-/* mdb.c - memory-mapped database library */
+/** @file mdb.c
+ * @brief memory-mapped database library
+ *
+ * A Btree-based database management library modeled loosely on the
+ * BerkeleyDB API, but much simplified.
+ */
/*
* Copyright 2011 Howard Chu, Symas Corp.
* All rights reserved.
*/
#include <sys/types.h>
#include <sys/stat.h>
-#include <sys/queue.h>
#include <sys/param.h>
+#ifdef _WIN32
+#include <windows.h>
+#else
#include <sys/uio.h>
#include <sys/mman.h>
#ifdef HAVE_SYS_FILE_H
#include <sys/file.h>
#endif
#include <fcntl.h>
+#endif
#include <assert.h>
#include <errno.h>
#include <string.h>
#include <time.h>
#include <unistd.h>
+
+#ifndef _WIN32
#include <pthread.h>
+#endif
#include "mdb.h"
+#include "midl.h"
-#define ULONG unsigned long
-typedef ULONG pgno_t;
+/** @defgroup internal MDB Internals
+ * @{
+ */
+/** @defgroup compat Windows Compatibility Macros
+ * @{
+ */
+#ifdef _WIN32
+#define pthread_t DWORD
+#define pthread_mutex_t HANDLE
+#define pthread_key_t DWORD
+#define pthread_self() GetCurrentThreadId()
+#define pthread_key_create(x,y) *(x) = TlsAlloc()
+#define pthread_key_delete(x) TlsFree(x)
+#define pthread_getspecific(x) TlsGetValue(x)
+#define pthread_setspecific(x,y) TlsSetValue(x,y)
+#define pthread_mutex_unlock(x) ReleaseMutex(x)
+#define pthread_mutex_lock(x) WaitForSingleObject(x, INFINITE)
+#define LOCK_MUTEX_R(env) pthread_mutex_lock(env->me_rmutex)
+#define UNLOCK_MUTEX_R(env) pthread_mutex_unlock(env->me_rmutex)
+#define LOCK_MUTEX_W(env) pthread_mutex_lock(env->me_wmutex)
+#define UNLOCK_MUTEX_W(env) pthread_mutex_unlock(env->me_wmutex)
+#define getpid() GetCurrentProcessId()
+#define fdatasync(fd) !FlushFileBuffers(fd)
+#define ErrCode() GetLastError()
+#define GetPageSize(x) {SYSTEM_INFO si; GetSystemInfo(&si); (x) = si.dwPageSize;}
+#define close(fd) CloseHandle(fd)
+#define munmap(ptr,len) UnmapViewOfFile(ptr)
+#else
+#define LOCK_MUTEX_R(env) pthread_mutex_lock(&env->me_txns->mti_mutex)
+#define UNLOCK_MUTEX_R(env) pthread_mutex_unlock(&env->me_txns->mti_mutex)
+#define LOCK_MUTEX_W(env) pthread_mutex_lock(&env->me_txns->mti_wmutex)
+#define UNLOCK_MUTEX_W(env) pthread_mutex_unlock(&env->me_txns->mti_wmutex)
+#define ErrCode() errno
+#define HANDLE int
+#define INVALID_HANDLE_VALUE -1
+#define GetPageSize(x) (x) = sysconf(_SC_PAGE_SIZE)
+#endif
-#include "midl.h"
+/** @} */
+
+#ifndef _WIN32
+/* Note: If O_DSYNC is undefined but exists in /usr/include,
+ * preferably set some compiler flag to get the definition.
+ * Otherwise compile with the less efficient -DMDB_DSYNC=O_SYNC.
+ */
+#ifndef MDB_DSYNC
+# define MDB_DSYNC O_DSYNC
+#endif
+#endif
+
+typedef ULONG pgno_t;
#ifndef DEBUG
-#define DEBUG 1
+#define DEBUG 0
#endif
-#if DEBUG && defined(__GNUC__)
-# define DPRINTF(fmt, ...) \
- fprintf(stderr, "%s:%d: " fmt "\n", __func__, __LINE__, ##__VA_ARGS__)
+#if !(__STDC_VERSION__ >= 199901L || defined(__GNUC__))
+# define DPRINTF (void) /* Vararg macros may be unsupported */
+#elif DEBUG
+# define DPRINTF(fmt, ...) /* Requires 2 or more args */ \
+ fprintf(stderr, "%s:%d:(%p) " fmt "\n", __func__, __LINE__, pthread_self(), __VA_ARGS__)
#else
-# define DPRINTF(...) ((void) 0)
+# define DPRINTF(fmt, ...) ((void) 0)
#endif
+#define DPUTS(arg) DPRINTF("%s", arg)
#define PAGESIZE 4096
-#define MDB_MINKEYS 4
+#define MDB_MINKEYS 2
#define MDB_MAGIC 0xBEEFC0DE
#define MDB_VERSION 1
#define MAXKEYSIZE 511
+#if DEBUG
+#define KBUF (MAXKEYSIZE*2+1)
+#define DKBUF char kbuf[KBUF]
+#define DKEY(x) mdb_dkey(x, kbuf)
+#else
+#define DKBUF
+#define DKEY(x)
+#endif
+
+/* The DB view is always consistent because all writes are wrapped in
+ * the wmutex. Finer-grained locks aren't necessary.
+ */
+#ifndef LAZY_LOCKS
+#define LAZY_LOCKS 1
+#endif
+#if LAZY_LOCKS
+#define LAZY_MUTEX_LOCK(x)
+#define LAZY_MUTEX_UNLOCK(x)
+#define LAZY_RWLOCK_UNLOCK(x)
+#define LAZY_RWLOCK_WRLOCK(x)
+#define LAZY_RWLOCK_RDLOCK(x)
+#define LAZY_RWLOCK_DEF(x)
+#define LAZY_RWLOCK_INIT(x,y)
+#define LAZY_RWLOCK_DESTROY(x)
+#else
+#define LAZY_MUTEX_LOCK(x) pthread_mutex_lock(x)
+#define LAZY_MUTEX_UNLOCK(x) pthread_mutex_unlock(x)
+#define LAZY_RWLOCK_UNLOCK(x) pthread_rwlock_unlock(x)
+#define LAZY_RWLOCK_WRLOCK(x) pthread_rwlock_wrlock(x)
+#define LAZY_RWLOCK_RDLOCK(x) pthread_rwlock_rdlock(x)
+#define LAZY_RWLOCK_DEF(x) pthread_rwlock_t x
+#define LAZY_RWLOCK_INIT(x,y) pthread_rwlock_init(x,y)
+#define LAZY_RWLOCK_DESTROY(x) pthread_rwlock_destroy(x)
+#endif
#define P_INVALID (~0UL)
typedef struct MDB_txbody {
uint32_t mtb_magic;
uint32_t mtb_version;
+/* For POSIX the actual mutexes reside in shared memory.
+ * On Windows, mutexes are allocated by the kernel; we store
+ * the name in shared memory so that other processes can
+ * grab them.
+ */
+#ifdef _WIN32
+ char mtb_rmname[32];
+#else
pthread_mutex_t mtb_mutex;
+#endif
ULONG mtb_txnid;
uint32_t mtb_numreaders;
uint32_t mtb_me_toggle;
#define mti_magic mt1.mtb.mtb_magic
#define mti_version mt1.mtb.mtb_version
#define mti_mutex mt1.mtb.mtb_mutex
+#define mti_rmname mt1.mtb.mtb_rmname
#define mti_txnid mt1.mtb.mtb_txnid
#define mti_numreaders mt1.mtb.mtb_numreaders
#define mti_me_toggle mt1.mtb.mtb_me_toggle
char pad[(sizeof(MDB_txbody)+CACHELINE-1) & ~(CACHELINE-1)];
} mt1;
union {
+#ifdef _WIN32
+ char mt2_wmname[32];
+#define mti_wmname mt2.mt2_wmname
+#else
pthread_mutex_t mt2_wmutex;
#define mti_wmutex mt2.mt2_wmutex
+#endif
char pad[(sizeof(pthread_mutex_t)+CACHELINE-1) & ~(CACHELINE-1)];
} mt2;
MDB_reader mti_readers[1];
indx_t pb_upper; /* upper bound of free space */
} pb;
uint32_t pb_pages; /* number of overflow pages */
- struct {
- indx_t pb_ksize; /* on a LEAF2 page */
- indx_t pb_numkeys;
- } pb2;
} mp_pb;
indx_t mp_ptrs[1]; /* dynamic size */
} MDB_page;
#define PAGEFILL(env, p) (1000L * ((env)->me_psize - PAGEHDRSZ - SIZELEFT(p)) / \
((env)->me_psize - PAGEHDRSZ))
#define IS_LEAF(p) F_ISSET((p)->mp_flags, P_LEAF)
+#define IS_LEAF2(p) F_ISSET((p)->mp_flags, P_LEAF2)
#define IS_BRANCH(p) F_ISSET((p)->mp_flags, P_BRANCH)
#define IS_OVERFLOW(p) F_ISSET((p)->mp_flags, P_OVERFLOW)
-#define OVPAGES(size, psize) (PAGEHDRSZ + size + psize - 1) / psize;
+#define OVPAGES(size, psize) ((PAGEHDRSZ-1 + (size)) / (psize) + 1)
typedef struct MDB_db {
- uint32_t md_pad;
+ uint32_t md_pad; /* also ksize for LEAF2 pages */
uint16_t md_flags;
uint16_t md_depth;
ULONG md_branch_pages;
unsigned mp_pi;
} MDB_pageparent;
-static MDB_dpage *mdb_alloc_page(MDB_txn *txn, MDB_page *parent, unsigned int parent_idx, int num);
-static int mdb_touch(MDB_txn *txn, MDB_pageparent *mp);
+static MDB_dpage *mdb_alloc_page(MDB_txn *txn, MDB_dbi dbi, MDB_page *parent, unsigned int parent_idx, int num);
+static int mdb_touch(MDB_txn *txn, MDB_dbi dbi, MDB_pageparent *mp);
typedef struct MDB_ppage { /* ordered list of pages */
- SLIST_ENTRY(MDB_ppage) mp_entry;
MDB_page *mp_page;
unsigned int mp_ki; /* cursor index on page */
} MDB_ppage;
-SLIST_HEAD(page_stack, MDB_ppage);
-/* FIXME: tree depth is mostly bounded, we should just
- * use a fixed array and avoid malloc/pointer chasing
- */
-#define CURSOR_EMPTY(c) SLIST_EMPTY(&(c)->mc_stack)
-#define CURSOR_TOP(c) SLIST_FIRST(&(c)->mc_stack)
-#define CURSOR_POP(c) SLIST_REMOVE_HEAD(&(c)->mc_stack, mp_entry)
-#define CURSOR_PUSH(c,p) SLIST_INSERT_HEAD(&(c)->mc_stack, p, mp_entry)
+#define CURSOR_TOP(c) (&(c)->mc_stack[(c)->mc_snum-1])
+#define CURSOR_PARENT(c) (&(c)->mc_stack[(c)->mc_snum-2])
+#define CURSOR_STACK 32
struct MDB_xcursor;
struct MDB_cursor {
MDB_txn *mc_txn;
- struct page_stack mc_stack; /* stack of parent pages */
+ MDB_ppage mc_stack[CURSOR_STACK]; /* stack of parent pages */
+ unsigned int mc_snum; /* number of pushed pages */
MDB_dbi mc_dbi;
short mc_initialized; /* 1 if initialized */
short mc_eof; /* 1 if end is reached */
struct MDB_xcursor *mc_xcursor;
};
-#define METADATA(p) ((void *)((char *)p + PAGEHDRSZ))
+#define METADATA(p) ((void *)((char *)(p) + PAGEHDRSZ))
typedef struct MDB_node {
#define mn_pgno mn_p.np_pgno
unsigned int mn_ksize:12; /* key size */
#define F_BIGDATA 0x01 /* data put on overflow page */
#define F_SUBDATA 0x02 /* data is a sub-database */
+#define F_DUPDATA 0x04 /* data has duplicates */
char mn_data[1];
} MDB_node;
struct MDB_txn {
pgno_t mt_next_pgno; /* next unallocated page */
ULONG mt_txnid;
- ULONG mt_oldest;
MDB_env *mt_env;
pgno_t *mt_free_pgs; /* this is an IDL */
union {
- MIDL2 *dirty_list; /* modified pages */
+ ID2L dirty_list; /* modified pages */
MDB_reader *reader;
} mt_u;
MDB_dbx *mt_dbxs; /* array */
#define MDB_TXN_RDONLY 0x01 /* read-only transaction */
#define MDB_TXN_ERROR 0x02 /* an error has occurred */
-#define MDB_TXN_METOGGLE 0x04 /* used meta page 1 */
unsigned int mt_flags;
+ unsigned int mt_toggle;
};
/* Context for sorted-dup records */
} MDB_xcursor;
struct MDB_env {
- int me_fd;
- int me_lfd;
- int me_mfd; /* just for writing the meta pages */
+ HANDLE me_fd;
+ HANDLE me_lfd;
+ HANDLE me_mfd; /* just for writing the meta pages */
#define MDB_FATAL_ERROR 0x80000000U
uint32_t me_flags;
uint32_t me_extrapad; /* unused for now */
char *me_map;
MDB_txninfo *me_txns;
MDB_meta *me_metas[2];
- MDB_meta *me_meta;
MDB_txn *me_txn; /* current write transaction */
size_t me_mapsize;
off_t me_size; /* current file size */
MDB_db *me_dbs[2];
MDB_oldpages *me_pghead;
pthread_key_t me_txkey; /* thread-key for readers */
+ MDB_dpage *me_dpages;
pgno_t me_free_pgs[MDB_IDL_UM_SIZE];
- MIDL2 me_dirty_list[MDB_IDL_DB_SIZE];
+ ID2 me_dirty_list[MDB_IDL_DB_SIZE];
+ LAZY_RWLOCK_DEF(me_dblock);
+#ifdef _WIN32
+ HANDLE me_rmutex; /* Windows mutexes don't reside in shared mem */
+ HANDLE me_wmutex;
+#endif
};
#define NODESIZE offsetof(MDB_node, mn_data)
#define NODEDATA(node) (void *)((char *)(node)->mn_data + (node)->mn_ksize)
#define NODEPGNO(node) ((node)->mn_pgno)
#define NODEDSZ(node) ((node)->mn_dsize)
+#define NODEKSZ(node) ((node)->mn_ksize)
+#define LEAF2KEY(p, i, ks) ((char *)(p) + PAGEHDRSZ + ((i)*(ks)))
+
+#define MDB_SET_KEY(node, key) if (key!=NULL) {(key)->mv_size = NODEKSZ(node); (key)->mv_data = NODEKEY(node);}
#define MDB_COMMIT_PAGES 64 /* max number of pages to write in one commit */
static int mdb_env_read_header(MDB_env *env, MDB_meta *meta);
static int mdb_env_read_meta(MDB_env *env, int *which);
static int mdb_env_write_meta(MDB_txn *txn);
-static MDB_page *mdb_get_page(MDB_txn *txn, pgno_t pgno);
+static int mdb_get_page(MDB_txn *txn, pgno_t pgno, MDB_page **mp);
static MDB_node *mdb_search_node(MDB_txn *txn, MDB_dbi dbi, MDB_page *mp,
MDB_val *key, int *exactp, unsigned int *kip);
static int mdb_add_node(MDB_txn *txn, MDB_dbi dbi, MDB_page *mp,
indx_t indx, MDB_val *key, MDB_val *data,
pgno_t pgno, uint8_t flags);
-static void mdb_del_node(MDB_page *mp, indx_t indx);
+static void mdb_del_node(MDB_page *mp, indx_t indx, int ksize);
static int mdb_del0(MDB_txn *txn, MDB_dbi dbi, unsigned int ki,
MDB_pageparent *mpp, MDB_node *leaf);
static int mdb_put0(MDB_txn *txn, MDB_dbi dbi,
static MDB_ppage *cursor_push_page(MDB_cursor *cursor,
MDB_page *mp);
-static int mdb_set_key(MDB_node *node, MDB_val *key);
static int mdb_sibling(MDB_cursor *cursor, int move_right);
static int mdb_cursor_next(MDB_cursor *cursor,
MDB_val *key, MDB_val *data, MDB_cursor_op op);
static int mdb_cursor_last(MDB_cursor *cursor,
MDB_val *key, MDB_val *data);
-static void mdb_xcursor_init0(MDB_txn *txn, MDB_dbi dbi, MDB_xcursor *mx);
-static void mdb_xcursor_init1(MDB_txn *txn, MDB_dbi dbi, MDB_xcursor *mx, MDB_node *node);
+static void mdb_xcursor_init0(MDB_cursor *mc);
+static void mdb_xcursor_init1(MDB_txn *txn, MDB_dbi dbi, MDB_xcursor *mx,
+ MDB_page *mp, MDB_node *node);
static void mdb_xcursor_fini(MDB_txn *txn, MDB_dbi dbi, MDB_xcursor *mx);
static size_t mdb_leaf_size(MDB_env *env, MDB_val *key,
static int memnrcmp(const void *s1, size_t n1,
const void *s2, size_t n2);
+#ifdef _WIN32
+static SECURITY_DESCRIPTOR mdb_null_sd;
+static SECURITY_ATTRIBUTES mdb_all_sa;
+static int mdb_sec_inited;
+#endif
+
static int
memncmp(const void *s1, size_t n1, const void *s2, size_t n2)
{
char *
mdb_version(int *maj, int *min, int *pat)
{
- *maj = MDB_VERSION_MAJOR;
- *min = MDB_VERSION_MINOR;
- *pat = MDB_VERSION_PATCH;
+ if (maj) *maj = MDB_VERSION_MAJOR;
+ if (min) *min = MDB_VERSION_MINOR;
+ if (pat) *pat = MDB_VERSION_PATCH;
return MDB_VERSION_STRING;
}
-static const char *errstr[] = {
+static char *const mdb_errstr[] = {
"MDB_KEYEXIST: Key/data pair already exists",
"MDB_NOTFOUND: No matching key/data pair found",
"MDB_PAGE_NOTFOUND: Requested page not found",
return ("Successful return: 0");
if (err >= MDB_KEYEXIST && err <= MDB_VERSION_MISMATCH)
- return (char *)errstr[err - MDB_KEYEXIST];
+ return mdb_errstr[err - MDB_KEYEXIST];
return strerror(err);
}
-int
-mdb_cmp(MDB_txn *txn, MDB_dbi dbi, const MDB_val *a, const MDB_val *b)
+#if DEBUG
+static char *
+mdb_dkey(MDB_val *key, char *buf)
{
- return txn->mt_dbxs[dbi].md_cmp(a, b);
+ char *ptr = buf;
+ unsigned char *c = key->mv_data;
+ unsigned int i;
+ if (key->mv_size > MAXKEYSIZE)
+ return "MAXKEYSIZE";
+ for (i=0; i<key->mv_size; i++)
+ ptr += sprintf(ptr, "%02x", *c++);
+ return buf;
}
+#endif
-static int
-_mdb_cmp(MDB_txn *txn, MDB_dbi dbi, const MDB_val *key1, const MDB_val *key2)
+int
+mdb_cmp(MDB_txn *txn, MDB_dbi dbi, const MDB_val *a, const MDB_val *b)
{
+ if (txn->mt_dbxs[dbi].md_cmp)
+ return txn->mt_dbxs[dbi].md_cmp(a, b);
+
if (txn->mt_dbs[dbi].md_flags & (MDB_REVERSEKEY
#if __BYTE_ORDER == __LITTLE_ENDIAN
|MDB_INTEGERKEY
#endif
))
- return memnrcmp(key1->mv_data, key1->mv_size, key2->mv_data, key2->mv_size);
+ return memnrcmp(a->mv_data, a->mv_size, b->mv_data, b->mv_size);
else
- return memncmp((char *)key1->mv_data, key1->mv_size, key2->mv_data, key2->mv_size);
+ return memncmp((char *)a->mv_data, a->mv_size, b->mv_data, b->mv_size);
+}
+
+int
+mdb_dcmp(MDB_txn *txn, MDB_dbi dbi, const MDB_val *a, const MDB_val *b)
+{
+ if (txn->mt_dbxs[dbi].md_dcmp)
+ return txn->mt_dbxs[dbi].md_dcmp(a, b);
+
+ if (txn->mt_dbs[dbi].md_flags & (0
+#if __BYTE_ORDER == __LITTLE_ENDIAN
+ |MDB_INTEGERDUP
+#endif
+ ))
+ return memnrcmp(a->mv_data, a->mv_size, b->mv_data, b->mv_size);
+ else
+ return memncmp((char *)a->mv_data, a->mv_size, b->mv_data, b->mv_size);
}
/* Allocate new page(s) for writing */
static MDB_dpage *
-mdb_alloc_page(MDB_txn *txn, MDB_page *parent, unsigned int parent_idx, int num)
+mdb_alloc_page(MDB_txn *txn, MDB_dbi dbi, MDB_page *parent, unsigned int parent_idx, int num)
{
MDB_dpage *dp;
pgno_t pgno = P_INVALID;
- ULONG oldest;
- MIDL2 mid;
+ ID2 mid;
if (txn->mt_txnid > 2) {
- oldest = txn->mt_txnid - 2;
- if (!txn->mt_env->me_pghead && txn->mt_dbs[FREE_DBI].md_root != P_INVALID) {
- /* See if there's anything in the free DB */
- MDB_pageparent mpp;
- MDB_node *leaf;
- ULONG *kptr;
-
- mpp.mp_parent = NULL;
- mpp.mp_pi = 0;
- mdb_search_page(txn, FREE_DBI, NULL, NULL, 0, &mpp);
- leaf = NODEPTR(mpp.mp_page, 0);
- kptr = (ULONG *)NODEKEY(leaf);
+ if (!txn->mt_env->me_pghead && dbi != FREE_DBI &&
+ txn->mt_dbs[FREE_DBI].md_root != P_INVALID) {
+ /* See if there's anything in the free DB */
+ MDB_pageparent mpp;
+ MDB_node *leaf;
+ ULONG *kptr, oldest;
- /* It's potentially usable, unless there are still
- * older readers outstanding. Grab it.
