linux/drivers/staging/lustre/lustre/include/lustre_fid.h
<<
>>
Prefs
   1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 * GPL HEADER START
   4 *
   5 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   6 *
   7 * This program is free software; you can redistribute it and/or modify
   8 * it under the terms of the GNU General Public License version 2 only,
   9 * as published by the Free Software Foundation.
  10 *
  11 * This program is distributed in the hope that it will be useful, but
  12 * WITHOUT ANY WARRANTY; without even the implied warranty of
  13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  14 * General Public License version 2 for more details (a copy is included
  15 * in the LICENSE file that accompanied this code).
  16 *
  17 * You should have received a copy of the GNU General Public License
  18 * version 2 along with this program; If not, see
  19 * http://www.gnu.org/licenses/gpl-2.0.html
  20 *
  21 * GPL HEADER END
  22 */
  23/*
  24 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  25 * Use is subject to license terms.
  26 *
  27 * Copyright (c) 2011, 2015, Intel Corporation.
  28 */
  29/*
  30 * This file is part of Lustre, http://www.lustre.org/
  31 * Lustre is a trademark of Sun Microsystems, Inc.
  32 *
  33 * lustre/include/lustre_fid.h
  34 *
  35 * Author: Yury Umanets <umka@clusterfs.com>
  36 */
  37
  38#ifndef __LUSTRE_FID_H
  39#define __LUSTRE_FID_H
  40
  41/** \defgroup fid fid
  42 *
  43 * @{
  44 *
  45 * http://wiki.lustre.org/index.php/Architecture_-_Interoperability_fids_zfs
  46 * describes the FID namespace and interoperability requirements for FIDs.
  47 * The important parts of that document are included here for reference.
  48 *
  49 * FID
  50 *   File IDentifier generated by client from range allocated by the SEQuence
  51 *   service and stored in struct lu_fid. The FID is composed of three parts:
  52 *   SEQuence, ObjectID, and VERsion.  The SEQ component is a filesystem
  53 *   unique 64-bit integer, and only one client is ever assigned any SEQ value.
  54 *   The first 0x400 FID_SEQ_NORMAL [2^33, 2^33 + 0x400] values are reserved
  55 *   for system use.  The OID component is a 32-bit value generated by the
  56 *   client on a per-SEQ basis to allow creating many unique FIDs without
  57 *   communication with the server.  The VER component is a 32-bit value that
  58 *   distinguishes between different FID instantiations, such as snapshots or
  59 *   separate subtrees within the filesystem.  FIDs with the same VER field
  60 *   are considered part of the same namespace.
  61 *
  62 * OLD filesystems are those upgraded from Lustre 1.x that predate FIDs, and
  63 *   MDTs use 32-bit ldiskfs internal inode/generation numbers (IGIFs), while
  64 *   OSTs use 64-bit Lustre object IDs and generation numbers.
  65 *
  66 * NEW filesystems are those formatted since the introduction of FIDs.
  67 *
  68 * IGIF
  69 *   Inode and Generation In FID, a surrogate FID used to globally identify
  70 *   an existing object on OLD formatted MDT file system. This would only be
  71 *   used on MDT0 in a DNE filesystem, because there cannot be more than one
  72 *   MDT in an OLD formatted filesystem. Belongs to sequence in [12, 2^32 - 1]
  73 *   range, where inode number is stored in SEQ, and inode generation is in OID.
  74 *   NOTE: This assumes no more than 2^32-1 inodes exist in the MDT filesystem,
  75 *   which is the maximum possible for an ldiskfs backend.  It also assumes
  76 *   that the reserved ext3/ext4/ldiskfs inode numbers [0-11] are never visible
  77 *   to clients, which has always been true.
