* BerkeleyDB API, but much simplified.
*/
/*
- * Copyright 2011 Howard Chu, Symas Corp.
+ * Copyright 2011-2012 Howard Chu, Symas Corp.
* All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
#include <errno.h>
#include <limits.h>
#include <stddef.h>
-#include <stdint.h>
+#include <inttypes.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#endif
#endif
+#ifdef USE_VALGRIND
+#include <valgrind/memcheck.h>
+#define VGMEMP_CREATE(h,r,z) VALGRIND_CREATE_MEMPOOL(h,r,z)
+#define VGMEMP_ALLOC(h,a,s) VALGRIND_MEMPOOL_ALLOC(h,a,s)
+#define VGMEMP_FREE(h,a) VALGRIND_MEMPOOL_FREE(h,a)
+#define VGMEMP_DESTROY(h) VALGRIND_DESTROY_MEMPOOL(h)
+#define VGMEMP_DEFINED(a,s) VALGRIND_MAKE_MEM_DEFINED(a,s)
+#else
+#define VGMEMP_CREATE(h,r,z)
+#define VGMEMP_ALLOC(h,a,s)
+#define VGMEMP_FREE(h,a)
+#define VGMEMP_DESTROY(h)
+#define VGMEMP_DEFINED(a,s)
+#endif
+
#ifndef BYTE_ORDER
-#define BYTE_ORDER __BYTE_ORDER
+# if (defined(_LITTLE_ENDIAN) || defined(_BIG_ENDIAN)) && !(defined(_LITTLE_ENDIAN) && defined(_BIG_ENDIAN))
+/* Solaris just defines one or the other */
+# define LITTLE_ENDIAN 1234
+# define BIG_ENDIAN 4321
+# ifdef _LITTLE_ENDIAN
+# define BYTE_ORDER LITTLE_ENDIAN
+# else
+# define BYTE_ORDER BIG_ENDIAN
+# endif
+# else
+# define BYTE_ORDER __BYTE_ORDER
+# endif
#endif
+
#ifndef LITTLE_ENDIAN
#define LITTLE_ENDIAN __LITTLE_ENDIAN
#endif
#define BIG_ENDIAN __BIG_ENDIAN
#endif
+#if defined(__i386) || defined(__x86_64)
+#define MISALIGNED_OK 1
+#endif
+
#include "mdb.h"
#include "midl.h"
#define UNLOCK_MUTEX_W(env) sem_post((env)->me_wmutex)
#define fdatasync(fd) fsync(fd)
#else
+#ifdef ANDROID
+#define fdatasync(fd) fsync(fd)
+#endif
/** Lock the reader mutex.
*/
#define LOCK_MUTEX_R(env) pthread_mutex_lock(&(env)->me_txns->mti_mutex)
#ifndef MDB_DSYNC
# define MDB_DSYNC O_DSYNC
#endif
+#endif
+
+/** Function for flushing the data of a file. Define this to fsync
+ * if fdatasync() is not supported.
+ */
+#ifndef MDB_FDATASYNC
+# define MDB_FDATASYNC fdatasync
#endif
/** A page number in the database.
/** @defgroup debug Debug Macros
* @{
*/
-#ifndef DEBUG
+#ifndef MDB_DEBUG
/** Enable debug output.
* Set this to 1 for copious tracing. Set to 2 to add dumps of all IDLs
* read from and written to the database (used for free space management).
*/
-#define DEBUG 0
+#define MDB_DEBUG 0
#endif
#if !(__STDC_VERSION__ >= 199901L || defined(__GNUC__))
# define DPRINTF (void) /* Vararg macros may be unsupported */
-#elif DEBUG
+#elif MDB_DEBUG
/** Print a debug message with printf formatting. */
# define DPRINTF(fmt, ...) /**< Requires 2 or more args */ \
fprintf(stderr, "%s:%d " fmt "\n", __func__, __LINE__, __VA_ARGS__)
* pressure from other processes is high. So until OSs have
* actual paging support for Huge pages, they're not viable.
*/
-#define PAGESIZE 4096
+#define MDB_PAGESIZE 4096
/** The minimum number of keys required in a database page.
* Setting this to a larger value will place a smaller bound on the
/** The maximum size of a key in the database.
* While data items have essentially unbounded size, we require that
* keys all fit onto a regular page. This limit could be raised a bit
- * further if needed; to something just under #PAGESIZE / #MDB_MINKEYS.
+ * further if needed; to something just under #MDB_PAGESIZE / #MDB_MINKEYS.
*/
#define MAXKEYSIZE 511
-#if DEBUG
+#if MDB_DEBUG
/** A key buffer.
* @ingroup debug
* This is used for printing a hex dump of a key's contents.
#define DKEY(x) mdb_dkey(x, kbuf)
#else
#define DKBUF typedef int dummy_kbuf /* so we can put ';' after */
-#define DKEY(x)
+#define DKEY(x) 0
#endif
/** @defgroup lazylock Lazy Locking
* For POSIX the actual mutexes reside in the shared memory of this
* mapped file. On Windows, mutexes are named objects allocated by the
* kernel; we store the mutex names in this mapped file so that other
- * processes can grab them. This same approach will also be used on
+ * processes can grab them. This same approach is also used on
* MacOSX/Darwin (using named semaphores) since MacOSX doesn't support
- * process-shared POSIX mutexes.
+ * process-shared POSIX mutexes. For these cases where a named object
+ * is used, the object name is derived from a 64 bit FNV hash of the
+ * environment pathname. As such, naming collisions are extremely
+ * unlikely. If a collision occurs, the results are unpredictable.
*/
typedef struct MDB_txbody {
/** Stamp identifying this as an MDB lock file. It must be set
#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 */
+
+/** valid flags for #mdb_node_add() */
+#define NODE_ADD_FLAGS (F_DUPDATA|F_SUBDATA|MDB_RESERVE|MDB_APPEND)
+
/** @} */
unsigned short mn_flags; /**< @ref mdb_node */
unsigned short mn_ksize; /**< key size */
/** The size of a key in a node */
#define NODEKSZ(node) ((node)->mn_ksize)
+ /** Copy a page number from src to dst */
+#ifdef MISALIGNED_OK
+#define COPY_PGNO(dst,src) dst = src
+#else
+#if SIZE_MAX > 4294967295UL
+#define COPY_PGNO(dst,src) do { \
+ unsigned short *s, *d; \
+ s = (unsigned short *)&(src); \
+ d = (unsigned short *)&(dst); \
+ *d++ = *s++; \
+ *d++ = *s++; \
+ *d++ = *s++; \
+ *d = *s; \
+} while (0)
+#else
+#define COPY_PGNO(dst,src) do { \
+ unsigned short *s, *d; \
+ s = (unsigned short *)&(src); \
+ d = (unsigned short *)&(dst); \
+ *d++ = *s++; \
+ *d = *s; \
+} while (0)
+#endif
+#endif
/** The address of a key in a LEAF2 page.
* LEAF2 pages are used for #MDB_DUPFIXED sorted-duplicate sub-DBs.
* There are no node headers, keys are stored contiguously.
