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) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
32 * lustre/include/lustre_fid.h
34 * Author: Yury Umanets <umka@clusterfs.com>
37 #ifndef __LUSTRE_FID_H
38 #define __LUSTRE_FID_H
44 * http://wiki.lustre.org/index.php/Architecture_-_Interoperability_fids_zfs
45 * describes the FID namespace and interoperability requirements for FIDs.
46 * The important parts of that document are included here for reference.
49 * File IDentifier generated by client from range allocated by the SEQuence
50 * service and stored in struct lu_fid. The FID is composed of three parts:
51 * SEQuence, ObjectID, and VERsion. The SEQ component is a filesystem
52 * unique 64-bit integer, and only one client is ever assigned any SEQ value.
53 * The first 0x400 FID_SEQ_NORMAL [2^33, 2^33 + 0x400] values are reserved
54 * for system use. The OID component is a 32-bit value generated by the
55 * client on a per-SEQ basis to allow creating many unique FIDs without
56 * communication with the server. The VER component is a 32-bit value that
57 * distinguishes between different FID instantiations, such as snapshots or
58 * separate subtrees within the filesystem. FIDs with the same VER field
59 * are considered part of the same namespace.
61 * OLD filesystems are those upgraded from Lustre 1.x that predate FIDs, and
62 * MDTs use 32-bit ldiskfs internal inode/generation numbers (IGIFs), while
63 * OSTs use 64-bit Lustre object IDs and generation numbers.
65 * NEW filesystems are those formatted since the introduction of FIDs.
68 * Inode and Generation In FID, a surrogate FID used to globally identify
69 * an existing object on OLD formatted MDT file system. This would only be
70 * used on MDT0 in a DNE filesystem, because there cannot be more than one
71 * MDT in an OLD formatted filesystem. Belongs to sequence in [12, 2^32 - 1]
72 * range, where inode number is stored in SEQ, and inode generation is in OID.
73 * NOTE: This assumes no more than 2^32-1 inodes exist in the MDT filesystem,
74 * which is the maximum possible for an ldiskfs backend. It also assumes
75 * that the reserved ext3/ext4/ldiskfs inode numbers [0-11] are never visible
76 * to clients, which has always been true.
79 * object ID In FID, a surrogate FID used to globally identify an existing
80 * OST object on OLD formatted OST file system. Belongs to a sequence in
81 * [2^32, 2^33 - 1]. Sequence number is calculated as:
83 * 1 << 32 | (ost_index << 16) | ((objid >> 32) & 0xffff)
85 * that is, SEQ consists of 16-bit OST index, and higher 16 bits of object
86 * ID. The generation of unique SEQ values per OST allows the IDIF FIDs to
87 * be identified in the FLD correctly. The OID field is calculated as:
91 * that is, it consists of lower 32 bits of object ID. For objects within
92 * the IDIF range, object ID extraction will be:
94 * o_id = (fid->f_seq & 0x7fff) << 16 | fid->f_oid;
95 * o_seq = 0; // formerly group number
97 * NOTE: This assumes that no more than 2^48-1 objects have ever been created
98 * on any OST, and that no more than 65535 OSTs are in use. Both are very
99 * reasonable assumptions, i.e. an IDIF can uniquely map all objects assuming
100 * a maximum creation rate of 1M objects per second for a maximum of 9 years,
101 * or combinations thereof.
104 * Surrogate FID used to identify an existing object on OLD formatted OST
105 * filesystem. Belongs to the reserved SEQuence 0, and is used prior to
106 * the introduction of FID-on-OST, at which point IDIF will be used to
107 * identify objects as residing on a specific OST.
110 * For Lustre Log objects the object sequence 1 is used. This is compatible
111 * with both OLD and NEW namespaces, as this SEQ number is in the
112 * ext3/ldiskfs reserved inode range and does not conflict with IGIF
116 * For testing OST IO performance the object sequence 2 is used. This is
117 * compatible with both OLD and NEW namespaces, as this SEQ number is in
118 * the ext3/ldiskfs reserved inode range and does not conflict with IGIF
121 * OST_MDT1 .. OST_MAX
122 * For testing with multiple MDTs the object sequence 3 through 9 is used,
123 * allowing direct mapping of MDTs 1 through 7 respectively, for a total
124 * of 8 MDTs including OST_MDT0. This matches the legacy CMD project "group"
125 * mappings. However, this SEQ range is only for testing prior to any
126 * production DNE release, as the objects in this range conflict across all
127 * OSTs, as the OST index is not part of the FID. For production DNE usage,
128 * OST objects created by MDT1+ will use FID_SEQ_NORMAL FIDs.