  78 *
  79 * IDIF
  80 *   object ID In FID, a surrogate FID used to globally identify an existing
  81 *   OST object on OLD formatted OST file system. Belongs to a sequence in
  82 *   [2^32, 2^33 - 1]. Sequence number is calculated as:
  83 *
  84 *      1 << 32 | (ost_index << 16) | ((objid >> 32) & 0xffff)
  85 *
  86 *   that is, SEQ consists of 16-bit OST index, and higher 16 bits of object
  87 *   ID. The generation of unique SEQ values per OST allows the IDIF FIDs to
  88 *   be identified in the FLD correctly. The OID field is calculated as:
  89 *
  90 *      objid & 0xffffffff
  91 *
  92 *   that is, it consists of lower 32 bits of object ID.  For objects within
  93 *   the IDIF range, object ID extraction will be:
  94 *
  95 *      o_id = (fid->f_seq & 0x7fff) << 16 | fid->f_oid;
  96 *      o_seq = 0;  // formerly group number
  97 *
  98 *   NOTE: This assumes that no more than 2^48-1 objects have ever been created
  99 *   on any OST, and that no more than 65535 OSTs are in use.  Both are very
 100 *   reasonable assumptions, i.e. an IDIF can uniquely map all objects assuming
 101 *   a maximum creation rate of 1M objects per second for a maximum of 9 years,
 102 *   or combinations thereof.
 103 *
 104 * OST_MDT0
 105 *   Surrogate FID used to identify an existing object on OLD formatted OST
 106 *   filesystem. Belongs to the reserved SEQuence 0, and is used prior to
 107 *   the introduction of FID-on-OST, at which point IDIF will be used to
 108 *   identify objects as residing on a specific OST.
 109 *
 110 * LLOG
 111 *   For Lustre Log objects the object sequence 1 is used. This is compatible
 112 *   with both OLD and NEW namespaces, as this SEQ number is in the
 113 *   ext3/ldiskfs reserved inode range and does not conflict with IGIF
 114 *   sequence numbers.
 115 *
 116 * ECHO
 117 *   For testing OST IO performance the object sequence 2 is used. This is
 118 *   compatible with both OLD and NEW namespaces, as this SEQ number is in
 119 *   the ext3/ldiskfs reserved inode range and does not conflict with IGIF
 120 *   sequence numbers.
 121 *
 122 * OST_MDT1 .. OST_MAX
 123 *   For testing with multiple MDTs the object sequence 3 through 9 is used,
 124 *   allowing direct mapping of MDTs 1 through 7 respectively, for a total
 125 *   of 8 MDTs including OST_MDT0. This matches the legacy CMD project "group"
 126 *   mappings. However, this SEQ range is only for testing prior to any
 127 *   production DNE release, as the objects in this range conflict across all
 128 *   OSTs, as the OST index is not part of the FID.  For production DNE usage,
 129 *   OST objects created by MDT1+ will use FID_SEQ_NORMAL FIDs.
 130 *
 131 * DLM OST objid to IDIF mapping
 132 *   For compatibility with existing OLD OST network protocol structures, the
 133 *   FID must map onto the o_id and o_seq in a manner that ensures existing
 134 *   objects are identified consistently for IO, as well as onto the LDLM
 135 *   namespace to ensure IDIFs there is only a single resource name for any
 136 *   object in the DLM.  The OLD OST object DLM resource mapping is:
 137 *
 138 *      resource[] = {o_id, o_seq, 0, 0}; // o_seq == 0 for production releases
 139 *
 140 *   The NEW OST object DLM resource mapping is the same for both MDT and OST:
 141 *
 142 *      resource[] = {SEQ, OID, VER, HASH};
 143 *
 144 *  NOTE: for mapping IDIF values to DLM resource names the o_id may be
 145 *  larger than the 2^33 reserved sequence numbers for IDIF, so it is possible
 146 *  for the o_id numbers to overlap FID SEQ numbers in the resource. However,
 147 *  in all production releases the OLD o_seq field is always zero, and all
 148 *  valid FID OID values are non-zero, so the lock resources will not collide.
 149 *  Even so, the MDT and OST resources are also in different LDLM namespaces.
