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) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2012, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/include/lustre_fid.h
38 * Author: Yury Umanets <umka@clusterfs.com>
48 * http://wiki.lustre.org/index.php/Architecture_-_Interoperability_fids_zfs
49 * describes the FID namespace and interoperability requirements for FIDs.
50 * The important parts of that document are included here for reference.
53 * File IDentifier generated by client from range allocated by the SEQuence
54 * service and stored in struct lu_fid. The FID is composed of three parts:
55 * SEQuence, ObjectID, and VERsion. The SEQ component is a filesystem
56 * unique 64-bit integer, and only one client is ever assigned any SEQ value.
57 * The first 0x400 FID_SEQ_NORMAL [2^33, 2^33 + 0x400] values are reserved
58 * for system use. The OID component is a 32-bit value generated by the
59 * client on a per-SEQ basis to allow creating many unique FIDs without
60 * communication with the server. The VER component is a 32-bit value that
61 * distinguishes between different FID instantiations, such as snapshots or
62 * separate subtrees within the filesystem. FIDs with the same VER field
63 * are considered part of the same namespace.
65 * OLD filesystems are those upgraded from Lustre 1.x that predate FIDs, and
66 * MDTs use 32-bit ldiskfs internal inode/generation numbers (IGIFs), while
67 * OSTs use 64-bit Lustre object IDs and generation numbers.
69 * NEW filesystems are those formatted since the introduction of FIDs.
72 * Inode and Generation In FID, a surrogate FID used to globally identify
73 * an existing object on OLD formatted MDT file system. This would only be
74 * used on MDT0 in a DNE filesystem, because there cannot be more than one
75 * MDT in an OLD formatted filesystem. Belongs to sequence in [12, 2^32 - 1]
76 * range, where inode number is stored in SEQ, and inode generation is in OID.
77 * NOTE: This assumes no more than 2^32-1 inodes exist in the MDT filesystem,
78 * which is the maximum possible for an ldiskfs backend. It also assumes
79 * that the reserved ext3/ext4/ldiskfs inode numbers [0-11] are never visible
80 * to clients, which has always been true.
83 * object ID In FID, a surrogate FID used to globally identify an existing
84 * OST object on OLD formatted OST file system. Belongs to a sequence in
85 * [2^32, 2^33 - 1]. Sequence number is calculated as:
87 * 1 << 32 | (ost_index << 16) | ((objid >> 32) & 0xffff)
89 * that is, SEQ consists of 16-bit OST index, and higher 16 bits of object
90 * ID. The generation of unique SEQ values per OST allows the IDIF FIDs to
91 * be identified in the FLD correctly. The OID field is calculated as:
95 * that is, it consists of lower 32 bits of object ID. For objects within
96 * the IDIF range, object ID extraction will be:
98 * o_id = (fid->f_seq & 0x7fff) << 16 | fid->f_oid;
99 * o_seq = 0; // formerly group number
101 * NOTE: This assumes that no more than 2^48-1 objects have ever been created
102 * on any OST, and that no more than 65535 OSTs are in use. Both are very
103 * reasonable assumptions, i.e. an IDIF can uniquely map all objects assuming
104 * a maximum creation rate of 1M objects per second for a maximum of 9 years,
105 * or combinations thereof.
108 * Surrogate FID used to identify an existing object on OLD formatted OST
109 * filesystem. Belongs to the reserved SEQuence 0, and is used prior to
110 * the introduction of FID-on-OST, at which point IDIF will be used to
111 * identify objects as residing on a specific OST.
114 * For Lustre Log objects the object sequence 1 is used. This is compatible
115 * with both OLD and NEW namespaces, as this SEQ number is in the
116 * ext3/ldiskfs reserved inode range and does not conflict with IGIF
120 * For testing OST IO performance the object sequence 2 is used. This is
121 * compatible with both OLD and NEW namespaces, as this SEQ number is in
122 * the ext3/ldiskfs reserved inode range and does not conflict with IGIF
125 * OST_MDT1 .. OST_MAX
126 * For testing with multiple MDTs the object sequence 3 through 9 is used,
127 * allowing direct mapping of MDTs 1 through 7 respectively, for a total
128 * of 8 MDTs including OST_MDT0. This matches the legacy CMD project "group"
129 * mappings. However, this SEQ range is only for testing prior to any
130 * production DNE release, as the objects in this range conflict across all
131 * OSTs, as the OST index is not part of the FID. For production DNE usage,
132 * OST objects created by MDT1+ will use FID_SEQ_NORMAL FIDs.
