linux/drivers/staging/lustre/lustre/include/lustre_fid.h
<<
>>
Prefs
   1/*
   2 * GPL HEADER START
   3 *
   4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5 *
   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.
   9 *
  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).
  15 *
  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
  19 *
  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
  22 * have any questions.
  23 *
  24 * GPL HEADER END
  25 */
  26/*
  27 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  28 * Use is subject to license terms.
  29 *
  30 * Copyright (c) 2011, 2012, Intel Corporation.
  31 */
  32/*
  33 * This file is part of Lustre, http://www.lustre.org/
  34 * Lustre is a trademark of Sun Microsystems, Inc.
  35 *
  36 * lustre/include/lustre_fid.h
  37 *
  38 * Author: Yury Umanets <umka@clusterfs.com>
  39 */
  40
  41#ifndef __LUSTRE_FID_H
  42#define __LUSTRE_FID_H
  43
  44/** \defgroup fid fid
  45 *
  46 * @{
  47 *
  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.
  51 *
  52 * FID
  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.
  64 *
  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.
  68 *
  69 * NEW filesystems are those formatted since the introduction of FIDs.
  70 *
  71 * IGIF
  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.
  81 *
  82 * IDIF
  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:
  86 *
  87 *      1 << 32 | (ost_index << 16) | ((objid >> 32) & 0xffff)
  88 *
  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:
  92 *
  93 *      objid & 0xffffffff
  94 *
  95 *   that is, it consists of lower 32 bits of object ID.  For objects within
  96 *   the IDIF range, object ID extraction will be:
  97 *
  98 *      o_id = (fid->f_seq & 0x7fff) << 16 | fid->f_oid;
  99 *      o_seq = 0;  // formerly group number
 100 *
 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.
 106 *
 107 * OST_MDT0
 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.
 112 *
 113 * LLOG
 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
 117 *   sequence numbers.
 118 *
 119 * ECHO
 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
 123 *   sequence numbers.
 124 *
 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.
 133 *
 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:
 140 *
 141 *      resource[] = {o_id, o_seq, 0, 0}; // o_seq == 0 for production releases
 142 *
 143 *   The NEW OST object DLM resource mapping is the same for both MDT and OST:
 144 *
 145 *      resource[] = {SEQ, OID, VER, HASH};
 146 *
 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.
 153 */
 154
 155#include <linux/libcfs/libcfs.h>
 156#include <lustre/lustre_idl.h>
 157
 158struct lu_env;
 159struct lu_site;
 160struct lu_context;
 161struct obd_device;
 162struct obd_export;
 163
 164/* Whole sequences space range and zero range definitions */
 165extern const struct lu_seq_range LUSTRE_SEQ_SPACE_RANGE;
 166extern const struct lu_seq_range LUSTRE_SEQ_ZERO_RANGE;
 167extern const struct lu_fid LUSTRE_BFL_FID;
 168extern const struct lu_fid LU_OBF_FID;
 169extern const struct lu_fid LU_DOT_LUSTRE_FID;
 170
 171enum {
 172        /*
 173         * This is how may metadata FIDs may be allocated in one sequence(128k)
 174         */
 175        LUSTRE_METADATA_SEQ_MAX_WIDTH = 0x0000000000020000ULL,
 176
 177        /*
 178         * This is how many data FIDs could be allocated in one sequence(4B - 1)
 179         */
 180        LUSTRE_DATA_SEQ_MAX_WIDTH = 0x00000000FFFFFFFFULL,
 181
 182        /*
 183         * How many sequences to allocate to a client at once.
 184         */
 185        LUSTRE_SEQ_META_WIDTH = 0x0000000000000001ULL,
 186
 187        /*
 188         * seq allocation pool size.
 189         */
 190        LUSTRE_SEQ_BATCH_WIDTH = LUSTRE_SEQ_META_WIDTH * 1000,
 191
 192        /*
 193         * This is how many sequences may be in one super-sequence allocated to
 194         * MDTs.