130 * DLM OST objid to IDIF mapping
131 * For compatibility with existing OLD OST network protocol structures, the
132 * FID must map onto the o_id and o_seq in a manner that ensures existing
133 * objects are identified consistently for IO, as well as onto the LDLM
134 * namespace to ensure IDIFs there is only a single resource name for any
135 * object in the DLM. The OLD OST object DLM resource mapping is:
137 * resource[] = {o_id, o_seq, 0, 0}; // o_seq == 0 for production releases
139 * The NEW OST object DLM resource mapping is the same for both MDT and OST:
141 * resource[] = {SEQ, OID, VER, HASH};
143 * NOTE: for mapping IDIF values to DLM resource names the o_id may be
144 * larger than the 2^33 reserved sequence numbers for IDIF, so it is possible
145 * for the o_id numbers to overlap FID SEQ numbers in the resource. However,
146 * in all production releases the OLD o_seq field is always zero, and all
147 * valid FID OID values are non-zero, so the lock resources will not collide.
148 * Even so, the MDT and OST resources are also in different LDLM namespaces.
151 #include <libcfs/libcfs.h>
152 #include <uapi/linux/lustre/lustre_fid.h>
153 #include <uapi/linux/lustre/lustre_idl.h>
154 #include <uapi/linux/lustre/lustre_ostid.h>
162 /* Whole sequences space range and zero range definitions */
163 extern const struct lu_seq_range LUSTRE_SEQ_SPACE_RANGE;
164 extern const struct lu_seq_range LUSTRE_SEQ_ZERO_RANGE;
165 extern const struct lu_fid LUSTRE_BFL_FID;
166 extern const struct lu_fid LU_OBF_FID;
167 extern const struct lu_fid LU_LPF_FID;
168 extern const struct lu_fid LU_DOT_LUSTRE_FID;
169 extern const struct lu_fid LU_BACKEND_LPF_FID;
173 * This is how may metadata FIDs may be allocated in one sequence(128k)
175 LUSTRE_METADATA_SEQ_MAX_WIDTH = 0x0000000000020000ULL,
178 * This is how many data FIDs could be allocated in one sequence(4B - 1)
180 LUSTRE_DATA_SEQ_MAX_WIDTH = 0x00000000FFFFFFFFULL,
183 * How many sequences to allocate to a client at once.
185 LUSTRE_SEQ_META_WIDTH = 0x0000000000000001ULL,
188 * seq allocation pool size.
190 LUSTRE_SEQ_BATCH_WIDTH = LUSTRE_SEQ_META_WIDTH * 1000,
193 * This is how many sequences may be in one super-sequence allocated to
196 LUSTRE_SEQ_SUPER_WIDTH = ((1ULL << 30ULL) * LUSTRE_SEQ_META_WIDTH)
199 /** special OID for local objects */
201 /** \see fld_mod_init */
203 /** \see fid_mod_init */
204 FID_SEQ_CTL_OID = 4UL,
205 FID_SEQ_SRV_OID = 5UL,
206 /** \see mdd_mod_init */
207 MDD_ROOT_INDEX_OID = 6UL, /* deprecated in 2.4 */
208 MDD_ORPHAN_OID = 7UL, /* deprecated in 2.4 */
209 MDD_LOV_OBJ_OID = 8UL,
210 MDD_CAPA_KEYS_OID = 9UL,
211 /** \see mdt_mod_init */
212 LAST_RECV_OID = 11UL,
213 OSD_FS_ROOT_OID = 13UL,
214 ACCT_USER_OID = 15UL,
215 ACCT_GROUP_OID = 16UL,
216 LFSCK_BOOKMARK_OID = 17UL,
217 OTABLE_IT_OID = 18UL,
219 REPLY_DATA_OID = 21UL,
220 ACCT_PROJECT_OID = 22UL,
221 INDEX_BACKUP_OID = 4116UL,
222 OFD_LAST_GROUP_OID = 4117UL,
223 LLOG_CATALOGS_OID = 4118UL,
224 MGS_CONFIGS_OID = 4119UL,
225 OFD_HEALTH_CHECK_OID = 4120UL,
226 MDD_LOV_OBJ_OSEQ = 4121UL,
227 LFSCK_NAMESPACE_OID = 4122UL,
228 REMOTE_PARENT_DIR_OID = 4123UL,
229 /* This definition is obsolete
230 * SLAVE_LLOG_CATALOGS_OID = 4124UL,