- */
- if (oldest > *kptr) {
- MDB_oldpages *mop;
- MDB_val data;
- pgno_t *idl;
-
- mdb_read_data(txn, leaf, &data);
- idl = (ULONG *)data.mv_data;
- mop = malloc(sizeof(MDB_oldpages) + MDB_IDL_SIZEOF(idl) - sizeof(pgno_t));
- mop->mo_next = txn->mt_env->me_pghead;
- mop->mo_txnid = *kptr;
- txn->mt_env->me_pghead = mop;
- memcpy(mop->mo_pages, idl, MDB_IDL_SIZEOF(idl));
+ mpp.mp_parent = NULL;
+ mpp.mp_pi = 0;
+ mdb_search_page(txn, FREE_DBI, NULL, NULL, 0, &mpp);
+ leaf = NODEPTR(mpp.mp_page, 0);
+ kptr = (ULONG *)NODEKEY(leaf);
-#if DEBUG > 1
{
unsigned int i;
- DPRINTF("IDL read txn %lu root %lu num %lu",
- mop->mo_txnid, txn->mt_dbs[FREE_DBI].md_root, idl[0]);
- for (i=0; i<idl[0]; i++) {
- DPRINTF("IDL %lu", idl[i+1]);
+ oldest = txn->mt_txnid - 1;
+ for (i=0; i<txn->mt_env->me_txns->mti_numreaders; i++) {
+ ULONG mr = txn->mt_env->me_txns->mti_readers[i].mr_txnid;
+ if (mr && mr < oldest)
+ oldest = mr;
}
}
+
+ if (oldest > *kptr) {
+ /* It's usable, grab it.
+ */
+ MDB_oldpages *mop;
+ MDB_val data;
+ pgno_t *idl;
+
+ mdb_read_data(txn, leaf, &data);
+ idl = (ULONG *)data.mv_data;
+ mop = malloc(sizeof(MDB_oldpages) + MDB_IDL_SIZEOF(idl) - sizeof(pgno_t));
+ mop->mo_next = txn->mt_env->me_pghead;
+ mop->mo_txnid = *kptr;
+ txn->mt_env->me_pghead = mop;
+ memcpy(mop->mo_pages, idl, MDB_IDL_SIZEOF(idl));
+
+#if DEBUG > 1
+ {
+ unsigned int i;
+ DPRINTF("IDL read txn %lu root %lu num %lu",
+ mop->mo_txnid, txn->mt_dbs[FREE_DBI].md_root, idl[0]);
+ for (i=0; i<idl[0]; i++) {
+ DPRINTF("IDL %lu", idl[i+1]);
+ }
+ }
#endif
- /* drop this IDL from the DB */
- mpp.mp_parent = NULL;
- mpp.mp_pi = 0;
- mdb_search_page(txn, FREE_DBI, NULL, NULL, 1, &mpp);
- leaf = NODEPTR(mpp.mp_page, 0);
- mdb_del0(txn, FREE_DBI, 0, &mpp, leaf);
- }
- }
- if (txn->mt_env->me_pghead) {
- unsigned int i;
- for (i=0; i<txn->mt_env->me_txns->mti_numreaders; i++) {
- ULONG mr = txn->mt_env->me_txns->mti_readers[i].mr_txnid;
- if (!mr) continue;
- if (mr < oldest)
- oldest = txn->mt_env->me_txns->mti_readers[i].mr_txnid;
+ /* drop this IDL from the DB */
+ mpp.mp_parent = NULL;
+ mpp.mp_pi = 0;
+ mdb_search_page(txn, FREE_DBI, NULL, NULL, 1, &mpp);
+ leaf = NODEPTR(mpp.mp_page, 0);
+ mdb_del0(txn, FREE_DBI, 0, &mpp, leaf);
+ }
}
- if (oldest > txn->mt_env->me_pghead->mo_txnid) {
+ if (txn->mt_env->me_pghead) {
MDB_oldpages *mop = txn->mt_env->me_pghead;
- txn->mt_oldest = oldest;
if (num > 1) {
/* FIXME: For now, always use fresh pages. We
* really ought to search the free list for a
}
}
}
- }
if (pgno == P_INVALID) {
/* DB size is maxed out */
if (txn->mt_next_pgno + num >= txn->mt_env->me_maxpg)
return NULL;
}
- if ((dp = malloc(txn->mt_env->me_psize * num + sizeof(MDB_dhead))) == NULL)
- return NULL;
+ if (txn->mt_env->me_dpages && num == 1) {
+ dp = txn->mt_env->me_dpages;
+ txn->mt_env->me_dpages = (MDB_dpage *)dp->h.md_parent;
+ } else {
+ if ((dp = malloc(txn->mt_env->me_psize * num + sizeof(MDB_dhead))) == NULL)
+ return NULL;
+ }
dp->h.md_num = num;
dp->h.md_parent = parent;
dp->h.md_pi = parent_idx;
}
mid.mid = dp->p.mp_pgno;
mid.mptr = dp;
- mdb_midl2_insert(txn->mt_u.dirty_list, &mid);
+ mdb_mid2l_insert(txn->mt_u.dirty_list, &mid);
return dp;
}
/* Touch a page: make it dirty and re-insert into tree with updated pgno.
*/
static int
-mdb_touch(MDB_txn *txn, MDB_pageparent *pp)
+mdb_touch(MDB_txn *txn, MDB_dbi dbi, MDB_pageparent *pp)
{
MDB_page *mp = pp->mp_page;
pgno_t pgno;
if (!F_ISSET(mp->mp_flags, P_DIRTY)) {
MDB_dpage *dp;
- if ((dp = mdb_alloc_page(txn, pp->mp_parent, pp->mp_pi, 1)) == NULL)
+ if ((dp = mdb_alloc_page(txn, dbi, pp->mp_parent, pp->mp_pi, 1)) == NULL)
return ENOMEM;
- DPRINTF("touched page %lu -> %lu", mp->mp_pgno, dp->p.mp_pgno);
+ DPRINTF("touched db %u page %lu -> %lu", dbi, mp->mp_pgno, dp->p.mp_pgno);
+ assert(mp->mp_pgno != dp->p.mp_pgno);
mdb_midl_insert(txn->mt_free_pgs, mp->mp_pgno);
pgno = dp->p.mp_pgno;
memcpy(&dp->p, mp, txn->mt_env->me_psize);
int rc = 0;
if (force || !F_ISSET(env->me_flags, MDB_NOSYNC)) {
if (fdatasync(env->me_fd))
- rc = errno;
+ rc = ErrCode();
}
return rc;
}
-int
-mdb_txn_begin(MDB_env *env, int rdonly, MDB_txn **ret)
-{
- MDB_txn *txn;
- int rc, toggle;
-
- if (env->me_flags & MDB_FATAL_ERROR) {
- DPRINTF("mdb_txn_begin: environment had fatal error, must shutdown!");
- return MDB_PANIC;
- }
- if ((txn = calloc(1, sizeof(MDB_txn))) == NULL) {
- DPRINTF("calloc: %s", strerror(errno));
- return ENOMEM;
- }
-
- if (rdonly) {
- txn->mt_flags |= MDB_TXN_RDONLY;
- } else {
- txn->mt_u.dirty_list = env->me_dirty_list;
- txn->mt_u.dirty_list[0].mid = 0;
- txn->mt_free_pgs = env->me_free_pgs;
- txn->mt_free_pgs[0] = 0;
+static inline void
+mdb_txn_reset0(MDB_txn *txn);
- pthread_mutex_lock(&env->me_txns->mti_wmutex);
- env->me_txns->mti_txnid++;
- }
+static inline int
+mdb_txn_renew0(MDB_txn *txn)
+{
+ MDB_env *env = txn->mt_env;
- txn->mt_txnid = env->me_txns->mti_txnid;
- if (rdonly) {
+ if (txn->mt_flags & MDB_TXN_RDONLY) {
MDB_reader *r = pthread_getspecific(env->me_txkey);
if (!r) {
unsigned int i;
- pthread_mutex_lock(&env->me_txns->mti_mutex);
+ pid_t pid = getpid();
+ pthread_t tid = pthread_self();
+
+ LOCK_MUTEX_R(env);
for (i=0; i<env->me_txns->mti_numreaders; i++)
if (env->me_txns->mti_readers[i].mr_pid == 0)
break;
if (i == env->me_maxreaders) {
- pthread_mutex_unlock(&env->me_txns->mti_mutex);
- return ENOSPC;
+ UNLOCK_MUTEX_R(env);
+ return ENOMEM;
}
- env->me_txns->mti_readers[i].mr_pid = getpid();
- env->me_txns->mti_readers[i].mr_tid = pthread_self();
- r = &env->me_txns->mti_readers[i];
- pthread_setspecific(env->me_txkey, r);
+ env->me_txns->mti_readers[i].mr_pid = pid;
+ env->me_txns->mti_readers[i].mr_tid = tid;
if (i >= env->me_txns->mti_numreaders)
env->me_txns->mti_numreaders = i+1;
- pthread_mutex_unlock(&env->me_txns->mti_mutex);
+ UNLOCK_MUTEX_R(env);
+ r = &env->me_txns->mti_readers[i];
+ pthread_setspecific(env->me_txkey, r);
}
+ txn->mt_txnid = env->me_txns->mti_txnid;
+ txn->mt_toggle = env->me_txns->mti_me_toggle;
r->mr_txnid = txn->mt_txnid;
txn->mt_u.reader = r;
} else {
- env->me_txn = txn;
- }
-
- txn->mt_env = env;
+ LOCK_MUTEX_W(env);
- toggle = env->me_txns->mti_me_toggle;
- if ((rc = mdb_env_read_meta(env, &toggle)) != MDB_SUCCESS) {
- mdb_txn_abort(txn);
- return rc;
+ txn->mt_txnid = env->me_txns->mti_txnid+1;
+ txn->mt_toggle = env->me_txns->mti_me_toggle;
+ txn->mt_u.dirty_list = env->me_dirty_list;
+ txn->mt_u.dirty_list[0].mid = 0;
+ txn->mt_free_pgs = env->me_free_pgs;
+ txn->mt_free_pgs[0] = 0;
+ txn->mt_next_pgno = env->me_metas[txn->mt_toggle]->mm_last_pg+1;
+ env->me_txn = txn;
}
/* Copy the DB arrays */
+ LAZY_RWLOCK_RDLOCK(&env->me_dblock);
txn->mt_numdbs = env->me_numdbs;
txn->mt_dbxs = env->me_dbxs; /* mostly static anyway */
- txn->mt_dbs = malloc(env->me_maxdbs * sizeof(MDB_db));
- memcpy(txn->mt_dbs, env->me_meta->mm_dbs, 2 * sizeof(MDB_db));
+ memcpy(txn->mt_dbs, env->me_metas[txn->mt_toggle]->mm_dbs, 2 * sizeof(MDB_db));
if (txn->mt_numdbs > 2)
memcpy(txn->mt_dbs+2, env->me_dbs[env->me_db_toggle]+2,
(txn->mt_numdbs - 2) * sizeof(MDB_db));
+ LAZY_RWLOCK_UNLOCK(&env->me_dblock);
- if (!rdonly) {
- if (toggle)
- txn->mt_flags |= MDB_TXN_METOGGLE;
- txn->mt_next_pgno = env->me_meta->mm_last_pg+1;
- }
+ return MDB_SUCCESS;
+}
+
+int
+mdb_txn_renew(MDB_txn *txn)
+{
+ int rc;
- DPRINTF("begin transaction %lu on mdbenv %p, root page %lu",
- txn->mt_txnid, (void *) env, txn->mt_dbs[MAIN_DBI].md_root);
+ if (!txn)
+ return EINVAL;
- *ret = txn;
- return MDB_SUCCESS;
+ if (txn->mt_env->me_flags & MDB_FATAL_ERROR) {
+ DPUTS("environment had fatal error, must shutdown!");
+ return MDB_PANIC;
+ }
+
+ rc = mdb_txn_renew0(txn);
+ if (rc == MDB_SUCCESS) {
+ DPRINTF("renew txn %lu%c %p on mdbenv %p, root page %lu",
+ txn->mt_txnid, (txn->mt_flags & MDB_TXN_RDONLY) ? 'r' : 'w', txn,
+ (void *)txn->mt_env, txn->mt_dbs[MAIN_DBI].md_root);
+ }
+ return rc;
}
-void
-mdb_txn_abort(MDB_txn *txn)
+int
+mdb_txn_begin(MDB_env *env, unsigned int flags, MDB_txn **ret)
{
- MDB_env *env;
+ MDB_txn *txn;
+ int rc;
- if (txn == NULL)
- return;
+ if (env->me_flags & MDB_FATAL_ERROR) {
+ DPUTS("environment had fatal error, must shutdown!");
+ return MDB_PANIC;
+ }
+ if ((txn = calloc(1, sizeof(MDB_txn) + env->me_maxdbs * sizeof(MDB_db))) == NULL) {
+ DPRINTF("calloc: %s", strerror(ErrCode()));
+ return ENOMEM;
+ }
+ txn->mt_dbs = (MDB_db *)(txn+1);
+ if (flags & MDB_RDONLY) {
+ txn->mt_flags |= MDB_TXN_RDONLY;
+ }
+ txn->mt_env = env;
- env = txn->mt_env;
- DPRINTF("abort transaction %lu on mdbenv %p, root page %lu",
- txn->mt_txnid, (void *) env, txn->mt_dbs[MAIN_DBI].md_root);
+ rc = mdb_txn_renew0(txn);
+ if (rc)
+ free(txn);
+ else {
+ *ret = txn;
+ DPRINTF("begin txn %lu%c %p on mdbenv %p, root page %lu",
+ txn->mt_txnid, (txn->mt_flags & MDB_TXN_RDONLY) ? 'r' : 'w', txn,
+ (void *) env, txn->mt_dbs[MAIN_DBI].md_root);
+ }
+
+ return rc;
+}
- free(txn->mt_dbs);
+static inline void
+mdb_txn_reset0(MDB_txn *txn)
+{
+ MDB_env *env = txn->mt_env;
if (F_ISSET(txn->mt_flags, MDB_TXN_RDONLY)) {
txn->mt_u.reader->mr_txnid = 0;
} else {
MDB_oldpages *mop;
+ MDB_dpage *dp;
unsigned int i;
- /* Discard all dirty pages. */
- for (i=1; i<=txn->mt_u.dirty_list[0].mid; i++)
- free(txn->mt_u.dirty_list[i].mptr);
+ /* return all dirty pages to dpage list */
+ for (i=1; i<=txn->mt_u.dirty_list[0].mid; i++) {
+ dp = txn->mt_u.dirty_list[i].mptr;
+ if (dp->h.md_num == 1) {
+ dp->h.md_parent = (MDB_page *)txn->mt_env->me_dpages;
+ txn->mt_env->me_dpages = dp;
+ } else {
+ /* large pages just get freed directly */
+ free(dp);
+ }
+ }
while ((mop = txn->mt_env->me_pghead)) {
txn->mt_env->me_pghead = mop->mo_next;
}
env->me_txn = NULL;
- env->me_txns->mti_txnid--;
for (i=2; i<env->me_numdbs; i++)
env->me_dbxs[i].md_dirty = 0;
- pthread_mutex_unlock(&env->me_txns->mti_wmutex);
+ UNLOCK_MUTEX_W(env);
}
+}
+
+void
+mdb_txn_reset(MDB_txn *txn)
+{
+ if (txn == NULL)
+ return;
+
+ DPRINTF("reset txn %lu%c %p on mdbenv %p, root page %lu",
+ txn->mt_txnid, (txn->mt_flags & MDB_TXN_RDONLY) ? 'r' : 'w', txn,
+ (void *)txn->mt_env, txn->mt_dbs[MAIN_DBI].md_root);
+
+ mdb_txn_reset0(txn);
+}
+
+void
+mdb_txn_abort(MDB_txn *txn)
+{
+ if (txn == NULL)
+ return;
+
+ DPRINTF("abort txn %lu%c %p on mdbenv %p, root page %lu",
+ txn->mt_txnid, (txn->mt_flags & MDB_TXN_RDONLY) ? 'r' : 'w', txn,
+ (void *)txn->mt_env, txn->mt_dbs[MAIN_DBI].md_root);
+ mdb_txn_reset0(txn);
free(txn);
}
MDB_dpage *dp;
MDB_env *env;
pgno_t next;
- struct iovec iov[MDB_COMMIT_PAGES];
assert(txn != NULL);
assert(txn->mt_env != NULL);
}
if (txn != env->me_txn) {
- DPRINTF("attempt to commit unknown transaction");
+ DPUTS("attempt to commit unknown transaction");
mdb_txn_abort(txn);
return EINVAL;
}
if (F_ISSET(txn->mt_flags, MDB_TXN_ERROR)) {
- DPRINTF("error flag is set, can't commit");
+ DPUTS("error flag is set, can't commit");
mdb_txn_abort(txn);
return EINVAL;
}
if (!txn->mt_u.dirty_list[0].mid)
goto done;
- DPRINTF("committing transaction %lu on mdbenv %p, root page %lu",
- txn->mt_txnid, (void *) env, txn->mt_dbs[MAIN_DBI].md_root);
+ DPRINTF("committing txn %lu %p on mdbenv %p, root page %lu",
+ txn->mt_txnid, txn, (void *)env, txn->mt_dbs[MAIN_DBI].md_root);
/* should only be one record now */
if (env->me_pghead) {
- MDB_val key, data;
- MDB_oldpages *mop;
+ MDB_pageparent mpp;
- mop = env->me_pghead;
- key.mv_size = sizeof(pgno_t);
- key.mv_data = (char *)&mop->mo_txnid;
- data.mv_size = MDB_IDL_SIZEOF(mop->mo_pages);
- data.mv_data = mop->mo_pages;
- mdb_put0(txn, FREE_DBI, &key, &data, 0);
- free(env->me_pghead);
- env->me_pghead = NULL;
+ /* make sure first page of freeDB is touched and on freelist */
+ mpp.mp_parent = NULL;
+ mpp.mp_pi = 0;
+ mdb_search_page(txn, FREE_DBI, NULL, NULL, 1, &mpp);
}
/* save to free list */
if (!MDB_IDL_IS_ZERO(txn->mt_free_pgs)) {
MDB_val key, data;
MDB_pageparent mpp;
+ ULONG i;
/* make sure last page of freeDB is touched and on freelist */
key.mv_size = MAXKEYSIZE+1;
/* write to last page of freeDB */
key.mv_size = sizeof(pgno_t);
key.mv_data = (char *)&txn->mt_txnid;
- data.mv_size = MDB_IDL_SIZEOF(txn->mt_free_pgs);
data.mv_data = txn->mt_free_pgs;
+ /* The free list can still grow during this call,
+ * despite the pre-emptive touches above. So check
+ * and make sure the entire thing got written.
