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, 2015, 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.old.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 "../../include/linux/libcfs/libcfs.h"
152 #include "lustre/lustre_idl.h"
160 /* Whole sequences space range and zero range definitions */
161 extern const struct lu_seq_range LUSTRE_SEQ_SPACE_RANGE;
162 extern const struct lu_seq_range LUSTRE_SEQ_ZERO_RANGE;
163 extern const struct lu_fid LUSTRE_BFL_FID;
164 extern const struct lu_fid LU_OBF_FID;
165 extern const struct lu_fid LU_DOT_LUSTRE_FID;
169 * This is how may metadata FIDs may be allocated in one sequence(128k)
171 LUSTRE_METADATA_SEQ_MAX_WIDTH = 0x0000000000020000ULL,
174 * This is how many data FIDs could be allocated in one sequence(4B - 1)
176 LUSTRE_DATA_SEQ_MAX_WIDTH = 0x00000000FFFFFFFFULL,
179 * How many sequences to allocate to a client at once.
181 LUSTRE_SEQ_META_WIDTH = 0x0000000000000001ULL,
184 * seq allocation pool size.
186 LUSTRE_SEQ_BATCH_WIDTH = LUSTRE_SEQ_META_WIDTH * 1000,
189 * This is how many sequences may be in one super-sequence allocated to
192 LUSTRE_SEQ_SUPER_WIDTH = ((1ULL << 30ULL) * LUSTRE_SEQ_META_WIDTH)
196 /** 2^6 FIDs for OI containers */
197 OSD_OI_FID_OID_BITS = 6,
198 /** reserve enough FIDs in case we want more in the future */
199 OSD_OI_FID_OID_BITS_MAX = 10,
202 /** special OID for local objects */
204 /** \see fld_mod_init */
206 /** \see fid_mod_init */
207 FID_SEQ_CTL_OID = 4UL,
208 FID_SEQ_SRV_OID = 5UL,
209 /** \see mdd_mod_init */
210 MDD_ROOT_INDEX_OID = 6UL, /* deprecated in 2.4 */
211 MDD_ORPHAN_OID = 7UL, /* deprecated in 2.4 */
212 MDD_LOV_OBJ_OID = 8UL,
213 MDD_CAPA_KEYS_OID = 9UL,
214 /** \see mdt_mod_init */
215 LAST_RECV_OID = 11UL,
216 OSD_FS_ROOT_OID = 13UL,
217 ACCT_USER_OID = 15UL,
218 ACCT_GROUP_OID = 16UL,
219 LFSCK_BOOKMARK_OID = 17UL,
220 OTABLE_IT_OID = 18UL,
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,
230 LFSCK_NAMESPACE_OID = 4122UL,
231 REMOTE_PARENT_DIR_OID = 4123UL,
232 SLAVE_LLOG_CATALOGS_OID = 4124UL,
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
253 static inline int fid_is_root(const struct lu_fid *fid)
255 return unlikely((fid_seq(fid) == FID_SEQ_ROOT &&
259 static inline int fid_is_dot_lustre(const struct lu_fid *fid)
261 return unlikely(fid_seq(fid) == FID_SEQ_DOT_LUSTRE &&
262 fid_oid(fid) == FID_OID_DOT_LUSTRE);
265 static inline int fid_is_obf(const struct lu_fid *fid)
267 return unlikely(fid_seq(fid) == FID_SEQ_DOT_LUSTRE &&
268 fid_oid(fid) == FID_OID_DOT_LUSTRE_OBF);
271 static inline int fid_is_otable_it(const struct lu_fid *fid)
273 return unlikely(fid_seq(fid) == FID_SEQ_LOCAL_FILE &&
274 fid_oid(fid) == OTABLE_IT_OID);
277 static inline int fid_is_acct(const struct lu_fid *fid)
279 return fid_seq(fid) == FID_SEQ_LOCAL_FILE &&
280 (fid_oid(fid) == ACCT_USER_OID ||
281 fid_oid(fid) == ACCT_GROUP_OID);
284 static inline int fid_is_quota(const struct lu_fid *fid)
286 return fid_seq(fid) == FID_SEQ_QUOTA ||
287 fid_seq(fid) == FID_SEQ_QUOTA_GLB;
290 static inline int fid_is_namespace_visible(const struct lu_fid *fid)
292 const __u64 seq = fid_seq(fid);
294 /* Here, we cannot distinguish whether the normal FID is for OST
295 * object or not. It is caller's duty to check more if needed.