232 BATCHID_COMMITTED_OID = 4125UL,
235 static inline void lu_local_obj_fid(struct lu_fid *fid, __u32 oid)
237 fid->f_seq = FID_SEQ_LOCAL_FILE;
242 static inline void lu_local_name_obj_fid(struct lu_fid *fid, __u32 oid)
244 fid->f_seq = FID_SEQ_LOCAL_NAME;
249 /* For new FS (>= 2.4), the root FID will be changed to
250 * [FID_SEQ_ROOT:1:0], for existing FS, (upgraded to 2.4),
251 * the root FID will still be IGIF */
252 static inline int fid_is_root(const struct lu_fid *fid)
254 return unlikely((fid_seq(fid) == FID_SEQ_ROOT &&
255 fid_oid(fid) == FID_OID_ROOT));
258 static inline int fid_is_dot_lustre(const struct lu_fid *fid)
260 return unlikely(fid_seq(fid) == FID_SEQ_DOT_LUSTRE &&
261 fid_oid(fid) == FID_OID_DOT_LUSTRE);
264 static inline int fid_is_obf(const struct lu_fid *fid)
266 return unlikely(fid_seq(fid) == FID_SEQ_DOT_LUSTRE &&
267 fid_oid(fid) == FID_OID_DOT_LUSTRE_OBF);
270 static inline int fid_is_otable_it(const struct lu_fid *fid)
272 return unlikely(fid_seq(fid) == FID_SEQ_LOCAL_FILE &&
273 fid_oid(fid) == OTABLE_IT_OID);
276 static inline int fid_oid_is_quota(const struct lu_fid *fid)
278 switch (fid_oid(fid)) {
281 case ACCT_PROJECT_OID:
288 static inline int fid_is_acct(const struct lu_fid *fid)
290 return fid_seq(fid) == FID_SEQ_LOCAL_FILE &&
291 fid_oid_is_quota(fid);
294 static inline int fid_is_quota(const struct lu_fid *fid)
296 return fid_seq(fid) == FID_SEQ_QUOTA ||
297 fid_seq(fid) == FID_SEQ_QUOTA_GLB;
300 static inline int fid_is_name_llog(const struct lu_fid *fid)
302 return fid_seq(fid) == FID_SEQ_LLOG_NAME;
305 static inline int fid_is_namespace_visible(const struct lu_fid *fid)
307 const __u64 seq = fid_seq(fid);
309 /* Here, we cannot distinguish whether the normal FID is for OST
310 * object or not. It is caller's duty to check more if needed. */
311 return (!fid_is_last_id(fid) &&
312 (fid_seq_is_norm(seq) || fid_seq_is_igif(seq))) ||
313 fid_is_root(fid) || fid_seq_is_dot(seq);
316 static inline int fid_seq_in_fldb(__u64 seq)
318 return fid_seq_is_igif(seq) || fid_seq_is_norm(seq) ||
319 fid_seq_is_root(seq) || fid_seq_is_dot(seq);
322 static inline void ost_layout_cpu_to_le(struct ost_layout *dst,
323 const struct ost_layout *src)
325 dst->ol_stripe_size = __cpu_to_le32(src->ol_stripe_size);
326 dst->ol_stripe_count = __cpu_to_le32(src->ol_stripe_count);
327 dst->ol_comp_start = __cpu_to_le64(src->ol_comp_start);
328 dst->ol_comp_end = __cpu_to_le64(src->ol_comp_end);
329 dst->ol_comp_id = __cpu_to_le32(src->ol_comp_id);
332 static inline void ost_layout_le_to_cpu(struct ost_layout *dst,
333 const struct ost_layout *src)
335 dst->ol_stripe_size = __le32_to_cpu(src->ol_stripe_size);
336 dst->ol_stripe_count = __le32_to_cpu(src->ol_stripe_count);
337 dst->ol_comp_start = __le64_to_cpu(src->ol_comp_start);
338 dst->ol_comp_end = __le64_to_cpu(src->ol_comp_end);
339 dst->ol_comp_id = __le32_to_cpu(src->ol_comp_id);
342 /* Both filter_fid_*cpu* functions not currently used */
343 static inline void filter_fid_cpu_to_le(struct filter_fid *dst,
344 const struct filter_fid *src, int size)
346 fid_cpu_to_le(&dst->ff_parent, &src->ff_parent);
348 if (size < sizeof(struct filter_fid)) {
