seq)) {
383 CERROR("%s: Can not find seq group "DFID"\n", osd_name(osd),
385 return PTR_ERR(osd_seq);
388 if (fid_is_last_id(fid)) {
391 rc = fid_to_ostid(fid, &osd_oti_get(env)->oti_ostid);
392 LASSERT(rc == 0); /* we should not get here with IGIF */
393 id = ostid_id(&osd_oti_get(env)->oti_ostid);
396 b = id % OSD_OST_MAP_SIZE;
397 LASSERT(osd_seq->os_compat_dirs[b]);
399 sprintf(buf, LPU64, id);
401 return osd_seq->os_compat_dirs[b];
404 /* XXX: f_ver is not counted, but may differ too */
405 static void osd_fid2str(char *buf, const struct lu_fid *fid)
407 sprintf(buf, DFID_NOBRACE, PFID(fid));
411 * Determine the zap object id which is being used as the OI for the
412 * given fid. The lowest N bits in the sequence ID are used as the
413 * index key. On failure 0 is returned which zfs treats internally
414 * as an invalid object id.
417 osd_get_idx_for_fid(struct osd_device *osd, const struct lu_fid *fid,
422 LASSERT(osd->od_oi_table != NULL);
423 oi = osd->od_oi_table[fid_seq(fid) & (osd->od_oi_count - 1)];
424 osd_fid2str(buf, fid);
429 uint64_t osd_get_name_n_idx(const struct lu_env *env, struct osd_device *osd,
430 const struct lu_fid *fid, char *buf)
437 if (fid_is_on_ost(env, osd, fid) == 1 || fid_seq(fid) == FID_SEQ_ECHO) {
438 zapid = osd_get_idx_for_ost_obj(env, osd, fid, buf);
439 } else if (unlikely(fid_seq(fid) == FID_SEQ_LOCAL_FILE)) {
440 /* special objects with fixed known fids get their name */
441 char *name = oid2name(fid_oid(fid));
444 zapid = osd->od_root;
446 if (fid_is_acct(fid))
447 zapid = MASTER_NODE_OBJ;
449 zapid = osd_get_idx_for_fid(osd, fid, buf);
452 zapid = osd_get_idx_for_fid(osd, fid, buf);
458 static inline int fid_is_fs_root(const struct lu_fid *fid)
460 /* Map root inode to special local object FID */
461 return fid_seq(fid) == FID_SEQ_LOCAL_FILE &&
462 fid_oid(fid) == OSD_FS_ROOT_OID;
465 int osd_fid_lookup(const struct lu_env *env, struct osd_device *dev,
466 const struct lu_fid *fid, uint64_t *oid)
468 struct osd_thread_info *info = osd_oti_get(env);
469 char *buf = info->oti_buf;
474 if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOENT))
477 if (unlikely(fid_is_acct(fid))) {
478 if (fid_oid(fid) == ACCT_USER_OID)
479 *oid = dev->od_iusr_oid;
481 *oid = dev->od_igrp_oid;
482 } else if (unlikely(fid_is_fs_root(fid))) {
485 zapid = osd_get_name_n_idx(env, dev, fid, buf);
487 rc = -zap_lookup(dev->od_objset.os, zapid, buf,
488 8, 1, &info->oti_zde);
491 *oid = info->oti_zde.lzd_reg.zde_dnode;
495 dmu_prefetch(dev->od_objset.os, *oid, 0, 0);
501 * Close an entry in a specific slot.
504 osd_oi_remove_table(const struct lu_env *env, struct osd_device *o, int key)
508 LASSERT(key < o->od_oi_count);
510 oi = o->od_oi_table[key];
513 o->od_oi_table[key] = NULL;
518 * Allocate and open a new entry in the specified unused slot.
521 osd_oi_add_table(const struct lu_env *env, struct osd_device *o,
527 LASSERT(key < o->od_oi_count);
528 LASSERT(o->od_oi_table[key] == NULL);
534 rc = osd_oi_lookup(env, o, o->od_root, name, oi);
540 o->od_oi_table[key] = oi;
546 * Depopulate the OI table.
549 osd_oi_close_table(const struct lu_env *env, struct osd_device *o)
553 for (i = 0; i < o->od_oi_count; i++)
554 osd_oi_remove_table(env, o, i);
558 * Populate the OI table based.