297 return (!fid_is_last_id(fid) &&
298 (fid_seq_is_norm(seq) || fid_seq_is_igif(seq))) ||
299 fid_is_root(fid) || fid_is_dot_lustre(fid);
302 static inline int fid_seq_in_fldb(__u64 seq)
304 return fid_seq_is_igif(seq) || fid_seq_is_norm(seq) ||
305 fid_seq_is_root(seq) || fid_seq_is_dot(seq);
308 static inline void lu_last_id_fid(struct lu_fid *fid, __u64 seq, __u32 ost_idx)
310 if (fid_seq_is_mdt0(seq)) {
311 fid->f_seq = fid_idif_seq(0, ost_idx);
313 LASSERTF(fid_seq_is_norm(seq) || fid_seq_is_echo(seq) ||
314 fid_seq_is_idif(seq), "%#llx\n", seq);
321 /* seq client type */
323 LUSTRE_SEQ_METADATA = 1,
329 LUSTRE_SEQ_CONTROLLER
332 /* Client sequence manager interface. */
333 struct lu_client_seq {
334 /* Sequence-controller export. */
335 struct obd_export *lcs_exp;
336 struct mutex lcs_mutex;
339 * Range of allowed for allocation sequences. When using lu_client_seq on
340 * clients, this contains meta-sequence range. And for servers this
341 * contains super-sequence range.
343 struct lu_seq_range lcs_space;
345 /* Seq related proc */
346 struct dentry *lcs_debugfs_entry;
348 /* This holds last allocated fid in last obtained seq */
349 struct lu_fid lcs_fid;
351 /* LUSTRE_SEQ_METADATA or LUSTRE_SEQ_DATA */
352 enum lu_cli_type lcs_type;
355 * Service uuid, passed from MDT + seq name to form unique seq name to
356 * use it with procfs.
358 char lcs_name[LUSTRE_MDT_MAXNAMELEN];
361 * Sequence width, that is how many objects may be allocated in one
362 * sequence. Default value for it is LUSTRE_SEQ_MAX_WIDTH.
366 /* wait queue for fid allocation and update indicator */
367 wait_queue_head_t lcs_waitq;
372 void seq_client_flush(struct lu_client_seq *seq);
374 int seq_client_alloc_fid(const struct lu_env *env, struct lu_client_seq *seq,
376 /* Fids common stuff */
377 int fid_is_local(const struct lu_env *env,
378 struct lu_site *site, const struct lu_fid *fid);
381 int client_fid_init(struct obd_device *obd, struct obd_export *exp,
382 enum lu_cli_type type);
383 int client_fid_fini(struct obd_device *obd);
387 struct ldlm_namespace;
390 * Build (DLM) resource name from FID.
392 * NOTE: until Lustre 1.8.7/2.1.1 the fid_ver() was packed into name[2],
393 * but was moved into name[1] along with the OID to avoid consuming the
394 * renaming name[2,3] fields that need to be used for the quota identifier.
397 fid_build_reg_res_name(const struct lu_fid *fid, struct ldlm_res_id *res)
399 memset(res, 0, sizeof(*res));
400 res->name[LUSTRE_RES_ID_SEQ_OFF] = fid_seq(fid);
401 res->name[LUSTRE_RES_ID_VER_OID_OFF] = fid_ver_oid(fid);
405 * Return true if resource is for object identified by FID.
407 static inline bool fid_res_name_eq(const struct lu_fid *fid,
408 const struct ldlm_res_id *res)
410 return res->name[LUSTRE_RES_ID_SEQ_OFF] == fid_seq(fid) &&
411 res->name[LUSTRE_RES_ID_VER_OID_OFF] == fid_ver_oid(fid);
415 * Extract FID from LDLM resource. Reverse of fid_build_reg_res_name().
418 fid_extract_from_res_name(struct lu_fid *fid, const struct ldlm_res_id *res)
420 fid->f_seq = res->name[LUSTRE_RES_ID_SEQ_OFF];
421 fid->f_oid = (__u32)(res->name[LUSTRE_RES_ID_VER_OID_OFF]);
422 fid->f_ver = (__u32)(res->name[LUSTRE_RES_ID_VER_OID_OFF] >> 32);
423 LASSERT(fid_res_name_eq(fid, res));
427 * Build (DLM) resource identifier from global quota FID and quota ID.