+ */
+ do {
+ i = txn->mt_free_pgs[0];
+ data.mv_size = MDB_IDL_SIZEOF(txn->mt_free_pgs);
+ rc = mdb_put0(txn, FREE_DBI, &key, &data, 0);
+ if (rc) {
+ mdb_txn_abort(txn);
+ return rc;
+ }
+ } while (i != txn->mt_free_pgs[0]);
+ }
+ /* should only be one record now */
+ if (env->me_pghead) {
+ MDB_val key, data;
+ MDB_oldpages *mop;
+
+ mop = env->me_pghead;
+ key.mv_size = sizeof(pgno_t);
+ key.mv_data = (char *)&mop->mo_txnid;
+ data.mv_size = MDB_IDL_SIZEOF(mop->mo_pages);
+ data.mv_data = mop->mo_pages;
mdb_put0(txn, FREE_DBI, &key, &data, 0);
+ free(env->me_pghead);
+ env->me_pghead = NULL;
}
/* Update DB root pointers. Their pages have already been
next = 0;
i = 1;
do {
+#ifdef _WIN32
+ /* Windows actually supports scatter/gather I/O, but only on
+ * unbuffered file handles. Since we're relying on the OS page
+ * cache for all our data, that's self-defeating. So we just
+ * write pages one at a time. We use the ov structure to set
+ * the write offset, to at least save the overhead of a Seek
+ * system call.
+ */
+ OVERLAPPED ov;
+ memset(&ov, 0, sizeof(ov));
+ for (; i<=txn->mt_u.dirty_list[0].mid; i++) {
+ dp = txn->mt_u.dirty_list[i].mptr;
+ DPRINTF("committing page %lu", dp->p.mp_pgno);
+ size = dp->p.mp_pgno * env->me_psize;
+ ov.Offset = size & 0xffffffff;
+ ov.OffsetHigh = size >> 16;
+ ov.OffsetHigh >>= 16;
+ /* clear dirty flag */
+ dp->p.mp_flags &= ~P_DIRTY;
+ rc = WriteFile(env->me_fd, &dp->p, env->me_psize * dp->h.md_num, NULL, &ov);
+ if (!rc) {
+ n = ErrCode();
+ DPRINTF("WriteFile: %d", n);
+ mdb_txn_abort(txn);
+ return n;
+ }
+ }
+ done = 1;;
+#else
+ struct iovec iov[MDB_COMMIT_PAGES];
n = 0;
done = 1;
size = 0;
DPRINTF("committing %u dirty pages", n);
rc = writev(env->me_fd, iov, n);
if (rc != size) {
- n = errno;
+ n = ErrCode();
if (rc > 0)
- DPRINTF("short write, filesystem full?");
+ DPUTS("short write, filesystem full?");
else
- DPRINTF("writev: %s", strerror(errno));
+ DPRINTF("writev: %s", strerror(n));
mdb_txn_abort(txn);
return n;
}
dp->p.mp_flags &= ~P_DIRTY;
if (++n >= MDB_COMMIT_PAGES) {
done = 0;
+ i++;
break;
}
}
DPRINTF("committing %u dirty pages", n);
rc = writev(env->me_fd, iov, n);
if (rc != size) {
- n = errno;
+ n = ErrCode();
if (rc > 0)
- DPRINTF("short write, filesystem full?");
+ DPUTS("short write, filesystem full?");
else
- DPRINTF("writev: %s", strerror(errno));
+ DPRINTF("writev: %s", strerror(n));
mdb_txn_abort(txn);
return n;
}
-
+#endif
} while (!done);
/* Drop the dirty pages.
*/
- for (i=1; i<=txn->mt_u.dirty_list[0].mid; i++)
- free(txn->mt_u.dirty_list[i].mptr);
+ for (i=1; i<=txn->mt_u.dirty_list[0].mid; i++) {
+ dp = txn->mt_u.dirty_list[i].mptr;
+ if (dp->h.md_num == 1) {
+ dp->h.md_parent = (MDB_page *)txn->mt_env->me_dpages;
+ txn->mt_env->me_dpages = dp;
+ } else {
+ free(dp);
+ }
+ txn->mt_u.dirty_list[i].mid = 0;
+ }
+ txn->mt_u.dirty_list[0].mid = 0;
if ((n = mdb_env_sync(env, 0)) != 0 ||
(n = mdb_env_write_meta(txn)) != MDB_SUCCESS) {
/* update the DB tables */
{
int toggle = !env->me_db_toggle;
+ MDB_db *ip, *jp;
- for (i = 2; i < env->me_numdbs; i++) {
- if (txn->mt_dbxs[i].md_dirty) {
- env->me_dbs[toggle][i] = txn->mt_dbs[i];
- txn->mt_dbxs[i].md_dirty = 0;
- }
+ ip = &env->me_dbs[toggle][2];
+ jp = &txn->mt_dbs[2];
+ LAZY_RWLOCK_WRLOCK(&env->me_dblock);
+ for (i = 2; i < txn->mt_numdbs; i++) {
+ if (ip->md_root != jp->md_root)
+ *ip = *jp;
+ ip++; jp++;
}
- for (i = env->me_numdbs; i < txn->mt_numdbs; i++) {
- txn->mt_dbxs[i].md_dirty = 0;
- env->me_dbxs[i] = txn->mt_dbxs[i];
- env->me_dbs[toggle][i] = txn->mt_dbs[i];
+
+ for (i = 2; i < txn->mt_numdbs; i++) {
+ if (txn->mt_dbxs[i].md_dirty)
+ txn->mt_dbxs[i].md_dirty = 0;
}
env->me_db_toggle = toggle;
env->me_numdbs = txn->mt_numdbs;
-
- free(txn->mt_dbs);
+ LAZY_RWLOCK_UNLOCK(&env->me_dblock);
}
- pthread_mutex_unlock(&env->me_txns->mti_wmutex);
+ UNLOCK_MUTEX_W(env);
free(txn);
return MDB_SUCCESS;
char page[PAGESIZE];
MDB_page *p;
MDB_meta *m;
- int rc;
-
- assert(env != NULL);
+ int rc, err;
/* We don't know the page size yet, so use a minimum value.
*/
- if ((rc = pread(env->me_fd, page, PAGESIZE, 0)) == 0) {
+#ifdef _WIN32
+ if (!ReadFile(env->me_fd, page, PAGESIZE, (DWORD *)&rc, NULL) || rc == 0)
+#else
+ if ((rc = read(env->me_fd, page, PAGESIZE)) == 0)
+#endif
+ {
return ENOENT;
- } else if (rc != PAGESIZE) {
+ }
+ else if (rc != PAGESIZE) {
+ err = ErrCode();
if (rc > 0)
- errno = EINVAL;
- DPRINTF("read: %s", strerror(errno));
- return errno;
+ err = EINVAL;
+ DPRINTF("read: %s", strerror(err));
+ return err;
}
p = (MDB_page *)page;
m = METADATA(p);
if (m->mm_magic != MDB_MAGIC) {
- DPRINTF("meta has invalid magic");
+ DPUTS("meta has invalid magic");
return EINVAL;
}
int rc;
unsigned int psize;
- DPRINTF("writing new meta page");
- psize = sysconf(_SC_PAGE_SIZE);
+ DPUTS("writing new meta page");
+
+ GetPageSize(psize);
meta->mm_magic = MDB_MAGIC;
meta->mm_version = MDB_VERSION;
m = METADATA(q);
memcpy(m, meta, sizeof(*meta));
+#ifdef _WIN32
+ {
+ DWORD len;
+ rc = WriteFile(env->me_fd, p, psize * 2, &len, NULL);
+ rc = (len == psize * 2) ? MDB_SUCCESS : ErrCode();
+ }
+#else
rc = write(env->me_fd, p, psize * 2);
+ rc = (rc == (int)psize * 2) ? MDB_SUCCESS : ErrCode();
+#endif
free(p);
- return (rc == (int)psize * 2) ? MDB_SUCCESS : errno;
+ return rc;
}
static int
off_t off;
int rc, len, toggle;
char *ptr;
+#ifdef _WIN32
+ OVERLAPPED ov;
+#endif
assert(txn != NULL);
assert(txn->mt_env != NULL);
- toggle = !F_ISSET(txn->mt_flags, MDB_TXN_METOGGLE);
+ toggle = !txn->mt_toggle;
DPRINTF("writing meta page %d for root page %lu",
toggle, txn->mt_dbs[MAIN_DBI].md_root);
off += PAGEHDRSZ;
/* Write to the SYNC fd */
+#ifdef _WIN32
+ {
+ memset(&ov, 0, sizeof(ov));
+ ov.Offset = off;
+ WriteFile(env->me_mfd, ptr, len, (DWORD *)&rc, &ov);
+ }
+#else
rc = pwrite(env->me_mfd, ptr, len, off);
+#endif
if (rc != len) {
- DPRINTF("write failed, disk error?");
+ int r2;
+ rc = ErrCode();
+ DPUTS("write failed, disk error?");
/* On a failure, the pagecache still contains the new data.
* Write some old data back, to prevent it from being used.
* Use the non-SYNC fd; we know it will fail anyway.
*/
meta.mm_last_pg = metab.mm_last_pg;
meta.mm_txnid = metab.mm_txnid;
- rc = pwrite(env->me_fd, ptr, len, off);
+#ifdef _WIN32
+ WriteFile(env->me_fd, ptr, len, NULL, &ov);
+#else
+ r2 = pwrite(env->me_fd, ptr, len, off);
+#endif
env->me_flags |= MDB_FATAL_ERROR;
- return errno;
+ return rc;
}
+ /* Memory ordering issues are irrelevant; since the entire writer
+ * is wrapped by wmutex, all of these changes will become visible
+ * after the wmutex is unlocked. Since the DB is multi-version,
+ * readers will get consistent data regardless of how fresh or
+ * how stale their view of these values is.
+ */
+ LAZY_MUTEX_LOCK(&env->me_txns->mti_mutex);
txn->mt_env->me_txns->mti_me_toggle = toggle;
+ txn->mt_env->me_txns->mti_txnid = txn->mt_txnid;
+ LAZY_MUTEX_UNLOCK(&env->me_txns->mti_mutex);
return MDB_SUCCESS;
}
assert(env != NULL);
- if (which)
- toggle = *which;
- else if (env->me_metas[0]->mm_txnid < env->me_metas[1]->mm_txnid)
+ if (env->me_metas[0]->mm_txnid < env->me_metas[1]->mm_txnid)
toggle = 1;
- if (env->me_meta != env->me_metas[toggle])
- env->me_meta = env->me_metas[toggle];
-
DPRINTF("Using meta page %d", toggle);
+ *which = toggle;
return MDB_SUCCESS;
}
e->me_maxreaders = DEFAULT_READERS;
e->me_maxdbs = 2;
- e->me_fd = -1;
- e->me_lfd = -1;
- e->me_mfd = -1;
+ e->me_fd = INVALID_HANDLE_VALUE;
+ e->me_lfd = INVALID_HANDLE_VALUE;
+ e->me_mfd = INVALID_HANDLE_VALUE;
*env = e;
return MDB_SUCCESS;
}
int
mdb_env_set_maxdbs(MDB_env *env, int dbs)
{
+ if (env->me_map)
+ return EINVAL;
env->me_maxdbs = dbs;
return MDB_SUCCESS;
}
int
mdb_env_set_maxreaders(MDB_env *env, int readers)
{
+ if (env->me_map)
+ return EINVAL;
env->me_maxreaders = readers;
return MDB_SUCCESS;
}
static int
mdb_env_open2(MDB_env *env, unsigned int flags)
{
- int i, newenv = 0;
+ int i, newenv = 0, toggle;
MDB_meta meta;
MDB_page *p;
if ((i = mdb_env_read_header(env, &meta)) != 0) {
if (i != ENOENT)
return i;
- DPRINTF("new mdbenv");
+ DPUTS("new mdbenv");
newenv = 1;
}
env->me_mapsize = newenv ? DEFAULT_MAPSIZE : meta.mm_mapsize;
}
+#ifdef _WIN32
+ {
+ HANDLE mh;
+ LONG sizelo, sizehi;
+ sizelo = env->me_mapsize & 0xffffffff;
+ sizehi = env->me_mapsize >> 16; /* pointless on WIN32, only needed on W64 */
+ sizehi >>= 16;
+ /* Windows won't create mappings for zero length files.
+ * Just allocate the maxsize right now.