561 osd_oi_open_table(const struct lu_env *env, struct osd_device *o, int count)
567 for (i = 0; i < count; i++) {
568 sprintf(name, "%s.%d", DMU_OSD_OI_NAME_BASE, i);
569 rc = osd_oi_add_table(env, o, name, i);
571 osd_oi_close_table(env, o);
580 * Determine if the type and number of OIs used by this file system.
583 osd_oi_probe(const struct lu_env *env, struct osd_device *o, int *count)
585 uint64_t root_oid = o->od_root;
592 * Check for multiple OIs and determine the count. There is no
593 * gap handling, if an OI is missing the wrong size can be returned.
594 * The only safeguard is that we know the number of OIs must be a
595 * power of two and this is checked for basic sanity.
597 for (*count = 0; *count < OSD_OI_FID_NR_MAX; (*count)++) {
598 sprintf(name, "%s.%d", DMU_OSD_OI_NAME_BASE, *count);
599 rc = osd_oi_lookup(env, o, root_oid, name, &oi);
607 if ((*count & (*count - 1)) != 0)
617 * No OIs exist, this must be a new filesystem.
624 static void osd_ost_seq_init(const struct lu_env *env, struct osd_device *osd)
626 struct osd_seq_list *osl = &osd->od_seq_list;
628 INIT_LIST_HEAD(&osl->osl_seq_list);
629 rwlock_init(&osl->osl_seq_list_lock);
630 sema_init(&osl->osl_seq_init_sem, 1);
633 static void osd_ost_seq_fini(const struct lu_env *env, struct osd_device *osd)
635 struct osd_seq_list *osl = &osd->od_seq_list;
636 struct osd_seq *osd_seq, *tmp;
638 write_lock(&osl->osl_seq_list_lock);
639 list_for_each_entry_safe(osd_seq, tmp, &osl->osl_seq_list,
641 list_del(&osd_seq->os_seq_list);
642 OBD_FREE(osd_seq->os_compat_dirs,
643 sizeof(uint64_t) * osd_seq->os_subdir_count);
644 OBD_FREE(osd_seq, sizeof(*osd_seq));
646 write_unlock(&osl->osl_seq_list_lock);
652 * Create /O subdirectory to map legacy OST objects for compatibility.
655 osd_oi_init_compat(const struct lu_env *env, struct osd_device *o)
661 rc = osd_oi_find_or_create(env, o, o->od_root, "O", &sdb);
667 osd_ost_seq_init(env, o);
668 /* Create on-disk indexes to maintain per-UID/GID inode usage.
669 * Those new indexes are created in the top-level ZAP outside the
670 * namespace in order not to confuse ZPL which might interpret those
671 * indexes as directories and assume the values are object IDs */
672 rc = osd_oi_find_or_create(env, o, MASTER_NODE_OBJ,
673 oid2name(ACCT_USER_OID), &odb);
676 o->od_iusr_oid = odb;
678 rc = osd_oi_find_or_create(env, o, MASTER_NODE_OBJ,
679 oid2name(ACCT_GROUP_OID), &odb);
682 o->od_igrp_oid = odb;
687 static char *root2convert = "ROOT";
689 * due to DNE requirements we have to change sequence of /ROOT object
690 * so that it doesn't belong to the local sequence FID_SEQ_LOCAL_FILE
691 * but a normal sequence living on MDS#0
692 * this is the sole purpose of this function.
694 * This is only needed for pre-production 2.4 ZFS filesystems, and
695 * can be removed in the future.