349 memset(&dst->ff_layout, 0, sizeof(dst->ff_layout));
351 ost_layout_cpu_to_le(&dst->ff_layout, &src->ff_layout);
352 dst->ff_layout_version = cpu_to_le32(src->ff_layout_version);
353 dst->ff_range = cpu_to_le32(src->ff_range);
356 /* XXX: Add more if filter_fid is enlarged in the future. */
359 static inline void filter_fid_le_to_cpu(struct filter_fid *dst,
360 const struct filter_fid *src, int size)
362 fid_le_to_cpu(&dst->ff_parent, &src->ff_parent);
364 if (size < sizeof(struct filter_fid)) {
365 memset(&dst->ff_layout, 0, sizeof(dst->ff_layout));
367 ost_layout_le_to_cpu(&dst->ff_layout, &src->ff_layout);
368 dst->ff_layout_version = le32_to_cpu(src->ff_layout_version);
369 dst->ff_range = le32_to_cpu(src->ff_range);
372 /* XXX: Add more if filter_fid is enlarged in the future. */
375 static inline void lu_last_id_fid(struct lu_fid *fid, __u64 seq, __u32 ost_idx)
377 if (fid_seq_is_mdt0(seq)) {
378 fid->f_seq = fid_idif_seq(0, ost_idx);
380 LASSERTF(fid_seq_is_norm(seq) || fid_seq_is_echo(seq) ||
381 fid_seq_is_idif(seq), "%#llx\n", seq);
388 static inline bool fid_is_md_operative(const struct lu_fid *fid)
390 return fid_is_mdt0(fid) || fid_is_igif(fid) ||
391 fid_is_norm(fid) || fid_is_root(fid);
394 /* seq client type */
396 LUSTRE_SEQ_METADATA = 1,
402 LUSTRE_SEQ_CONTROLLER
405 struct lu_server_seq;
407 /* Client sequence manager interface. */
408 struct lu_client_seq {
409 /* Sequence-controller export. */
410 struct obd_export *lcs_exp;
411 struct mutex lcs_mutex;
414 * Range of allowed for allocation sequeces. When using lu_client_seq on
415 * clients, this contains meta-sequence range. And for servers this
416 * contains super-sequence range.
418 struct lu_seq_range lcs_space;
420 /* Seq related debugfs */
421 struct dentry *lcs_debugfs_entry;
423 /* This holds last allocated fid in last obtained seq */
424 struct lu_fid lcs_fid;
426 /* LUSTRE_SEQ_METADATA or LUSTRE_SEQ_DATA */
427 enum lu_cli_type lcs_type;
430 * Service uuid, passed from MDT + seq name to form unique seq name to
431 * use it with debugfs.
436 * Sequence width, that is how many objects may be allocated in one
437 * sequence. Default value for it is LUSTRE_SEQ_MAX_WIDTH.
441 /* Seq-server for direct talking */
442 struct lu_server_seq *lcs_srv;
444 /* wait queue for fid allocation and update indicator */
445 wait_queue_head_t lcs_waitq;
449 /* server sequence manager interface */
450 struct lu_server_seq {
451 /* Available sequences space */
452 struct lu_seq_range lss_space;
454 /* keeps highwater in lsr_end for seq allocation algorithm */
455 struct lu_seq_range lss_lowater_set;
456 struct lu_seq_range lss_hiwater_set;
459 * Device for server side seq manager needs (saving sequences to backing
462 struct dt_device *lss_dev;
464 /* /seq file object device */
465 struct dt_object *lss_obj;
467 /* Seq related debugfs */
468 struct dentry *lss_debugfs_entry;
470 /* LUSTRE_SEQ_SERVER or LUSTRE_SEQ_CONTROLLER */
471 enum lu_mgr_type lss_type;
473 /* Client interface to request controller */
474 struct lu_client_seq *lss_cli;
476 /* Mutex for protecting allocation */
477 struct mutex lss_mutex;
480 * Service uuid, passed from MDT + seq name to form unique seq name to
481 * use it with debugfs.