 150 */
 151
 152#include <linux/libcfs/libcfs.h>
 153#include <uapi/linux/lustre/lustre_fid.h>
 154#include <uapi/linux/lustre/lustre_idl.h>
 155#include <uapi/linux/lustre/lustre_ostid.h>
 156
 157struct lu_env;
 158struct lu_site;
 159struct lu_context;
 160struct obd_device;
 161struct obd_export;
 162
 163/* Whole sequences space range and zero range definitions */
 164extern const struct lu_seq_range LUSTRE_SEQ_SPACE_RANGE;
 165extern const struct lu_seq_range LUSTRE_SEQ_ZERO_RANGE;
 166extern const struct lu_fid LUSTRE_BFL_FID;
 167extern const struct lu_fid LU_OBF_FID;
 168extern const struct lu_fid LU_DOT_LUSTRE_FID;
 169
 170enum {
 171        /*
 172         * This is how may metadata FIDs may be allocated in one sequence(128k)
 173         */
 174        LUSTRE_METADATA_SEQ_MAX_WIDTH = 0x0000000000020000ULL,
 175
 176        /*
 177         * This is how many data FIDs could be allocated in one sequence(4B - 1)
 178         */
 179        LUSTRE_DATA_SEQ_MAX_WIDTH = 0x00000000FFFFFFFFULL,
 180
 181        /*
 182         * How many sequences to allocate to a client at once.
 183         */
 184        LUSTRE_SEQ_META_WIDTH = 0x0000000000000001ULL,
 185
 186        /*
 187         * seq allocation pool size.
 188         */
 189        LUSTRE_SEQ_BATCH_WIDTH = LUSTRE_SEQ_META_WIDTH * 1000,
 190
 191        /*
 192         * This is how many sequences may be in one super-sequence allocated to
 193         * MDTs.
 194         */
 195        LUSTRE_SEQ_SUPER_WIDTH = ((1ULL << 30ULL) * LUSTRE_SEQ_META_WIDTH)
 196};
 197
 198enum {
 199        /** 2^6 FIDs for OI containers */
 200        OSD_OI_FID_OID_BITS     = 6,
 201        /** reserve enough FIDs in case we want more in the future */
 202        OSD_OI_FID_OID_BITS_MAX = 10,
 203};
 204
 205/** special OID for local objects */
 206enum local_oid {
 207        /** \see fld_mod_init */
 208        FLD_INDEX_OID           = 3UL,
 209        /** \see fid_mod_init */
 210        FID_SEQ_CTL_OID         = 4UL,
 211        FID_SEQ_SRV_OID         = 5UL,
 212        /** \see mdd_mod_init */
 213        MDD_ROOT_INDEX_OID      = 6UL, /* deprecated in 2.4 */
 214        MDD_ORPHAN_OID          = 7UL, /* deprecated in 2.4 */
 215        MDD_LOV_OBJ_OID         = 8UL,
 216        MDD_CAPA_KEYS_OID       = 9UL,
 217        /** \see mdt_mod_init */
 218        LAST_RECV_OID           = 11UL,
 219        OSD_FS_ROOT_OID         = 13UL,
 220        ACCT_USER_OID           = 15UL,
 221        ACCT_GROUP_OID          = 16UL,
 222        LFSCK_BOOKMARK_OID      = 17UL,
 223        OTABLE_IT_OID           = 18UL,
 224        /* These two definitions are obsolete
 225         * OFD_GROUP0_LAST_OID     = 20UL,
 226         * OFD_GROUP4K_LAST_OID    = 20UL+4096,
 227         */
 228        OFD_LAST_GROUP_OID      = 4117UL,
 229        LLOG_CATALOGS_OID       = 4118UL,
 230        MGS_CONFIGS_OID         = 4119UL,
 231        OFD_HEALTH_CHECK_OID    = 4120UL,
 232        MDD_LOV_OBJ_OSEQ        = 4121UL,
 233        LFSCK_NAMESPACE_OID     = 4122UL,
 234        REMOTE_PARENT_DIR_OID   = 4123UL,
 235        SLAVE_LLOG_CATALOGS_OID = 4124UL,
 236};
