X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Finclude%2Flustre_fid.h;h=bb664b2ffefe5b7a9b80ab8999f24b929648d413;hp=6e952b5925de05d9e65892587c722956d22d74ea;hb=c356794cbe6c89281d1287da280d85301629ed9f;hpb=64218d0b47731c039234b16b8343d8ee3b407281 diff --git a/lustre/include/lustre_fid.h b/lustre/include/lustre_fid.h index 6e952b5..bb664b2 100644 --- a/lustre/include/lustre_fid.h +++ b/lustre/include/lustre_fid.h @@ -1,6 +1,4 @@ -/* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*- - * vim:expandtab:shiftwidth=8:tabstop=8: - * +/* * GPL HEADER START * * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. @@ -26,8 +24,10 @@ * GPL HEADER END */ /* - * Copyright 2008 Sun Microsystems, Inc. All rights reserved + * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved. * Use is subject to license terms. + * + * Copyright (c) 2011, 2012, Whamcloud, Inc. */ /* * This file is part of Lustre, http://www.lustre.org/ @@ -41,14 +41,19 @@ #ifndef __LINUX_FID_H #define __LINUX_FID_H +/** \defgroup fid fid + * + * @{ + */ + /* * struct lu_fid */ +#include #include #include #include -#include struct lu_site; struct lu_context; @@ -62,43 +67,36 @@ extern const struct lu_fid LU_DOT_LUSTRE_FID; enum { /* - * This is how may FIDs may be allocated in one sequence. 16384 for - * now. + * This is how may FIDs may be allocated in one sequence(128k) */ - LUSTRE_SEQ_MAX_WIDTH = 0x0000000000000400ULL, + LUSTRE_SEQ_MAX_WIDTH = 0x0000000000020000ULL, /* - * How many sequences may be allocate for meta-sequence (this is 128 - * sequences). + * How many sequences to allocate to a client at once. */ - /* changed to 16 to avoid overflow in test11 */ - LUSTRE_SEQ_META_WIDTH = 0x0000000000000010ULL, + LUSTRE_SEQ_META_WIDTH = 0x0000000000000001ULL, + + /* + * seq allocation pool size. + */ + LUSTRE_SEQ_BATCH_WIDTH = LUSTRE_SEQ_META_WIDTH * 1000, /* * This is how many sequences may be in one super-sequence allocated to * MDTs. */ - LUSTRE_SEQ_SUPER_WIDTH = (LUSTRE_SEQ_META_WIDTH * LUSTRE_SEQ_META_WIDTH) + LUSTRE_SEQ_SUPER_WIDTH = ((1ULL << 30ULL) * LUSTRE_SEQ_META_WIDTH) }; -/** special fid seq: used for local object create. */ -#define FID_SEQ_LOCAL_FILE (FID_SEQ_START + 1) - -/** special fid seq: used for .lustre objects. */ -#define LU_DOT_LUSTRE_SEQ (FID_SEQ_START + 0x02ULL) - -/* Note that reserved SEQ numbers below 12 will conflict with ldiskfs - * inodes in the IGIF namespace, so these reserved SEQ numbers must be - * used sparingly until ldiskfs-based MDT backends and/or IGIF FIDs - * have been completely removed. */ - -/** fid sequence for distributed fs objects */ -#define FID_SEQ_DISTRIBUTED_START (FID_SEQ_START + 0x400ULL) +enum { + /** 2^6 FIDs for OI containers */ + OSD_OI_FID_OID_BITS = 6, + /** reserve enough FIDs in case we want more in the future */ + OSD_OI_FID_OID_BITS_MAX = 10, +}; /** special OID for local objects */ -enum { - /** \see osd_oi_index_create */ - OSD_OI_FID_16_OID = 2UL, +enum local_oid { /** \see fld_mod_init */ FLD_INDEX_OID = 3UL, /** \see fid_mod_init */ @@ -109,11 +107,20 @@ enum { MDD_ORPHAN_OID = 7UL, MDD_LOV_OBJ_OID = 8UL, MDD_CAPA_KEYS_OID = 9UL, - MDD_OBJECTS_OID = 10UL, /** \see mdt_mod_init */ MDT_LAST_RECV_OID = 11UL, - /** \see osd_mod_init */ - OSD_REM_OBJ_DIR_OID = 12UL, + OSD_FS_ROOT_OID = 13UL, + ACCT_USER_OID = 15UL, + ACCT_GROUP_OID = 16UL, + LFSCK_BOOKMARK_OID = 17UL, + OTABLE_IT_OID = 18UL, + OFD_LAST_RECV_OID = 19UL, + OFD_GROUP0_LAST_OID = 20UL, + OFD_GROUP4K_LAST_OID = 20UL+4096, + OFD_LAST_GROUP_OID = 4117UL, + LLOG_CATALOGS_OID = 4118UL, + MGS_CONFIGS_OID = 4119UL, + OFD_HEALTH_CHECK_OID = 4120UL, }; static inline void lu_local_obj_fid(struct lu_fid *fid, __u32 oid) @@ -123,6 +130,19 @@ static inline void lu_local_obj_fid(struct lu_fid *fid, __u32 oid) fid->f_ver = 