X-Git-Url: https://git.ozlabs.org/?p=ccan;a=blobdiff_plain;f=ccan%2Ftdb2%2Fprivate.h;h=b60f33a0a2012224d9e3192090cad82dd1d76d64;hp=387e49f61182af8fe33028840ebc3be073fea70a;hb=a56db4a5cbf4346ec8a74322b5267d47b2590ed7;hpb=dfae76fd82d4bbd8989264dadc2c3c9cde7e5af7

diff --git a/ccan/tdb2/private.h b/ccan/tdb2/private.h
index 387e49f6..b60f33a0 100644
--- a/ccan/tdb2/private.h
+++ b/ccan/tdb2/private.h
@@ -36,6 +36,7 @@
 #include "config.h"
 #include <ccan/tdb2/tdb2.h>
 #include <ccan/likely/likely.h>
+#include <ccan/compiler/compiler.h>
 #ifdef HAVE_BYTESWAP_H
 #include <byteswap.h>
 #endif
@@ -65,7 +66,6 @@ typedef uint64_t tdb_off_t;
 #define TDB_VERSION ((uint64_t)(0x26011967 + 7))
 #define TDB_MAGIC ((uint64_t)0x1999)
 #define TDB_FREE_MAGIC ((uint64_t)0xFE)
-#define TDB_COALESCING_MAGIC ((uint64_t)0xFD)
 #define TDB_HASH_MAGIC (0xA1ABE11A01092008ULL)
 #define TDB_RECOVERY_MAGIC (0xf53bc0e7ad124589ULL)
 #define TDB_RECOVERY_INVALID_MAGIC (0x0ULL)
@@ -91,20 +91,22 @@ typedef uint64_t tdb_off_t;
 #define TDB_SUBLEVEL_HASH_BITS 6
 /* And 8 entries in each group, ie 8 groups per sublevel. */
 #define TDB_HASH_GROUP_BITS 3
+/* This is currently 10: beyond this we chain. */
+#define TDB_MAX_LEVELS (1+(64-TDB_TOPLEVEL_HASH_BITS) / TDB_SUBLEVEL_HASH_BITS)
 
-/* Extend file by least 32 times larger than needed. */
-#define TDB_EXTENSION_FACTOR 32
+/* Extend file by least 100 times larger than needed. */
+#define TDB_EXTENSION_FACTOR 100
 
 /* We steal bits from the offsets to store hash info. */
 #define TDB_OFF_HASH_GROUP_MASK ((1ULL << TDB_HASH_GROUP_BITS) - 1)
 /* We steal this many upper bits, giving a maximum offset of 64 exabytes. */
 #define TDB_OFF_UPPER_STEAL 8
 #define   TDB_OFF_UPPER_STEAL_EXTRA 7
-#define   TDB_OFF_UPPER_STEAL_TRUNCBIT 1
-/* If this is set, hash is truncated (only 1 bit is valid). */
-#define TDB_OFF_HASH_TRUNCATED_BIT 56
-/* The bit number where we store next level of hash. */
+/* The bit number where we store extra hash bits. */
 #define TDB_OFF_HASH_EXTRA_BIT 57
+#define TDB_OFF_UPPER_STEAL_SUBHASH_BIT 56
+
+/* The bit number where we store the extra hash bits. */
 /* Convenience mask to get actual offset. */
 #define TDB_OFF_MASK \
 	(((1ULL << (64 - TDB_OFF_UPPER_STEAL)) - 1) - TDB_OFF_HASH_GROUP_MASK)
@@ -116,6 +118,9 @@ typedef uint64_t tdb_off_t;
 #define TDB_MIN_DATA_LEN	\
 	(sizeof(struct tdb_free_record) - sizeof(struct tdb_used_record))
 
+/* Indicates this entry is not on an flist (can happen during coalescing) */
+#define TDB_FLIST_NONE ((1ULL << TDB_OFF_UPPER_STEAL) - 1)
+
 #if !HAVE_BSWAP_64
 static inline uint64_t bswap_64(uint64_t x)
 {
@@ -173,20 +178,30 @@ static inline uint16_t rec_magic(const struct tdb_used_record *r)
 }
 
 struct tdb_free_record {
-        uint64_t magic_and_meta; /* TDB_OFF_UPPER_STEAL bits of magic */
-        uint64_t data_len; /* Not counting these two fields. */
-	/* This is why the minimum record size is 16 bytes.  */
-	uint64_t next, prev;
+        uint64_t magic_and_prev; /* TDB_OFF_UPPER_STEAL bits magic, then prev */
+        uint64_t flist_and_len; /* Len not counting these two fields. */
+	/* This is why the minimum record size is 8 bytes.  */
+	uint64_t next;
 };
 
+static inline uint64_t frec_prev(const struct tdb_free_record *f)
+{
+	return f->magic_and_prev & ((1ULL << (64 - TDB_OFF_UPPER_STEAL)) - 1);
+}
+
 static inline uint64_t frec_magic(const struct tdb_free_record *f)
 {
-	return f->magic_and_meta >> (64 - TDB_OFF_UPPER_STEAL);
+	return f->magic_and_prev >> (64 - TDB_OFF_UPPER_STEAL);
+}
+
+static inline uint64_t frec_len(const struct tdb_free_record *f)
+{
+	return f->flist_and_len & ((1ULL << (64 - TDB_OFF_UPPER_STEAL))-1);
 }
 