134 * DLM OST objid to IDIF mapping
135 * For compatibility with existing OLD OST network protocol structures, the
136 * FID must map onto the o_id and o_seq in a manner that ensures existing
137 * objects are identified consistently for IO, as well as onto the LDLM
138 * namespace to ensure IDIFs there is only a single resource name for any
139 * object in the DLM. The OLD OST object DLM resource mapping is:
141 * resource[] = {o_id, o_seq, 0, 0}; // o_seq == 0 for production releases
143 * The NEW OST object DLM resource mapping is the same for both MDT and OST:
145 * resource[] = {SEQ, OID, VER, HASH};
147 * NOTE: for mapping IDIF values to DLM resource names the o_id may be
148 * larger than the 2^33 reserved sequence numbers for IDIF, so it is possible
149 * for the o_id numbers to overlap FID SEQ numbers in the resource. However,
150 * in all production releases the OLD o_seq field is always zero, and all
151 * valid FID OID values are non-zero, so the lock resources will not collide.
152 * Even so, the MDT and OST resources are also in different LDLM namespaces.
155 #include <libcfs/libcfs.h>
156 #include <lustre/lustre_idl.h>
157 #include <lustre_req_layout.h>
158 #include <lustre_mdt.h>
164 /* Whole sequences space range and zero range definitions */
165 extern const struct lu_seq_range LUSTRE_SEQ_SPACE_RANGE;
166 extern const struct lu_seq_range LUSTRE_SEQ_ZERO_RANGE;
167 extern const struct lu_fid LUSTRE_BFL_FID;
168 extern const struct lu_fid LU_OBF_FID;
169 extern const struct lu_fid LU_DOT_LUSTRE_FID;
173 * This is how may FIDs may be allocated in one sequence(128k)
175 LUSTRE_SEQ_MAX_WIDTH = 0x0000000000020000ULL,
178 * How many sequences to allocate to a client at once.
180 LUSTRE_SEQ_META_WIDTH = 0x0000000000000001ULL,
183 * seq allocation pool size.
185 LUSTRE_SEQ_BATCH_WIDTH = LUSTRE_SEQ_META_WIDTH * 1000,
188 * This is how many sequences may be in one super-sequence allocated to
191 LUSTRE_SEQ_SUPER_WIDTH = ((1ULL << 30ULL) * LUSTRE_SEQ_META_WIDTH)
195 /** 2^6 FIDs for OI containers */
196 OSD_OI_FID_OID_BITS = 6,
197 /** reserve enough FIDs in case we want more in the future */
198 OSD_OI_FID_OID_BITS_MAX = 10,
201 /** special OID for local objects */
203 /** \see fld_mod_init */
205 /** \see fid_mod_init */
206 FID_SEQ_CTL_OID = 4UL,
207 FID_SEQ_SRV_OID = 5UL,
208 /** \see mdd_mod_init */
209 MDD_ROOT_INDEX_OID = 6UL,
210 MDD_ORPHAN_OID = 7UL,
211 MDD_LOV_OBJ_OID = 8UL,
212 MDD_CAPA_KEYS_OID = 9UL,
213 /** \see mdt_mod_init */
214 MDT_LAST_RECV_OID = 11UL,
215 OSD_FS_ROOT_OID = 13UL,
216 ACCT_USER_OID = 15UL,
217 ACCT_GROUP_OID = 16UL,
218 LFSCK_BOOKMARK_OID = 17UL,
219 OTABLE_IT_OID = 18UL,
220 OFD_LAST_RECV_OID = 19UL,
221 /* These two definitions are obsolete
222 * OFD_GROUP0_LAST_OID = 20UL,
223 * OFD_GROUP4K_LAST_OID = 20UL+4096,
225 OFD_LAST_GROUP_OID = 4117UL,
226 LLOG_CATALOGS_OID = 4118UL,
227 MGS_CONFIGS_OID = 4119UL,
228 OFD_HEALTH_CHECK_OID = 4120UL,
229 MDD_LOV_OBJ_OSEQ = 4121UL,
232 static inline void lu_local_obj_fid(struct lu_fid *fid, __u32 oid)
234 fid->f_seq = FID_SEQ_LOCAL_FILE;