697 int osd_convert_root_to_new_seq(const struct lu_env *env,
698 struct osd_device *o)
700 struct luz_direntry *lze = &osd_oti_get(env)->oti_zde;
701 char *buf = osd_oti_get(env)->oti_str;
702 struct lu_fid newfid;
709 if (strstr(o->od_svname, "MDT") == NULL)
713 rc = -zap_lookup(o->od_objset.os, o->od_root, root2convert, 8,
714 sizeof(*lze) / 8, (void *)lze);
715 /* doesn't exist or let actual user to handle the error */
719 CDEBUG(D_OTHER, "%s: /ROOT -> "DFID" -> "LPU64"\n", o->od_svname,
720 PFID(&lze->lzd_fid), (long long int) lze->lzd_reg.zde_dnode);
722 /* already right one? */
723 if (fid_seq(&lze->lzd_fid) == FID_SEQ_ROOT)
726 tx = dmu_tx_create(o->od_objset.os);
730 dmu_tx_hold_bonus(tx, o->od_root);
732 /* declare delete/insert of the name */
733 dmu_tx_hold_zap(tx, o->od_root, TRUE, root2convert);
734 dmu_tx_hold_zap(tx, o->od_root, FALSE, root2convert);
736 /* declare that we'll remove object from fid-dnode mapping */
737 zapid = osd_get_name_n_idx(env, o, &lze->lzd_fid, buf);
738 dmu_tx_hold_bonus(tx, zapid);
739 dmu_tx_hold_zap(tx, zapid, FALSE, buf);
741 /* declare that we'll add object to fid-dnode mapping */
742 newfid.f_seq = FID_SEQ_ROOT;
745 zapid = osd_get_name_n_idx(env, o, &newfid, buf);
746 dmu_tx_hold_bonus(tx, zapid);
747 dmu_tx_hold_zap(tx, zapid, TRUE, buf);
749 rc = -dmu_tx_assign(tx, TXG_WAIT);
753 rc = -zap_remove(o->od_objset.os, o->od_root, root2convert, tx);
758 zapid = osd_get_name_n_idx(env, o, &lze->lzd_fid, buf);
759 rc = -zap_remove(o->od_objset.os, zapid, buf, tx);
763 lze->lzd_fid = newfid;
764 rc = -zap_add(o->od_objset.os, o->od_root, root2convert,
765 8, sizeof(*lze) / 8, (void *)lze, tx);
769 /* add to OI with the new fid */
770 zapid = osd_get_name_n_idx(env, o, &newfid, buf);
771 rc = -zap_add(o->od_objset.os, zapid, buf, 8, 1, &lze->lzd_reg, tx);
776 /* LMA will be updated in mdd_compat_fixes */
784 CERROR("%s: can't convert to new fid: rc = %d\n", o->od_svname, rc);
789 * Initialize the OIs by either opening or creating them as needed.
791 int osd_oi_init(const struct lu_env *env, struct osd_device *o)
793 char *key = osd_oti_get(env)->oti_buf;
794 int i, rc, count = 0;
797 rc = osd_oi_probe(env, o, &count);
804 count = osd_oi_count;
807 for (i = 0; i < count; i++) {
808 sprintf(key, "%s.%d", DMU_OSD_OI_NAME_BASE, i);
809 rc = osd_oi_find_or_create(env, o, odb, key, &sdb);
815 rc = osd_oi_init_compat(env, o);
819 LASSERT((count & (count - 1)) == 0);
820 o->od_oi_count = count;
821 OBD_ALLOC(o->od_oi_table, sizeof(struct osd_oi *) * count);
822 if (o->od_oi_table == NULL)
825 rc = osd_oi_open_table(env, o, count);
827 OBD_FREE(o->od_oi_table, sizeof(struct osd_oi *) * count);
828 o->od_oi_table = NULL;
834 void osd_oi_fini(const struct lu_env *env, struct osd_device *o)
838 osd_ost_seq_fini(env, o);
840 if (o->od_oi_table != NULL) {
841 (void) osd_oi_close_table(env, o);
842 OBD_FREE(o->od_oi_table,
843 sizeof(struct osd_oi *) * o->od_oi_count);
844 o->od_oi_table = NULL;
851 int osd_options_init(void)
853 /* osd_oi_count - Default number of OIs, 128 works well for ZFS */
854 if (osd_oi_count == 0 || osd_oi_count > OSD_OI_FID_NR_MAX)
855 osd_oi_count = OSD_OI_FID_NR;
857 if ((osd_oi_count & (osd_oi_count - 1)) != 0) {
858 LCONSOLE_WARN("Round up osd_oi_count %d to power2 %d\n",
859 osd_oi_count, size_roundup_power2(osd_oi_count));
860 osd_oi_count = size_roundup_power2(osd_oi_count);