#define DB_DIRTY 0x01 /**< DB was written in this txn */
#define DB_STALE 0x02 /**< DB record is older than txnID */
/** @} */
+ /** Array of cursors for each DB */
+ MDB_cursor **mt_cursors;
/** Array of flags for each DB */
unsigned char *mt_dbflags;
/** Number of DB records in use. This number only ever increments;
/** Cursors are used for all DB operations */
struct MDB_cursor {
+ /** Next cursor on this DB in this txn */
+ MDB_cursor *mc_next;
+ /** Original cursor if this is a shadow */
+ MDB_cursor *mc_orig;
/** Context used for databases with #MDB_DUPSORT, otherwise NULL */
struct MDB_xcursor *mc_xcursor;
/** The transaction that owns this cursor */
*/
#define C_INITIALIZED 0x01 /**< cursor has been initialized and is valid */
#define C_EOF 0x02 /**< No more data */
+#define C_SUB 0x04 /**< Cursor is a sub-cursor */
+#define C_SHADOW 0x08 /**< Cursor is a dup from a parent txn */
+#define C_ALLOCD 0x10 /**< Cursor was malloc'd */
/** @} */
unsigned int mc_flags; /**< @ref mdb_cursor */
MDB_page *mc_pg[CURSOR_STACK]; /**< stack of pushed pages */
MDB_val *key, int modify);
static int mdb_page_merge(MDB_cursor *csrc, MDB_cursor *cdst);
static int mdb_page_split(MDB_cursor *mc, MDB_val *newkey, MDB_val *newdata,
- pgno_t newpgno);
+ pgno_t newpgno, unsigned int nflags);
static int mdb_env_read_header(MDB_env *env, MDB_meta *meta);
static int mdb_env_read_meta(MDB_env *env, int *which);
static MDB_node *mdb_node_search(MDB_cursor *mc, MDB_val *key, int *exactp);
static int mdb_node_add(MDB_cursor *mc, indx_t indx,
- MDB_val *key, MDB_val *data, pgno_t pgno, uint8_t flags);
+ MDB_val *key, MDB_val *data, pgno_t pgno, unsigned int flags);
static void mdb_node_del(MDB_page *mp, indx_t indx, int ksize);
static void mdb_node_shrink(MDB_page *mp, indx_t indx);
static int mdb_node_move(MDB_cursor *csrc, MDB_cursor *cdst);
return strerror(err);
}
-#if DEBUG
+#if MDB_DEBUG
/** Display a key in hexadecimal and return the address of the result.
* @param[in] key the key to display
* @param[in] buf the buffer to write into. Should always be #DKBUF.
static MDB_page *
mdb_page_malloc(MDB_cursor *mc) {
MDB_page *ret;
+ size_t sz = mc->mc_txn->mt_env->me_psize;
if (mc->mc_txn->mt_env->me_dpages) {
ret = mc->mc_txn->mt_env->me_dpages;
+ VGMEMP_ALLOC(mc->mc_txn->mt_env, ret, sz);
+ VGMEMP_DEFINED(ret, sizeof(ret->mp_next));
mc->mc_txn->mt_env->me_dpages = ret->mp_next;
} else {
- ret = malloc(mc->mc_txn->mt_env->me_psize);
+ ret = malloc(sz);
+ VGMEMP_ALLOC(mc->mc_txn->mt_env, ret, sz);
}
return ret;
}
txn->mt_env->me_pghead = mop;
memcpy(mop->mo_pages, idl, MDB_IDL_SIZEOF(idl));
-#if DEBUG > 1
+#if MDB_DEBUG > 1
{
unsigned int i;
DPRINTF("IDL read txn %zu root %zu num %zu",
if (pgno == P_INVALID) {
/* DB size is maxed out */
if (txn->mt_next_pgno + num >= txn->mt_env->me_maxpg) {
- assert(txn->mt_next_pgno + num < txn->mt_env->me_maxpg);
+ DPUTS("DB size maxed out");
return NULL;
}
}
if (txn->mt_env->me_dpages && num == 1) {
np = txn->mt_env->me_dpages;
+ VGMEMP_ALLOC(txn->mt_env, np, txn->mt_env->me_psize);
+ VGMEMP_DEFINED(np, sizeof(np->mp_next));
txn->mt_env->me_dpages = np->mp_next;
} else {
- if ((np = malloc(txn->mt_env->me_psize * num )) == NULL)
+ size_t sz = txn->mt_env->me_psize * num;
+ if ((np = malloc(sz)) == NULL)
return NULL;
+ VGMEMP_ALLOC(txn->mt_env, np, sz);
}
if (pgno == P_INVALID) {
np->mp_pgno = txn->mt_next_pgno;
mp->mp_flags |= P_DIRTY;
finish:
+ /* Adjust other cursors pointing to mp */
+ if (mc->mc_flags & C_SUB) {
+ MDB_cursor *m2, *m3;
+ MDB_dbi dbi = mc->mc_dbi-1;
+
+ for (m2 = mc->mc_txn->mt_cursors[dbi]; m2; m2=m2->mc_next) {
+ if (m2 == mc) continue;
+ m3 = &m2->mc_xcursor->mx_cursor;
+ if (m3->mc_snum < mc->mc_snum) continue;
+ if (m3->mc_pg[mc->mc_top] == mc->mc_pg[mc->mc_top]) {
+ m3->mc_pg[mc->mc_top] = mp;
+ }
+ }
+ } else {
+ MDB_cursor *m2;
+
+ for (m2 = mc->mc_txn->mt_cursors[mc->mc_dbi]; m2; m2=m2->mc_next) {
+ if (m2 == mc || m2->mc_snum < mc->mc_snum) continue;
+ if (m2->mc_pg[mc->mc_top] == mc->mc_pg[mc->mc_top]) {
+ m2->mc_pg[mc->mc_top] = mp;
+ }
+ }
+ }
mc->mc_pg[mc->mc_top] = mp;
/** If this page has a parent, update the parent to point to
* this new page.
{
int rc = 0;
if (force || !F_ISSET(env->me_flags, MDB_NOSYNC)) {
- if (fdatasync(env->me_fd))
+ if (MDB_FDATASYNC(env->me_fd))
rc = ErrCode();
}
return rc;
}
+/** Make shadow copies of all of parent txn's cursors */
+static int
+mdb_cursor_shadow(MDB_txn *src, MDB_txn *dst)
+{
+ MDB_cursor *mc, *m2;
+ unsigned int i, j, size;
+
+ for (i=0;i<src->mt_numdbs; i++) {
+ if (src->mt_cursors[i]) {
+ size = sizeof(MDB_cursor);
+ if (src->mt_cursors[i]->mc_xcursor)
+ size += sizeof(MDB_xcursor);
+ for (m2 = src->mt_cursors[i]; m2; m2=m2->mc_next) {
+ mc = malloc(size);
+ if (!mc)
+ return ENOMEM;
+ mc->mc_orig = m2;
+ mc->mc_txn = dst;
+ mc->mc_dbi = i;
+ mc->mc_db = &dst->mt_dbs[i];
+ mc->mc_dbx = m2->mc_dbx;
+ mc->mc_dbflag = &dst->mt_dbflags[i];
+ mc->mc_snum = m2->mc_snum;
+ mc->mc_top = m2->mc_top;
+ mc->mc_flags = m2->mc_flags | C_SHADOW;
+ for (j=0; j<mc->mc_snum; j++) {
+ mc->mc_pg[j] = m2->mc_pg[j];