486 * Allocation chunks for super and meta sequences. Default values are
487 * LUSTRE_SEQ_SUPER_WIDTH and LUSTRE_SEQ_META_WIDTH.
492 * minimum lss_alloc_set size that should be allocated from
497 /* sync is needed for update operation */
501 * Pointer to site object, required to access site fld.
503 struct seq_server_site *lss_site;
506 struct seq_server_site {
507 struct lu_site *ss_lu;
509 * mds number of this site.
513 * Fid location database
515 struct lu_server_fld *ss_server_fld;
516 struct lu_client_fld *ss_client_fld;
521 struct lu_server_seq *ss_server_seq;
524 * Controller Seq Manager
526 struct lu_server_seq *ss_control_seq;
527 struct obd_export *ss_control_exp;
532 struct lu_client_seq *ss_client_seq;
537 int seq_server_init(const struct lu_env *env,
538 struct lu_server_seq *seq,
539 struct dt_device *dev,
541 enum lu_mgr_type type,
542 struct seq_server_site *ss);
544 void seq_server_fini(struct lu_server_seq *seq,
545 const struct lu_env *env);
547 int seq_server_alloc_super(struct lu_server_seq *seq,
548 struct lu_seq_range *out,
549 const struct lu_env *env);
551 int seq_server_alloc_meta(struct lu_server_seq *seq,
552 struct lu_seq_range *out,
553 const struct lu_env *env);
555 int seq_server_set_cli(const struct lu_env *env,
556 struct lu_server_seq *seq,
557 struct lu_client_seq *cli);
559 int seq_server_check_and_alloc_super(const struct lu_env *env,
560 struct lu_server_seq *seq);
562 int seq_client_init(struct lu_client_seq *seq,
563 struct obd_export *exp,
564 enum lu_cli_type type,
566 struct lu_server_seq *srv);
568 void seq_client_fini(struct lu_client_seq *seq);
570 void seq_client_flush(struct lu_client_seq *seq);
572 int seq_client_alloc_fid(const struct lu_env *env, struct lu_client_seq *seq,
574 int seq_client_get_seq(const struct lu_env *env, struct lu_client_seq *seq,
576 int seq_site_fini(const struct lu_env *env, struct seq_server_site *ss);
577 /* Fids common stuff */
578 int fid_is_local(const struct lu_env *env,
579 struct lu_site *site, const struct lu_fid *fid);
582 int client_fid_init(struct obd_device *obd, struct obd_export *exp,
583 enum lu_cli_type type);
584 int client_fid_fini(struct obd_device *obd);
588 struct ldlm_namespace;
591 * Build (DLM) resource name from FID.
593 * NOTE: until Lustre 1.8.7/2.1.1 the fid_ver() was packed into name[2],
594 * but was moved into name[1] along with the OID to avoid consuming the
595 * renaming name[2,3] fields that need to be used for the quota identifier.
598 fid_build_reg_res_name(const struct lu_fid *fid, struct ldlm_res_id *res)
600 memset(res, 0, sizeof(*res));
601 res->name[LUSTRE_RES_ID_SEQ_OFF] = fid_seq(fid);
602 res->name[LUSTRE_RES_ID_VER_OID_OFF] = fid_ver_oid(fid);
606 * Return true if resource is for object identified by FID.
608 static inline int fid_res_name_eq(const struct lu_fid *fid,
609 const struct ldlm_res_id *res)
611 return res->name[LUSTRE_RES_ID_SEQ_OFF] == fid_seq(fid) &&
612 res->name[LUSTRE_RES_ID_VER_OID_OFF] == fid_ver_oid(fid);
616 * Extract FID from LDLM resource. Reverse of fid_build_reg_res_name().