 237
 238static inline void lu_local_obj_fid(struct lu_fid *fid, __u32 oid)
 239{
 240        fid->f_seq = FID_SEQ_LOCAL_FILE;
 241        fid->f_oid = oid;
 242        fid->f_ver = 0;
 243}
 244
 245static inline void lu_local_name_obj_fid(struct lu_fid *fid, __u32 oid)
 246{
 247        fid->f_seq = FID_SEQ_LOCAL_NAME;
 248        fid->f_oid = oid;
 249        fid->f_ver = 0;
 250}
 251
 252/* For new FS (>= 2.4), the root FID will be changed to
 253 * [FID_SEQ_ROOT:1:0], for existing FS, (upgraded to 2.4),
 254 * the root FID will still be IGIF
 255 */
 256static inline int fid_is_root(const struct lu_fid *fid)
 257{
 258        return unlikely((fid_seq(fid) == FID_SEQ_ROOT &&
 259                         fid_oid(fid) == 1));
 260}
 261
 262static inline int fid_is_dot_lustre(const struct lu_fid *fid)
 263{
 264        return unlikely(fid_seq(fid) == FID_SEQ_DOT_LUSTRE &&
 265                        fid_oid(fid) == FID_OID_DOT_LUSTRE);
 266}
 267
 268static inline int fid_is_obf(const struct lu_fid *fid)
 269{
 270        return unlikely(fid_seq(fid) == FID_SEQ_DOT_LUSTRE &&
 271                        fid_oid(fid) == FID_OID_DOT_LUSTRE_OBF);
 272}
 273
 274static inline int fid_is_otable_it(const struct lu_fid *fid)
 275{
 276        return unlikely(fid_seq(fid) == FID_SEQ_LOCAL_FILE &&
 277                        fid_oid(fid) == OTABLE_IT_OID);
 278}
 279
 280static inline int fid_is_acct(const struct lu_fid *fid)
 281{
 282        return fid_seq(fid) == FID_SEQ_LOCAL_FILE &&
 283               (fid_oid(fid) == ACCT_USER_OID ||
 284                fid_oid(fid) == ACCT_GROUP_OID);
 285}
 286
 287static inline int fid_is_quota(const struct lu_fid *fid)
 288{
 289        return fid_seq(fid) == FID_SEQ_QUOTA ||
 290               fid_seq(fid) == FID_SEQ_QUOTA_GLB;
 291}
 292
 293static inline int fid_seq_in_fldb(__u64 seq)
 294{
 295        return fid_seq_is_igif(seq) || fid_seq_is_norm(seq) ||
 296               fid_seq_is_root(seq) || fid_seq_is_dot(seq);
 297}
 298
 299static inline void lu_last_id_fid(struct lu_fid *fid, __u64 seq, __u32 ost_idx)
 300{
 301        if (fid_seq_is_mdt0(seq)) {
 302                fid->f_seq = fid_idif_seq(0, ost_idx);
 303        } else {
 304                LASSERTF(fid_seq_is_norm(seq) || fid_seq_is_echo(seq) ||
 305                         fid_seq_is_idif(seq), "%#llx\n", seq);
 306                fid->f_seq = seq;
 307        }
 308        fid->f_oid = 0;
 309        fid->f_ver = 0;
 310}
 311
 312/* seq client type */
 313enum lu_cli_type {
 314        LUSTRE_SEQ_METADATA = 1,
 315        LUSTRE_SEQ_DATA
 316};
 317
 318enum lu_mgr_type {
 319        LUSTRE_SEQ_SERVER,
 320        LUSTRE_SEQ_CONTROLLER
 321};
 322
 323/* Client sequence manager interface. */
 324struct lu_client_seq {
 325        /* Sequence-controller export. */
 326        struct obd_export      *lcs_exp;
 327        spinlock_t              lcs_lock;
 328
 329        /*
 330         * Range of allowed for allocation sequences. When using lu_client_seq on
 331         * clients, this contains meta-sequence range. And for servers this
 332         * contains super-sequence range.