+ */
+ if (newenv) {
+ SetFilePointer(env->me_fd, sizelo, sizehi ? &sizehi : NULL, 0);
+ if (!SetEndOfFile(env->me_fd))
+ return ErrCode();
+ SetFilePointer(env->me_fd, 0, NULL, 0);
+ }
+ mh = CreateFileMapping(env->me_fd, NULL, PAGE_READONLY,
+ sizehi, sizelo, NULL);
+ if (!mh)
+ return ErrCode();
+ env->me_map = MapViewOfFileEx(mh, FILE_MAP_READ, 0, 0, env->me_mapsize,
+ meta.mm_address);
+ CloseHandle(mh);
+ if (!env->me_map)
+ return ErrCode();
+ }
+#else
i = MAP_SHARED;
if (meta.mm_address && (flags & MDB_FIXEDMAP))
i |= MAP_FIXED;
env->me_map = mmap(meta.mm_address, env->me_mapsize, PROT_READ, i,
env->me_fd, 0);
if (env->me_map == MAP_FAILED)
- return errno;
+ return ErrCode();
+#endif
if (newenv) {
meta.mm_mapsize = env->me_mapsize;
env->me_metas[0] = METADATA(p);
env->me_metas[1] = (MDB_meta *)((char *)env->me_metas[0] + meta.mm_psize);
- if ((i = mdb_env_read_meta(env, NULL)) != 0)
+ if ((i = mdb_env_read_meta(env, &toggle)) != 0)
return i;
DPRINTF("opened database version %u, pagesize %u",
- env->me_meta->mm_version, env->me_psize);
- DPRINTF("depth: %u", env->me_meta->mm_dbs[MAIN_DBI].md_depth);
- DPRINTF("entries: %lu", env->me_meta->mm_dbs[MAIN_DBI].md_entries);
- DPRINTF("branch pages: %lu", env->me_meta->mm_dbs[MAIN_DBI].md_branch_pages);
- DPRINTF("leaf pages: %lu", env->me_meta->mm_dbs[MAIN_DBI].md_leaf_pages);
- DPRINTF("overflow pages: %lu", env->me_meta->mm_dbs[MAIN_DBI].md_overflow_pages);
- DPRINTF("root: %lu", env->me_meta->mm_dbs[MAIN_DBI].md_root);
+ env->me_metas[toggle]->mm_version, env->me_psize);
+ DPRINTF("depth: %u", env->me_metas[toggle]->mm_dbs[MAIN_DBI].md_depth);
+ DPRINTF("entries: %lu", env->me_metas[toggle]->mm_dbs[MAIN_DBI].md_entries);
+ DPRINTF("branch pages: %lu", env->me_metas[toggle]->mm_dbs[MAIN_DBI].md_branch_pages);
+ DPRINTF("leaf pages: %lu", env->me_metas[toggle]->mm_dbs[MAIN_DBI].md_leaf_pages);
+ DPRINTF("overflow pages: %lu", env->me_metas[toggle]->mm_dbs[MAIN_DBI].md_overflow_pages);
+ DPRINTF("root: %lu", env->me_metas[toggle]->mm_dbs[MAIN_DBI].md_root);
return MDB_SUCCESS;
}
+#ifndef _WIN32
+/* Windows doesn't support destructor callbacks for thread-specific storage */
static void
mdb_env_reader_dest(void *ptr)
{
reader->mr_pid = 0;
reader->mr_tid = 0;
}
+#endif
/* downgrade the exclusive lock on the region back to shared */
static void
mdb_env_share_locks(MDB_env *env)
{
- struct flock lock_info;
+ int toggle = 0;
- env->me_txns->mti_txnid = env->me_meta->mm_txnid;
if (env->me_metas[0]->mm_txnid < env->me_metas[1]->mm_txnid)
- env->me_txns->mti_me_toggle = 1;
+ toggle = 1;
+ env->me_txns->mti_me_toggle = toggle;
+ env->me_txns->mti_txnid = env->me_metas[toggle]->mm_txnid;
- memset((void *)&lock_info, 0, sizeof(lock_info));
- lock_info.l_type = F_RDLCK;
- lock_info.l_whence = SEEK_SET;
- lock_info.l_start = 0;
- lock_info.l_len = 1;
- fcntl(env->me_lfd, F_SETLK, &lock_info);
+#ifdef _WIN32
+ {
+ OVERLAPPED ov;
+ /* First acquire a shared lock. The Unlock will
+ * then release the existing exclusive lock.
+ */
+ memset(&ov, 0, sizeof(ov));
+ LockFileEx(env->me_lfd, 0, 0, 1, 0, &ov);
+ UnlockFile(env->me_lfd, 0, 0, 1, 0);
+ }
+#else
+ {
+ struct flock lock_info;
+ /* The shared lock replaces the existing lock */
+ memset((void *)&lock_info, 0, sizeof(lock_info));
+ lock_info.l_type = F_RDLCK;
+ lock_info.l_whence = SEEK_SET;
+ lock_info.l_start = 0;
+ lock_info.l_len = 1;
+ fcntl(env->me_lfd, F_SETLK, &lock_info);
+ }
+#endif
}
static int
{
int rc;
off_t size, rsize;
- struct flock lock_info;
*excl = 0;
+#ifdef _WIN32
+ if ((env->me_lfd = CreateFile(lpath, GENERIC_READ|GENERIC_WRITE,
+ FILE_SHARE_READ|FILE_SHARE_WRITE, NULL, OPEN_ALWAYS,
+ FILE_ATTRIBUTE_NORMAL, NULL)) == INVALID_HANDLE_VALUE) {
+ rc = ErrCode();
+ return rc;
+ }
+ /* Try to get exclusive lock. If we succeed, then
+ * nobody is using the lock region and we should initialize it.
+ */
+ {
+ if (LockFile(env->me_lfd, 0, 0, 1, 0)) {
+ *excl = 1;
+ } else {
+ OVERLAPPED ov;
+ memset(&ov, 0, sizeof(ov));
+ if (!LockFileEx(env->me_lfd, 0, 0, 1, 0, &ov)) {
+ rc = ErrCode();
+ goto fail;
+ }
+ }
+ }
+ size = GetFileSize(env->me_lfd, NULL);
+#else
if ((env->me_lfd = open(lpath, O_RDWR|O_CREAT, mode)) == -1) {
- rc = errno;
+ rc = ErrCode();
return rc;
}
/* Try to get exclusive lock. If we succeed, then
* nobody is using the lock region and we should initialize it.
*/
- memset((void *)&lock_info, 0, sizeof(lock_info));
- lock_info.l_type = F_WRLCK;
- lock_info.l_whence = SEEK_SET;
- lock_info.l_start = 0;
- lock_info.l_len = 1;
- rc = fcntl(env->me_lfd, F_SETLK, &lock_info);
- if (rc == 0) {
- *excl = 1;
- } else {
- lock_info.l_type = F_RDLCK;
+ {
+ struct flock lock_info;
+ memset((void *)&lock_info, 0, sizeof(lock_info));
+ lock_info.l_type = F_WRLCK;
+ lock_info.l_whence = SEEK_SET;
+ lock_info.l_start = 0;
+ lock_info.l_len = 1;
rc = fcntl(env->me_lfd, F_SETLK, &lock_info);
- if (rc) {
- rc = errno;
- goto fail;
+ if (rc == 0) {
+ *excl = 1;
+ } else {
+ lock_info.l_type = F_RDLCK;
+ rc = fcntl(env->me_lfd, F_SETLKW, &lock_info);
+ if (rc) {
+ rc = ErrCode();
+ goto fail;
+ }
}
}
size = lseek(env->me_lfd, 0, SEEK_END);
+#endif
rsize = (env->me_maxreaders-1) * sizeof(MDB_reader) + sizeof(MDB_txninfo);
if (size < rsize && *excl) {
+#ifdef _WIN32
+ SetFilePointer(env->me_lfd, rsize, NULL, 0);
+ if (!SetEndOfFile(env->me_lfd)) {
+ rc = ErrCode();
+ goto fail;
+ }
+#else
if (ftruncate(env->me_lfd, rsize) != 0) {
- rc = errno;
+ rc = ErrCode();
goto fail;
}
+#endif
} else {
rsize = size;
size = rsize - sizeof(MDB_txninfo);
env->me_maxreaders = size/sizeof(MDB_reader) + 1;
}
+#ifdef _WIN32
+ {
+ HANDLE mh;
+ mh = CreateFileMapping(env->me_lfd, NULL, PAGE_READWRITE,
+ 0, 0, NULL);
+ if (!mh) {
+ rc = ErrCode();
+ goto fail;
+ }
+ env->me_txns = MapViewOfFileEx(mh, FILE_MAP_WRITE, 0, 0, rsize, NULL);
+ CloseHandle(mh);
+ if (!env->me_txns) {
+ rc = ErrCode();
+ goto fail;
+ }
+ }
+#else
env->me_txns = mmap(0, rsize, PROT_READ|PROT_WRITE, MAP_SHARED,
env->me_lfd, 0);
if (env->me_txns == MAP_FAILED) {
- rc = errno;
+ rc = ErrCode();
goto fail;
}
+#endif
if (*excl) {
+#ifdef _WIN32
+ char *ptr;
+ if (!mdb_sec_inited) {
+ InitializeSecurityDescriptor(&mdb_null_sd,
+ SECURITY_DESCRIPTOR_REVISION);
+ SetSecurityDescriptorDacl(&mdb_null_sd, TRUE, 0, FALSE);
+ mdb_all_sa.nLength = sizeof(SECURITY_ATTRIBUTES);
+ mdb_all_sa.bInheritHandle = FALSE;
+ mdb_all_sa.lpSecurityDescriptor = &mdb_null_sd;
+ mdb_sec_inited = 1;
+ }
+ /* FIXME: only using up to 20 characters of the env path here,
+ * probably not enough to assure uniqueness...
+ */
+ sprintf(env->me_txns->mti_rmname, "Global\\MDBr%.20s", lpath);
+ ptr = env->me_txns->mti_rmname + sizeof("Global\\MDBr");
+ while ((ptr = strchr(ptr, '\\')))
+ *ptr++ = '/';
+ env->me_rmutex = CreateMutex(&mdb_all_sa, FALSE, env->me_txns->mti_rmname);
+ if (!env->me_rmutex) {
+ rc = ErrCode();
+ goto fail;
+ }
+ sprintf(env->me_txns->mti_rmname, "Global\\MDBw%.20s", lpath);
+ ptr = env->me_txns->mti_rmname + sizeof("Global\\MDBw");
+ while ((ptr = strchr(ptr, '\\')))
+ *ptr++ = '/';
+ env->me_wmutex = CreateMutex(&mdb_all_sa, FALSE, env->me_txns->mti_rmname);
+ if (!env->me_wmutex) {
+ rc = ErrCode();
+ goto fail;
+ }
+#else
pthread_mutexattr_t mattr;
pthread_mutexattr_init(&mattr);
}
pthread_mutex_init(&env->me_txns->mti_mutex, &mattr);
pthread_mutex_init(&env->me_txns->mti_wmutex, &mattr);
+#endif
env->me_txns->mti_version = MDB_VERSION;
env->me_txns->mti_magic = MDB_MAGIC;
env->me_txns->mti_txnid = 0;
} else {
if (env->me_txns->mti_magic != MDB_MAGIC) {
- DPRINTF("lock region has invalid magic");
+ DPUTS("lock region has invalid magic");
rc = EINVAL;
goto fail;
}
rc = MDB_VERSION_MISMATCH;
goto fail;
}
- if (errno != EACCES && errno != EAGAIN) {
- rc = errno;
+ rc = ErrCode();
+ if (rc != EACCES && rc != EAGAIN) {
goto fail;
}
+#ifdef _WIN32
+ env->me_rmutex = OpenMutex(SYNCHRONIZE, FALSE, env->me_txns->mti_rmname);
+ if (!env->me_rmutex) {
+ rc = ErrCode();
+ goto fail;
+ }
+ env->me_wmutex = OpenMutex(SYNCHRONIZE, FALSE, env->me_txns->mti_wmname);
+ if (!env->me_wmutex) {
+ rc = ErrCode();
+ goto fail;
+ }
+#endif
}
return MDB_SUCCESS;
fail:
close(env->me_lfd);
- env->me_lfd = -1;
+ env->me_lfd = INVALID_HANDLE_VALUE;
return rc;
}
if (rc)
goto leave;
+#ifdef _WIN32
+ if (F_ISSET(flags, MDB_RDONLY)) {
+ oflags = GENERIC_READ;
+ len = OPEN_EXISTING;
+ } else {
+ oflags = GENERIC_READ|GENERIC_WRITE;
+ len = OPEN_ALWAYS;
+ }
+ mode = FILE_ATTRIBUTE_NORMAL;
+ if ((env->me_fd = CreateFile(dpath, oflags, FILE_SHARE_READ|FILE_SHARE_WRITE,
+ NULL, len, mode, NULL)) == INVALID_HANDLE_VALUE) {
+ rc = ErrCode();
+ goto leave;
+ }
+#else
if (F_ISSET(flags, MDB_RDONLY))
oflags = O_RDONLY;
else
oflags = O_RDWR | O_CREAT;
if ((env->me_fd = open(dpath, oflags, mode)) == -1) {
- rc = errno;
+ rc = ErrCode();
goto leave;
}
+#endif
if ((rc = mdb_env_open2(env, flags)) == MDB_SUCCESS) {
/* synchronous fd for meta writes */
+#ifdef _WIN32
if (!(flags & (MDB_RDONLY|MDB_NOSYNC)))
- oflags |= O_DSYNC;
+ mode |= FILE_FLAG_WRITE_THROUGH;
+ if ((env->me_mfd = CreateFile(dpath, oflags, FILE_SHARE_READ|FILE_SHARE_WRITE,
+ NULL, len, mode, NULL)) == INVALID_HANDLE_VALUE) {
+ rc = ErrCode();
+ goto leave;
+ }
+#else
+ if (!(flags & (MDB_RDONLY|MDB_NOSYNC)))
+ oflags |= MDB_DSYNC;
if ((env->me_mfd = open(dpath, oflags, mode)) == -1) {
- rc = errno;
+ rc = ErrCode();
goto leave;
}
-
+#endif
env->me_path = strdup(path);
DPRINTF("opened dbenv %p", (void *) env);
pthread_key_create(&env->me_txkey, mdb_env_reader_dest);
+ LAZY_RWLOCK_INIT(&env->me_dblock, NULL);
if (excl)
mdb_env_share_locks(env);
env->me_dbxs = calloc(env->me_maxdbs, sizeof(MDB_dbx));
leave:
if (rc) {
- if (env->me_fd >= 0) {
+ if (env->me_fd != INVALID_HANDLE_VALUE) {
close(env->me_fd);
- env->me_fd = -1;
+ env->me_fd = INVALID_HANDLE_VALUE;
}
- if (env->me_lfd >= 0) {
+ if (env->me_lfd != INVALID_HANDLE_VALUE) {
close(env->me_lfd);
- env->me_lfd = -1;
+ env->me_lfd = INVALID_HANDLE_VALUE;
}
}
free(lpath);
void
mdb_env_close(MDB_env *env)
{
+ MDB_dpage *dp;
+
if (env == NULL)
return;
+ while (env->me_dpages) {
+ dp = env->me_dpages;
+ env->me_dpages = (MDB_dpage *)dp->h.md_parent;
+ free(dp);
+ }
+
free(env->me_dbs[1]);
free(env->me_dbs[0]);
free(env->me_dbxs);
free(env->me_path);
+ LAZY_RWLOCK_DESTROY(&env->me_dblock);
pthread_key_delete(env->me_txkey);
if (env->me_map) {
close(env->me_fd);
if (env->me_txns) {
pid_t pid = getpid();
- size_t size = (env->me_maxreaders-1) * sizeof(MDB_reader) + sizeof(MDB_txninfo);
- int i;
+ unsigned int i;
for (i=0; i<env->me_txns->mti_numreaders; i++)
if (env->me_txns->mti_readers[i].mr_pid == pid)
env->me_txns->mti_readers[i].mr_pid = 0;
- munmap(env->me_txns, size);
+ munmap(env->me_txns, (env->me_maxreaders-1)*sizeof(MDB_reader)+sizeof(MDB_txninfo));
}
close(env->me_lfd);
free(env);
unsigned int i = 0;
int low, high;
int rc = 0;
- MDB_node *node;
+ MDB_node *node = NULL;
MDB_val nodekey;
+ DKBUF;
DPRINTF("searching %u keys in %s page %lu",
NUMKEYS(mp),
high = NUMKEYS(mp) - 1;
while (low <= high) {
i = (low + high) >> 1;
- node = NODEPTR(mp, i);
- nodekey.mv_size = node->mn_ksize;
- nodekey.mv_data = NODEKEY(node);
+ if (IS_LEAF2(mp)) {
+ nodekey.mv_size = txn->mt_dbs[dbi].md_pad;
+ nodekey.mv_data = LEAF2KEY(mp, i, nodekey.mv_size);
+ } else {
+ node = NODEPTR(mp, i);
- if (txn->mt_dbxs[dbi].md_cmp)
- rc = txn->mt_dbxs[dbi].md_cmp(key, &nodekey);
- else
- rc = _mdb_cmp(txn, dbi, key, &nodekey);
+ nodekey.mv_size = node->mn_ksize;
+ nodekey.mv_data = NODEKEY(node);
+ }
+
+ rc = mdb_cmp(txn, dbi, key, &nodekey);
if (IS_LEAF(mp))
- DPRINTF("found leaf index %u [%.*s], rc = %i",
- i, (int)nodekey.mv_size, (char *)nodekey.mv_data, rc);
+ DPRINTF("found leaf index %u [%s], rc = %i",
+ i, DKEY(&nodekey), rc);
else
- DPRINTF("found branch index %u [%.*s -> %lu], rc = %i",
- i, (int)node->mn_ksize, (char *)NODEKEY(node),
- node->mn_pgno, rc);
+ DPRINTF("found branch index %u [%s -> %lu], rc = %i",
+ i, DKEY(&nodekey), NODEPGNO(node), rc);
if (rc == 0)
break;
if (kip) /* Store the key index if requested. */
*kip = i;
- return NODEPTR(mp, i);
+ /* nodeptr is fake for LEAF2 */
+ return IS_LEAF2(mp) ? NODEPTR(mp, 0) : NODEPTR(mp, i);
}
static void
{
MDB_ppage *top;
- top = CURSOR_TOP(cursor);
- CURSOR_POP(cursor);
-
- DPRINTF("popped page %lu off cursor %p", top->mp_page->mp_pgno, (void *) cursor);
+ if (cursor->mc_snum) {
+ top = CURSOR_TOP(cursor);
+ cursor->mc_snum--;
- free(top);
+ DPRINTF("popped page %lu off db %u cursor %p", top->mp_page->mp_pgno,
+ cursor->mc_dbi, (void *) cursor);
+ }
}
static MDB_ppage *
{
MDB_ppage *ppage;
- DPRINTF("pushing page %lu on cursor %p", mp->mp_pgno, (void *) cursor);
+ DPRINTF("pushing page %lu on db %u cursor %p", mp->mp_pgno,
+ cursor->mc_dbi, (void *) cursor);
- if ((ppage = calloc(1, sizeof(MDB_ppage))) == NULL)
- return NULL;
+ assert(cursor->mc_snum < CURSOR_STACK);
+
+ ppage = &cursor->mc_stack[cursor->mc_snum++];
ppage->mp_page = mp;
- CURSOR_PUSH(cursor, ppage);
+ ppage->mp_ki = 0;
return ppage;
}
-static MDB_page *
-mdb_get_page(MDB_txn *txn, pgno_t pgno)
+static int
+mdb_get_page(MDB_txn *txn, pgno_t pgno, MDB_page **ret)
{
MDB_page *p = NULL;
- int found = 0;
if (!F_ISSET(txn->mt_flags, MDB_TXN_RDONLY) && txn->mt_u.dirty_list[0].mid) {
MDB_dpage *dp;
- MIDL2 id;
unsigned x;
- id.mid = pgno;
- x = mdb_midl2_search(txn->mt_u.dirty_list, &id);
+ x = mdb_mid2l_search(txn->mt_u.dirty_list, pgno);
if (x <= txn->mt_u.dirty_list[0].mid && txn->mt_u.dirty_list[x].mid == pgno) {
dp = txn->mt_u.dirty_list[x].mptr;
p = &dp->p;
- found = 1;
}
}
- if (!found) {
- if (pgno > txn->mt_env->me_meta->mm_last_pg)
- return NULL;
- p = (MDB_page *)(txn->mt_env->me_map + txn->mt_env->me_psize * pgno);
+ if (!p) {
+ if (pgno <= txn->mt_env->me_metas[txn->mt_toggle]->mm_last_pg)
+ p = (MDB_page *)(txn->mt_env->me_map + txn->mt_env->me_psize * pgno);
}
- return p;
+ *ret = p;
+ if (!p) {
+ DPRINTF("page %lu not found", pgno);
+ assert(p != NULL);
+ }
+ return (p != NULL) ? MDB_SUCCESS : MDB_PAGE_NOTFOUND;
}
static int
MDB_cursor *cursor, int modify, MDB_pageparent *mpp)
{
MDB_page *mp = mpp->mp_page;
+ DKBUF;
int rc;
if (cursor && cursor_push_page(cursor, mp) == NULL)
}
if (key)
- DPRINTF("following index %u for key %.*s",
- i, (int)key->mv_size, (char *)key->mv_data);
+ DPRINTF("following index %u for key [%s]",
+ i, DKEY(key));
assert(i < NUMKEYS(mp));
node = NODEPTR(mp, i);
CURSOR_TOP(cursor)->mp_ki = i;
mpp->mp_parent = mp;
- if ((mp = mdb_get_page(txn, NODEPGNO(node))) == NULL)
- return MDB_PAGE_NOTFOUND;
+ if ((rc = mdb_get_page(txn, NODEPGNO(node), &mp)))
+ return rc;
mpp->mp_pi = i;
mpp->mp_page = mp;
if (modify) {
MDB_dhead *dh = ((MDB_dhead *)mp)-1;
- if ((rc = mdb_touch(txn, mpp)) != 0)
+ if ((rc = mdb_touch(txn, dbi, mpp)) != 0)
return rc;
dh = ((MDB_dhead *)mpp->mp_page)-1;
dh->md_parent = mpp->mp_parent;
return MDB_CORRUPTED;
}
- DPRINTF("found leaf page %lu for key %.*s", mp->mp_pgno,
- key ? (int)key->mv_size : 0, key ? (char *)key->mv_data : NULL);
+ DPRINTF("found leaf page %lu for key [%s]", mp->mp_pgno,
+ key ? DKEY(key) : NULL);
return MDB_SUCCESS;
}
* committed root page.