 195         */
 196        LUSTRE_SEQ_SUPER_WIDTH = ((1ULL << 30ULL) * LUSTRE_SEQ_META_WIDTH)
 197};
 198
 199enum {
 200        /** 2^6 FIDs for OI containers */
 201        OSD_OI_FID_OID_BITS     = 6,
 202        /** reserve enough FIDs in case we want more in the future */
 203        OSD_OI_FID_OID_BITS_MAX = 10,
 204};
 205
 206/** special OID for local objects */
 207enum local_oid {
 208        /** \see fld_mod_init */
 209        FLD_INDEX_OID           = 3UL,
 210        /** \see fid_mod_init */
 211        FID_SEQ_CTL_OID         = 4UL,
 212        FID_SEQ_SRV_OID         = 5UL,
 213        /** \see mdd_mod_init */
 214        MDD_ROOT_INDEX_OID      = 6UL, /* deprecated in 2.4 */
 215        MDD_ORPHAN_OID          = 7UL, /* deprecated in 2.4 */
 216        MDD_LOV_OBJ_OID         = 8UL,
 217        MDD_CAPA_KEYS_OID       = 9UL,
 218        /** \see mdt_mod_init */
 219        LAST_RECV_OID           = 11UL,
 220        OSD_FS_ROOT_OID         = 13UL,
 221        ACCT_USER_OID           = 15UL,
 222        ACCT_GROUP_OID          = 16UL,
 223        LFSCK_BOOKMARK_OID      = 17UL,
 224        OTABLE_IT_OID           = 18UL,
 225        /* These two definitions are obsolete
 226         * OFD_GROUP0_LAST_OID     = 20UL,
 227         * OFD_GROUP4K_LAST_OID    = 20UL+4096,
 228         */
 229        OFD_LAST_GROUP_OID      = 4117UL,
 230        LLOG_CATALOGS_OID       = 4118UL,
 231        MGS_CONFIGS_OID         = 4119UL,
 232        OFD_HEALTH_CHECK_OID    = 4120UL,
 233        MDD_LOV_OBJ_OSEQ        = 4121UL,
 234        LFSCK_NAMESPACE_OID     = 4122UL,
 235        REMOTE_PARENT_DIR_OID   = 4123UL,
 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 */
 255static inline int fid_is_root(const struct lu_fid *fid)
 256{
 257        return unlikely((fid_seq(fid) == FID_SEQ_ROOT &&
 258                         fid_oid(fid) == 1));
 259}
 260
 261static inline int fid_is_dot_lustre(const struct lu_fid *fid)
 262{
 263        return unlikely(fid_seq(fid) == FID_SEQ_DOT_LUSTRE &&
 264                        fid_oid(fid) == FID_OID_DOT_LUSTRE);
 265}
 266
 267static inline int fid_is_obf(const struct lu_fid *fid)
 268{
 269        return unlikely(fid_seq(fid) == FID_SEQ_DOT_LUSTRE &&
 270                        fid_oid(fid) == FID_OID_DOT_LUSTRE_OBF);
 271}
 272
 273static inline int fid_is_otable_it(const struct lu_fid *fid)
 274{
 275        return unlikely(fid_seq(fid) == FID_SEQ_LOCAL_FILE &&
 276                        fid_oid(fid) == OTABLE_IT_OID);
 277}
 278
 279static inline int fid_is_acct(const struct lu_fid *fid)
 280{
 281        return fid_seq(fid) == FID_SEQ_LOCAL_FILE &&
 282               (fid_oid(fid) == ACCT_USER_OID ||
 283                fid_oid(fid) == ACCT_GROUP_OID);
 284}
 285
 286static inline int fid_is_quota(const struct lu_fid *fid)
 287{
 288        return fid_seq(fid) == FID_SEQ_QUOTA ||
 289               fid_seq(fid) == FID_SEQ_QUOTA_GLB;
 290}
 291
 292static inline int fid_is_namespace_visible(const struct lu_fid *fid)
 293{
 294        const __u64 seq = fid_seq(fid);
 295
 296        /* Here, we cannot distinguish whether the normal FID is for OST
 297         * object or not. It is caller's duty to check more if needed. */
 298        return (!fid_is_last_id(fid) &&
 299                (fid_seq_is_norm(seq) || fid_seq_is_igif(seq))) ||
 300               fid_is_root(fid) || fid_is_dot_lustre(fid);
 301}
 302
 303static inline int fid_seq_in_fldb(__u64 seq)
 304{
 305        return fid_seq_is_igif(seq) || fid_seq_is_norm(seq) ||