+ mc->mc_ki[j] = m2->mc_ki[j];
+ }
+ if (m2->mc_xcursor) {
+ MDB_xcursor *mx, *mx2;
+ mx = (MDB_xcursor *)(mc+1);
+ mc->mc_xcursor = mx;
+ mx2 = m2->mc_xcursor;
+ mx->mx_db = mx2->mx_db;
+ mx->mx_dbx = mx2->mx_dbx;
+ mx->mx_dbflag = mx2->mx_dbflag;
+ mx->mx_cursor.mc_txn = dst;
+ mx->mx_cursor.mc_dbi = mx2->mx_cursor.mc_dbi;
+ mx->mx_cursor.mc_db = &mx->mx_db;
+ mx->mx_cursor.mc_dbx = &mx->mx_dbx;
+ mx->mx_cursor.mc_dbflag = &mx->mx_dbflag;
+ mx->mx_cursor.mc_snum = mx2->mx_cursor.mc_snum;
+ mx->mx_cursor.mc_top = mx2->mx_cursor.mc_top;
+ mx->mx_cursor.mc_flags = mx2->mx_cursor.mc_flags | C_SHADOW;
+ for (j=0; j<mx2->mx_cursor.mc_snum; j++) {
+ mx->mx_cursor.mc_pg[j] = mx2->mx_cursor.mc_pg[j];
+ mx->mx_cursor.mc_ki[j] = mx2->mx_cursor.mc_ki[j];
+ }
+ } else {
+ mc->mc_xcursor = NULL;
+ }
+ mc->mc_next = dst->mt_cursors[i];
+ dst->mt_cursors[i] = mc;
+ }
+ }
+ }
+ return MDB_SUCCESS;
+}
+
+/** Merge shadow cursors back into parent's */
+static void
+mdb_cursor_merge(MDB_txn *txn)
+{
+ MDB_dbi i;
+ for (i=0; i<txn->mt_numdbs; i++) {
+ if (txn->mt_cursors[i]) {
+ MDB_cursor *mc;
+ while ((mc = txn->mt_cursors[i])) {
+ txn->mt_cursors[i] = mc->mc_next;
+ if (mc->mc_flags & C_SHADOW) {
+ MDB_cursor *m2 = mc->mc_orig;
+ unsigned int j;
+ m2->mc_snum = mc->mc_snum;
+ m2->mc_top = mc->mc_top;
+ for (j=0; j<mc->mc_snum; j++) {
+ m2->mc_pg[j] = mc->mc_pg[j];
+ m2->mc_ki[j] = mc->mc_ki[j];
+ }
+ }
+ if (mc->mc_flags & C_ALLOCD)
+ free(mc);
+ }
+ }
+ }
+}
+
static void
mdb_txn_reset0(MDB_txn *txn);
mdb_txn_begin(MDB_env *env, MDB_txn *parent, unsigned int flags, MDB_txn **ret)
{
MDB_txn *txn;
- int rc;
+ int rc, size;
if (env->me_flags & MDB_FATAL_ERROR) {
DPUTS("environment had fatal error, must shutdown!");
return EINVAL;
}
}
- if ((txn = calloc(1, sizeof(MDB_txn) +
- env->me_maxdbs * (sizeof(MDB_db)+1))) == NULL) {
+ size = sizeof(MDB_txn) + env->me_maxdbs * (sizeof(MDB_db)+1);
+ if (!(flags & MDB_RDONLY))
+ size += env->me_maxdbs * sizeof(MDB_cursor *);
+
+ if ((txn = calloc(1, size)) == 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_dbflags = (unsigned char *)(txn->mt_dbs + env->me_maxdbs);
+ } else {
+ txn->mt_cursors = (MDB_cursor **)(txn->mt_dbs + env->me_maxdbs);
+ txn->mt_dbflags = (unsigned char *)(txn->mt_cursors + env->me_maxdbs);
}
- txn->mt_dbflags = (unsigned char *)(txn->mt_dbs + env->me_maxdbs);
txn->mt_env = env;
if (parent) {
txn->mt_dbxs = parent->mt_dbxs;
memcpy(txn->mt_dbs, parent->mt_dbs, txn->mt_numdbs * sizeof(MDB_db));
memcpy(txn->mt_dbflags, parent->mt_dbflags, txn->mt_numdbs);
+ mdb_cursor_shadow(parent, txn);
rc = 0;
} else {
rc = mdb_txn_renew0(txn);
MDB_page *dp;
unsigned int i;
+ /* close(free) all cursors */
+ for (i=0; i<txn->mt_numdbs; i++) {
+ if (txn->mt_cursors[i]) {
+ MDB_cursor *mc;
+ while ((mc = txn->mt_cursors[i])) {
+ txn->mt_cursors[i] = mc->mc_next;
+ if (mc->mc_flags & C_ALLOCD)
+ free(mc);
+ }
+ }
+ }
+
/* 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 (!IS_OVERFLOW(dp) || dp->mp_pages == 1) {
dp->mp_next = txn->mt_env->me_dpages;
+ VGMEMP_FREE(txn->mt_env, dp);
txn->mt_env->me_dpages = dp;
} else {
/* large pages just get freed directly */
+ VGMEMP_FREE(txn->mt_env, dp);
free(dp);
}
}
return MDB_SUCCESS;
}
-
if (F_ISSET(txn->mt_flags, MDB_TXN_ERROR)) {
DPUTS("error flag is set, can't commit");
if (txn->mt_parent)
return EINVAL;
}
+ /* Merge (and close) our cursors with parent's */
+ mdb_cursor_merge(txn);
+
if (txn->mt_parent) {
MDB_db *ip, *jp;
MDB_dbi i;
src = txn->mt_u.dirty_list;
x = mdb_mid2l_search(dst, src[1].mid);
for (y=1; y<=src[0].mid; y++) {
- MDB_page *mp;
while (x <= dst[0].mid && dst[x].mid != src[y].mid) x++;
if (x > dst[0].mid)
break;
- mp = dst[x].mptr;
- dp = src[y].mptr;
- mp->mp_lower = dp->mp_lower;
- mp->mp_upper = dp->mp_upper;
- if (IS_LEAF2(mp)) {
- memcpy(METADATA(mp), METADATA(dp), NUMKEYS(mp) * mp->mp_pad);
- } else {
- memcpy((char *)mp+mp->mp_upper, (char *)dp+dp->mp_upper,
- txn->mt_env->me_psize - mp->mp_upper);
- }
- free(src[y].mptr);
+ free(dst[x].mptr);
+ dst[x].mptr = src[y].mptr;
}
x = dst[0].mid;
for (; y<=src[0].mid; y++) {
- if (++x >= MDB_IDL_UM_MAX)
+ if (++x >= MDB_IDL_UM_MAX) {
+ mdb_txn_abort(txn);
return ENOMEM;
+ }
dst[x] = src[y];
}
dst[0].mid = x;
mdb_page_search(&mc, &key, 1);
mdb_midl_sort(txn->mt_free_pgs);
-#if DEBUG > 1
+#if MDB_DEBUG > 1
{
unsigned int i;
ID *idl = txn->mt_free_pgs;
MDB_oldpages *mop;
mop = env->me_pghead;
+ env->me_pghead = NULL;
key.mv_size = sizeof(pgno_t);
key.mv_data = &mop->mo_txnid;
data.mv_size = MDB_IDL_SIZEOF(mop->mo_pages);
data.mv_data = mop->mo_pages;
mdb_cursor_put(&mc, &key, &data, 0);
- free(env->me_pghead);
- env->me_pghead = NULL;
+ free(mop);
}
/* Update DB root pointers. Their pages have already been
DPRINTF("committing page %zu", dp->mp_pgno);
iov[n].iov_len = env->me_psize;
if (IS_OVERFLOW(dp)) iov[n].iov_len *= dp->mp_pages;
- iov[n].iov_base = dp;
+ iov[n].iov_base = (char *)dp;
size += iov[n].iov_len;
next = dp->mp_pgno + (IS_OVERFLOW(dp) ? dp->mp_pages : 1);
/* clear dirty flag */
dp = txn->mt_u.dirty_list[i].mptr;
if (!IS_OVERFLOW(dp) || dp->mp_pages == 1) {
