X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Finclude%2Flustre_fid.h;h=68947353acda24865b61a7be83ec2df48810527e;hb=e99739d9abf08ca8d76df281e481924fc2ae6849;hp=d00b9af44828527954f66ad1bb7b34d3ce2fefe3;hpb=4f3d5c73fdb2891826d4e7789cbaec39cf4bbf3d;p=fs%2Flustre-release.git diff --git a/lustre/include/lustre_fid.h b/lustre/include/lustre_fid.h index d00b9af..6894735 100644 --- a/lustre/include/lustre_fid.h +++ b/lustre/include/lustre_fid.h @@ -26,7 +26,7 @@ * 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. */ /* @@ -49,11 +49,11 @@ /* * struct lu_fid */ +#include #include #include #include -#include struct lu_site; struct lu_context; @@ -67,17 +67,14 @@ 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. @@ -91,22 +88,8 @@ enum { 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) - /** special OID for local objects */ -enum { +enum local_oid { /** \see osd_oi_index_create */ OSD_OI_FID_16_OID = 2UL, /** \see fld_mod_init */ @@ -181,6 +164,10 @@ 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 */ @@ -330,9 +317,66 @@ 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" @@ -343,28 +387,32 @@ 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 */