*/
if (F_ISSET(txn->mt_flags, MDB_TXN_ERROR)) {
- DPRINTF("transaction has failed, must abort");
+ DPUTS("transaction has failed, must abort");
return EINVAL;
} else
root = txn->mt_dbs[dbi].md_root;
if (root == P_INVALID) { /* Tree is empty. */
- DPRINTF("tree is empty");
+ DPUTS("tree is empty");
return MDB_NOTFOUND;
}
- if ((mpp->mp_page = mdb_get_page(txn, root)) == NULL)
- return MDB_PAGE_NOTFOUND;
+ if ((rc = mdb_get_page(txn, root, &mpp->mp_page)))
+ return rc;
- DPRINTF("root page has flags 0x%X", mpp->mp_page->mp_flags);
+ DPRINTF("db %u root page %lu has flags 0x%X",
+ dbi, root, mpp->mp_page->mp_flags);
if (modify) {
/* For sub-databases, update main root first */
if (!F_ISSET(mpp->mp_page->mp_flags, P_DIRTY)) {
mpp->mp_parent = NULL;
mpp->mp_pi = 0;
- if ((rc = mdb_touch(txn, mpp)))
+ if ((rc = mdb_touch(txn, dbi, mpp)))
return rc;
txn->mt_dbs[dbi].md_root = mpp->mp_page->mp_pgno;
}
{
MDB_page *omp; /* overflow mpage */
pgno_t pgno;
+ int rc;
if (!F_ISSET(leaf->mn_flags, F_BIGDATA)) {
data->mv_size = leaf->mn_dsize;
*/
data->mv_size = leaf->mn_dsize;
memcpy(&pgno, NODEDATA(leaf), sizeof(pgno));
- if ((omp = mdb_get_page(txn, pgno)) == NULL) {
+ if ((rc = mdb_get_page(txn, pgno, &omp))) {
DPRINTF("read overflow page %lu failed", pgno);
- return MDB_PAGE_NOTFOUND;
+ return rc;
}
data->mv_data = METADATA(omp);
mdb_get(MDB_txn *txn, MDB_dbi dbi,
MDB_val *key, MDB_val *data)
{
- int rc, exact;
- MDB_node *leaf;
- MDB_pageparent mpp;
+ MDB_cursor mc;
+ MDB_xcursor mx;
+ int exact;
+ DKBUF;
assert(key);
assert(data);
- DPRINTF("===> get key [%.*s]", (int)key->mv_size, (char *)key->mv_data);
+ DPRINTF("===> get db %u key [%s]", dbi, DKEY(key));
if (txn == NULL || !dbi || dbi >= txn->mt_numdbs)
return EINVAL;
return EINVAL;
}
- if ((rc = mdb_search_page(txn, dbi, key, NULL, 0, &mpp)) != MDB_SUCCESS)
- return rc;
-
- leaf = mdb_search_node(txn, dbi, mpp.mp_page, key, &exact, NULL);
- if (leaf && exact) {
- /* Return first duplicate data item */
- if (F_ISSET(txn->mt_dbs[dbi].md_flags, MDB_DUPSORT)) {
- MDB_xcursor mx;
-
- mdb_xcursor_init0(txn, dbi, &mx);
- mdb_xcursor_init1(txn, dbi, &mx, leaf);
- rc = mdb_search_page(&mx.mx_txn, mx.mx_cursor.mc_dbi, NULL, NULL, 0, &mpp);
- if (rc != MDB_SUCCESS)
- return rc;
- leaf = NODEPTR(mpp.mp_page, 0);
- }
- rc = mdb_read_data(txn, leaf, data);
+ mc.mc_txn = txn;
+ mc.mc_dbi = dbi;
+ mc.mc_initialized = 0;
+ if (txn->mt_dbs[dbi].md_flags & MDB_DUPSORT) {
+ mc.mc_xcursor = &mx;
+ mdb_xcursor_init0(&mc);
} else {
- rc = MDB_NOTFOUND;
+ mc.mc_xcursor = NULL;
}
-
- return rc;
+ return mdb_cursor_set(&mc, key, data, MDB_SET, &exact);
}
static int
{
int rc;
MDB_node *indx;
- MDB_ppage *parent, *top;
+ MDB_ppage *parent;
MDB_page *mp;
- top = CURSOR_TOP(cursor);
- if ((parent = SLIST_NEXT(top, mp_entry)) == NULL) {
+ if (cursor->mc_snum < 2) {
return MDB_NOTFOUND; /* root has no siblings */
}
+ parent = CURSOR_PARENT(cursor);
DPRINTF("parent page is page %lu, index %u",
parent->mp_page->mp_pgno, parent->mp_ki);
assert(IS_BRANCH(parent->mp_page));
indx = NODEPTR(parent->mp_page, parent->mp_ki);
- if ((mp = mdb_get_page(cursor->mc_txn, indx->mn_pgno)) == NULL)
- return MDB_PAGE_NOTFOUND;
+ if ((rc = mdb_get_page(cursor->mc_txn, NODEPGNO(indx), &mp)))
+ return rc;;
#if 0
mp->parent = parent->mp_page;
mp->parent_index = parent->mp_ki;
return MDB_SUCCESS;
}
-static int
-mdb_set_key(MDB_node *node, MDB_val *key)
-{
- if (key == NULL)
- return 0;
-
- key->mv_size = node->mn_ksize;
- key->mv_data = NODEKEY(node);
-
- return 0;
-}
-
static int
mdb_cursor_next(MDB_cursor *cursor, MDB_val *key, MDB_val *data, MDB_cursor_op op)
{
assert(cursor->mc_initialized);
+ top = CURSOR_TOP(cursor);
+ mp = top->mp_page;
+
if (cursor->mc_txn->mt_dbs[cursor->mc_dbi].md_flags & MDB_DUPSORT) {
- if (op == MDB_NEXT || op == MDB_NEXT_DUP) {
- rc = mdb_cursor_next(&cursor->mc_xcursor->mx_cursor, data, NULL, MDB_NEXT);
- if (op != MDB_NEXT || rc == MDB_SUCCESS)
- return rc;
+ leaf = NODEPTR(mp, top->mp_ki);
+ if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
+ if (op == MDB_NEXT || op == MDB_NEXT_DUP) {
+ rc = mdb_cursor_next(&cursor->mc_xcursor->mx_cursor, data, NULL, MDB_NEXT);
+ if (op != MDB_NEXT || rc == MDB_SUCCESS)
+ return rc;
+ }
+ } else {
+ cursor->mc_xcursor->mx_cursor.mc_initialized = 0;
+ if (op == MDB_NEXT_DUP)
+ return MDB_NOTFOUND;
}
}
- top = CURSOR_TOP(cursor);
- mp = top->mp_page;
-
DPRINTF("cursor_next: top page is %lu in cursor %p", mp->mp_pgno, (void *) cursor);
if (top->mp_ki + 1 >= NUMKEYS(mp)) {
- DPRINTF("=====> move to next sibling page");
+ DPUTS("=====> move to next sibling page");
if (mdb_sibling(cursor, 1) != MDB_SUCCESS) {
cursor->mc_eof = 1;
return MDB_NOTFOUND;
DPRINTF("==> cursor points to page %lu with %u keys, key index %u",
mp->mp_pgno, NUMKEYS(mp), top->mp_ki);
+ if (IS_LEAF2(mp)) {
+ key->mv_size = cursor->mc_txn->mt_dbs[cursor->mc_dbi].md_pad;
+ key->mv_data = LEAF2KEY(mp, top->mp_ki, key->mv_size);
+ return MDB_SUCCESS;
+ }
+
assert(IS_LEAF(mp));
leaf = NODEPTR(mp, top->mp_ki);
+ if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
+ mdb_xcursor_init1(cursor->mc_txn, cursor->mc_dbi, cursor->mc_xcursor, mp, leaf);
+ }
if (data) {
if ((rc = mdb_read_data(cursor->mc_txn, leaf, data) != MDB_SUCCESS))
return rc;
- if (cursor->mc_txn->mt_dbs[cursor->mc_dbi].md_flags & MDB_DUPSORT) {
- mdb_xcursor_init1(cursor->mc_txn, cursor->mc_dbi, cursor->mc_xcursor, leaf);
+ if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
rc = mdb_cursor_first(&cursor->mc_xcursor->mx_cursor, data, NULL);
if (rc != MDB_SUCCESS)
return rc;
}
}
- return mdb_set_key(leaf, key);
+ MDB_SET_KEY(leaf, key);
+ return MDB_SUCCESS;
}
static int
assert(cursor->mc_initialized);
+ top = CURSOR_TOP(cursor);
+ mp = top->mp_page;
+
if (cursor->mc_txn->mt_dbs[cursor->mc_dbi].md_flags & MDB_DUPSORT) {
+ leaf = NODEPTR(mp, top->mp_ki);
if (op == MDB_PREV || op == MDB_PREV_DUP) {
- rc = mdb_cursor_prev(&cursor->mc_xcursor->mx_cursor, data, NULL, MDB_PREV);
- if (op != MDB_PREV || rc == MDB_SUCCESS)
- return rc;
+ if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
+ rc = mdb_cursor_prev(&cursor->mc_xcursor->mx_cursor, data, NULL, MDB_PREV);
+ if (op != MDB_PREV || rc == MDB_SUCCESS)
+ return rc;
+ } else {
+ cursor->mc_xcursor->mx_cursor.mc_initialized = 0;
+ if (op == MDB_PREV_DUP)
+ return MDB_NOTFOUND;
+ }
}
}
- top = CURSOR_TOP(cursor);
- mp = top->mp_page;
-
DPRINTF("cursor_prev: top page is %lu in cursor %p", mp->mp_pgno, (void *) cursor);
if (top->mp_ki == 0) {
- DPRINTF("=====> move to prev sibling page");
+ DPUTS("=====> move to prev sibling page");
if (mdb_sibling(cursor, 0) != MDB_SUCCESS) {
+ cursor->mc_initialized = 0;
return MDB_NOTFOUND;
}
top = CURSOR_TOP(cursor);
DPRINTF("==> cursor points to page %lu with %u keys, key index %u",
mp->mp_pgno, NUMKEYS(mp), top->mp_ki);
+ if (IS_LEAF2(mp)) {
+ key->mv_size = cursor->mc_txn->mt_dbs[cursor->mc_dbi].md_pad;
+ key->mv_data = LEAF2KEY(mp, top->mp_ki, key->mv_size);
+ return MDB_SUCCESS;
+ }
+
assert(IS_LEAF(mp));
leaf = NODEPTR(mp, top->mp_ki);
+ if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
+ mdb_xcursor_init1(cursor->mc_txn, cursor->mc_dbi, cursor->mc_xcursor, mp, leaf);
+ }
if (data) {
if ((rc = mdb_read_data(cursor->mc_txn, leaf, data) != MDB_SUCCESS))
return rc;
- if (cursor->mc_txn->mt_dbs[cursor->mc_dbi].md_flags & MDB_DUPSORT) {
- mdb_xcursor_init1(cursor->mc_txn, cursor->mc_dbi, cursor->mc_xcursor, leaf);
+ if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
rc = mdb_cursor_last(&cursor->mc_xcursor->mx_cursor, data, NULL);
if (rc != MDB_SUCCESS)
return rc;
}
}
- return mdb_set_key(leaf, key);
+ MDB_SET_KEY(leaf, key);
+ return MDB_SUCCESS;
}
static int
MDB_node *leaf;
MDB_ppage *top;
MDB_pageparent mpp;
+ DKBUF;
assert(cursor);
assert(key);
assert(key->mv_size > 0);
- while (CURSOR_TOP(cursor) != NULL)
- cursor_pop_page(cursor);
+ /* See if we're already on the right page */
+ if (cursor->mc_initialized) {
+ MDB_val nodekey;
+ top = CURSOR_TOP(cursor);
+ /* Don't try this for LEAF2 pages. Maybe support that later. */
+ if ((top->mp_page->mp_flags & (P_LEAF|P_LEAF2)) == P_LEAF) {
+ leaf = NODEPTR(top->mp_page, 0);
+ MDB_SET_KEY(leaf, &nodekey);
+ rc = mdb_cmp(cursor->mc_txn, cursor->mc_dbi, key, &nodekey);
+ if (rc >= 0) {
+ leaf = NODEPTR(top->mp_page, NUMKEYS(top->mp_page)-1);
+ MDB_SET_KEY(leaf, &nodekey);
+ rc = mdb_cmp(cursor->mc_txn, cursor->mc_dbi, key, &nodekey);
+ if (rc <= 0) {
+ /* we're already on the right page */
+ mpp.mp_page = top->mp_page;
+ rc = 0;
+ goto set2;
+ }
+ }
+ }
+ }
+ cursor->mc_snum = 0;
rc = mdb_search_page(cursor->mc_txn, cursor->mc_dbi, key, cursor, 0, &mpp);
if (rc != MDB_SUCCESS)
return rc;
+
assert(IS_LEAF(mpp.mp_page));
top = CURSOR_TOP(cursor);
+set2:
leaf = mdb_search_node(cursor->mc_txn, cursor->mc_dbi, mpp.mp_page, key, exactp, &top->mp_ki);
if (exactp != NULL && !*exactp) {
/* MDB_SET specified and not an exact match. */
}
if (leaf == NULL) {
- DPRINTF("===> inexact leaf not found, goto sibling");
+ DPUTS("===> inexact leaf not found, goto sibling");
if ((rc = mdb_sibling(cursor, 1)) != MDB_SUCCESS)
return rc; /* no entries matched */
top = CURSOR_TOP(cursor);
cursor->mc_initialized = 1;
cursor->mc_eof = 0;
- if (data) {
- MDB_val d2;
- if ((rc = mdb_read_data(cursor->mc_txn, leaf, &d2)) != MDB_SUCCESS)
- return rc;
+ if (IS_LEAF2(mpp.mp_page)) {
+ key->mv_size = cursor->mc_txn->mt_dbs[cursor->mc_dbi].md_pad;
+ key->mv_data = LEAF2KEY(mpp.mp_page, top->mp_ki, key->mv_size);
+ return MDB_SUCCESS;
+ }
- if (cursor->mc_txn->mt_dbs[cursor->mc_dbi].md_flags & MDB_DUPSORT) {
- mdb_xcursor_init1(cursor->mc_txn, cursor->mc_dbi, cursor->mc_xcursor, leaf);
+ if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
+ mdb_xcursor_init1(cursor->mc_txn, cursor->mc_dbi, cursor->mc_xcursor, mpp.mp_page, leaf);
+ }
+ if (data) {
+ if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
if (op == MDB_SET || op == MDB_SET_RANGE) {
rc = mdb_cursor_first(&cursor->mc_xcursor->mx_cursor, data, NULL);
} else {
int ex2, *ex2p;
- MDB_cursor_op op2;
if (op == MDB_GET_BOTH) {
ex2p = &ex2;
- op2 = MDB_SET;
+ ex2 = 0;
} else {
ex2p = NULL;
- op2 = MDB_SET_RANGE;
}
- rc = mdb_cursor_set(&cursor->mc_xcursor->mx_cursor, data, NULL, op2, ex2p);
+ rc = mdb_cursor_set(&cursor->mc_xcursor->mx_cursor, data, NULL, MDB_SET_RANGE, ex2p);
if (rc != MDB_SUCCESS)
return rc;
}
+ } else if (op == MDB_GET_BOTH || op == MDB_GET_BOTH_RANGE) {
+ MDB_val d2;
+ if ((rc = mdb_read_data(cursor->mc_txn, leaf, &d2)) != MDB_SUCCESS)
+ return rc;
+ rc = mdb_dcmp(cursor->mc_txn, cursor->mc_dbi, data, &d2);
+ if (rc) {
+ if (op == MDB_GET_BOTH || rc > 0)
+ return MDB_NOTFOUND;
+ }
+
} else {
- *data = d2;
+ if ((rc = mdb_read_data(cursor->mc_txn, leaf, data)) != MDB_SUCCESS)
+ return rc;
}
-
}
- rc = mdb_set_key(leaf, key);
- if (rc == MDB_SUCCESS) {
- DPRINTF("==> cursor placed on key %.*s",
- (int)key->mv_size, (char *)key->mv_data);
- ;
- }
+ /* The key already matches in all other cases */
+ if (op == MDB_SET_RANGE)
+ MDB_SET_KEY(leaf, key);
+ DPRINTF("==> cursor placed on key [%s]", DKEY(key));
return rc;
}
MDB_pageparent mpp;
MDB_node *leaf;
- while (CURSOR_TOP(cursor) != NULL)
- cursor_pop_page(cursor);
+ cursor->mc_snum = 0;
rc = mdb_search_page(cursor->mc_txn, cursor->mc_dbi, NULL, cursor, 0, &mpp);
if (rc != MDB_SUCCESS)
cursor->mc_initialized = 1;
cursor->mc_eof = 0;
- if (data) {
- if ((rc = mdb_read_data(cursor->mc_txn, leaf, data)) != MDB_SUCCESS)
- return rc;
+ if (IS_LEAF2(mpp.mp_page)) {
+ key->mv_size = cursor->mc_txn->mt_dbs[cursor->mc_dbi].md_pad;
+ key->mv_data = LEAF2KEY(mpp.mp_page, 0, key->mv_size);
+ return MDB_SUCCESS;
+ }
- if (cursor->mc_txn->mt_dbs[cursor->mc_dbi].md_flags & MDB_DUPSORT) {
- mdb_xcursor_init1(cursor->mc_txn, cursor->mc_dbi, cursor->mc_xcursor, leaf);
+ if (data) {
+ if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