 306               fid_seq_is_root(seq) || fid_seq_is_dot(seq);
 307}
 308
 309static inline void lu_last_id_fid(struct lu_fid *fid, __u64 seq)
 310{
 311        if (fid_seq_is_mdt0(seq)) {
 312                fid->f_seq = fid_idif_seq(0, 0);
 313        } else {
 314                LASSERTF(fid_seq_is_norm(seq) || fid_seq_is_echo(seq) ||
 315                         fid_seq_is_idif(seq), LPX64"\n", seq);
 316                fid->f_seq = seq;
 317        }
 318        fid->f_oid = 0;
 319        fid->f_ver = 0;
 320}
 321
 322/* seq client type */
 323enum lu_cli_type {
 324        LUSTRE_SEQ_METADATA = 1,
 325        LUSTRE_SEQ_DATA
 326};
 327
 328enum lu_mgr_type {
 329        LUSTRE_SEQ_SERVER,
 330        LUSTRE_SEQ_CONTROLLER
 331};
 332
 333struct lu_server_seq;
 334
 335/* Client sequence manager interface. */
 336struct lu_client_seq {
 337        /* Sequence-controller export. */
 338        struct obd_export      *lcs_exp;
 339        struct mutex            lcs_mutex;
 340
 341        /*
 342         * Range of allowed for allocation sequeces. When using lu_client_seq on
 343         * clients, this contains meta-sequence range. And for servers this
 344         * contains super-sequence range.
 345         */
 346        struct lu_seq_range      lcs_space;
 347
 348        /* Seq related proc */
 349        struct proc_dir_entry   *lcs_proc_dir;
 350
 351        /* This holds last allocated fid in last obtained seq */
 352        struct lu_fid      lcs_fid;
 353
 354        /* LUSTRE_SEQ_METADATA or LUSTRE_SEQ_DATA */
 355        enum lu_cli_type        lcs_type;
 356
 357        /*
 358         * Service uuid, passed from MDT + seq name to form unique seq name to
 359         * use it with procfs.
 360         */
 361        char                lcs_name[80];
 362
 363        /*
 364         * Sequence width, that is how many objects may be allocated in one
 365         * sequence. Default value for it is LUSTRE_SEQ_MAX_WIDTH.
 366         */
 367        __u64              lcs_width;
 368
 369        /* Seq-server for direct talking */
 370        struct lu_server_seq   *lcs_srv;
 371
 372        /* wait queue for fid allocation and update indicator */
 373        wait_queue_head_t            lcs_waitq;
 374        int                  lcs_update;
 375};
 376
 377/* server sequence manager interface */
 378struct lu_server_seq {
 379        /* Available sequences space */
 380        struct lu_seq_range      lss_space;
 381
 382        /* keeps highwater in lsr_end for seq allocation algorithm */
 383        struct lu_seq_range      lss_lowater_set;
 384        struct lu_seq_range      lss_hiwater_set;
 385
 386        /*
 387         * Device for server side seq manager needs (saving sequences to backing
 388         * store).
 389         */
 390        struct dt_device       *lss_dev;
 391
 392        /* /seq file object device */
 393        struct dt_object       *lss_obj;
 394
 395        /* Seq related proc */
 396        struct proc_dir_entry   *lss_proc_dir;
 397
 398        /* LUSTRE_SEQ_SERVER or LUSTRE_SEQ_CONTROLLER */
 399        enum lu_mgr_type       lss_type;
 400
 401        /* Client interafce to request controller */
 402        struct lu_client_seq   *lss_cli;
 403
 404        /* Mutex for protecting allocation */
 405        struct mutex            lss_mutex;
 406
 407        /*
 408         * Service uuid, passed from MDT + seq name to form unique seq name to
 409         * use it with procfs.