dp->mp_next = txn->mt_env->me_dpages;
+ VGMEMP_FREE(txn->mt_env, dp);
txn->mt_env->me_dpages = dp;
} else {
+ VGMEMP_FREE(txn->mt_env, dp);
free(dp);
}
txn->mt_u.dirty_list[i].mid = 0;
static int
mdb_env_read_header(MDB_env *env, MDB_meta *meta)
{
- char page[PAGESIZE];
+ char page[MDB_PAGESIZE];
MDB_page *p;
MDB_meta *m;
int rc, err;
*/
#ifdef _WIN32
- if (!ReadFile(env->me_fd, page, PAGESIZE, (DWORD *)&rc, NULL) || rc == 0)
+ if (!ReadFile(env->me_fd, page, MDB_PAGESIZE, (DWORD *)&rc, NULL) || rc == 0)
#else
- if ((rc = read(env->me_fd, page, PAGESIZE)) == 0)
+ if ((rc = read(env->me_fd, page, MDB_PAGESIZE)) == 0)
#endif
{
return ENOENT;
}
- else if (rc != PAGESIZE) {
+ else if (rc != MDB_PAGESIZE) {
err = ErrCode();
if (rc > 0)
err = EINVAL;
e->me_fd = INVALID_HANDLE_VALUE;
e->me_lfd = INVALID_HANDLE_VALUE;
e->me_mfd = INVALID_HANDLE_VALUE;
+ VGMEMP_CREATE(e,0,0);
*env = e;
return MDB_SUCCESS;
}
}
}
#else
- env->me_txns = mmap(0, rsize, PROT_READ|PROT_WRITE, MAP_SHARED,
+ env->me_txns = (MDB_txninfo *)mmap(0, rsize, PROT_READ|PROT_WRITE, MAP_SHARED,
env->me_lfd, 0);
if (env->me_txns == MAP_FAILED) {
rc = ErrCode();
#define LOCKNAME "/lock.mdb"
/** The name of the data file in the DB environment */
#define DATANAME "/data.mdb"
+ /** The suffix of the lock file when no subdir is used */
+#define LOCKSUFF "-lock"
+
int
mdb_env_open(MDB_env *env, const char *path, unsigned int flags, mode_t mode)
{
char *lpath, *dpath;
len = strlen(path);
- lpath = malloc(len + sizeof(LOCKNAME) + len + sizeof(DATANAME));
+ if (flags & MDB_NOSUBDIR) {
+ rc = len + sizeof(LOCKSUFF) + len + 1;
+ } else {
+ rc = len + sizeof(LOCKNAME) + len + sizeof(DATANAME);
+ }
+ lpath = malloc(rc);
if (!lpath)
return ENOMEM;
- dpath = lpath + len + sizeof(LOCKNAME);
- sprintf(lpath, "%s" LOCKNAME, path);
- sprintf(dpath, "%s" DATANAME, path);
+ if (flags & MDB_NOSUBDIR) {
+ dpath = lpath + len + sizeof(LOCKSUFF);
+ sprintf(lpath, "%s" LOCKSUFF, path);
+ strcpy(dpath, path);
+ } else {
+ dpath = lpath + len + sizeof(LOCKNAME);
+ sprintf(lpath, "%s" LOCKNAME, path);
+ sprintf(dpath, "%s" DATANAME, path);
+ }
rc = mdb_env_setup_locks(env, lpath, mode, &excl);
if (rc)
if (env == NULL)
return;
+ VGMEMP_DESTROY(env);
while (env->me_dpages) {
dp = env->me_dpages;
+ VGMEMP_DEFINED(&dp->mp_next, sizeof(dp->mp_next));
env->me_dpages = dp->mp_next;
free(dp);
}
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, (env->me_maxreaders-1)*sizeof(MDB_reader)+sizeof(MDB_txninfo));
+ munmap((void *)env->me_txns, (env->me_maxreaders-1)*sizeof(MDB_reader)+sizeof(MDB_txninfo));
}
close(env->me_lfd);
mdb_midl_free(env->me_free_pgs);
nkeys = NUMKEYS(mp);
+#if MDB_DEBUG
+ {
+ pgno_t pgno;
+ COPY_PGNO(pgno, mp->mp_pgno);
DPRINTF("searching %u keys in %s %spage %zu",
nkeys, IS_LEAF(mp) ? "leaf" : "branch", IS_SUBP(mp) ? "sub-" : "",
- mp->mp_pgno);
+ pgno);
+ }
+#endif
assert(nkeys > 0);
nodekey.mv_data = NODEKEY(node);
rc = cmp(key, &nodekey);
-#if DEBUG
+#if MDB_DEBUG
if (IS_LEAF(mp))
DPRINTF("found leaf index %u [%s], rc = %i",
i, DKEY(&nodekey), rc);
return node;
}
+#if 0
+static void
+mdb_cursor_adjust(MDB_cursor *mc, func)
+{
+ MDB_cursor *m2;
+
+ for (m2 = mc->mc_txn->mt_cursors[mc->mc_dbi]; m2; m2=m2->mc_next) {
+ if (m2->mc_pg[m2->mc_top] == mc->mc_pg[mc->mc_top]) {
+ func(mc, m2);
+ }
+ }
+}
+#endif
+
/** Pop a page off the top of the cursor's stack. */
static void
mdb_cursor_pop(MDB_cursor *mc)
}
}
+ assert(root > 1);
if ((rc = mdb_page_get(mc->mc_txn, root, &mc->mc_pg[0])))
return rc;
return rc;
}
} else {
- mc->mc_xcursor->mx_cursor.mc_flags = 0;
+ mc->mc_xcursor->mx_cursor.mc_flags &= ~C_INITIALIZED;
if (op == MDB_NEXT_DUP)
return MDB_NOTFOUND;
}
if (op != MDB_PREV || rc == MDB_SUCCESS)
return rc;
} else {
- mc->mc_xcursor->mx_cursor.mc_flags = 0;
+ mc->mc_xcursor->mx_cursor.mc_flags &= ~C_INITIALIZED;
if (op == MDB_PREV_DUP)
return MDB_NOTFOUND;
}
} else {
if (mc->mc_xcursor)
- mc->mc_xcursor->mx_cursor.mc_flags = 0;
+ mc->mc_xcursor->mx_cursor.mc_flags &= ~C_INITIALIZED;
if ((rc = mdb_node_read(mc->mc_txn, leaf, data)) != MDB_SUCCESS)
return rc;
}
return rc;
} else {
if (mc->mc_xcursor)
- mc->mc_xcursor->mx_cursor.mc_flags = 0;
+ mc->mc_xcursor->mx_cursor.mc_flags &= ~C_INITIALIZED;
if ((rc = mdb_node_read(mc->mc_txn, leaf, data)) != MDB_SUCCESS)
return rc;
}
return rc;
} else {
if (mc->mc_xcursor)
- mc->mc_xcursor->mx_cursor.mc_flags = 0;
+ mc->mc_xcursor->mx_cursor.mc_flags &= ~C_INITIALIZED;
if ((rc = mdb_node_read(mc->mc_txn, leaf, data)) != MDB_SUCCESS)
return rc;
}
MDB_page *fp;
MDB_db dummy;
int do_sub = 0;
+ unsigned int mcount = 0;
size_t nsize;
int rc, rc2;
- char pbuf[PAGESIZE];
+ char pbuf[MDB_PAGESIZE];
char dbuf[MAXKEYSIZE+1];
+ unsigned int nflags;
DKBUF;
if (F_ISSET(mc->mc_txn->mt_flags, MDB_TXN_RDONLY))
int exact = 0;
MDB_val d2;
rc = mdb_cursor_set(mc, key, &d2, MDB_SET, &exact);
- if (flags == MDB_NOOVERWRITE && rc == 0) {
+ if ((flags & MDB_NOOVERWRITE) && rc == 0) {
DPRINTF("duplicate key [%s]", DKEY(key));
*data = d2;
return MDB_KEYEXIST;
/* DB has dups? */
if (F_ISSET(mc->mc_db->md_flags, MDB_DUPSORT)) {