 333         */
 334        struct lu_seq_range      lcs_space;
 335
 336        /* Seq related proc */
 337        struct dentry           *lcs_debugfs_entry;
 338
 339        /* This holds last allocated fid in last obtained seq */
 340        struct lu_fid      lcs_fid;
 341
 342        /* LUSTRE_SEQ_METADATA or LUSTRE_SEQ_DATA */
 343        enum lu_cli_type        lcs_type;
 344
 345        /*
 346         * Service uuid, passed from MDT + seq name to form unique seq name to
 347         * use it with procfs.
 348         */
 349        char                lcs_name[LUSTRE_MDT_MAXNAMELEN];
 350
 351        /*
 352         * Sequence width, that is how many objects may be allocated in one
 353         * sequence. Default value for it is LUSTRE_SEQ_MAX_WIDTH.
 354         */
 355        __u64              lcs_width;
 356
 357        /* wait queue for fid allocation and update indicator */
 358        wait_queue_head_t            lcs_waitq;
 359        int                  lcs_update;
 360};
 361
 362/* Client methods */
 363void seq_client_flush(struct lu_client_seq *seq);
 364
 365int seq_client_alloc_fid(const struct lu_env *env, struct lu_client_seq *seq,
 366                         struct lu_fid *fid);
 367/* Fids common stuff */
 368int fid_is_local(const struct lu_env *env,
 369                 struct lu_site *site, const struct lu_fid *fid);
 370
 371enum lu_cli_type;
 372int client_fid_init(struct obd_device *obd, struct obd_export *exp,
 373                    enum lu_cli_type type);
 374int client_fid_fini(struct obd_device *obd);
 375
 376/* fid locking */
 377
 378struct ldlm_namespace;
 379
 380/*
 381 * Build (DLM) resource name from FID.
 382 *
 383 * NOTE: until Lustre 1.8.7/2.1.1 the fid_ver() was packed into name[2],
 384 * but was moved into name[1] along with the OID to avoid consuming the
 385 * renaming name[2,3] fields that need to be used for the quota identifier.
 386 */
 387static inline void
 388fid_build_reg_res_name(const struct lu_fid *fid, struct ldlm_res_id *res)
 389{
 390        memset(res, 0, sizeof(*res));
 391        res->name[LUSTRE_RES_ID_SEQ_OFF] = fid_seq(fid);
 392        res->name[LUSTRE_RES_ID_VER_OID_OFF] = fid_ver_oid(fid);
 393}
 394
 395/*
 396 * Return true if resource is for object identified by FID.
 397 */
 398static inline bool fid_res_name_eq(const struct lu_fid *fid,
 399                                   const struct ldlm_res_id *res)
 400{
 401        return res->name[LUSTRE_RES_ID_SEQ_OFF] == fid_seq(fid) &&
 402               res->name[LUSTRE_RES_ID_VER_OID_OFF] == fid_ver_oid(fid);
 403}
 404
 405/*
 406 * Extract FID from LDLM resource. Reverse of fid_build_reg_res_name().
 407 */
 408static inline void
 409fid_extract_from_res_name(struct lu_fid *fid, const struct ldlm_res_id *res)
 410{
 411        fid->f_seq = res->name[LUSTRE_RES_ID_SEQ_OFF];
 412        fid->f_oid = (__u32)(res->name[LUSTRE_RES_ID_VER_OID_OFF]);
 413        fid->f_ver = (__u32)(res->name[LUSTRE_RES_ID_VER_OID_OFF] >> 32);
 414        LASSERT(fid_res_name_eq(fid, res));
 415}
 416
 417/*
 418 * Build (DLM) resource identifier from global quota FID and quota ID.