+ mdb_xcursor_init1(cursor->mc_txn, cursor->mc_dbi, cursor->mc_xcursor, mpp.mp_page, leaf);
rc = mdb_cursor_first(&cursor->mc_xcursor->mx_cursor, data, NULL);
if (rc)
return rc;
+ } else {
+ if (cursor->mc_xcursor)
+ cursor->mc_xcursor->mx_cursor.mc_initialized = 0;
+ if ((rc = mdb_read_data(cursor->mc_txn, leaf, data)) != MDB_SUCCESS)
+ return rc;
}
}
- return mdb_set_key(leaf, key);
+ MDB_SET_KEY(leaf, key);
+ return MDB_SUCCESS;
}
static int
MDB_node *leaf;
MDB_val lkey;
- while (CURSOR_TOP(cursor) != NULL)
- cursor_pop_page(cursor);
+ cursor->mc_snum = 0;
lkey.mv_size = MAXKEYSIZE+1;
lkey.mv_data = NULL;
top = CURSOR_TOP(cursor);
top->mp_ki = NUMKEYS(top->mp_page) - 1;
- if (data) {
- if ((rc = mdb_read_data(cursor->mc_txn, leaf, data)) != MDB_SUCCESS)
- return rc;
+ if (IS_LEAF2(mpp.mp_page)) {
+ key->mv_size = cursor->mc_txn->mt_dbs[cursor->mc_dbi].md_pad;
+ key->mv_data = LEAF2KEY(mpp.mp_page, top->mp_ki, key->mv_size);
+ return MDB_SUCCESS;
+ }
- if (cursor->mc_txn->mt_dbs[cursor->mc_dbi].md_flags & MDB_DUPSORT) {
- mdb_xcursor_init1(cursor->mc_txn, cursor->mc_dbi, cursor->mc_xcursor, leaf);
+ if (data) {
+ if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
+ mdb_xcursor_init1(cursor->mc_txn, cursor->mc_dbi, cursor->mc_xcursor, mpp.mp_page, leaf);
rc = mdb_cursor_last(&cursor->mc_xcursor->mx_cursor, data, NULL);
if (rc)
return rc;
+ } else {
+ if ((rc = mdb_read_data(cursor->mc_txn, leaf, data)) != MDB_SUCCESS)
+ return rc;
}
}
- return mdb_set_key(leaf, key);
+ MDB_SET_KEY(leaf, key);
+ return MDB_SUCCESS;
}
int
switch (op) {
case MDB_GET_BOTH:
case MDB_GET_BOTH_RANGE:
- if (data == NULL) {
+ if (data == NULL || cursor->mc_xcursor == NULL) {
rc = EINVAL;
break;
}
case MDB_SET_RANGE:
if (key == NULL || key->mv_size == 0 || key->mv_size > MAXKEYSIZE) {
rc = EINVAL;
- } else if (op != MDB_SET_RANGE)
+ } else if (op == MDB_SET_RANGE)
rc = mdb_cursor_set(cursor, key, data, op, NULL);
else
rc = mdb_cursor_set(cursor, key, data, op, &exact);
break;
+ case MDB_GET_MULTIPLE:
+ if (data == NULL ||
+ !(cursor->mc_txn->mt_dbs[cursor->mc_dbi].md_flags & MDB_DUPFIXED) ||
+ !cursor->mc_initialized) {
+ rc = EINVAL;
+ break;
+ }
+ rc = MDB_SUCCESS;
+ if (!cursor->mc_xcursor->mx_cursor.mc_initialized || cursor->mc_xcursor->mx_cursor.mc_eof)
+ break;
+ goto fetchm;
+ case MDB_NEXT_MULTIPLE:
+ if (data == NULL ||
+ !(cursor->mc_txn->mt_dbs[cursor->mc_dbi].md_flags & MDB_DUPFIXED)) {
+ rc = EINVAL;
+ break;
+ }
+ if (!cursor->mc_initialized)
+ rc = mdb_cursor_first(cursor, key, data);
+ else
+ rc = mdb_cursor_next(cursor, key, data, MDB_NEXT_DUP);
+ if (rc == MDB_SUCCESS) {
+ if (cursor->mc_xcursor->mx_cursor.mc_initialized) {
+ MDB_ppage *top;
+fetchm:
+ top = CURSOR_TOP(&cursor->mc_xcursor->mx_cursor);
+ data->mv_size = NUMKEYS(top->mp_page) *
+ cursor->mc_xcursor->mx_txn.mt_dbs[cursor->mc_xcursor->mx_cursor.mc_dbi].md_pad;
+ data->mv_data = METADATA(top->mp_page);
+ top->mp_ki = NUMKEYS(top->mp_page)-1;
+ } else {
+ rc = MDB_NOTFOUND;
+ }
+ }
+ break;
case MDB_NEXT:
case MDB_NEXT_DUP:
case MDB_NEXT_NODUP:
case MDB_FIRST:
rc = mdb_cursor_first(cursor, key, data);
break;
+ case MDB_FIRST_DUP:
+ if (data == NULL ||
+ !(cursor->mc_txn->mt_dbs[cursor->mc_dbi].md_flags & MDB_DUPSORT) ||
+ !cursor->mc_initialized ||
+ !cursor->mc_xcursor->mx_cursor.mc_initialized) {
+ rc = EINVAL;
+ break;
+ }
+ rc = mdb_cursor_first(&cursor->mc_xcursor->mx_cursor, data, NULL);
+ break;
case MDB_LAST:
rc = mdb_cursor_last(cursor, key, data);
break;
+ case MDB_LAST_DUP:
+ if (data == NULL ||
+ !(cursor->mc_txn->mt_dbs[cursor->mc_dbi].md_flags & MDB_DUPSORT) ||
+ !cursor->mc_initialized ||
+ !cursor->mc_xcursor->mx_cursor.mc_initialized) {
+ rc = EINVAL;
+ break;
+ }
+ rc = mdb_cursor_last(&cursor->mc_xcursor->mx_cursor, data, NULL);
+ break;
default:
DPRINTF("unhandled/unimplemented cursor operation %u", op);
rc = EINVAL;
return rc;
}
+static int
+mdb_cursor_touch(MDB_cursor *cursor)
+{
+ MDB_pageparent mpp;
+ unsigned int i;
+ int rc;
+
+ mpp.mp_parent = NULL;
+ mpp.mp_pi = 0;
+ for(i=0; i<cursor->mc_snum; i++) {
+ mpp.mp_page = cursor->mc_stack[i].mp_page;
+ if (!F_ISSET(mpp.mp_page->mp_flags, P_DIRTY)) {
+ rc = mdb_touch(cursor->mc_txn, cursor->mc_dbi, &mpp);
+ if (rc) return rc;
+ cursor->mc_stack[i].mp_page = mpp.mp_page;
+ }
+ mpp.mp_parent = mpp.mp_page;
+ mpp.mp_pi = cursor->mc_stack[i].mp_ki;
+ }
+ if (cursor->mc_xcursor)
+ return mdb_cursor_touch(&cursor->mc_xcursor->mx_cursor);
+ return MDB_SUCCESS;
+}
+
+int
+mdb_cursor_put(MDB_cursor *cursor, MDB_val *key, MDB_val *data,
+ unsigned int flags)
+{
+ MDB_ppage *top;
+ MDB_node *leaf;
+ int rc, exact;
+
+ if (flags == MDB_NODUPDATA) {
+ rc = mdb_cursor_set(cursor, key, data, MDB_GET_BOTH, &exact);
+ if (rc == 0)
+ return MDB_KEYEXIST;
+ if (rc != MDB_NOTFOUND)
+ return rc;
+ } else if (flags != MDB_CURRENT) {
+ rc = mdb_cursor_set(cursor, key, NULL, MDB_SET, &exact);
+ if (flags == MDB_NOOVERWRITE && rc == 0)
+ return MDB_KEYEXIST;
+ if (rc != MDB_NOTFOUND)
+ return rc;
+ } else if (!cursor->mc_initialized) {
+ return EINVAL;
+ }
+ /* Cursor is positioned, now make sure all pages are writable */
+ rc = mdb_cursor_touch(cursor);
+ if (rc) return rc;
+
+ top = CURSOR_TOP(cursor);
+
+}
+
/* Allocate a page and initialize it
*/
static MDB_dpage *
{
MDB_dpage *dp;
- if ((dp = mdb_alloc_page(txn, NULL, 0, num)) == NULL)
+ if ((dp = mdb_alloc_page(txn, dbi, NULL, 0, num)) == NULL)
return NULL;
DPRINTF("allocated new mpage %lu, page size %u",
dp->p.mp_pgno, txn->mt_env->me_psize);
indx_t ofs;
MDB_node *node;
MDB_dpage *ofp = NULL; /* overflow page */
+ DKBUF;
assert(mp->mp_upper >= mp->mp_lower);
- DPRINTF("add node [%.*s] to %s page %lu at index %i, key size %zu",
- key ? (int)key->mv_size : 0, key ? (char *)key->mv_data : NULL,
+ DPRINTF("add to %s page %lu index %i, data size %zu key size %zu [%s]",
IS_LEAF(mp) ? "leaf" : "branch",
- mp->mp_pgno, indx, key ? key->mv_size : 0);
+ mp->mp_pgno, indx, data ? data->mv_size : 0,
+ key ? key->mv_size : 0, key ? DKEY(key) : NULL);
+
+ if (IS_LEAF2(mp)) {
+ /* Move higher keys up one slot. */
+ int ksize = txn->mt_dbs[dbi].md_pad, dif;
+ char *ptr = LEAF2KEY(mp, indx, ksize);
+ dif = NUMKEYS(mp) - indx;
+ if (dif > 0)
+ memmove(ptr+ksize, ptr, dif*ksize);
+ /* insert new key */
+ memcpy(ptr, key->mv_data, ksize);
+
+ /* Just using these for counting */
+ mp->mp_lower += sizeof(indx_t);
+ mp->mp_upper -= ksize - sizeof(indx_t);
+ return MDB_SUCCESS;
+ }
if (key != NULL)
node_size += key->mv_size;
if (IS_LEAF(mp))
node->mn_dsize = data->mv_size;
else
- node->mn_pgno = pgno;
+ NODEPGNO(node) = pgno;
if (key)
memcpy(NODEKEY(node), key->mv_data, key->mv_size);
}
static void
-mdb_del_node(MDB_page *mp, indx_t indx)
+mdb_del_node(MDB_page *mp, indx_t indx, int ksize)
{
unsigned int sz;
indx_t i, j, numkeys, ptr;
IS_LEAF(mp) ? "leaf" : "branch", mp->mp_pgno);
assert(indx < NUMKEYS(mp));
+ if (IS_LEAF2(mp)) {
+ int x = NUMKEYS(mp) - 1 - indx;
+ base = LEAF2KEY(mp, indx, ksize);
+ if (x)
+ memmove(base, base + ksize, x * ksize);
+ mp->mp_lower -= sizeof(indx_t);
+ mp->mp_upper += ksize - sizeof(indx_t);
+ return;
+ }
+
node = NODEPTR(mp, indx);
sz = NODESIZE + node->mn_ksize;
if (IS_LEAF(mp)) {
}
static void
-mdb_xcursor_init0(MDB_txn *txn, MDB_dbi dbi, MDB_xcursor *mx)
+mdb_xcursor_init0(MDB_cursor *mc)
{
+ MDB_xcursor *mx = mc->mc_xcursor;
MDB_dbi dbn;
- mx->mx_txn = *txn;
+ mx->mx_txn = *mc->mc_txn;
mx->mx_txn.mt_dbxs = mx->mx_dbxs;
mx->mx_txn.mt_dbs = mx->mx_dbs;
- mx->mx_dbxs[0] = txn->mt_dbxs[0];
- mx->mx_dbxs[1] = txn->mt_dbxs[1];
- if (dbi > 1) {
- mx->mx_dbxs[2] = txn->mt_dbxs[dbi];
+ mx->mx_dbxs[0] = mc->mc_txn->mt_dbxs[0];
+ mx->mx_dbxs[1] = mc->mc_txn->mt_dbxs[1];
+ if (mc->mc_dbi > 1) {
+ mx->mx_dbxs[2] = mc->mc_txn->mt_dbxs[mc->mc_dbi];
dbn = 2;
} else {
dbn = 1;
mx->mx_dbxs[dbn+1].md_dirty = 0;
mx->mx_txn.mt_numdbs = dbn+2;
- SLIST_INIT(&mx->mx_cursor.mc_stack);
+ mx->mx_cursor.mc_snum = 0;
mx->mx_cursor.mc_txn = &mx->mx_txn;
mx->mx_cursor.mc_dbi = dbn+1;
}
static void
-mdb_xcursor_init1(MDB_txn *txn, MDB_dbi dbi, MDB_xcursor *mx, MDB_node *node)
+mdb_xcursor_init1(MDB_txn *txn, MDB_dbi dbi, MDB_xcursor *mx, MDB_page *mp, MDB_node *node)
{
MDB_db *db = NODEDATA(node);
MDB_dbi dbn;
} else {
dbn = 2;
}
+ DPRINTF("Sub-db %u for db %u root page %lu", dbn, dbi, db->md_root);
mx->mx_dbs[dbn] = *db;
+ if (F_ISSET(mp->mp_flags, P_DIRTY))
+ mx->mx_dbxs[dbn].md_dirty = 1;
mx->mx_dbxs[dbn].md_name.mv_data = NODEKEY(node);
mx->mx_dbxs[dbn].md_name.mv_size = node->mn_ksize;
mx->mx_txn.mt_next_pgno = txn->mt_next_pgno;
- mx->mx_txn.mt_oldest = txn->mt_oldest;
mx->mx_txn.mt_u = txn->mt_u;
+ mx->mx_cursor.mc_initialized = 0;
+ mx->mx_cursor.mc_eof = 0;
}
static void
mdb_xcursor_fini(MDB_txn *txn, MDB_dbi dbi, MDB_xcursor *mx)
{
txn->mt_next_pgno = mx->mx_txn.mt_next_pgno;
- txn->mt_oldest = mx->mx_txn.mt_oldest;
txn->mt_u = mx->mx_txn.mt_u;
txn->mt_dbs[0] = mx->mx_dbs[0];
txn->mt_dbs[1] = mx->mx_dbs[1];
size += sizeof(MDB_xcursor);
if ((cursor = calloc(1, size)) != NULL) {
- SLIST_INIT(&cursor->mc_stack);
cursor->mc_dbi = dbi;
cursor->mc_txn = txn;
if (txn->mt_dbs[dbi].md_flags & MDB_DUPSORT) {
MDB_xcursor *mx = (MDB_xcursor *)(cursor + 1);
cursor->mc_xcursor = mx;
- mdb_xcursor_init0(txn, dbi, mx);
+ mdb_xcursor_init0(cursor);
}
} else {
return ENOMEM;
int
mdb_cursor_count(MDB_cursor *mc, unsigned long *countp)
{
+ MDB_ppage *top;
+ MDB_node *leaf;
+
if (mc == NULL || countp == NULL)
return EINVAL;
if (!(mc->mc_txn->mt_dbs[mc->mc_dbi].md_flags & MDB_DUPSORT))
return EINVAL;
- if (!mc->mc_xcursor->mx_cursor.mc_initialized)
- return EINVAL;
+ top = CURSOR_TOP(mc);
+ leaf = NODEPTR(top->mp_page, top->mp_ki);
+ if (!F_ISSET(leaf->mn_flags, F_DUPDATA)) {
+ *countp = 1;
+ } else {
+ if (!mc->mc_xcursor->mx_cursor.mc_initialized)
+ return EINVAL;
- *countp = mc->mc_xcursor->mx_txn.mt_dbs[mc->mc_xcursor->mx_cursor.mc_dbi].md_entries;
+ *countp = mc->mc_xcursor->mx_txn.mt_dbs[mc->mc_xcursor->mx_cursor.mc_dbi].md_entries;
+ }
return MDB_SUCCESS;
}
mdb_cursor_close(MDB_cursor *cursor)
{
if (cursor != NULL) {
- while(!CURSOR_EMPTY(cursor))
- cursor_pop_page(cursor);
- if (cursor->mc_txn->mt_dbs[cursor->mc_dbi].md_flags & MDB_DUPSORT) {
- while(!CURSOR_EMPTY(&cursor->mc_xcursor->mx_cursor))
- cursor_pop_page(&cursor->mc_xcursor->mx_cursor);
- }
-
free(cursor);
}
}
size_t len;
MDB_node *node;
char *base;
+ DKBUF;
node = NODEPTR(mp, indx);
ptr = mp->mp_ptrs[indx];
- DPRINTF("update key %u (ofs %u) [%.*s] to [%.*s] on page %lu",
+ DPRINTF("update key %u (ofs %u) [%.*s] to [%s] on page %lu",
indx, ptr,
(int)node->mn_ksize, (char *)NODEKEY(node),
- (int)key->mv_size, (char *)key->mv_data,
+ DKEY(key),
mp->mp_pgno);
delta = key->mv_size - node->mn_ksize;
int rc;
MDB_node *srcnode;
MDB_val key, data;
+ DKBUF;
+
+ /* Mark src and dst as dirty. */
+ if ((rc = mdb_touch(txn, dbi, src)) ||
+ (rc = mdb_touch(txn, dbi, dst)))
+ return rc;;
- srcnode = NODEPTR(src->mp_page, srcindx);
- DPRINTF("moving %s node %u [%.*s] on page %lu to node %u on page %lu",
+ if (IS_LEAF2(src->mp_page)) {
+ srcnode = NODEPTR(src->mp_page, 0); /* fake */
+ key.mv_size = txn->mt_dbs[dbi].md_pad;
+ key.mv_data = LEAF2KEY(src->mp_page, srcindx, key.mv_size);
+ data.mv_size = 0;
+ data.mv_data = NULL;
+ } else {
+ srcnode = NODEPTR(src->mp_page, srcindx);
+ key.mv_size = NODEKSZ(srcnode);
+ key.mv_data = NODEKEY(srcnode);
+ data.mv_size = NODEDSZ(srcnode);
+ data.mv_data = NODEDATA(srcnode);
+ }
+ DPRINTF("moving %s node %u [%s] on page %lu to node %u on page %lu",
IS_LEAF(src->mp_page) ? "leaf" : "branch",
srcindx,
- (int)srcnode->mn_ksize, (char *)NODEKEY(srcnode),
+ DKEY(&key),
src->mp_page->mp_pgno,
dstindx, dst->mp_page->mp_pgno);
- /* Mark src and dst as dirty. */
- if ((rc = mdb_touch(txn, src)) ||
- (rc = mdb_touch(txn, dst)))
- return rc;;
-
/* Add the node to the destination page.