/* Was a single item before, must convert now */
+more:
if (!F_ISSET(leaf->mn_flags, F_DUPDATA)) {
/* Just overwrite the current item */
if (flags == MDB_CURRENT)
goto current;
- /* create a fake page for the dup items */
dkey.mv_size = NODEDSZ(leaf);
dkey.mv_data = NODEDATA(leaf);
- /* data matches, ignore it */
+#if UINT_MAX < SIZE_MAX
+ if (mc->mc_dbx->md_dcmp == mdb_cmp_int && dkey.mv_size == sizeof(size_t))
+#ifdef MISALIGNED_OK
+ mc->mc_dbx->md_dcmp = mdb_cmp_long;
+#else
+ mc->mc_dbx->md_dcmp = mdb_cmp_cint;
+#endif
+#endif
+ /* if data matches, ignore it */
if (!mc->mc_dbx->md_dcmp(data, &dkey))
return (flags == MDB_NODUPDATA) ? MDB_KEYEXIST : MDB_SUCCESS;
+
+ /* create a fake page for the dup items */
memcpy(dbuf, dkey.mv_data, dkey.mv_size);
dkey.mv_data = dbuf;
fp = (MDB_page *)pbuf;
fp = NODEDATA(leaf);
if (flags == MDB_CURRENT) {
fp->mp_flags |= P_DIRTY;
- fp->mp_pgno = mc->mc_pg[mc->mc_top]->mp_pgno;
+ COPY_PGNO(fp->mp_pgno, mc->mc_pg[mc->mc_top]->mp_pgno);
mc->mc_xcursor->mx_cursor.mc_pg[0] = fp;
flags |= F_DUPDATA;
goto put_sub;
/* same size, just replace it */
if (!F_ISSET(leaf->mn_flags, F_BIGDATA) &&
NODEDSZ(leaf) == data->mv_size) {
- memcpy(NODEDATA(leaf), data->mv_data, data->mv_size);
+ if (F_ISSET(flags, MDB_RESERVE))
+ data->mv_data = NODEDATA(leaf);
+ else
+ memcpy(NODEDATA(leaf), data->mv_data, data->mv_size);
goto done;
}
mdb_node_del(mc->mc_pg[mc->mc_top], mc->mc_ki[mc->mc_top], 0);
rdata = data;
new_sub:
+ nflags = flags & NODE_ADD_FLAGS;
nsize = IS_LEAF2(mc->mc_pg[mc->mc_top]) ? key->mv_size : mdb_leaf_size(mc->mc_txn->mt_env, key, rdata);
if (SIZELEFT(mc->mc_pg[mc->mc_top]) < nsize) {
- rc = mdb_page_split(mc, key, rdata, P_INVALID);
+ if (( flags & (F_DUPDATA|F_SUBDATA)) == F_DUPDATA )
+ nflags &= ~MDB_APPEND;
+ rc = mdb_page_split(mc, key, rdata, P_INVALID, nflags);
} else {
/* There is room already in this leaf page. */
- rc = mdb_node_add(mc, mc->mc_ki[mc->mc_top], key, rdata, 0, 0);
+ rc = mdb_node_add(mc, mc->mc_ki[mc->mc_top], key, rdata, 0, nflags);
+ if (rc == 0 && !do_sub) {
+ /* Adjust other cursors pointing to mp */
+ MDB_cursor *m2, *m3;
+ MDB_dbi dbi = mc->mc_dbi;
+ unsigned i = mc->mc_top;
+ MDB_page *mp = mc->mc_pg[i];
+
+ if (mc->mc_flags & C_SUB)
+ dbi--;
+
+ for (m2 = mc->mc_txn->mt_cursors[dbi]; m2; m2=m2->mc_next) {
+ if (mc->mc_flags & C_SUB)
+ m3 = &m2->mc_xcursor->mx_cursor;
+ else
+ m3 = m2;
+ if (m3 == mc || m3->mc_snum < mc->mc_snum) continue;
+ if (m3->mc_pg[i] == mp && m3->mc_ki[i] >= mc->mc_ki[i]) {
+ m3->mc_ki[i]++;
+ }
+ }
+ }
}
if (rc != MDB_SUCCESS)
mc->mc_txn->mt_flags |= MDB_TXN_ERROR;
else {
- /* Remember if we just added a subdatabase */
- if (flags & (F_SUBDATA|F_DUPDATA)) {
- leaf = NODEPTR(mc->mc_pg[mc->mc_top], mc->mc_ki[mc->mc_top]);
- leaf->mn_flags |= (flags & (F_SUBDATA|F_DUPDATA));
- }
-
/* Now store the actual data in the child DB. Note that we're
* storing the user data in the keys field, so there are strict
* size limits on dupdata. The actual data fields of the child
put_sub:
xdata.mv_size = 0;
xdata.mv_data = "";
+ leaf = NODEPTR(mc->mc_pg[mc->mc_top], mc->mc_ki[mc->mc_top]);
if (flags & MDB_CURRENT) {
xflags = MDB_CURRENT;
} else {
rc = mdb_cursor_put(&mc->mc_xcursor->mx_cursor, &dkey, &xdata, xflags);
if (rc)
return rc;
+ {
+ /* Adjust other cursors pointing to mp */
+ MDB_cursor *m2;
+ unsigned i = mc->mc_top;
+ MDB_page *mp = mc->mc_pg[i];
+
+ for (m2 = mc->mc_txn->mt_cursors[mc->mc_dbi]; m2; m2=m2->mc_next) {
+ if (m2 == mc || m2->mc_snum < mc->mc_snum) continue;
+ if (m2->mc_pg[i] == mp && m2->mc_ki[i] == mc->mc_ki[i]) {
+ mdb_xcursor_init1(m2, leaf);
+ }
+ }
+ }
}
+ xflags |= (flags & MDB_APPEND);
rc = mdb_cursor_put(&mc->mc_xcursor->mx_cursor, data, &xdata, xflags);
if (flags & F_SUBDATA) {
db = NODEDATA(leaf);
*/
if (!rc && !(flags & MDB_CURRENT))
mc->mc_db->md_entries++;
+ if (flags & MDB_MULTIPLE) {
+ mcount++;
+ if (mcount < data[1].mv_size) {
+ data[0].mv_data = (char *)data[0].mv_data + data[0].mv_size;
+ leaf = NODEPTR(mc->mc_pg[mc->mc_top], mc->mc_ki[mc->mc_top]);
+ goto more;
+ }
+ }
}
done:
return rc;
size_t sz;
sz = LEAFSIZE(key, data);
- if (data->mv_size >= env->me_psize / MDB_MINKEYS) {
+ if (sz >= env->me_psize / MDB_MINKEYS) {
/* put on overflow page */
sz -= data->mv_size - sizeof(pgno_t);
}
*/
static int
mdb_node_add(MDB_cursor *mc, indx_t indx,
- MDB_val *key, MDB_val *data, pgno_t pgno, uint8_t flags)
+ MDB_val *key, MDB_val *data, pgno_t pgno, unsigned int flags)
{
unsigned int i;
size_t node_size = NODESIZE;
if (F_ISSET(flags, F_BIGDATA)) {
/* Data already on overflow page. */
node_size += sizeof(pgno_t);
- } else if (data->mv_size >= mc->mc_txn->mt_env->me_psize / MDB_MINKEYS) {
+ } else if (node_size + data->mv_size >= mc->mc_txn->mt_env->me_psize / MDB_MINKEYS) {
int ovpages = OVPAGES(data->mv_size, mc->mc_txn->mt_env->me_psize);
/* Put data on overflow page. */
- DPRINTF("data size is %zu, put on overflow page",
- data->mv_size);
+ DPRINTF("data size is %zu, node would be %zu, put data on overflow page",
+ data->mv_size, node_size+data->mv_size);
node_size += sizeof(pgno_t);
if ((ofp = mdb_page_new(mc, P_OVERFLOW, ovpages)) == NULL)