0; } +static inline int fid_is_otable_it(const struct lu_fid *fid) +{ + return unlikely(fid_seq(fid) == FID_SEQ_LOCAL_FILE && + fid_oid(fid) == OTABLE_IT_OID); +} + +static inline int fid_is_acct(const struct lu_fid *fid) +{ + return fid_seq(fid) == FID_SEQ_LOCAL_FILE && + (fid_oid(fid) == ACCT_USER_OID || + fid_oid(fid) == ACCT_GROUP_OID); +} + enum lu_mgr_type { LUSTRE_SEQ_SERVER, LUSTRE_SEQ_CONTROLLER @@ -139,7 +159,7 @@ struct lu_server_seq; struct lu_client_seq { /* Sequence-controller export. */ struct obd_export *lcs_exp; - cfs_semaphore_t lcs_sem; + cfs_mutex_t lcs_mutex; /* * Range of allowed for allocation sequeces. When using lu_client_seq on @@ -171,6 +191,10 @@ struct lu_client_seq { /* Seq-server for direct talking */ struct lu_server_seq *lcs_srv; + + /* wait queue for fid allocation and update indicator */ + cfs_waitq_t lcs_waitq; + int lcs_update; }; /* server sequence manager interface */ @@ -178,6 +202,10 @@ struct lu_server_seq { /* Available sequences space */ struct lu_seq_range lss_space; + /* keeps highwater in lsr_end for seq allocation algorithm */ + struct lu_seq_range lss_lowater_set; + struct lu_seq_range lss_hiwater_set; + /* * Device for server side seq manager needs (saving sequences to backing * store). @@ -196,8 +224,8 @@ struct lu_server_seq { /* Client interafce to request controller */ struct lu_client_seq *lss_cli; - /* Semaphore for protecting allocation */ - cfs_semaphore_t lss_sem; + /* Mutex for protecting allocation */ + cfs_mutex_t lss_mutex; /* * Service uuid, passed from MDT + seq name to form unique seq name to @@ -211,6 +239,14 @@ struct lu_server_seq { */ __u64 lss_width; + /* + * minimum lss_alloc_set size that should be allocated from + * lss_space + */ + __u64 lss_set_width; + + /* sync is needed for update operation */ + __u32 lss_need_sync; /** * Pointer to site object, required to access site fld. */ @@ -231,12 +267,10 @@ void seq_server_fini(struct lu_server_seq *seq, const struct lu_env *env); int seq_server_alloc_super(struct lu_server_seq *seq, - struct lu_seq_range *in, struct lu_seq_range *out, const struct lu_env *env); int seq_server_alloc_meta(struct lu_server_seq *seq, - struct lu_seq_range *in, struct lu_seq_range *out, const struct lu_env *env); @@ -255,8 +289,10 @@ void seq_client_fini(struct lu_client_seq *seq); void seq_client_flush(struct lu_client_seq *seq); -int seq_client_alloc_fid(struct lu_client_seq *seq, +int seq_client_alloc_fid(const struct lu_env *env, struct lu_client_seq *seq, struct lu_fid *fid); +int seq_client_get_seq(const struct lu_env *env, struct lu_client_seq *seq, + seqno_t *seqnr); /* Fids common stuff */ int fid_is_local(const struct lu_env *env, @@ -266,15 +302,12 @@ int fid_is_local(const struct lu_env *env, struct ldlm_namespace; -enum { - LUSTRE_RES_ID_SEQ_OFF = 0, - LUSTRE_RES_ID_OID_OFF = 1, - LUSTRE_RES_ID_VER_OFF = 2, - LUSTRE_RES_ID_HSH_OFF = 3 -}; - /* - * Build (DLM) resource name from fid. + * Build (DLM) resource name from FID. + * + * NOTE: until Lustre 1.8.7/2.1.1 the fid_ver() was packed into name[2], + * but was moved into name[1] along with the OID to avoid consuming the + * renaming name[2,3] fields that need to be used for the quota identifier. */ static inline struct ldlm_res_id * fid_build_reg_res_name(const struct lu_fid *f, @@ -282,8 +315,7 @@ fid_build_reg_res_name(const struct lu_fid *f, { memset(name, 0, sizeof *name); name->name[LUSTRE_RES_ID_SEQ_OFF] = fid_seq(f); - name->name[LUSTRE_RES_ID_OID_OFF] = fid_oid(f); - name->name[LUSTRE_RES_ID_VER_OFF] = fid_ver(f); + name->name[LUSTRE_RES_ID_VER_OID_OFF] = fid_ver_oid(f); return name; } @@ -293,10 +325,8 @@ fid_build_reg_res_name(const struct lu_fid *f, static inline int fid_res_name_eq(const struct lu_fid *f, const struct ldlm_res_id *name) { - return - name->name[LUSTRE_RES_ID_SEQ_OFF] == fid_seq(f) && - name->name[LUSTRE_RES_ID_OID_OFF] == fid_oid(f) && - name->name[LUSTRE_RES_ID_VER_OFF] == fid_ver(f); + return name->name[LUSTRE_RES_ID_SEQ_OFF] == fid_seq(f) && + name->name[LUSTRE_RES_ID_VER_OID_OFF] == fid_ver_oid(f); } @@ -310,9 +340,66 @@ fid_build_pdo_res_name(const struct lu_fid *f, return name; } + +/** + * Flatten 128-bit FID values into a 64-bit value for use as an inode number. + * For non-IGIF FIDs this starts just over 2^32, and continues without + * conflict until 2^64, at which point we wrap the high 24 bits of the SEQ + * into the range where there may not be many OID values in use, to minimize + * the risk of conflict. + * + * Suppose LUSTRE_SEQ_MAX_WIDTH less than (1 << 24) which is currently true, + * the time between re-used inode numbers is very long - 2^40 SEQ numbers, + * or about 2^40 client mounts, if clients create less than 2^24 files/mount. + */ static inline __u64 fid_flatten(const struct lu_fid *fid) { - return (fid_seq(fid) - 1) * LUSTRE_SEQ_MAX_WIDTH + fid_oid(fid); + __u64 ino; + __u64 seq; + + if (fid_is_igif(fid)) { + ino = lu_igif_ino(fid); + RETURN(ino); + } + + seq = fid_seq(fid); + + ino = (seq << 24) + ((seq >> 24) & 0xffffff0000ULL) + fid_oid(fid); + + RETURN(ino ? ino : fid_oid(fid)); +} + +static inline __u32 fid_hash(const struct lu_fid *f, int bits) +{ + /* all objects with same id and different versions will belong to same + * collisions list. */ + return cfs_hash_long(fid_flatten(f), bits); +} + +/** + * map fid to 32 bit value for ino on 32bit systems. */ +static inline __u32 fid_flatten32(const struct lu_fid *fid) +{ + __u32 ino; + __u64 seq; + + if (fid_is_igif(fid)) { + ino = lu_igif_ino(fid); + RETURN(ino); + } + + seq = fid_seq(fid) - FID_SEQ_START; + + /* Map the high bits of the OID into higher bits of the inode number so + * that inodes generated at about the same time have a reduced chance + * of collisions. This will give a period of 2^12 = 1024 unique clients + * (from SEQ) and up to min(LUSTRE_SEQ_MAX_WIDTH, 2^20) = 128k objects + * (from OID), or up to 128M inodes without collisions for new files. */ + ino = ((seq & 0x000fffffULL) << 12) + ((seq >> 8) & 0xfffff000) + + (seq >> (64 - (40-8)) & 0xffffff00) + + (fid_oid(fid) & 0xff000fff) + ((fid_oid(fid) & 0x00fff000) << 8); + + RETURN(ino ? ino : fid_oid(fid)); } #define LUSTRE_SEQ_SRV_NAME "seq_srv" @@ -323,28 +410,34 @@ static inline void range_cpu_to_le(struct lu_seq_range *dst, const struct lu_seq { dst->lsr_start = cpu_to_le64(src->lsr_start); dst->lsr_end = cpu_to_le64(src->lsr_end); - dst->lsr_mdt = cpu_to_le32(src->lsr_mdt); + dst->lsr_index = cpu_to_le32(src->lsr_index); + dst->lsr_flags = cpu_to_le32(src->lsr_flags); } static inline void range_le_to_cpu(struct lu_seq_range *dst, const struct lu_seq_range *src) { dst->lsr_start = le64_to_cpu(src->lsr_start); dst->lsr_end = le64_to_cpu(src->lsr_end); - dst->lsr_mdt = le32_to_cpu(src->lsr_mdt); + dst->lsr_index = le32_to_cpu(src->lsr_index); + dst->lsr_flags = le32_to_cpu(src->lsr_flags); } static inline void range_cpu_to_be(struct lu_seq_range *dst, const struct lu_seq_range *src) { dst->lsr_start = cpu_to_be64(src->lsr_start); dst->lsr_end = cpu_to_be64(src->lsr_end); - dst->lsr_mdt = cpu_to_be32(src->lsr_mdt); + dst->lsr_index = cpu_to_be32(src->lsr_index); + dst->lsr_flags = cpu_to_be32(src->lsr_flags); } static inline void range_be_to_cpu(struct lu_seq_range *dst, const struct lu_seq_range *src) { dst->lsr_start = be64_to_cpu(src->lsr_start); dst->lsr_end = be64_to_cpu(src->lsr_end); - dst->lsr_mdt = be32_to_cpu(src->lsr_mdt); + dst->lsr_index = be32_to_cpu(src->lsr_index); + dst->lsr_flags = be32_to_cpu(src->lsr_flags); } +/** @} fid */ + #endif /* __LINUX_FID_H */