 410         */
 411        char                lss_name[80];
 412
 413        /*
 414         * Allocation chunks for super and meta sequences. Default values are
 415         * LUSTRE_SEQ_SUPER_WIDTH and LUSTRE_SEQ_META_WIDTH.
 416         */
 417        __u64              lss_width;
 418
 419        /*
 420         * minimum lss_alloc_set size that should be allocated from
 421         * lss_space
 422         */
 423        __u64              lss_set_width;
 424
 425        /* sync is needed for update operation */
 426        __u32              lss_need_sync;
 427
 428        /**
 429         * Pointer to site object, required to access site fld.
 430         */
 431        struct seq_server_site  *lss_site;
 432};
 433
 434struct com_thread_info;
 435int seq_query(struct com_thread_info *info);
 436
 437struct ptlrpc_request;
 438int seq_handle(struct ptlrpc_request *req);
 439
 440/* Server methods */
 441
 442int seq_server_init(struct lu_server_seq *seq,
 443                    struct dt_device *dev,
 444                    const char *prefix,
 445                    enum lu_mgr_type type,
 446                    struct seq_server_site *ss,
 447                    const struct lu_env *env);
 448
 449void seq_server_fini(struct lu_server_seq *seq,
 450                     const struct lu_env *env);
 451
 452int seq_server_alloc_super(struct lu_server_seq *seq,
 453                           struct lu_seq_range *out,
 454                           const struct lu_env *env);
 455
 456int seq_server_alloc_meta(struct lu_server_seq *seq,
 457                          struct lu_seq_range *out,
 458                          const struct lu_env *env);
 459
 460int seq_server_set_cli(struct lu_server_seq *seq,
 461                       struct lu_client_seq *cli,
 462                       const struct lu_env *env);
 463
 464/* Client methods */
 465int seq_client_init(struct lu_client_seq *seq,
 466                    struct obd_export *exp,
 467                    enum lu_cli_type type,
 468                    const char *prefix,
 469                    struct lu_server_seq *srv);
 470
 471void seq_client_fini(struct lu_client_seq *seq);
 472
 473void seq_client_flush(struct lu_client_seq *seq);
 474
 475int seq_client_alloc_fid(const struct lu_env *env, struct lu_client_seq *seq,
 476                         struct lu_fid *fid);
 477int seq_client_get_seq(const struct lu_env *env, struct lu_client_seq *seq,
 478                       seqno_t *seqnr);
 479int seq_site_fini(const struct lu_env *env, struct seq_server_site *ss);
 480/* Fids common stuff */
 481int fid_is_local(const struct lu_env *env,
 482                 struct lu_site *site, const struct lu_fid *fid);
 483
 484enum lu_cli_type;
 485int client_fid_init(struct obd_device *obd, struct obd_export *exp,
 486                    enum lu_cli_type type);
 487int client_fid_fini(struct obd_device *obd);
 488
 489/* fid locking */
 490
 491struct ldlm_namespace;
 492
 493/*
 494 * Build (DLM) resource name from FID.
 495 *
 496 * NOTE: until Lustre 1.8.7/2.1.1 the fid_ver() was packed into name[2],
 497 * but was moved into name[1] along with the OID to avoid consuming the
 498 * renaming name[2,3] fields that need to be used for the quota identifier.
 499 */
 500static inline struct ldlm_res_id *
 501fid_build_reg_res_name(const struct lu_fid *fid, struct ldlm_res_id *res)
 502{
 503        memset(res, 0, sizeof(*res));
 504        res->name[LUSTRE_RES_ID_SEQ_OFF] = fid_seq(fid);
 505        res->name[LUSTRE_RES_ID_VER_OID_OFF] = fid_ver_oid(fid);
 506
 507        return res;
 508}
 509
 510/*
 511 * Return true if resource is for object identified by FID.
 512 */
 513static inline int fid_res_name_eq(const struct lu_fid *fid,
 514                                  const struct ldlm_res_id *res)
 515{
 516        return res->name[LUSTRE_RES_ID_SEQ_OFF] == fid_seq(fid) &&
 517               res->name[LUSTRE_RES_ID_VER_OID_OFF] == fid_ver_oid(fid);
 518}
 519
 520/*
 521 * Extract FID from LDLM resource. Reverse of fid_build_reg_res_name().