return ENOMEM;
if (F_ISSET(flags, F_BIGDATA))
memcpy(node->mn_data + key->mv_size, data->mv_data,
sizeof(pgno_t));
+ else if (F_ISSET(flags, MDB_RESERVE))
+ data->mv_data = node->mn_data + key->mv_size;
else
memcpy(node->mn_data + key->mv_size, data->mv_data,
data->mv_size);
} else {
memcpy(node->mn_data + key->mv_size, &ofp->mp_pgno,
sizeof(pgno_t));
- memcpy(METADATA(ofp), data->mv_data, data->mv_size);
+ if (F_ISSET(flags, MDB_RESERVE))
+ data->mv_data = METADATA(ofp);
+ else
+ memcpy(METADATA(ofp), data->mv_data, data->mv_size);
}
}
MDB_node *node;
char *base;
+#if MDB_DEBUG
+ {
+ pgno_t pgno;
+ COPY_PGNO(pgno, mp->mp_pgno);
DPRINTF("delete node %u on %s page %zu", indx,
- IS_LEAF(mp) ? "leaf" : "branch", mp->mp_pgno);
+ IS_LEAF(mp) ? "leaf" : "branch", pgno);
+ }
+#endif
assert(indx < NUMKEYS(mp));
if (IS_LEAF2(mp)) {
xp->mp_lower = sp->mp_lower;
xp->mp_flags = sp->mp_flags;
xp->mp_pad = sp->mp_pad;
- xp->mp_pgno = mp->mp_pgno;
+ COPY_PGNO(xp->mp_pgno, mp->mp_pgno);
/* shift lower nodes upward */
ptr = mp->mp_ptrs[indx];
mx->mx_cursor.mc_dbx = &mx->mx_dbx;
mx->mx_cursor.mc_dbi = mc->mc_dbi+1;
mx->mx_cursor.mc_dbflag = &mx->mx_dbflag;
+ mx->mx_cursor.mc_snum = 0;
+ mx->mx_cursor.mc_flags = C_SUB;
mx->mx_dbx.md_cmp = mc->mc_dbx->md_dcmp;
mx->mx_dbx.md_dcmp = NULL;
mx->mx_dbx.md_rel = mc->mc_dbx->md_rel;
MDB_db *db = NODEDATA(node);
mx->mx_db = *db;
mx->mx_cursor.mc_snum = 0;
- mx->mx_cursor.mc_flags = 0;
+ mx->mx_cursor.mc_flags = C_SUB;
} else {
MDB_page *fp = NODEDATA(node);
mx->mx_db.md_pad = mc->mc_pg[mc->mc_top]->mp_pad;
mx->mx_db.md_leaf_pages = 1;
mx->mx_db.md_overflow_pages = 0;
mx->mx_db.md_entries = NUMKEYS(fp);
- mx->mx_db.md_root = fp->mp_pgno;
+ COPY_PGNO(mx->mx_db.md_root, fp->mp_pgno);
mx->mx_cursor.mc_snum = 1;
- mx->mx_cursor.mc_flags = C_INITIALIZED;
+ mx->mx_cursor.mc_flags = C_INITIALIZED|C_SUB;
mx->mx_cursor.mc_top = 0;
mx->mx_cursor.mc_pg[0] = fp;
mx->mx_cursor.mc_ki[0] = 0;
DB_DIRTY : 0;
mx->mx_dbx.md_name.mv_data = NODEKEY(node);
mx->mx_dbx.md_name.mv_size = node->mn_ksize;
+#if UINT_MAX < SIZE_MAX
if (mx->mx_dbx.md_cmp == mdb_cmp_int && mx->mx_db.md_pad == sizeof(size_t))
+#ifdef MISALIGNED_OK
mx->mx_dbx.md_cmp = mdb_cmp_long;
+#else
+ mx->mx_dbx.md_cmp = mdb_cmp_cint;
+#endif
+#endif
}
/** Initialize a cursor for a given transaction and database. */
static void
mdb_cursor_init(MDB_cursor *mc, MDB_txn *txn, MDB_dbi dbi, MDB_xcursor *mx)
{
+ mc->mc_orig = NULL;
mc->mc_dbi = dbi;
mc->mc_txn = txn;
mc->mc_db = &txn->mt_dbs[dbi];
mc->mc_dbx = &txn->mt_dbxs[dbi];
mc->mc_dbflag = &txn->mt_dbflags[dbi];
mc->mc_snum = 0;
+ mc->mc_top = 0;
mc->mc_flags = 0;
if (txn->mt_dbs[dbi].md_flags & MDB_DUPSORT) {
assert(mx != NULL);
mx = (MDB_xcursor *)(mc + 1);
}
mdb_cursor_init(mc, txn, dbi, mx);
+ if (txn->mt_cursors) {
+ mc->mc_next = txn->mt_cursors[dbi];
+ txn->mt_cursors[dbi] = mc;
+ }
+ mc->mc_flags |= C_ALLOCD;
} else {
return ENOMEM;
}
mdb_cursor_close(MDB_cursor *mc)
{
if (mc != NULL) {
- free(mc);
+ /* remove from txn, if tracked */
+ if (mc->mc_txn->mt_cursors) {
+ MDB_cursor **prev = &mc->mc_txn->mt_cursors[mc->mc_dbi];
+ while (*prev && *prev != mc) prev = &(*prev)->mc_next;
+ if (*prev == mc)
+ *prev = mc->mc_next;
+ }
+ if (mc->mc_flags & C_ALLOCD)
+ free(mc);
}
}
+MDB_txn *
+mdb_cursor_txn(MDB_cursor *mc)
+{
+ if (!mc) return NULL;
+ return mc->mc_txn;
+}
+
+MDB_dbi
+mdb_cursor_dbi(MDB_cursor *mc)
+{
+ if (!mc) return 0;
+ return mc->mc_dbi;
+}
+
/** Replace the key for a node with a new key.
* @param[in] mp The page containing the node to operate on.
* @param[in] indx The index of the node to operate on.
*/
mdb_node_del(csrc->mc_pg[csrc->mc_top], csrc->mc_ki[csrc->mc_top], key.mv_size);
+ {
+ /* Adjust other cursors pointing to mp */
+ MDB_cursor *m2, *m3;
+ MDB_dbi dbi = csrc->mc_dbi;
+ MDB_page *mp = csrc->mc_pg[csrc->mc_top];
+
+ if (csrc->mc_flags & C_SUB)
+ dbi--;
+
+ for (m2 = csrc->mc_txn->mt_cursors[dbi]; m2; m2=m2->mc_next) {
+ if (m2 == csrc) continue;
+ if (csrc->mc_flags & C_SUB)
+ m3 = &m2->mc_xcursor->mx_cursor;
+ else
+ m3 = m2;
+ if (m3->mc_pg[csrc->mc_top] == mp && m3->mc_ki[csrc->mc_top] ==
+ csrc->mc_ki[csrc->mc_top]) {
+ m3->mc_pg[csrc->mc_top] = cdst->mc_pg[cdst->mc_top];
+ m3->mc_ki[csrc->mc_top] = cdst->mc_ki[cdst->mc_top];
+ }
+ }
+ }
+
/* Update the parent separators.
*/
if (csrc->mc_ki[csrc->mc_top] == 0) {
if (IS_BRANCH(csrc->mc_pg[csrc->mc_top])) {
MDB_val nullkey;
nullkey.mv_size = 0;
- assert(mdb_update_key(csrc->mc_pg[csrc->mc_top], 0, &nullkey) == MDB_SUCCESS);
+ rc = mdb_update_key(csrc->mc_pg[csrc->mc_top], 0, &nullkey);
+ assert(rc == MDB_SUCCESS);
}
}
if (IS_BRANCH(cdst->mc_pg[cdst->mc_top])) {
MDB_val nullkey;
nullkey.mv_size = 0;
- assert(mdb_update_key(cdst->mc_pg[cdst->mc_top], 0, &nullkey) == MDB_SUCCESS);
+ rc = mdb_update_key(cdst->mc_pg[cdst->mc_top], 0, &nullkey);
+ assert(rc == MDB_SUCCESS);
}
}
indx_t i, j;
MDB_node *srcnode;
MDB_val key, data;
+ unsigned nkeys;
DPRINTF("merging page %zu into %zu", csrc->mc_pg[csrc->mc_top]->mp_pgno,
cdst->mc_pg[cdst->mc_top]->mp_pgno);
/* Move all nodes from src to dst.
