* 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.
*/
/*
#ifndef __LINUX_FID_H
#define __LINUX_FID_H
+/** \defgroup fid fid
+ *
+ * @{
+ */
+
/*
* struct lu_fid
*/
+#include <libcfs/libcfs.h>
#include <lustre/lustre_idl.h>
#include <lustre_req_layout.h>
#include <lustre_mdt.h>
-#include <libcfs/libcfs.h>
struct lu_site;
struct lu_context;
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)
-
/** special OID for local objects */
-enum {
+enum local_oid {
/** \see osd_oi_index_create */
OSD_OI_FID_16_OID = 2UL,
/** \see fld_mod_init */
/* 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 */
/* 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).
*/
__u64 lss_width;
+ /*
+ * minimum lss_alloc_set size that should be allocated from
+ * lss_space
+ */
+ __u64 lss_set_width;
+
+ /* transaction no of seq update write operation */
+ __u64 lss_set_transno;
/**
* Pointer to site object, required to access site fld.
*/
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);
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"
{
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 */