 419 */
 420static inline void
 421fid_build_quota_res_name(const struct lu_fid *glb_fid, union lquota_id *qid,
 422                         struct ldlm_res_id *res)
 423{
 424        fid_build_reg_res_name(glb_fid, res);
 425        res->name[LUSTRE_RES_ID_QUOTA_SEQ_OFF] = fid_seq(&qid->qid_fid);
 426        res->name[LUSTRE_RES_ID_QUOTA_VER_OID_OFF] = fid_ver_oid(&qid->qid_fid);
 427}
 428
 429/*
 430 * Extract global FID and quota ID from resource name
 431 */
 432static inline void fid_extract_from_quota_res(struct lu_fid *glb_fid,
 433                                              union lquota_id *qid,
 434                                              const struct ldlm_res_id *res)
 435{
 436        fid_extract_from_res_name(glb_fid, res);
 437        qid->qid_fid.f_seq = res->name[LUSTRE_RES_ID_QUOTA_SEQ_OFF];
 438        qid->qid_fid.f_oid = (__u32)res->name[LUSTRE_RES_ID_QUOTA_VER_OID_OFF];
 439        qid->qid_fid.f_ver =
 440                (__u32)(res->name[LUSTRE_RES_ID_QUOTA_VER_OID_OFF] >> 32);
 441}
 442
 443static inline void
 444fid_build_pdo_res_name(const struct lu_fid *fid, unsigned int hash,
 445                       struct ldlm_res_id *res)
 446{
 447        fid_build_reg_res_name(fid, res);
 448        res->name[LUSTRE_RES_ID_HSH_OFF] = hash;
 449}
 450
 451/**
 452 * Build DLM resource name from object id & seq, which will be removed
 453 * finally, when we replace ost_id with FID in data stack.
 454 *
 455 * Currently, resid from the old client, whose res[0] = object_id,
 456 * res[1] = object_seq, is just opposite with Metatdata
 457 * resid, where, res[0] = fid->f_seq, res[1] = fid->f_oid.
 458 * To unify the resid identification, we will reverse the data
 459 * resid to keep it same with Metadata resid, i.e.
 460 *
 461 * For resid from the old client,
 462 *    res[0] = objid,  res[1] = 0, still keep the original order,
 463 *    for compatibility.
 464 *
 465 * For new resid
 466 *    res will be built from normal FID directly, i.e. res[0] = f_seq,
 467 *    res[1] = f_oid + f_ver.
 468 */
 469static inline void ostid_build_res_name(const struct ost_id *oi,
 470                                        struct ldlm_res_id *name)
 471{
 472        memset(name, 0, sizeof(*name));
 473        if (fid_seq_is_mdt0(ostid_seq(oi))) {
 474                name->name[LUSTRE_RES_ID_SEQ_OFF] = ostid_id(oi);
 475                name->name[LUSTRE_RES_ID_VER_OID_OFF] = ostid_seq(oi);
 476        } else {
 477                fid_build_reg_res_name(&oi->oi_fid, name);
 478        }
 479}
 480
 481/**
 482 * Return true if the resource is for the object identified by this id & group.
 483 */
 484static inline int ostid_res_name_eq(const struct ost_id *oi,
 485                                    const struct ldlm_res_id *name)
 486{
 487        /* Note: it is just a trick here to save some effort, probably the
 488         * correct way would be turn them into the FID and compare
 489         */
 490        if (fid_seq_is_mdt0(ostid_seq(oi))) {
 491                return name->name[LUSTRE_RES_ID_SEQ_OFF] == ostid_id(oi) &&
 492                       name->name[LUSTRE_RES_ID_VER_OID_OFF] == ostid_seq(oi);
 493        } else {
 494                return name->name[LUSTRE_RES_ID_SEQ_OFF] == ostid_seq(oi) &&
 495                       name->name[LUSTRE_RES_ID_VER_OID_OFF] == ostid_id(oi);
 496        }
 497}
 498
 499/**
 500 * Note: we need check oi_seq to decide where to set oi_id,
 501 * so oi_seq should always be set ahead of oi_id.