*/
- key.mv_size = srcnode->mn_ksize;
- key.mv_data = NODEKEY(srcnode);
- data.mv_size = NODEDSZ(srcnode);
- data.mv_data = NODEDATA(srcnode);
rc = mdb_add_node(txn, dbi, dst->mp_page, dstindx, &key, &data, NODEPGNO(srcnode),
srcnode->mn_flags);
if (rc != MDB_SUCCESS)
/* Delete the node from the source page.
*/
- mdb_del_node(src->mp_page, srcindx);
+ mdb_del_node(src->mp_page, srcindx, key.mv_size);
- /* Update the parent separators.
+ /* The key value just changed due to del_node, find it again.
*/
- if (srcindx == 0 && src->mp_pi != 0) {
- DPRINTF("update separator for source page %lu to [%.*s]",
- src->mp_page->mp_pgno, (int)key.mv_size, (char *)key.mv_data);
- if ((rc = mdb_update_key(src->mp_parent, src->mp_pi,
- &key)) != MDB_SUCCESS)
- return rc;
- }
-
- if (srcindx == 0 && IS_BRANCH(src->mp_page)) {
- MDB_val nullkey;
- nullkey.mv_size = 0;
- assert(mdb_update_key(src->mp_page, 0, &nullkey) == MDB_SUCCESS);
+ if (!IS_LEAF2(src->mp_page)) {
+ srcnode = NODEPTR(src->mp_page, srcindx);
+ key.mv_data = NODEKEY(srcnode);
}
- if (dstindx == 0 && dst->mp_pi != 0) {
- DPRINTF("update separator for destination page %lu to [%.*s]",
- dst->mp_page->mp_pgno, (int)key.mv_size, (char *)key.mv_data);
- if ((rc = mdb_update_key(dst->mp_parent, dst->mp_pi,
- &key)) != MDB_SUCCESS)
- return rc;
+ /* Update the parent separators.
+ */
+ if (srcindx == 0) {
+ if (src->mp_pi != 0) {
+ DPRINTF("update separator for source page %lu to [%s]",
+ src->mp_page->mp_pgno, DKEY(&key));
+ if ((rc = mdb_update_key(src->mp_parent, src->mp_pi,
+ &key)) != MDB_SUCCESS)
+ return rc;
+ }
+ if (IS_BRANCH(src->mp_page)) {
+ MDB_val nullkey;
+ nullkey.mv_size = 0;
+ assert(mdb_update_key(src->mp_page, 0, &nullkey) == MDB_SUCCESS);
+ }
}
- if (dstindx == 0 && IS_BRANCH(dst->mp_page)) {
- MDB_val nullkey;
- nullkey.mv_size = 0;
- assert(mdb_update_key(dst->mp_page, 0, &nullkey) == MDB_SUCCESS);
+ if (dstindx == 0) {
+ if (dst->mp_pi != 0) {
+ DPRINTF("update separator for destination page %lu to [%s]",
+ dst->mp_page->mp_pgno, DKEY(&key));
+ if ((rc = mdb_update_key(dst->mp_parent, dst->mp_pi,
+ &key)) != MDB_SUCCESS)
+ return rc;
+ }
+ if (IS_BRANCH(dst->mp_page)) {
+ MDB_val nullkey;
+ nullkey.mv_size = 0;
+ assert(mdb_update_key(dst->mp_page, 0, &nullkey) == MDB_SUCCESS);
+ }
}
return MDB_SUCCESS;
assert(dst->mp_parent);
/* Mark src and dst as dirty. */
- if ((rc = mdb_touch(txn, src)) ||
- (rc = mdb_touch(txn, dst)))
+ if ((rc = mdb_touch(txn, dbi, src)) ||
+ (rc = mdb_touch(txn, dbi, dst)))
return rc;
/* Move all nodes from src to dst.
*/
- for (i = 0; i < NUMKEYS(src->mp_page); i++) {
- srcnode = NODEPTR(src->mp_page, i);
-
- key.mv_size = srcnode->mn_ksize;
- key.mv_data = NODEKEY(srcnode);
- data.mv_size = NODEDSZ(srcnode);
- data.mv_data = NODEDATA(srcnode);
- rc = mdb_add_node(txn, dbi, dst->mp_page, NUMKEYS(dst->mp_page), &key,
- &data, NODEPGNO(srcnode), srcnode->mn_flags);
- if (rc != MDB_SUCCESS)
- return rc;
+ if (IS_LEAF2(src->mp_page)) {
+ key.mv_size = txn->mt_dbs[dbi].md_pad;
+ key.mv_data = METADATA(src->mp_page);
+ for (i = 0; i < NUMKEYS(src->mp_page); i++) {
+ rc = mdb_add_node(txn, dbi, dst->mp_page, NUMKEYS(dst->mp_page), &key,
+ NULL, 0, 0);
+ if (rc != MDB_SUCCESS)
+ return rc;
+ key.mv_data = (char *)key.mv_data + key.mv_size;
+ }
+ } else {
+ for (i = 0; i < NUMKEYS(src->mp_page); i++) {
+ srcnode = NODEPTR(src->mp_page, i);
+
+ key.mv_size = srcnode->mn_ksize;
+ key.mv_data = NODEKEY(srcnode);
+ data.mv_size = NODEDSZ(srcnode);
+ data.mv_data = NODEDATA(srcnode);
+ rc = mdb_add_node(txn, dbi, dst->mp_page, NUMKEYS(dst->mp_page), &key,
+ &data, NODEPGNO(srcnode), srcnode->mn_flags);
+ if (rc != MDB_SUCCESS)
+ return rc;
+ }
}
DPRINTF("dst page %lu now has %u keys (%.1f%% filled)",
/* Unlink the src page from parent.
*/
- mdb_del_node(src->mp_parent, src->mp_pi);
+ mdb_del_node(src->mp_parent, src->mp_pi, 0);
if (src->mp_pi == 0) {
key.mv_size = 0;
if ((rc = mdb_update_key(src->mp_parent, 0, &key)) != MDB_SUCCESS)
MDB_page *root;
MDB_pageparent npp;
indx_t si = 0, di = 0;
+ int rc;
assert(txn != NULL);
assert(mpp != NULL);
if (mpp->mp_parent == NULL) {
if (NUMKEYS(mpp->mp_page) == 0) {
- DPRINTF("tree is completely empty");
+ DPUTS("tree is completely empty");
txn->mt_dbs[dbi].md_root = P_INVALID;
- txn->mt_dbs[dbi].md_depth--;
- txn->mt_dbs[dbi].md_leaf_pages--;
+ txn->mt_dbs[dbi].md_depth = 0;
+ txn->mt_dbs[dbi].md_leaf_pages = 0;
} else if (IS_BRANCH(mpp->mp_page) && NUMKEYS(mpp->mp_page) == 1) {
- DPRINTF("collapsing root page!");
+ DPUTS("collapsing root page!");
txn->mt_dbs[dbi].md_root = NODEPGNO(NODEPTR(mpp->mp_page, 0));
- if ((root = mdb_get_page(txn, txn->mt_dbs[dbi].md_root)) == NULL)
- return MDB_PAGE_NOTFOUND;
+ if ((rc = mdb_get_page(txn, txn->mt_dbs[dbi].md_root, &root)))
+ return rc;
txn->mt_dbs[dbi].md_depth--;
txn->mt_dbs[dbi].md_branch_pages--;
} else
- DPRINTF("root page doesn't need rebalancing");
+ DPUTS("root page doesn't need rebalancing");
return MDB_SUCCESS;
}
if (mpp->mp_pi == 0) {
/* We're the leftmost leaf in our parent.
*/
- DPRINTF("reading right neighbor");
+ DPUTS("reading right neighbor");
node = NODEPTR(mpp->mp_parent, mpp->mp_pi + 1);
- if ((npp.mp_page = mdb_get_page(txn, NODEPGNO(node))) == NULL)
- return MDB_PAGE_NOTFOUND;
+ if ((rc = mdb_get_page(txn, NODEPGNO(node), &npp.mp_page)))
+ return rc;
npp.mp_pi = mpp->mp_pi + 1;
si = 0;
di = NUMKEYS(mpp->mp_page);
} else {
/* There is at least one neighbor to the left.
*/
- DPRINTF("reading left neighbor");
+ DPUTS("reading left neighbor");
node = NODEPTR(mpp->mp_parent, mpp->mp_pi - 1);
- if ((npp.mp_page = mdb_get_page(txn, NODEPGNO(node))) == NULL)
- return MDB_PAGE_NOTFOUND;
+ if ((rc = mdb_get_page(txn, NODEPGNO(node), &npp.mp_page)))
+ return rc;
npp.mp_pi = mpp->mp_pi - 1;
si = NUMKEYS(npp.mp_page) - 1;
di = 0;
int rc;
/* add overflow pages to free list */
- if (F_ISSET(leaf->mn_flags, F_BIGDATA)) {
+ if (!IS_LEAF2(mpp->mp_page) && F_ISSET(leaf->mn_flags, F_BIGDATA)) {
int i, ovpages;
pgno_t pg;
pg++;
}
}
- mdb_del_node(mpp->mp_page, ki);
+ mdb_del_node(mpp->mp_page, ki, txn->mt_dbs[dbi].md_pad);
txn->mt_dbs[dbi].md_entries--;
rc = mdb_rebalance(txn, dbi, mpp);
if (rc != MDB_SUCCESS)
int
mdb_del(MDB_txn *txn, MDB_dbi dbi,
- MDB_val *key, MDB_val *data,
- unsigned int flags)
+ MDB_val *key, MDB_val *data)
{
int rc, exact;
unsigned int ki;
MDB_node *leaf;
MDB_pageparent mpp;
-
- DPRINTF("========> delete key %.*s", (int)key->mv_size, (char *)key->mv_data);
+ DKBUF;
assert(key != NULL);
+ DPRINTF("====> delete db %u key [%s]", dbi, DKEY(key));
+
if (txn == NULL || !dbi || dbi >= txn->mt_numdbs)
return EINVAL;
if (F_ISSET(txn->mt_flags, MDB_TXN_RDONLY)) {
- return EINVAL;
+ return EACCES;
}
if (key->mv_size == 0 || key->mv_size > MAXKEYSIZE) {
return MDB_NOTFOUND;
}
- if (F_ISSET(txn->mt_dbs[dbi].md_flags, MDB_DUPSORT)) {
+ if (!IS_LEAF2(mpp.mp_page) && F_ISSET(leaf->mn_flags, F_DUPDATA)) {
+ MDB_cursor mc;
MDB_xcursor mx;
MDB_pageparent mp2;
- mdb_xcursor_init0(txn, dbi, &mx);
- mdb_xcursor_init1(txn, dbi, &mx, leaf);
- if (flags == MDB_DEL_DUP) {
- rc = mdb_del(&mx.mx_txn, mx.mx_cursor.mc_dbi, data, NULL, 0);
+ mc.mc_txn = txn;
+ mc.mc_dbi = dbi;
+ mc.mc_initialized = 0;
+ mc.mc_xcursor = &mx;
+ mdb_xcursor_init0(&mc);
+ mdb_xcursor_init1(txn, dbi, &mx, mpp.mp_page, leaf);
+ if (data) {
+ rc = mdb_del(&mx.mx_txn, mx.mx_cursor.mc_dbi, data, NULL);
mdb_xcursor_fini(txn, dbi, &mx);
- if (rc != MDB_SUCCESS)
- return rc;
/* If sub-DB still has entries, we're done */
if (mx.mx_txn.mt_dbs[mx.mx_cursor.mc_dbi].md_root != P_INVALID) {
memcpy(NODEDATA(leaf), &mx.mx_txn.mt_dbs[mx.mx_cursor.mc_dbi],
sizeof(MDB_db));
+ txn->mt_dbs[dbi].md_entries--;
return rc;
}
/* otherwise fall thru and delete the sub-DB */
unsigned int i;
cursor_pop_page(&mx.mx_cursor);
- top = CURSOR_TOP(&mx.mx_cursor);
- if (top != NULL) {
- parent = SLIST_NEXT(top, mp_entry);
- while (parent != NULL) {
+ if (mx.mx_cursor.mc_snum) {
+ top = CURSOR_TOP(&mx.mx_cursor);
+ while (mx.mx_cursor.mc_snum > 1) {
+ parent = CURSOR_PARENT(&mx.mx_cursor);
for (i=0; i<NUMKEYS(top->mp_page); i++) {
ni = NODEPTR(top->mp_page, i);
- mdb_midl_insert(txn->mt_free_pgs, ni->mn_pgno);
+ mdb_midl_insert(txn->mt_free_pgs, NODEPGNO(ni));
}
- if (parent) {
- parent->mp_ki++;
- if (parent->mp_ki >= NUMKEYS(parent->mp_page)) {
- cursor_pop_page(&mx.mx_cursor);
- top = CURSOR_TOP(&mx.mx_cursor);
- parent = SLIST_NEXT(top, mp_entry);
- } else {
- ni = NODEPTR(parent->mp_page, parent->mp_ki);
- top->mp_page = mdb_get_page(&mx.mx_txn, ni->mn_pgno);
- }
+ parent->mp_ki++;
+ if (parent->mp_ki >= NUMKEYS(parent->mp_page)) {
+ cursor_pop_page(&mx.mx_cursor);
+ top = parent;
+ } else {
+ ni = NODEPTR(parent->mp_page, parent->mp_ki);
+ rc = mdb_get_page(&mx.mx_txn, NODEPGNO(ni), &top->mp_page);
}
}
}
}
}
- if (data && (rc = mdb_read_data(txn, leaf, data)) != MDB_SUCCESS)
- return rc;
-
return mdb_del0(txn, dbi, ki, &mpp, leaf);
}
int rc = MDB_SUCCESS, ins_new = 0;
indx_t newindx;
pgno_t pgno = 0;
- unsigned int i, j, split_indx;
+ unsigned int i, j, split_indx, nkeys, pmax;
MDB_node *node;
MDB_val sepkey, rkey, rdata;
- MDB_page *copy;
+ MDB_page *copy, *cptr;
MDB_dpage *mdp, *rdp, *pdp;
MDB_dhead *dh;
+ DKBUF;
assert(txn != NULL);
mdp = (MDB_dpage *)dh;
newindx = *newindxp;
- DPRINTF("-----> splitting %s page %lu and adding [%.*s] at index %i",
+ DPRINTF("-----> splitting %s page %lu and adding [%s] at index %i",
IS_LEAF(&mdp->p) ? "leaf" : "branch", mdp->p.mp_pgno,
- (int)newkey->mv_size, (char *)newkey->mv_data, *newindxp);
+ DKEY(newkey), *newindxp);
if (mdp->h.md_parent == NULL) {
if ((pdp = mdb_new_page(txn, dbi, P_BRANCH, 1)) == NULL)
rdp->h.md_pi = mdp->h.md_pi + 1;
DPRINTF("new right sibling: page %lu", rdp->p.mp_pgno);
- /* Move half of the keys to the right sibling. */
- if ((copy = malloc(txn->mt_env->me_psize)) == NULL)
- return ENOMEM;
- memcpy(copy, &mdp->p, txn->mt_env->me_psize);
- memset(&mdp->p.mp_ptrs, 0, txn->mt_env->me_psize - PAGEHDRSZ);
- mdp->p.mp_lower = PAGEHDRSZ;
- mdp->p.mp_upper = txn->mt_env->me_psize;
+ nkeys = NUMKEYS(&mdp->p);
+ split_indx = nkeys / 2 + 1;
+
+ if (IS_LEAF2(&rdp->p)) {
+ char *split, *ins;
+ int x;
+ unsigned int lsize, rsize, ksize;
+ /* Move half of the keys to the right sibling */
+ copy = NULL;
+ x = *newindxp - split_indx;
+ ksize = txn->mt_dbs[dbi].md_pad;
+ split = LEAF2KEY(&mdp->p, split_indx, ksize);
+ rsize = (nkeys - split_indx) * ksize;
+ lsize = (nkeys - split_indx) * sizeof(indx_t);
+ mdp->p.mp_lower -= lsize;
+ rdp->p.mp_lower += lsize;
+ mdp->p.mp_upper += rsize - lsize;
+ rdp->p.mp_upper -= rsize - lsize;
+ sepkey.mv_size = ksize;
+ if (newindx == split_indx) {
+ sepkey.mv_data = newkey->mv_data;
+ } else {
+ sepkey.mv_data = split;
+ }
+ if (x<0) {
+ ins = LEAF2KEY(&mdp->p, *newindxp, ksize);
+ memcpy(&rdp->p.mp_ptrs, split, rsize);
+ sepkey.mv_data = &rdp->p.mp_ptrs;
+ memmove(ins+ksize, ins, (split_indx - *newindxp) * ksize);
+ memcpy(ins, newkey->mv_data, ksize);
+ mdp->p.mp_lower += sizeof(indx_t);
+ mdp->p.mp_upper -= ksize - sizeof(indx_t);
+ } else {
+ if (x)
+ memcpy(&rdp->p.mp_ptrs, split, x * ksize);
+ ins = LEAF2KEY(&rdp->p, x, ksize);
+ memcpy(ins, newkey->mv_data, ksize);
+ memcpy(ins+ksize, split + x * ksize, rsize - x * ksize);
+ rdp->p.mp_lower += sizeof(indx_t);
+ rdp->p.mp_upper -= ksize - sizeof(indx_t);
+ *newindxp = x;
+ *mpp = &rdp->p;
+ }
+ goto newsep;
+ }
- split_indx = NUMKEYS(copy) / 2 + 1;
+ /* For leaf pages, check the split point based on what
+ * fits where, since otherwise add_node can fail.