 522 */
 523static inline struct lu_fid *
 524fid_extract_from_res_name(struct lu_fid *fid, const struct ldlm_res_id *res)
 525{
 526        fid->f_seq = res->name[LUSTRE_RES_ID_SEQ_OFF];
 527        fid->f_oid = (__u32)(res->name[LUSTRE_RES_ID_VER_OID_OFF]);
 528        fid->f_ver = (__u32)(res->name[LUSTRE_RES_ID_VER_OID_OFF] >> 32);
 529        LASSERT(fid_res_name_eq(fid, res));
 530
 531        return fid;
 532}
 533
 534/*
 535 * Build (DLM) resource identifier from global quota FID and quota ID.
 536 */
 537static inline struct ldlm_res_id *
 538fid_build_quota_res_name(const struct lu_fid *glb_fid, union lquota_id *qid,
 539                      struct ldlm_res_id *res)
 540{
 541        fid_build_reg_res_name(glb_fid, res);
 542        res->name[LUSTRE_RES_ID_QUOTA_SEQ_OFF] = fid_seq(&qid->qid_fid);
 543        res->name[LUSTRE_RES_ID_QUOTA_VER_OID_OFF] = fid_ver_oid(&qid->qid_fid);
 544
 545        return res;
 546}
 547
 548/*
 549 * Extract global FID and quota ID from resource name
 550 */
 551static inline void fid_extract_from_quota_res(struct lu_fid *glb_fid,
 552                                              union lquota_id *qid,
 553                                              const struct ldlm_res_id *res)
 554{
 555        fid_extract_from_res_name(glb_fid, res);
 556        qid->qid_fid.f_seq = res->name[LUSTRE_RES_ID_QUOTA_SEQ_OFF];
 557        qid->qid_fid.f_oid = (__u32)res->name[LUSTRE_RES_ID_QUOTA_VER_OID_OFF];
 558        qid->qid_fid.f_ver =
 559                (__u32)(res->name[LUSTRE_RES_ID_QUOTA_VER_OID_OFF] >> 32);
 560}
 561
 562static inline struct ldlm_res_id *
 563fid_build_pdo_res_name(const struct lu_fid *fid, unsigned int hash,
 564                       struct ldlm_res_id *res)
 565{
 566        fid_build_reg_res_name(fid, res);
 567        res->name[LUSTRE_RES_ID_HSH_OFF] = hash;
 568
 569        return res;
 570}
 571
 572/**
 573 * Build DLM resource name from object id & seq, which will be removed
 574 * finally, when we replace ost_id with FID in data stack.
 575 *
 576 * Currently, resid from the old client, whose res[0] = object_id,
 577 * res[1] = object_seq, is just oposite with Metatdata
 578 * resid, where, res[0] = fid->f_seq, res[1] = fid->f_oid.
 579 * To unifiy the resid identification, we will reverse the data
 580 * resid to keep it same with Metadata resid, i.e.
 581 *
 582 * For resid from the old client,
 583 *    res[0] = objid,  res[1] = 0, still keep the original order,
 584 *    for compatiblity.
 585 *
 586 * For new resid
 587 *    res will be built from normal FID directly, i.e. res[0] = f_seq,
 588 *    res[1] = f_oid + f_ver.
 589 */
 590static inline void ostid_build_res_name(struct ost_id *oi,
 591                                        struct ldlm_res_id *name)
 592{
 593        memset(name, 0, sizeof(*name));
 594        if (fid_seq_is_mdt0(ostid_seq(oi))) {
 595                name->name[LUSTRE_RES_ID_SEQ_OFF] = ostid_id(oi);
 596                name->name[LUSTRE_RES_ID_VER_OID_OFF] = ostid_seq(oi);
 597        } else {
 598                fid_build_reg_res_name(&oi->oi_fid, name);
 599        }
 600}
 601
 602static inline void ostid_res_name_to_id(struct ost_id *oi,
 603                                        struct ldlm_res_id *name)
 604{
 605        if (fid_seq_is_mdt0(name->name[LUSTRE_RES_ID_SEQ_OFF])) {
 606                /* old resid */
 607                ostid_set_seq(oi, name->name[LUSTRE_RES_ID_VER_OID_OFF]);
 608                ostid_set_id(oi, name->name[LUSTRE_RES_ID_SEQ_OFF]);
 609        } else {
 610                /* new resid */
 611                fid_extract_from_res_name(&oi->oi_fid, name);
 612        }
 613}
 614
 615/**
 616 * Return true if the resource is for the object identified by this id & group.