239 static inline void lu_local_name_obj_fid(struct lu_fid *fid, __u32 oid)
241 fid->f_seq = FID_SEQ_LOCAL_NAME;
246 static inline int fid_is_otable_it(const struct lu_fid *fid)
248 return unlikely(fid_seq(fid) == FID_SEQ_LOCAL_FILE &&
249 fid_oid(fid) == OTABLE_IT_OID);
252 static inline int fid_is_acct(const struct lu_fid *fid)
254 return fid_seq(fid) == FID_SEQ_LOCAL_FILE &&
255 (fid_oid(fid) == ACCT_USER_OID ||
256 fid_oid(fid) == ACCT_GROUP_OID);
259 static inline int fid_is_quota(const struct lu_fid *fid)
261 return fid_seq(fid) == FID_SEQ_QUOTA ||
262 fid_seq(fid) == FID_SEQ_QUOTA_GLB;
265 static inline void lu_last_id_fid(struct lu_fid *fid, __u64 seq)
267 if (fid_seq_is_mdt0(seq)) {
268 fid->f_seq = fid_idif_seq(0, 0);
270 LASSERTF(fid_seq_is_norm(seq) || fid_seq_is_echo(seq) ||
271 fid_seq_is_idif(seq), LPX64"\n", seq);
280 LUSTRE_SEQ_CONTROLLER
288 struct lu_server_seq;
290 /* Client sequence manager interface. */
291 struct lu_client_seq {
292 /* Sequence-controller export. */
293 struct obd_export *lcs_exp;
294 struct mutex lcs_mutex;
297 * Range of allowed for allocation sequeces. When using lu_client_seq on
298 * clients, this contains meta-sequence range. And for servers this
299 * contains super-sequence range.
301 struct lu_seq_range lcs_space;
303 /* Seq related proc */
304 cfs_proc_dir_entry_t *lcs_proc_dir;
306 /* This holds last allocated fid in last obtained seq */
307 struct lu_fid lcs_fid;
309 /* LUSTRE_SEQ_METADATA or LUSTRE_SEQ_DATA */
310 enum lu_cli_type lcs_type;
313 * Service uuid, passed from MDT + seq name to form unique seq name to
314 * use it with procfs.
319 * Sequence width, that is how many objects may be allocated in one
320 * sequence. Default value for it is LUSTRE_SEQ_MAX_WIDTH.
324 /* Seq-server for direct talking */
325 struct lu_server_seq *lcs_srv;
327 /* wait queue for fid allocation and update indicator */
328 cfs_waitq_t lcs_waitq;
332 /* server sequence manager interface */
333 struct lu_server_seq {
334 /* Available sequences space */
335 struct lu_seq_range lss_space;
337 /* keeps highwater in lsr_end for seq allocation algorithm */
338 struct lu_seq_range lss_lowater_set;
339 struct lu_seq_range lss_hiwater_set;
342 * Device for server side seq manager needs (saving sequences to backing
345 struct dt_device *lss_dev;
347 /* /seq file object device */
348 struct dt_object *lss_obj;
350 /* Seq related proc */
351 cfs_proc_dir_entry_t *lss_proc_dir;
353 /* LUSTRE_SEQ_SERVER or LUSTRE_SEQ_CONTROLLER */
354 enum lu_mgr_type lss_type;
356 /* Client interafce to request controller */
357 struct lu_client_seq *lss_cli;
359 /* Mutex for protecting allocation */
360 struct mutex lss_mutex;
363 * Service uuid, passed from MDT + seq name to form unique seq name to
364 * use it with procfs.
369 * Allocation chunks for super and meta sequences. Default values are
370 * LUSTRE_SEQ_SUPER_WIDTH and LUSTRE_SEQ_META_WIDTH.
375 * minimum lss_alloc_set size that should be allocated from
380 /* sync is needed for update operation */
383 * Pointer to site object, required to access site fld.