+ */
+ if (IS_LEAF(&mdp->p)) {
+ unsigned int psize, nsize;
+ /* Maximum free space in an empty page */
+ pmax = txn->mt_env->me_psize - PAGEHDRSZ;
+ nsize = mdb_leaf_size(txn->mt_env, newkey, newdata);
+ if (newindx < split_indx) {
+ psize = nsize;
+ for (i=0; i<split_indx; i++) {
+ node = NODEPTR(&mdp->p, i);
+ psize += NODESIZE + NODEKSZ(node) + sizeof(indx_t);
+ if (F_ISSET(node->mn_flags, F_BIGDATA))
+ psize += sizeof(pgno_t);
+ else
+ psize += NODEDSZ(node);
+ if (psize > pmax) {
+ split_indx = i;
+ break;
+ }
+ }
+ } else {
+ psize = nsize;
+ for (i=nkeys-1; i>=split_indx; i--) {
+ node = NODEPTR(&mdp->p, i);
+ psize += NODESIZE + NODEKSZ(node) + sizeof(indx_t);
+ if (F_ISSET(node->mn_flags, F_BIGDATA))
+ psize += sizeof(pgno_t);
+ else
+ psize += NODEDSZ(node);
+ if (psize > pmax) {
+ split_indx = i+1;
+ break;
+ }
+ }
+ }
+ }
/* First find the separating key between the split pages.
*/
sepkey.mv_size = newkey->mv_size;
sepkey.mv_data = newkey->mv_data;
} else {
- node = NODEPTR(copy, split_indx);
+ node = NODEPTR(&mdp->p, split_indx);
sepkey.mv_size = node->mn_ksize;
sepkey.mv_data = NODEKEY(node);
}
- DPRINTF("separator is [%.*s]", (int)sepkey.mv_size, (char *)sepkey.mv_data);
+newsep:
+ DPRINTF("separator is [%s]", DKEY(&sepkey));
/* Copy separator key to the parent.
*/
rc = mdb_add_node(txn, dbi, rdp->h.md_parent, rdp->h.md_pi,
&sepkey, NULL, rdp->p.mp_pgno, 0);
}
+ if (IS_LEAF2(&rdp->p)) {
+ return rc;
+ }
if (rc != MDB_SUCCESS) {
- free(copy);
return rc;
}
- for (i = j = 0; i <= NUMKEYS(copy); j++) {
- if (i < split_indx) {
- /* Re-insert in left sibling. */
- pdp = mdp;
- } else {
- /* Insert in right sibling. */
- if (i == split_indx)
- /* Reset insert index for right sibling. */
- j = (i == newindx && ins_new);
- pdp = rdp;
+ /* Move half of the keys to the right sibling. */
+ if ((copy = malloc(txn->mt_env->me_psize)) == NULL)
+ return ENOMEM;
+
+ copy->mp_pgno = mdp->p.mp_pgno;
+ copy->mp_flags = mdp->p.mp_flags;
+ copy->mp_lower = PAGEHDRSZ;
+ copy->mp_upper = txn->mt_env->me_psize;
+ cptr = copy;
+ for (i = j = 0; i <= nkeys; j++) {
+ if (i == split_indx) {
+ /* Insert in right sibling. */
+ /* Reset insert index for right sibling. */
+ j = (i == newindx && ins_new);
+ cptr = &rdp->p;
}
if (i == newindx && !ins_new) {
/* Update page and index for the new key. */
*newindxp = j;
- *mpp = &pdp->p;
- } else if (i == NUMKEYS(copy)) {
+ if (cptr == &rdp->p)
+ *mpp = cptr;
+ } else if (i == nkeys) {
break;
} else {
- node = NODEPTR(copy, i);
+ node = NODEPTR(&mdp->p, i);
rkey.mv_data = NODEKEY(node);
rkey.mv_size = node->mn_ksize;
if (IS_LEAF(&mdp->p)) {
rdata.mv_data = NODEDATA(node);
rdata.mv_size = node->mn_dsize;
} else
- pgno = node->mn_pgno;
+ pgno = NODEPGNO(node);
flags = node->mn_flags;
i++;
rkey.mv_size = 0;
}
- rc = mdb_add_node(txn, dbi, &pdp->p, j, &rkey, &rdata, pgno,flags);
+ rc = mdb_add_node(txn, dbi, cptr, j, &rkey, &rdata, pgno, flags);
}
+ nkeys = NUMKEYS(copy);
+ for (i=0; i<nkeys; i++)
+ mdp->p.mp_ptrs[i] = copy->mp_ptrs[i];
+ mdp->p.mp_lower = copy->mp_lower;
+ mdp->p.mp_upper = copy->mp_upper;
+ memcpy(NODEPTR(&mdp->p, nkeys-1), NODEPTR(copy, nkeys-1),
+ txn->mt_env->me_psize - copy->mp_upper);
free(copy);
return rc;
unsigned int ki;
MDB_node *leaf;
MDB_pageparent mpp;
- MDB_val xdata, *rdata;
+ MDB_val xdata, *rdata, dkey;
MDB_db dummy;
+ char dbuf[PAGESIZE];
+ int do_sub = 0;
+ size_t nsize;
+ DKBUF;
- DPRINTF("==> put key %.*s, size %zu, data size %zu",
- (int)key->mv_size, (char *)key->mv_data, key->mv_size, data->mv_size);
+ DPRINTF("==> put db %u key [%s], size %zu, data size %zu",
+ dbi, DKEY(key), key->mv_size, data->mv_size);
+ dkey.mv_size = 0;
mpp.mp_parent = NULL;
mpp.mp_pi = 0;
rc = mdb_search_page(txn, dbi, key, NULL, 1, &mpp);
if (rc == MDB_SUCCESS) {
leaf = mdb_search_node(txn, dbi, mpp.mp_page, key, &exact, &ki);
if (leaf && exact) {
- if (F_ISSET(txn->mt_dbs[dbi].md_flags, MDB_DUPSORT)) {
- goto put_sub;
- }
if (flags == MDB_NOOVERWRITE) {
- DPRINTF("duplicate key %.*s",
- (int)key->mv_size, (char *)key->mv_data);
+ DPRINTF("duplicate key [%s]", DKEY(key));
return MDB_KEYEXIST;
}
+ /* there's only a key anyway, so this is a no-op */
+ if (IS_LEAF2(mpp.mp_page))
+ return MDB_SUCCESS;
+
+ if (F_ISSET(txn->mt_dbs[dbi].md_flags, MDB_DUPSORT)) {
+ /* Was a single item before, must convert now */
+ if (!F_ISSET(leaf->mn_flags, F_DUPDATA)) {
+ dkey.mv_size = NODEDSZ(leaf);
+ dkey.mv_data = dbuf;
+ memcpy(dbuf, NODEDATA(leaf), dkey.mv_size);
+ /* data matches, ignore it */
+ if (!mdb_dcmp(txn, dbi, data, &dkey))
+ return (flags == MDB_NODUPDATA) ? MDB_KEYEXIST : MDB_SUCCESS;
+ memset(&dummy, 0, sizeof(dummy));
+ if (txn->mt_dbs[dbi].md_flags & MDB_DUPFIXED) {
+ dummy.md_pad = data->mv_size;
+ dummy.md_flags = MDB_DUPFIXED;
+ if (txn->mt_dbs[dbi].md_flags & MDB_INTEGERDUP)
+ dummy.md_flags |= MDB_INTEGERKEY;
+ }
+ dummy.md_root = P_INVALID;
+ if (dkey.mv_size == sizeof(MDB_db)) {
+ memcpy(NODEDATA(leaf), &dummy, sizeof(dummy));
+ goto put_sub;
+ }
+ mdb_del_node(mpp.mp_page, ki, 0);
+ do_sub = 1;
+ rdata = &xdata;
+ xdata.mv_size = sizeof(MDB_db);
+ xdata.mv_data = &dummy;
+ goto new_sub;
+ }
+ goto put_sub;
+ }
/* same size, just replace it */
- if (NODEDSZ(leaf) == data->mv_size) {
+ if (!F_ISSET(leaf->mn_flags, F_BIGDATA) &&
+ NODEDSZ(leaf) == data->mv_size) {
memcpy(NODEDATA(leaf), data->mv_data, data->mv_size);
goto done;
}
- mdb_del_node(mpp.mp_page, ki);
+ mdb_del_node(mpp.mp_page, ki, 0);
}
if (leaf == NULL) { /* append if not found */
ki = NUMKEYS(mpp.mp_page);
} else if (rc == MDB_NOTFOUND) {
MDB_dpage *dp;
/* new file, just write a root leaf page */
- DPRINTF("allocating new root leaf page");
+ DPUTS("allocating new root leaf page");
if ((dp = mdb_new_page(txn, dbi, P_LEAF, 1)) == NULL) {
return ENOMEM;
}
txn->mt_dbs[dbi].md_root = mpp.mp_page->mp_pgno;
txn->mt_dbs[dbi].md_depth++;
txn->mt_dbxs[dbi].md_dirty = 1;
+ if ((txn->mt_dbs[dbi].md_flags & (MDB_DUPSORT|MDB_DUPFIXED)) == MDB_DUPFIXED)
+ mpp.mp_page->mp_flags |= P_LEAF2;
ki = 0;
}
else
DPRINTF("there are %u keys, should insert new key at index %i",
NUMKEYS(mpp.mp_page), ki);
- /* For sorted dups, the data item at this level is a DB record
- * for a child DB; the actual data elements are stored as keys
- * in the child DB.
- */
- if (F_ISSET(txn->mt_dbs[dbi].md_flags, MDB_DUPSORT)) {
- rdata = &xdata;
- xdata.mv_size = sizeof(MDB_db);
- xdata.mv_data = &dummy;
- memset(&dummy, 0, sizeof(dummy));
- dummy.md_root = P_INVALID;
- } else {
- rdata = data;
- }
+ rdata = data;
- if (SIZELEFT(mpp.mp_page) < mdb_leaf_size(txn->mt_env, key, rdata)) {
+new_sub:
+ nsize = IS_LEAF2(mpp.mp_page) ? key->mv_size : mdb_leaf_size(txn->mt_env, key, rdata);
+ if (SIZELEFT(mpp.mp_page) < nsize) {
rc = mdb_split(txn, dbi, &mpp.mp_page, &ki, key, rdata, P_INVALID);
} else {
/* There is room already in this leaf page. */
if (rc != MDB_SUCCESS)
txn->mt_flags |= MDB_TXN_ERROR;
else {
- txn->mt_dbs[dbi].md_entries++;
-
/* Remember if we just added a subdatabase */
if (flags & F_SUBDATA) {
leaf = NODEPTR(mpp.mp_page, ki);
* size limits on dupdata. The actual data fields of the child
* DB are all zero size.
*/
- if (F_ISSET(txn->mt_dbs[dbi].md_flags, MDB_DUPSORT)) {
+ if (do_sub) {
+ MDB_cursor mc;
MDB_xcursor mx;
leaf = NODEPTR(mpp.mp_page, ki);
put_sub:
- mdb_xcursor_init0(txn, dbi, &mx);
- mdb_xcursor_init1(txn, dbi, &mx, leaf);
+ mc.mc_txn = txn;
+ mc.mc_dbi = dbi;
+ mc.mc_initialized = 0;
+ mc.mc_xcursor = &mx;
+ mdb_xcursor_init0(&mc);
+ mdb_xcursor_init1(txn, dbi, &mx, mpp.mp_page, leaf);
xdata.mv_size = 0;
xdata.mv_data = "";
if (flags == MDB_NODUPDATA)
flags = MDB_NOOVERWRITE;
+ /* converted, write the original data first */
+ if (dkey.mv_size) {
+ rc = mdb_put0(&mx.mx_txn, mx.mx_cursor.mc_dbi, &dkey, &xdata, flags);
+ if (rc) return rc;
+ leaf->mn_flags |= F_DUPDATA;
+ }
rc = mdb_put0(&mx.mx_txn, mx.mx_cursor.mc_dbi, data, &xdata, flags);
mdb_xcursor_fini(txn, dbi, &mx);
memcpy(NODEDATA(leaf), &mx.mx_txn.mt_dbs[mx.mx_cursor.mc_dbi],
sizeof(MDB_db));
}
+ txn->mt_dbs[dbi].md_entries++;
}
done:
mdb_put(MDB_txn *txn, MDB_dbi dbi,
MDB_val *key, MDB_val *data, unsigned int flags)
{
+ MDB_cursor mc;
+ MDB_xcursor mx;
+ int rc;
+
assert(key != NULL);
assert(data != NULL);
return EINVAL;
if (F_ISSET(txn->mt_flags, MDB_TXN_RDONLY)) {
- return EINVAL;
+ return EACCES;
}
if (key->mv_size == 0 || key->mv_size > MAXKEYSIZE) {
if ((flags & (MDB_NOOVERWRITE|MDB_NODUPDATA)) != flags)
return EINVAL;
- return mdb_put0(txn, dbi, key, data, flags);
+ mc.mc_txn = txn;
+ mc.mc_dbi = dbi;
+ mc.mc_snum = 0;
+ mc.mc_initialized = 0;
+ mc.mc_eof = 0;
+ if (txn->mt_dbs[dbi].md_flags & MDB_DUPSORT) {
+ mc.mc_xcursor = &mx;
+ mdb_xcursor_init0(&mc);
+ } else {
+ mc.mc_xcursor = NULL;
+ }
+ rc = mdb_cursor_put(&mc, key, data, flags);
+ if (mc.mc_xcursor) {
+ MDB_ppage *top;
+ MDB_node *leaf;
+ mdb_xcursor_fini(txn, dbi, &mx);
+ top = CURSOR_TOP(&mc);
+ leaf = NODEPTR(top->mp_page, top->mp_ki);
+ memcpy(NODEDATA(leaf), &mx.mx_txn.mt_dbs[mx.mx_cursor.mc_dbi],
+ sizeof(MDB_db));
+ }
+ return rc;
}
int
int
mdb_env_stat(MDB_env *env, MDB_stat *arg)
{
+ int toggle;
+
if (env == NULL || arg == NULL)
return EINVAL;
- return mdb_stat0(env, &env->me_meta->mm_dbs[MAIN_DBI], arg);
+ mdb_env_read_meta(env, &toggle);
+
+ return mdb_stat0(env, &env->me_metas[toggle]->mm_dbs[MAIN_DBI], arg);
}
int mdb_open(MDB_txn *txn, const char *name, unsigned int flags, MDB_dbi *dbi)
txn->mt_dbxs[dbi].md_rel = rel;
return MDB_SUCCESS;
}
+
+/** @} */
projects / openldap / blobdiff