*/
- j = NUMKEYS(cdst->mc_pg[cdst->mc_top]);
+ j = nkeys = NUMKEYS(cdst->mc_pg[cdst->mc_top]);
if (IS_LEAF2(csrc->mc_pg[csrc->mc_top])) {
key.mv_size = csrc->mc_db->md_pad;
key.mv_data = METADATA(csrc->mc_pg[csrc->mc_top]);
csrc->mc_db->md_leaf_pages--;
else
csrc->mc_db->md_branch_pages--;
+ {
+ /* Adjust other cursors pointing to mp */
+ MDB_cursor *m2, *m3;
+ MDB_dbi dbi = csrc->mc_dbi;
+ MDB_page *mp = cdst->mc_pg[cdst->mc_top];
+
+ if (csrc->mc_flags & C_SUB)
+ dbi--;
+
+ for (m2 = csrc->mc_txn->mt_cursors[dbi]; m2; m2=m2->mc_next) {
+ if (m2 == csrc) continue;
+ if (csrc->mc_flags & C_SUB)
+ m3 = &m2->mc_xcursor->mx_cursor;
+ else
+ m3 = m2;
+ if (m3->mc_snum < csrc->mc_snum) continue;
+ if (m3->mc_pg[csrc->mc_top] == csrc->mc_pg[csrc->mc_top]) {
+ m3->mc_pg[csrc->mc_top] = mp;
+ m3->mc_ki[csrc->mc_top] += nkeys;
+ }
+ }
+ }
mdb_cursor_pop(csrc);
return mdb_rebalance(csrc);
unsigned int ptop;
MDB_cursor mn;
+#if MDB_DEBUG
+ {
+ pgno_t pgno;
+ COPY_PGNO(pgno, mc->mc_pg[mc->mc_top]->mp_pgno);
DPRINTF("rebalancing %s page %zu (has %u keys, %.1f%% full)",
IS_LEAF(mc->mc_pg[mc->mc_top]) ? "leaf" : "branch",
- mc->mc_pg[mc->mc_top]->mp_pgno, NUMKEYS(mc->mc_pg[mc->mc_top]), (float)PAGEFILL(mc->mc_txn->mt_env, mc->mc_pg[mc->mc_top]) / 10);
+ pgno, NUMKEYS(mc->mc_pg[mc->mc_top]), (float)PAGEFILL(mc->mc_txn->mt_env, mc->mc_pg[mc->mc_top]) / 10);
+ }
+#endif
if (PAGEFILL(mc->mc_txn->mt_env, mc->mc_pg[mc->mc_top]) >= FILL_THRESHOLD) {
+#if MDB_DEBUG
+ pgno_t pgno;
+ COPY_PGNO(pgno, mc->mc_pg[mc->mc_top]->mp_pgno);
DPRINTF("no need to rebalance page %zu, above fill threshold",
- mc->mc_pg[mc->mc_top]->mp_pgno);
+ pgno);
+#endif
return MDB_SUCCESS;
}
if (mc->mc_snum < 2) {
- if (NUMKEYS(mc->mc_pg[mc->mc_top]) == 0) {
+ MDB_page *mp = mc->mc_pg[0];
+ if (NUMKEYS(mp) == 0) {
DPUTS("tree is completely empty");
mc->mc_db->md_root = P_INVALID;
mc->mc_db->md_depth = 0;
mc->mc_db->md_leaf_pages = 0;
- mdb_midl_append(&mc->mc_txn->mt_free_pgs, mc->mc_pg[mc->mc_top]->mp_pgno);
+ mdb_midl_append(&mc->mc_txn->mt_free_pgs, mp->mp_pgno);
mc->mc_snum = 0;
- } else if (IS_BRANCH(mc->mc_pg[mc->mc_top]) && NUMKEYS(mc->mc_pg[mc->mc_top]) == 1) {
+ mc->mc_top = 0;
+ {
+ /* Adjust other cursors pointing to mp */
+ MDB_cursor *m2, *m3;
+ MDB_dbi dbi = mc->mc_dbi;
+
+ if (mc->mc_flags & C_SUB)
+ dbi--;
+
+ for (m2 = mc->mc_txn->mt_cursors[dbi]; m2; m2=m2->mc_next) {
+ if (m2 == mc) continue;
+ if (mc->mc_flags & C_SUB)
+ m3 = &m2->mc_xcursor->mx_cursor;
+ else
+ m3 = m2;
+ if (m3->mc_snum < mc->mc_snum) continue;
+ if (m3->mc_pg[0] == mp) {
+ m3->mc_snum = 0;
+ m3->mc_top = 0;
+ }
+ }
+ }
+ } else if (IS_BRANCH(mp) && NUMKEYS(mp) == 1) {
DPUTS("collapsing root page!");
- mdb_midl_append(&mc->mc_txn->mt_free_pgs, mc->mc_pg[mc->mc_top]->mp_pgno);
- mc->mc_db->md_root = NODEPGNO(NODEPTR(mc->mc_pg[mc->mc_top], 0));
+ mdb_midl_append(&mc->mc_txn->mt_free_pgs, mp->mp_pgno);
+ mc->mc_db->md_root = NODEPGNO(NODEPTR(mp, 0));
if ((rc = mdb_page_get(mc->mc_txn, mc->mc_db->md_root,
- &mc->mc_pg[mc->mc_top])))
+ &mc->mc_pg[0])))
return rc;
mc->mc_db->md_depth--;
mc->mc_db->md_branch_pages--;
+ {
+ /* Adjust other cursors pointing to mp */
+ MDB_cursor *m2, *m3;
+ MDB_dbi dbi = mc->mc_dbi;
+
+ if (mc->mc_flags & C_SUB)
+ dbi--;
+
+ for (m2 = mc->mc_txn->mt_cursors[dbi]; m2; m2=m2->mc_next) {
+ if (m2 == mc) continue;
+ if (mc->mc_flags & C_SUB)
+ m3 = &m2->mc_xcursor->mx_cursor;
+ else
+ m3 = m2;
+ if (m3->mc_snum < mc->mc_snum) continue;
+ if (m3->mc_pg[0] == mp) {
+ m3->mc_pg[0] = mc->mc_pg[0];
+ }
+ }
+ }
} else
DPUTS("root page doesn't need rebalancing");
return MDB_SUCCESS;
* @return 0 on success, non-zero on failure.