619 fid_extract_from_res_name(struct lu_fid *fid, const struct ldlm_res_id *res)
621 fid->f_seq = res->name[LUSTRE_RES_ID_SEQ_OFF];
622 fid->f_oid = (__u32)(res->name[LUSTRE_RES_ID_VER_OID_OFF]);
623 fid->f_ver = (__u32)(res->name[LUSTRE_RES_ID_VER_OID_OFF] >> 32);
624 LASSERT(fid_res_name_eq(fid, res));
628 * Build (DLM) resource identifier from global quota FID and quota ID.
631 fid_build_quota_res_name(const struct lu_fid *glb_fid, union lquota_id *qid,
632 struct ldlm_res_id *res)
634 fid_build_reg_res_name(glb_fid, res);
635 res->name[LUSTRE_RES_ID_QUOTA_SEQ_OFF] = fid_seq(&qid->qid_fid);
636 res->name[LUSTRE_RES_ID_QUOTA_VER_OID_OFF] = fid_ver_oid(&qid->qid_fid);
640 * Extract global FID and quota ID from resource name
642 static inline void fid_extract_from_quota_res(struct lu_fid *glb_fid,
643 union lquota_id *qid,
644 const struct ldlm_res_id *res)
646 fid_extract_from_res_name(glb_fid, res);
647 qid->qid_fid.f_seq = res->name[LUSTRE_RES_ID_QUOTA_SEQ_OFF];
648 qid->qid_fid.f_oid = (__u32)res->name[LUSTRE_RES_ID_QUOTA_VER_OID_OFF];
650 (__u32)(res->name[LUSTRE_RES_ID_QUOTA_VER_OID_OFF] >> 32);
654 fid_build_pdo_res_name(const struct lu_fid *fid, unsigned int hash,
655 struct ldlm_res_id *res)
657 fid_build_reg_res_name(fid, res);
658 res->name[LUSTRE_RES_ID_HSH_OFF] = hash;
662 * Build DLM resource name from object id & seq, which will be removed
663 * finally, when we replace ost_id with FID in data stack.
665 * Currently, resid from the old client, whose res[0] = object_id,
666 * res[1] = object_seq, is just oposite with Metatdata
667 * resid, where, res[0] = fid->f_seq, res[1] = fid->f_oid.
668 * To unifiy the resid identification, we will reverse the data
669 * resid to keep it same with Metadata resid, i.e.
671 * For resid from the old client,
672 * res[0] = objid, res[1] = 0, still keep the original order,
676 * res will be built from normal FID directly, i.e. res[0] = f_seq,
677 * res[1] = f_oid + f_ver.
679 static inline void ostid_build_res_name(const struct ost_id *oi,
680 struct ldlm_res_id *name)
682 memset(name, 0, sizeof *name);
683 if (fid_seq_is_mdt0(ostid_seq(oi))) {
684 name->name[LUSTRE_RES_ID_SEQ_OFF] = ostid_id(oi);
685 name->name[LUSTRE_RES_ID_VER_OID_OFF] = ostid_seq(oi);
687 fid_build_reg_res_name(&oi->oi_fid, name);
692 * Return true if the resource is for the object identified by this id & group.
694 static inline bool ostid_res_name_eq(const struct ost_id *oi,
695 const struct ldlm_res_id *name)
697 /* Note: it is just a trick here to save some effort, probably the
698 * correct way would be turn them into the FID and compare */
699 if (fid_seq_is_mdt0(ostid_seq(oi))) {
700 return name->name[LUSTRE_RES_ID_SEQ_OFF] == ostid_id(oi) &&
701 name->name[LUSTRE_RES_ID_VER_OID_OFF] == ostid_seq(oi);
703 return name->name[LUSTRE_RES_ID_SEQ_OFF] == ostid_seq(oi) &&
704 name->name[LUSTRE_RES_ID_VER_OID_OFF] == ostid_id(oi);
709 * Note: we need check oi_seq to decide where to set oi_id,
710 * so oi_seq should always be set ahead of oi_id.