430 fid_build_quota_res_name(const struct lu_fid *glb_fid, union lquota_id *qid,
431 struct ldlm_res_id *res)
433 fid_build_reg_res_name(glb_fid, res);
434 res->name[LUSTRE_RES_ID_QUOTA_SEQ_OFF] = fid_seq(&qid->qid_fid);
435 res->name[LUSTRE_RES_ID_QUOTA_VER_OID_OFF] = fid_ver_oid(&qid->qid_fid);
439 * Extract global FID and quota ID from resource name
441 static inline void fid_extract_from_quota_res(struct lu_fid *glb_fid,
442 union lquota_id *qid,
443 const struct ldlm_res_id *res)
445 fid_extract_from_res_name(glb_fid, res);
446 qid->qid_fid.f_seq = res->name[LUSTRE_RES_ID_QUOTA_SEQ_OFF];
447 qid->qid_fid.f_oid = (__u32)res->name[LUSTRE_RES_ID_QUOTA_VER_OID_OFF];
449 (__u32)(res->name[LUSTRE_RES_ID_QUOTA_VER_OID_OFF] >> 32);
453 fid_build_pdo_res_name(const struct lu_fid *fid, unsigned int hash,
454 struct ldlm_res_id *res)
456 fid_build_reg_res_name(fid, res);
457 res->name[LUSTRE_RES_ID_HSH_OFF] = hash;
461 * Build DLM resource name from object id & seq, which will be removed
462 * finally, when we replace ost_id with FID in data stack.
464 * Currently, resid from the old client, whose res[0] = object_id,
465 * res[1] = object_seq, is just opposite with Metatdata
466 * resid, where, res[0] = fid->f_seq, res[1] = fid->f_oid.
467 * To unify the resid identification, we will reverse the data
468 * resid to keep it same with Metadata resid, i.e.
470 * For resid from the old client,
471 * res[0] = objid, res[1] = 0, still keep the original order,
475 * res will be built from normal FID directly, i.e. res[0] = f_seq,
476 * res[1] = f_oid + f_ver.
478 static inline void ostid_build_res_name(const struct ost_id *oi,
479 struct ldlm_res_id *name)
481 memset(name, 0, sizeof(*name));
482 if (fid_seq_is_mdt0(ostid_seq(oi))) {
483 name->name[LUSTRE_RES_ID_SEQ_OFF] = ostid_id(oi);
484 name->name[LUSTRE_RES_ID_VER_OID_OFF] = ostid_seq(oi);
486 fid_build_reg_res_name(&oi->oi_fid, name);
491 * Return true if the resource is for the object identified by this id & group.
493 static inline int ostid_res_name_eq(const struct ost_id *oi,
494 const struct ldlm_res_id *name)
496 /* Note: it is just a trick here to save some effort, probably the
497 * correct way would be turn them into the FID and compare
499 if (fid_seq_is_mdt0(ostid_seq(oi))) {
500 return name->name[LUSTRE_RES_ID_SEQ_OFF] == ostid_id(oi) &&
501 name->name[LUSTRE_RES_ID_VER_OID_OFF] == ostid_seq(oi);
503 return name->name[LUSTRE_RES_ID_SEQ_OFF] == ostid_seq(oi) &&
504 name->name[LUSTRE_RES_ID_VER_OID_OFF] == ostid_id(oi);
508 /* The same as osc_build_res_name() */
509 static inline void ost_fid_build_resid(const struct lu_fid *fid,
510 struct ldlm_res_id *resname)
512 if (fid_is_mdt0(fid) || fid_is_idif(fid)) {
515 oi.oi.oi_id = 0; /* gcc 4.7.2 complains otherwise */
516 if (fid_to_ostid(fid, &oi) != 0)
518 ostid_build_res_name(&oi, resname);
520 fid_build_reg_res_name(fid, resname);
524 static inline void ost_fid_from_resid(struct lu_fid *fid,
525 const struct ldlm_res_id *name,
528 if (fid_seq_is_mdt0(name->name[LUSTRE_RES_ID_VER_OID_OFF])) {
532 ostid_set_seq(&oi, name->name[LUSTRE_RES_ID_VER_OID_OFF]);
533 ostid_set_id(&oi, name->name[LUSTRE_RES_ID_SEQ_OFF]);
534 ostid_to_fid(fid, &oi, ost_idx);
537 fid_extract_from_res_name(fid, name);
542 * Flatten 128-bit FID values into a 64-bit value for use as an inode number.