 617 */
 618static inline int ostid_res_name_eq(struct ost_id *oi,
 619                                    struct ldlm_res_id *name)
 620{
 621        /* Note: it is just a trick here to save some effort, probably the
 622         * correct way would be turn them into the FID and compare */
 623        if (fid_seq_is_mdt0(ostid_seq(oi))) {
 624                return name->name[LUSTRE_RES_ID_SEQ_OFF] == ostid_id(oi) &&
 625                       name->name[LUSTRE_RES_ID_VER_OID_OFF] == ostid_seq(oi);
 626        } else {
 627                return name->name[LUSTRE_RES_ID_SEQ_OFF] == ostid_seq(oi) &&
 628                       name->name[LUSTRE_RES_ID_VER_OID_OFF] == ostid_id(oi);
 629        }
 630}
 631
 632/* The same as osc_build_res_name() */
 633static inline void ost_fid_build_resid(const struct lu_fid *fid,
 634                                       struct ldlm_res_id *resname)
 635{
 636        if (fid_is_mdt0(fid) || fid_is_idif(fid)) {
 637                struct ost_id oi;
 638                oi.oi.oi_id = 0; /* gcc 4.7.2 complains otherwise */
 639                if (fid_to_ostid(fid, &oi) != 0)
 640                        return;
 641                ostid_build_res_name(&oi, resname);
 642        } else {
 643                fid_build_reg_res_name(fid, resname);
 644        }
 645}
 646
 647static inline void ost_fid_from_resid(struct lu_fid *fid,
 648                                      const struct ldlm_res_id *name)
 649{
 650        if (fid_seq_is_mdt0(name->name[LUSTRE_RES_ID_VER_OID_OFF])) {
 651                /* old resid */
 652                struct ost_id oi;
 653                ostid_set_seq(&oi, name->name[LUSTRE_RES_ID_VER_OID_OFF]);
 654                ostid_set_id(&oi, name->name[LUSTRE_RES_ID_SEQ_OFF]);
 655                ostid_to_fid(fid, &oi, 0);
 656        } else {
 657                /* new resid */
 658                fid_extract_from_res_name(fid, name);
 659        }
 660}
 661
 662/**
 663 * Flatten 128-bit FID values into a 64-bit value for use as an inode number.
 664 * For non-IGIF FIDs this starts just over 2^32, and continues without
 665 * conflict until 2^64, at which point we wrap the high 24 bits of the SEQ
 666 * into the range where there may not be many OID values in use, to minimize
 667 * the risk of conflict.
 668 *
 669 * Suppose LUSTRE_SEQ_MAX_WIDTH less than (1 << 24) which is currently true,
 670 * the time between re-used inode numbers is very long - 2^40 SEQ numbers,
 671 * or about 2^40 client mounts, if clients create less than 2^24 files/mount.