712 static inline int ostid_set_id(struct ost_id *oi, __u64 oid)
714 if (fid_seq_is_mdt0(oi->oi.oi_seq)) {
715 if (oid >= IDIF_MAX_OID)
718 } else if (fid_is_idif(&oi->oi_fid)) {
719 if (oid >= IDIF_MAX_OID)
721 oi->oi_fid.f_seq = fid_idif_seq(oid,
722 fid_idif_ost_idx(&oi->oi_fid));
723 oi->oi_fid.f_oid = oid;
724 oi->oi_fid.f_ver = oid >> 48;
726 if (oid >= OBIF_MAX_OID)
728 oi->oi_fid.f_oid = oid;
733 /* pack any OST FID into an ostid (id/seq) for the wire/disk */
734 static inline int fid_to_ostid(const struct lu_fid *fid, struct ost_id *ostid)
738 if (fid_seq_is_igif(fid->f_seq))
741 if (fid_is_idif(fid)) {
742 ostid_set_seq_mdt0(ostid);
743 rc = ostid_set_id(ostid, fid_idif_id(fid_seq(fid),
744 fid_oid(fid), fid_ver(fid)));
746 ostid->oi_fid = *fid;
752 /* The same as osc_build_res_name() */
753 static inline void ost_fid_build_resid(const struct lu_fid *fid,
754 struct ldlm_res_id *resname)
756 if (fid_is_mdt0(fid) || fid_is_idif(fid)) {
758 oi.oi.oi_id = 0; /* gcc 4.7.2 complains otherwise */
759 if (fid_to_ostid(fid, &oi) != 0)
761 ostid_build_res_name(&oi, resname);
763 fid_build_reg_res_name(fid, resname);
767 static inline void ost_fid_from_resid(struct lu_fid *fid,
768 const struct ldlm_res_id *name,
771 if (fid_seq_is_mdt0(name->name[LUSTRE_RES_ID_VER_OID_OFF])) {
775 memset(&oi, 0, sizeof(oi));
776 ostid_set_seq(&oi, name->name[LUSTRE_RES_ID_VER_OID_OFF]);
777 if (ostid_set_id(&oi, name->name[LUSTRE_RES_ID_SEQ_OFF])) {
778 CERROR("Bad %llu to set " DOSTID "\n",
779 name->name[LUSTRE_RES_ID_SEQ_OFF], POSTID(&oi));
781 ostid_to_fid(fid, &oi, ost_idx);
784 fid_extract_from_res_name(fid, name);
789 * Flatten 128-bit FID values into a 64-bit value for use as an inode number.
790 * For non-IGIF FIDs this starts just over 2^32, and continues without
791 * conflict until 2^64, at which point we wrap the high 24 bits of the SEQ
792 * into the range where there may not be many OID values in use, to minimize
793 * the risk of conflict.
795 * Suppose LUSTRE_SEQ_MAX_WIDTH less than (1 << 24) which is currently true,
796 * the time between re-used inode numbers is very long - 2^40 SEQ numbers,
797 * or about 2^40 client mounts, if clients create less than 2^24 files/mount.
799 static inline __u64 fid_flatten(const struct lu_fid *fid)
804 if (fid_is_igif(fid)) {
805 ino = lu_igif_ino(fid);
811 ino = (seq << 24) + ((seq >> 24) & 0xffffff0000ULL) + fid_oid(fid);
813 return ino ?: fid_oid(fid);
816 static inline __u32 fid_hash(const struct lu_fid *f, int bits)
818 /* all objects with same id and different versions will belong to same
819 * collisions list. */
820 return hash_long(fid_flatten(f), bits);
824 * map fid to 32 bit value for ino on 32bit systems. */
825 static inline __u32 fid_flatten32(const struct lu_fid *fid)
830 if (fid_is_igif(fid)) {
831 ino = lu_igif_ino(fid);
835 seq = fid_seq(fid) - FID_SEQ_START;
837 /* Map the high bits of the OID into higher bits of the inode number so
838 * that inodes generated at about the same time have a reduced chance
839 * of collisions. This will give a period of 2^12 = 1024 unique clients