385 struct md_site *lss_site;
388 int seq_query(struct com_thread_info *info);
391 int seq_server_init(struct lu_server_seq *seq,
392 struct dt_device *dev,
394 enum lu_mgr_type type,
396 const struct lu_env *env);
398 void seq_server_fini(struct lu_server_seq *seq,
399 const struct lu_env *env);
401 int seq_server_alloc_super(struct lu_server_seq *seq,
402 struct lu_seq_range *out,
403 const struct lu_env *env);
405 int seq_server_alloc_meta(struct lu_server_seq *seq,
406 struct lu_seq_range *out,
407 const struct lu_env *env);
409 int seq_server_set_cli(struct lu_server_seq *seq,
410 struct lu_client_seq *cli,
411 const struct lu_env *env);
414 int seq_client_init(struct lu_client_seq *seq,
415 struct obd_export *exp,
416 enum lu_cli_type type,
418 struct lu_server_seq *srv);
420 void seq_client_fini(struct lu_client_seq *seq);
422 void seq_client_flush(struct lu_client_seq *seq);
424 int seq_client_alloc_fid(const struct lu_env *env, struct lu_client_seq *seq,
426 int seq_client_get_seq(const struct lu_env *env, struct lu_client_seq *seq,
429 /* Fids common stuff */
430 int fid_is_local(const struct lu_env *env,
431 struct lu_site *site, const struct lu_fid *fid);
435 struct ldlm_namespace;
438 * Build (DLM) resource name from FID.
440 * NOTE: until Lustre 1.8.7/2.1.1 the fid_ver() was packed into name[2],
441 * but was moved into name[1] along with the OID to avoid consuming the
442 * renaming name[2,3] fields that need to be used for the quota identifier.
444 static inline struct ldlm_res_id *
445 fid_build_reg_res_name(const struct lu_fid *f,
446 struct ldlm_res_id *name)
448 memset(name, 0, sizeof *name);
449 name->name[LUSTRE_RES_ID_SEQ_OFF] = fid_seq(f);
450 name->name[LUSTRE_RES_ID_VER_OID_OFF] = fid_ver_oid(f);
455 * Build (DLM) resource identifier from global quota FID and quota ID.
457 static inline struct ldlm_res_id *
458 fid_build_quota_resid(const struct lu_fid *glb_fid, union lquota_id *qid,
459 struct ldlm_res_id *res)
461 fid_build_reg_res_name(glb_fid, res);
462 res->name[LUSTRE_RES_ID_QUOTA_SEQ_OFF] = fid_seq(&qid->qid_fid);
463 res->name[LUSTRE_RES_ID_QUOTA_VER_OID_OFF] = fid_ver_oid(&qid->qid_fid);
468 * Extract global FID and quota ID from resource name
470 static inline void fid_extract_quota_resid(struct ldlm_res_id *res,
471 struct lu_fid *glb_fid,
472 union lquota_id *qid)
474 glb_fid->f_seq = res->name[LUSTRE_RES_ID_SEQ_OFF];
475 glb_fid->f_oid = (__u32)res->name[LUSTRE_RES_ID_VER_OID_OFF];
476 glb_fid->f_ver = (__u32)(res->name[LUSTRE_RES_ID_VER_OID_OFF] >> 32);
478 qid->qid_fid.f_seq = res->name[LUSTRE_RES_ID_QUOTA_SEQ_OFF];
479 qid->qid_fid.f_oid = (__u32)res->name[LUSTRE_RES_ID_QUOTA_VER_OID_OFF];
481 (__u32)(res->name[LUSTRE_RES_ID_QUOTA_VER_OID_OFF] >> 32);
485 * Return true if resource is for object identified by fid.
487 static inline int fid_res_name_eq(const struct lu_fid *f,
488 const struct ldlm_res_id *name)
490 return name->name[LUSTRE_RES_ID_SEQ_OFF] == fid_seq(f) &&
491 name->name[LUSTRE_RES_ID_VER_OID_OFF] == fid_ver_oid(f);
495 static inline struct ldlm_res_id *
496 fid_build_pdo_res_name(const struct lu_fid *f,
498 struct ldlm_res_id *name)
500 fid_build_reg_res_name(f, name);
501 name->name[LUSTRE_RES_ID_HSH_OFF] = hash;
507 * Flatten 128-bit FID values into a 64-bit value for use as an inode number.