-static inline uint64_t frec_flist(const struct tdb_free_record *f)
+static inline unsigned frec_flist(const struct tdb_free_record *f)
 {
-	return f->magic_and_meta & ((1ULL << (64 - TDB_OFF_UPPER_STEAL)) - 1);
+	return f->flist_and_len >> (64 - TDB_OFF_UPPER_STEAL);
 }
 
 struct tdb_recovery_record {
@@ -199,6 +214,12 @@ struct tdb_recovery_record {
 	uint64_t eof;
 };
 
+/* If we bottom out of the subhashes, we chain. */
+struct tdb_chain {
+	tdb_off_t rec[1 << TDB_HASH_GROUP_BITS];
+	tdb_off_t next;
+};
+
 /* this is stored at the front of every database */
 struct tdb_header {
 	char magic_food[64]; /* for /etc/magic */
@@ -246,7 +267,7 @@ struct traverse_info {
 		/* We ignore groups here, and treat it as a big array. */
 		unsigned entry;
 		unsigned int total_buckets;
-	} levels[64 / TDB_SUBLEVEL_HASH_BITS];
+	} levels[TDB_MAX_LEVELS + 1];
 	unsigned int num_levels;
 	unsigned int toplevel_group;
 	/* This makes delete-everything-inside-traverse work as expected. */
@@ -298,8 +319,8 @@ struct tdb_context {
 	uint32_t flags;
 
 	/* Logging function */
-	tdb_logfn_t log;
-	void *log_priv;
+	tdb_logfn_t logfn;
+	void *log_private;
 
 	/* Hash function. */
 	tdb_hashfn_t khash;
@@ -311,15 +332,18 @@ struct tdb_context {
 	
 	/* What freelist are we using? */
 	uint64_t flist_off;
+	unsigned int flist;
 
 	/* IO methods: changes for transactions. */
 	const struct tdb_methods *methods;
 
 	/* Lock information */
 	struct tdb_lock_type allrecord_lock;
-	uint64_t num_lockrecs;
+	size_t num_lockrecs;
 	struct tdb_lock_type *lockrecs;
 
+	struct tdb_attribute_stats *stats;
+
 	/* Single list of all TDBs, to avoid multiple opens. */
 	struct tdb_context *next;
 	dev_t device;	
@@ -331,7 +355,7 @@ struct tdb_methods {
 	int (*write)(struct tdb_context *, tdb_off_t, const void *, tdb_len_t);
 	int (*oob)(struct tdb_context *, tdb_off_t, bool);
 	int (*expand_file)(struct tdb_context *, tdb_len_t);
-	void *(*direct)(struct tdb_context *, tdb_off_t, size_t);
+	void *(*direct)(struct tdb_context *, tdb_off_t, size_t, bool);
 };
 
 /*
@@ -405,16 +429,15 @@ void *tdb_convert(const struct tdb_context *tdb, void *buf, tdb_len_t size);
 void tdb_munmap(struct tdb_context *tdb);
 void tdb_mmap(struct tdb_context *tdb);
 
-/* Either make a copy into pad and return that, or return ptr into mmap.
- * Converts endian (ie. will use pad in that case). */
-void *tdb_get(struct tdb_context *tdb, tdb_off_t off, void *pad, size_t len);
-
 /* Either alloc a copy, or give direct access.  Release frees or noop. */
 const void *tdb_access_read(struct tdb_context *tdb,
 			    tdb_off_t off, tdb_len_t len, bool convert);
 void *tdb_access_write(struct tdb_context *tdb,
 		       tdb_off_t off, tdb_len_t len, bool convert);
 
+/* Is this pointer direct?  (Otherwise it's malloced) */
+bool is_direct(const struct tdb_context *tdb, const void *p);
+
 /* Release result of tdb_access_read/write. */
 void tdb_access_release(struct tdb_context *tdb, const void *p);
 /* Commit result of tdb_acces_write. */
@@ -455,6 +478,13 @@ int tdb_write_convert(struct tdb_context *tdb, tdb_off_t off,
 int tdb_read_convert(struct tdb_context *tdb, tdb_off_t off,
 		     void *rec, size_t len);
 
+/* Adds a stat, if it's in range. */
+void add_stat_(struct tdb_context *tdb, uint64_t *stat, size_t val);
+#define add_stat(tdb, statname, val)					\
+	do {								\
+		if (unlikely((tdb)->stats))				\
+			add_stat_((tdb), &(tdb)->stats->statname, (val)); \
+	} while (0)
 
 /* lock.c: */
 void tdb_lock_init(struct tdb_context *tdb);
@@ -510,6 +540,12 @@ int next_in_hash(struct tdb_context *tdb, int ltype,
 int tdb_transaction_recover(struct tdb_context *tdb);
 bool tdb_needs_recovery(struct tdb_context *tdb);
 
+/* tdb.c: */
+void COLD tdb_logerr(struct tdb_context *tdb,
+		     enum TDB_ERROR ecode,
+		     enum tdb_debug_level level,
+		     const char *fmt, ...);
+
 #ifdef TDB_TRACE
 void tdb_trace(struct tdb_context *tdb, const char *op);
 void tdb_trace_seqnum(struct tdb_context *tdb, uint32_t seqnum, const char *op);