840 * (from SEQ) and up to min(LUSTRE_SEQ_MAX_WIDTH, 2^20) = 128k objects
841 * (from OID), or up to 128M inodes without collisions for new files. */
842 ino = ((seq & 0x000fffffULL) << 12) + ((seq >> 8) & 0xfffff000) +
843 (seq >> (64 - (40-8)) & 0xffffff00) +
844 (fid_oid(fid) & 0xff000fff) + ((fid_oid(fid) & 0x00fff000) << 8);
846 return ino ?: fid_oid(fid);
850 lu_fid_diff(const struct lu_fid *fid1, const struct lu_fid *fid2)
852 LASSERTF(fid_seq(fid1) == fid_seq(fid2), "fid1:"DFID", fid2:"DFID"\n",
853 PFID(fid1), PFID(fid2));
855 if (fid_is_idif(fid1) && fid_is_idif(fid2))
856 return fid_idif_id(fid1->f_seq, fid1->f_oid, fid1->f_ver) -
857 fid_idif_id(fid2->f_seq, fid2->f_oid, fid2->f_ver);
859 return fid_oid(fid1) - fid_oid(fid2);
862 static inline int fid_set_id(struct lu_fid *fid, u64 oid)
864 if (unlikely(fid_seq_is_igif(fid->f_seq))) {
865 CERROR("bad IGIF, "DFID"\n", PFID(fid));
869 if (fid_is_idif(fid)) {
870 if (oid >= IDIF_MAX_OID) {
871 CERROR("Too large OID %#llx to set IDIF "DFID"\n",
872 (unsigned long long)oid, PFID(fid));
875 fid->f_seq = fid_idif_seq(oid, fid_idif_ost_idx(fid));
877 fid->f_ver = oid >> 48;
879 if (oid > OBIF_MAX_OID) {
880 CERROR("Too large OID %#llx to set REG "DFID"\n",
881 (unsigned long long)oid, PFID(fid));
889 #define LUSTRE_SEQ_SRV_NAME "seq_srv"
890 #define LUSTRE_SEQ_CTL_NAME "seq_ctl"
892 /* Range common stuff */
894 range_cpu_to_le(struct lu_seq_range *dst, const struct lu_seq_range *src)
896 dst->lsr_start = cpu_to_le64(src->lsr_start);
897 dst->lsr_end = cpu_to_le64(src->lsr_end);
898 dst->lsr_index = cpu_to_le32(src->lsr_index);
899 dst->lsr_flags = cpu_to_le32(src->lsr_flags);
903 range_le_to_cpu(struct lu_seq_range *dst, const struct lu_seq_range *src)
905 dst->lsr_start = le64_to_cpu(src->lsr_start);
906 dst->lsr_end = le64_to_cpu(src->lsr_end);
907 dst->lsr_index = le32_to_cpu(src->lsr_index);
908 dst->lsr_flags = le32_to_cpu(src->lsr_flags);
912 range_cpu_to_be(struct lu_seq_range *dst, const struct lu_seq_range *src)
914 dst->lsr_start = cpu_to_be64(src->lsr_start);
915 dst->lsr_end = cpu_to_be64(src->lsr_end);
916 dst->lsr_index = cpu_to_be32(src->lsr_index);
917 dst->lsr_flags = cpu_to_be32(src->lsr_flags);
921 range_be_to_cpu(struct lu_seq_range *dst, const struct lu_seq_range *src)
923 dst->lsr_start = be64_to_cpu(src->lsr_start);
924 dst->lsr_end = be64_to_cpu(src->lsr_end);
925 dst->lsr_index = be32_to_cpu(src->lsr_index);
926 dst->lsr_flags = be32_to_cpu(src->lsr_flags);
929 static inline void range_array_cpu_to_le(struct lu_seq_range_array *dst,
930 const struct lu_seq_range_array *src)
934 for (i = 0; i < src->lsra_count; i++)
935 range_cpu_to_le(&dst->lsra_lsr[i], &src->lsra_lsr[i]);
937 dst->lsra_count = cpu_to_le32(src->lsra_count);
940 static inline void range_array_le_to_cpu(struct lu_seq_range_array *dst,
941 const struct lu_seq_range_array *src)
945 dst->lsra_count = le32_to_cpu(src->lsra_count);
946 for (i = 0; i < dst->lsra_count; i++)
947 range_le_to_cpu(&dst->lsra_lsr[i], &src->lsra_lsr[i]);
952 #endif /* __LUSTRE_FID_H */