 672 */
 673static inline __u64 fid_flatten(const struct lu_fid *fid)
 674{
 675        __u64 ino;
 676        __u64 seq;
 677
 678        if (fid_is_igif(fid)) {
 679                ino = lu_igif_ino(fid);
 680                return ino;
 681        }
 682
 683        seq = fid_seq(fid);
 684
 685        ino = (seq << 24) + ((seq >> 24) & 0xffffff0000ULL) + fid_oid(fid);
 686
 687        return ino ? ino : fid_oid(fid);
 688}
 689
 690static inline __u32 fid_hash(const struct lu_fid *f, int bits)
 691{
 692        /* all objects with same id and different versions will belong to same
 693         * collisions list. */
 694        return cfs_hash_long(fid_flatten(f), bits);
 695}
 696
 697/**
 698 * map fid to 32 bit value for ino on 32bit systems. */
 699static inline __u32 fid_flatten32(const struct lu_fid *fid)
 700{
 701        __u32 ino;
 702        __u64 seq;
 703
 704        if (fid_is_igif(fid)) {
 705                ino = lu_igif_ino(fid);
 706                return ino;
 707        }
 708
 709        seq = fid_seq(fid) - FID_SEQ_START;
 710
 711        /* Map the high bits of the OID into higher bits of the inode number so
 712         * that inodes generated at about the same time have a reduced chance
 713         * of collisions. This will give a period of 2^12 = 1024 unique clients
 714         * (from SEQ) and up to min(LUSTRE_SEQ_MAX_WIDTH, 2^20) = 128k objects
 715         * (from OID), or up to 128M inodes without collisions for new files. */
 716        ino = ((seq & 0x000fffffULL) << 12) + ((seq >> 8) & 0xfffff000) +
 717               (seq >> (64 - (40-8)) & 0xffffff00) +
 718               (fid_oid(fid) & 0xff000fff) + ((fid_oid(fid) & 0x00fff000) << 8);
 719
 720        return ino ? ino : fid_oid(fid);
 721}
 722
 723static inline int lu_fid_diff(struct lu_fid *fid1, struct lu_fid *fid2)
 724{
 725        LASSERTF(fid_seq(fid1) == fid_seq(fid2), "fid1:"DFID", fid2:"DFID"\n",
 726                 PFID(fid1), PFID(fid2));
 727
 728        if (fid_is_idif(fid1) && fid_is_idif(fid2))
 729                return fid_idif_id(fid1->f_seq, fid1->f_oid, fid1->f_ver) -
 730                       fid_idif_id(fid2->f_seq, fid2->f_oid, fid2->f_ver);
 731
 732        return fid_oid(fid1) - fid_oid(fid2);
 733}
 734
 735#define LUSTRE_SEQ_SRV_NAME "seq_srv"
 736#define LUSTRE_SEQ_CTL_NAME "seq_ctl"
 737
 738/* Range common stuff */
 739static inline void range_cpu_to_le(struct lu_seq_range *dst, const struct lu_seq_range *src)
 740{
 741        dst->lsr_start = cpu_to_le64(src->lsr_start);
 742        dst->lsr_end = cpu_to_le64(src->lsr_end);
 743        dst->lsr_index = cpu_to_le32(src->lsr_index);
 744        dst->lsr_flags = cpu_to_le32(src->lsr_flags);
 745}
 746
 747static inline void range_le_to_cpu(struct lu_seq_range *dst, const struct lu_seq_range *src)
 748{
 749        dst->lsr_start = le64_to_cpu(src->lsr_start);
 750        dst->lsr_end = le64_to_cpu(src->lsr_end);
 751        dst->lsr_index = le32_to_cpu(src->lsr_index);
 752        dst->lsr_flags = le32_to_cpu(src->lsr_flags);
 753}
 754
 755static inline void range_cpu_to_be(struct lu_seq_range *dst, const struct lu_seq_range *src)
 756{
 757        dst->lsr_start = cpu_to_be64(src->lsr_start);
 758        dst->lsr_end = cpu_to_be64(src->lsr_end);
 759        dst->lsr_index = cpu_to_be32(src->lsr_index);
 760        dst->lsr_flags = cpu_to_be32(src->lsr_flags);
 761}
 762
 763static inline void range_be_to_cpu(struct lu_seq_range *dst, const struct lu_seq_range *src)
 764{
 765        dst->lsr_start = be64_to_cpu(src->lsr_start);
 766        dst->lsr_end = be64_to_cpu(src->lsr_end);
 767        dst->lsr_index = be32_to_cpu(src->lsr_index);
 768        dst->lsr_flags = be32_to_cpu(src->lsr_flags);
 769}
 770
 771/** @} fid */
 772
 773#endif /* __LUSTRE_FID_H */
 774