 502 */
 503static inline int ostid_set_id(struct ost_id *oi, __u64 oid)
 504{
 505        if (fid_seq_is_mdt0(oi->oi.oi_seq)) {
 506                if (oid >= IDIF_MAX_OID)
 507                        return -E2BIG;
 508                oi->oi.oi_id = oid;
 509        } else if (fid_is_idif(&oi->oi_fid)) {
 510                if (oid >= IDIF_MAX_OID)
 511                        return -E2BIG;
 512                oi->oi_fid.f_seq = fid_idif_seq(oid,
 513                                                fid_idif_ost_idx(&oi->oi_fid));
 514                oi->oi_fid.f_oid = oid;
 515                oi->oi_fid.f_ver = oid >> 48;
 516        } else {
 517                if (oid >= OBIF_MAX_OID)
 518                        return -E2BIG;
 519                oi->oi_fid.f_oid = oid;
 520        }
 521        return 0;
 522}
 523
 524/* pack any OST FID into an ostid (id/seq) for the wire/disk */
 525static inline int fid_to_ostid(const struct lu_fid *fid, struct ost_id *ostid)
 526{
 527        int rc = 0;
 528
 529        if (fid_seq_is_igif(fid->f_seq))
 530                return -EBADF;
 531
 532        if (fid_is_idif(fid)) {
 533                u64 objid = fid_idif_id(fid_seq(fid), fid_oid(fid),
 534                                        fid_ver(fid));
 535
 536                ostid_set_seq_mdt0(ostid);
 537                rc = ostid_set_id(ostid, objid);
 538        } else {
 539                ostid->oi_fid = *fid;
 540        }
 541
 542        return rc;
 543}
 544
 545/* The same as osc_build_res_name() */
 546static inline void ost_fid_build_resid(const struct lu_fid *fid,
 547                                       struct ldlm_res_id *resname)
 548{
 549        if (fid_is_mdt0(fid) || fid_is_idif(fid)) {
 550                struct ost_id oi;
 551
 552                oi.oi.oi_id = 0; /* gcc 4.7.2 complains otherwise */
 553                if (fid_to_ostid(fid, &oi) != 0)
 554                        return;
 555                ostid_build_res_name(&oi, resname);
 556        } else {
 557                fid_build_reg_res_name(fid, resname);
 558        }
 559}
 560
 561/**
 562 * Flatten 128-bit FID values into a 64-bit value for use as an inode number.
 563 * For non-IGIF FIDs this starts just over 2^32, and continues without
 564 * conflict until 2^64, at which point we wrap the high 24 bits of the SEQ
 565 * into the range where there may not be many OID values in use, to minimize
 566 * the risk of conflict.
 567 *
 568 * Suppose LUSTRE_SEQ_MAX_WIDTH less than (1 << 24) which is currently true,
 569 * the time between re-used inode numbers is very long - 2^40 SEQ numbers,
 570 * or about 2^40 client mounts, if clients create less than 2^24 files/mount.
 571 */
 572static inline __u64 fid_flatten(const struct lu_fid *fid)
 573{
 574        __u64 ino;
 575        __u64 seq;
 576
 577        if (fid_is_igif(fid)) {
 578                ino = lu_igif_ino(fid);
 579                return ino;
 580        }
 581
 582        seq = fid_seq(fid);
 583
 584        ino = (seq << 24) + ((seq >> 24) & 0xffffff0000ULL) + fid_oid(fid);
 585
 586        return ino ? ino : fid_oid(fid);
 587}
 588
 589static inline __u32 fid_hash(const struct lu_fid *f, int bits)
 590{
 591        /* all objects with same id and different versions will belong to same
 592         * collisions list.
 593         */
 594        return hash_long(fid_flatten(f), bits);
 595}
 596
 597/**
 598 * map fid to 32 bit value for ino on 32bit systems.