508 * For non-IGIF FIDs this starts just over 2^32, and continues without
509 * conflict until 2^64, at which point we wrap the high 24 bits of the SEQ
510 * into the range where there may not be many OID values in use, to minimize
511 * the risk of conflict.
513 * Suppose LUSTRE_SEQ_MAX_WIDTH less than (1 << 24) which is currently true,
514 * the time between re-used inode numbers is very long - 2^40 SEQ numbers,
515 * or about 2^40 client mounts, if clients create less than 2^24 files/mount.
517 static inline __u64 fid_flatten(const struct lu_fid *fid)
522 if (fid_is_igif(fid)) {
523 ino = lu_igif_ino(fid);
529 ino = (seq << 24) + ((seq >> 24) & 0xffffff0000ULL) + fid_oid(fid);
531 RETURN(ino ? ino : fid_oid(fid));
534 static inline __u32 fid_hash(const struct lu_fid *f, int bits)
536 /* all objects with same id and different versions will belong to same
537 * collisions list. */
538 return cfs_hash_long(fid_flatten(f), bits);
542 * map fid to 32 bit value for ino on 32bit systems. */
543 static inline __u32 fid_flatten32(const struct lu_fid *fid)
548 if (fid_is_igif(fid)) {
549 ino = lu_igif_ino(fid);
553 seq = fid_seq(fid) - FID_SEQ_START;
555 /* Map the high bits of the OID into higher bits of the inode number so
556 * that inodes generated at about the same time have a reduced chance
557 * of collisions. This will give a period of 2^12 = 1024 unique clients
558 * (from SEQ) and up to min(LUSTRE_SEQ_MAX_WIDTH, 2^20) = 128k objects
559 * (from OID), or up to 128M inodes without collisions for new files. */
560 ino = ((seq & 0x000fffffULL) << 12) + ((seq >> 8) & 0xfffff000) +
561 (seq >> (64 - (40-8)) & 0xffffff00) +
562 (fid_oid(fid) & 0xff000fff) + ((fid_oid(fid) & 0x00fff000) << 8);
564 RETURN(ino ? ino : fid_oid(fid));
567 #define LUSTRE_SEQ_SRV_NAME "seq_srv"
568 #define LUSTRE_SEQ_CTL_NAME "seq_ctl"
570 /* Range common stuff */
571 static inline void range_cpu_to_le(struct lu_seq_range *dst, const struct lu_seq_range *src)
573 dst->lsr_start = cpu_to_le64(src->lsr_start);
574 dst->lsr_end = cpu_to_le64(src->lsr_end);
575 dst->lsr_index = cpu_to_le32(src->lsr_index);
576 dst->lsr_flags = cpu_to_le32(src->lsr_flags);
579 static inline void range_le_to_cpu(struct lu_seq_range *dst, const struct lu_seq_range *src)
581 dst->lsr_start = le64_to_cpu(src->lsr_start);
582 dst->lsr_end = le64_to_cpu(src->lsr_end);
583 dst->lsr_index = le32_to_cpu(src->lsr_index);
584 dst->lsr_flags = le32_to_cpu(src->lsr_flags);
587 static inline void range_cpu_to_be(struct lu_seq_range *dst, const struct lu_seq_range *src)
589 dst->lsr_start = cpu_to_be64(src->lsr_start);
590 dst->lsr_end = cpu_to_be64(src->lsr_end);
591 dst->lsr_index = cpu_to_be32(src->lsr_index);
592 dst->lsr_flags = cpu_to_be32(src->lsr_flags);
595 static inline void range_be_to_cpu(struct lu_seq_range *dst, const struct lu_seq_range *src)
597 dst->lsr_start = be64_to_cpu(src->lsr_start);
598 dst->lsr_end = be64_to_cpu(src->lsr_end);
599 dst->lsr_index = be32_to_cpu(src->lsr_index);
600 dst->lsr_flags = be32_to_cpu(src->lsr_flags);
605 #endif /* __LINUX_FID_H */