543 * For non-IGIF FIDs this starts just over 2^32, and continues without
544 * conflict until 2^64, at which point we wrap the high 24 bits of the SEQ
545 * into the range where there may not be many OID values in use, to minimize
546 * the risk of conflict.
548 * Suppose LUSTRE_SEQ_MAX_WIDTH less than (1 << 24) which is currently true,
549 * the time between re-used inode numbers is very long - 2^40 SEQ numbers,
550 * or about 2^40 client mounts, if clients create less than 2^24 files/mount.
552 static inline __u64 fid_flatten(const struct lu_fid *fid)
557 if (fid_is_igif(fid)) {
558 ino = lu_igif_ino(fid);
564 ino = (seq << 24) + ((seq >> 24) & 0xffffff0000ULL) + fid_oid(fid);
566 return ino ? ino : fid_oid(fid);
569 static inline __u32 fid_hash(const struct lu_fid *f, int bits)
571 /* all objects with same id and different versions will belong to same
574 return hash_long(fid_flatten(f), bits);
578 * map fid to 32 bit value for ino on 32bit systems.
580 static inline __u32 fid_flatten32(const struct lu_fid *fid)
585 if (fid_is_igif(fid)) {
586 ino = lu_igif_ino(fid);
590 seq = fid_seq(fid) - FID_SEQ_START;
592 /* Map the high bits of the OID into higher bits of the inode number so
593 * that inodes generated at about the same time have a reduced chance
594 * of collisions. This will give a period of 2^12 = 1024 unique clients
595 * (from SEQ) and up to min(LUSTRE_SEQ_MAX_WIDTH, 2^20) = 128k objects
596 * (from OID), or up to 128M inodes without collisions for new files.
598 ino = ((seq & 0x000fffffULL) << 12) + ((seq >> 8) & 0xfffff000) +
599 (seq >> (64 - (40 - 8)) & 0xffffff00) +
600 (fid_oid(fid) & 0xff000fff) + ((fid_oid(fid) & 0x00fff000) << 8);
602 return ino ? ino : fid_oid(fid);
605 static inline int lu_fid_diff(const struct lu_fid *fid1,
606 const struct lu_fid *fid2)
608 LASSERTF(fid_seq(fid1) == fid_seq(fid2), "fid1:"DFID", fid2:"DFID"\n",
609 PFID(fid1), PFID(fid2));
611 if (fid_is_idif(fid1) && fid_is_idif(fid2))
612 return fid_idif_id(fid1->f_seq, fid1->f_oid, fid1->f_ver) -
613 fid_idif_id(fid2->f_seq, fid2->f_oid, fid2->f_ver);
615 return fid_oid(fid1) - fid_oid(fid2);
618 #define LUSTRE_SEQ_SRV_NAME "seq_srv"
619 #define LUSTRE_SEQ_CTL_NAME "seq_ctl"
621 /* Range common stuff */
622 static inline void range_cpu_to_le(struct lu_seq_range *dst, const struct lu_seq_range *src)
624 dst->lsr_start = cpu_to_le64(src->lsr_start);
625 dst->lsr_end = cpu_to_le64(src->lsr_end);
626 dst->lsr_index = cpu_to_le32(src->lsr_index);
627 dst->lsr_flags = cpu_to_le32(src->lsr_flags);
630 static inline void range_le_to_cpu(struct lu_seq_range *dst, const struct lu_seq_range *src)
632 dst->lsr_start = le64_to_cpu(src->lsr_start);
633 dst->lsr_end = le64_to_cpu(src->lsr_end);
634 dst->lsr_index = le32_to_cpu(src->lsr_index);
635 dst->lsr_flags = le32_to_cpu(src->lsr_flags);
638 static inline void range_cpu_to_be(struct lu_seq_range *dst, const struct lu_seq_range *src)
640 dst->lsr_start = cpu_to_be64(src->lsr_start);
641 dst->lsr_end = cpu_to_be64(src->lsr_end);
642 dst->lsr_index = cpu_to_be32(src->lsr_index);
643 dst->lsr_flags = cpu_to_be32(src->lsr_flags);
646 static inline void range_be_to_cpu(struct lu_seq_range *dst, const struct lu_seq_range *src)
648 dst->lsr_start = be64_to_cpu(src->lsr_start);
649 dst->lsr_end = be64_to_cpu(src->lsr_end);
650 dst->lsr_index = be32_to_cpu(src->lsr_index);
651 dst->lsr_flags = be32_to_cpu(src->lsr_flags);
656 #endif /* __LUSTRE_FID_H */