 599 */
 600static inline __u32 fid_flatten32(const struct lu_fid *fid)
 601{
 602        __u32 ino;
 603        __u64 seq;
 604
 605        if (fid_is_igif(fid)) {
 606                ino = lu_igif_ino(fid);
 607                return ino;
 608        }
 609
 610        seq = fid_seq(fid) - FID_SEQ_START;
 611
 612        /* Map the high bits of the OID into higher bits of the inode number so
 613         * that inodes generated at about the same time have a reduced chance
 614         * of collisions. This will give a period of 2^12 = 1024 unique clients
 615         * (from SEQ) and up to min(LUSTRE_SEQ_MAX_WIDTH, 2^20) = 128k objects
 616         * (from OID), or up to 128M inodes without collisions for new files.
 617         */
 618        ino = ((seq & 0x000fffffULL) << 12) + ((seq >> 8) & 0xfffff000) +
 619               (seq >> (64 - (40 - 8)) & 0xffffff00) +
 620               (fid_oid(fid) & 0xff000fff) + ((fid_oid(fid) & 0x00fff000) << 8);
 621
 622        return ino ? ino : fid_oid(fid);
 623}
 624
 625static inline int lu_fid_diff(const struct lu_fid *fid1,
 626                              const struct lu_fid *fid2)
 627{
 628        LASSERTF(fid_seq(fid1) == fid_seq(fid2), "fid1:" DFID ", fid2:" DFID "\n",
 629                 PFID(fid1), PFID(fid2));
 630
 631        if (fid_is_idif(fid1) && fid_is_idif(fid2))
 632                return fid_idif_id(fid1->f_seq, fid1->f_oid, fid1->f_ver) -
 633                       fid_idif_id(fid2->f_seq, fid2->f_oid, fid2->f_ver);
 634
 635        return fid_oid(fid1) - fid_oid(fid2);
 636}
 637
 638#define LUSTRE_SEQ_SRV_NAME "seq_srv"
 639#define LUSTRE_SEQ_CTL_NAME "seq_ctl"
 640
 641/* Range common stuff */
 642static inline void range_cpu_to_le(struct lu_seq_range *dst, const struct lu_seq_range *src)
 643{
 644        dst->lsr_start = cpu_to_le64(src->lsr_start);
 645        dst->lsr_end = cpu_to_le64(src->lsr_end);
 646        dst->lsr_index = cpu_to_le32(src->lsr_index);
 647        dst->lsr_flags = cpu_to_le32(src->lsr_flags);
 648}
 649
 650static inline void range_le_to_cpu(struct lu_seq_range *dst, const struct lu_seq_range *src)
 651{
 652        dst->lsr_start = le64_to_cpu(src->lsr_start);
 653        dst->lsr_end = le64_to_cpu(src->lsr_end);
 654        dst->lsr_index = le32_to_cpu(src->lsr_index);
 655        dst->lsr_flags = le32_to_cpu(src->lsr_flags);
 656}
 657
 658static inline void range_cpu_to_be(struct lu_seq_range *dst, const struct lu_seq_range *src)
 659{
 660        dst->lsr_start = cpu_to_be64(src->lsr_start);
 661        dst->lsr_end = cpu_to_be64(src->lsr_end);
 662        dst->lsr_index = cpu_to_be32(src->lsr_index);
 663        dst->lsr_flags = cpu_to_be32(src->lsr_flags);
 664}
 665
 666static inline void range_be_to_cpu(struct lu_seq_range *dst, const struct lu_seq_range *src)
 667{
 668        dst->lsr_start = be64_to_cpu(src->lsr_start);
 669        dst->lsr_end = be64_to_cpu(src->lsr_end);
 670        dst->lsr_index = be32_to_cpu(src->lsr_index);
 671        dst->lsr_flags = be32_to_cpu(src->lsr_flags);
 672}
 673
 674/** @} fid */
 675
 676#endif /* __LUSTRE_FID_H */
 677