*/
static int
-mdb_page_split(MDB_cursor *mc, MDB_val *newkey, MDB_val *newdata, pgno_t newpgno)
+mdb_page_split(MDB_cursor *mc, MDB_val *newkey, MDB_val *newdata, pgno_t newpgno,
+ unsigned int nflags)
{
- uint8_t flags;
- int rc = MDB_SUCCESS, ins_new = 0;
+ unsigned int flags;
+ int rc = MDB_SUCCESS, ins_new = 0, new_root = 0;
indx_t newindx;
pgno_t pgno = 0;
unsigned int i, j, split_indx, nkeys, pmax;
MDB_node *node;
- MDB_val sepkey, rkey, rdata;
+ MDB_val sepkey, rkey, xdata, *rdata = &xdata;
MDB_page *copy;
MDB_page *mp, *rp, *pp;
unsigned int ptop;
mc->mc_db->md_root = pp->mp_pgno;
DPRINTF("root split! new root = %zu", pp->mp_pgno);
mc->mc_db->md_depth++;
+ new_root = 1;
/* Add left (implicit) pointer. */
if ((rc = mdb_node_add(mc, 0, NULL, NULL, mp->mp_pgno, 0)) != MDB_SUCCESS) {
/* Create a right sibling. */
if ((rp = mdb_page_new(mc, mp->mp_flags, 1)) == NULL)
return ENOMEM;
+ DPRINTF("new right sibling: page %zu", rp->mp_pgno);
+
mdb_cursor_copy(mc, &mn);
mn.mc_pg[mn.mc_top] = rp;
mn.mc_ki[ptop] = mc->mc_ki[ptop]+1;
- DPRINTF("new right sibling: page %zu", rp->mp_pgno);
+
+ if (nflags & MDB_APPEND) {
+ mn.mc_ki[mn.mc_top] = 0;
+ sepkey = *newkey;
+ nkeys = 0;
+ split_indx = 0;
+ goto newsep;
+ }
nkeys = NUMKEYS(mp);
split_indx = nkeys / 2 + 1;
if (SIZELEFT(mn.mc_pg[ptop]) < mdb_branch_size(mc->mc_txn->mt_env, &sepkey)) {
mn.mc_snum--;
mn.mc_top--;
- rc = mdb_page_split(&mn, &sepkey, NULL, rp->mp_pgno);
+ rc = mdb_page_split(&mn, &sepkey, NULL, rp->mp_pgno, 0);
/* Right page might now have changed parent.
* Check if left page also changed parent.
rc = mdb_node_add(&mn, mn.mc_ki[ptop], &sepkey, NULL, rp->mp_pgno, 0);
mn.mc_top++;
}
- if (IS_LEAF2(rp)) {
- return rc;
- }
if (rc != MDB_SUCCESS) {
return rc;
}
+ if (nflags & MDB_APPEND) {
+ mc->mc_pg[mc->mc_top] = rp;
+ mc->mc_ki[mc->mc_top] = 0;
+ return mdb_node_add(mc, 0, newkey, newdata, newpgno, nflags);
+ }
+ if (IS_LEAF2(rp)) {
+ goto done;
+ }
/* Move half of the keys to the right sibling. */
rkey.mv_data = newkey->mv_data;
rkey.mv_size = newkey->mv_size;
if (IS_LEAF(mp)) {
- rdata.mv_data = newdata->mv_data;
- rdata.mv_size = newdata->mv_size;
+ rdata = newdata;
} else
pgno = newpgno;
- flags = 0;
+ flags = nflags;
ins_new = 1;
/* Update page and index for the new key. */
+ if (!newindx)
+ mc->mc_pg[mc->mc_top] = copy;
mc->mc_ki[mc->mc_top] = j;
} else if (i == nkeys) {
break;
rkey.mv_data = NODEKEY(node);
rkey.mv_size = node->mn_ksize;
if (IS_LEAF(mp)) {
- rdata.mv_data = NODEDATA(node);
- rdata.mv_size = NODEDSZ(node);
+ xdata.mv_data = NODEDATA(node);
+ xdata.mv_size = NODEDSZ(node);
+ rdata = &xdata;
} else
pgno = NODEPGNO(node);
flags = node->mn_flags;
rkey.mv_size = 0;
}
- rc = mdb_node_add(mc, j, &rkey, &rdata, pgno, flags);
+ rc = mdb_node_add(mc, j, &rkey, rdata, pgno, flags);
}
- /* reset back to original page */
- if (newindx < split_indx)
- mc->mc_pg[mc->mc_top] = mp;
-
nkeys = NUMKEYS(copy);
for (i=0; i<nkeys; i++)
mp->mp_ptrs[i] = copy->mp_ptrs[i];
memcpy(NODEPTR(mp, nkeys-1), NODEPTR(copy, nkeys-1),
mc->mc_txn->mt_env->me_psize - copy->mp_upper);
+ /* reset back to original page */
+ if (!newindx || (newindx < split_indx)) {
+ mc->mc_pg[mc->mc_top] = mp;
+ if (nflags & MDB_RESERVE) {
+ node = NODEPTR(mp, mc->mc_ki[mc->mc_top]);
+ if (!(node->mn_flags & F_BIGDATA))
+ newdata->mv_data = NODEDATA(node);
+ }
+ }
+
/* return tmp page to freelist */
copy->mp_next = mc->mc_txn->mt_env->me_dpages;
+ VGMEMP_FREE(mc->mc_txn->mt_env, copy);
mc->mc_txn->mt_env->me_dpages = copy;
+done:
+ {
+ /* Adjust other cursors pointing to mp */
+ MDB_cursor *m2, *m3;
+ MDB_dbi dbi = mc->mc_dbi;
+
+ if (mc->mc_flags & C_SUB)
+ dbi--;
+
+ for (m2 = mc->mc_txn->mt_cursors[dbi]; m2; m2=m2->mc_next) {
+ if (m2 == mc) continue;
+ if (mc->mc_flags & C_SUB)
+ m3 = &m2->mc_xcursor->mx_cursor;
+ else
+ m3 = m2;
+ if (!(m3->mc_flags & C_INITIALIZED))
+ continue;
+ if (new_root) {
+ /* root split */
+ for (i=m3->mc_top; i>0; i--) {
+ m3->mc_ki[i+1] = m3->mc_ki[i];
+ m3->mc_pg[i+1] = m3->mc_pg[i];
+ }
+ m3->mc_ki[0] = mc->mc_ki[0];
+ m3->mc_pg[0] = mc->mc_pg[0];
+ m3->mc_snum++;
+ m3->mc_top++;
+ }
+ if (m3->mc_pg[mc->mc_top] == mp) {
+ if (m3->mc_ki[m3->mc_top] >= split_indx) {
+ m3->mc_pg[m3->mc_top] = rp;
+ m3->mc_ki[m3->mc_top] -= split_indx;
+ }
+ }
+ }
+ }
return rc;
}
return EINVAL;
}
- if ((flags & (MDB_NOOVERWRITE|MDB_NODUPDATA)) != flags)
+ if ((flags & (MDB_NOOVERWRITE|MDB_NODUPDATA|MDB_RESERVE|MDB_APPEND)) != flags)
return EINVAL;
mdb_cursor_init(&mc, txn, dbi, &mx);
{
MDB_val key, data;
MDB_dbi i;
- int rc, dbflag = 0;
+ MDB_cursor mc;
+ int rc, dbflag, exact;
size_t len;
if (txn->mt_dbxs[FREE_DBI].md_cmp == NULL) {
return ENFILE;
/* Find the DB info */
+ dbflag = 0;
+ exact = 0;
key.mv_size = len;
key.mv_data = (void *)name;
- rc = mdb_get(txn, MAIN_DBI, &key, &data);
-
- /* Create if requested */
- if (rc == MDB_NOTFOUND && (flags & MDB_CREATE)) {
- MDB_cursor mc;
+ mdb_cursor_init(&mc, txn, MAIN_DBI, NULL);
+ rc = mdb_cursor_set(&mc, &key, &data, MDB_SET, &exact);
+ if (rc == MDB_SUCCESS) {
+ /* make sure this is actually a DB */
+ MDB_node *node = NODEPTR(mc.mc_pg[mc.mc_top], mc.mc_ki[mc.mc_top]);
+ if (!(node->mn_flags & F_SUBDATA))
+ return EINVAL;
+ } else if (rc == MDB_NOTFOUND && (flags & MDB_CREATE)) {
+ /* Create if requested */
MDB_db dummy;
data.mv_size = sizeof(MDB_db);
data.mv_data = &dummy;
memset(&dummy, 0, sizeof(dummy));
dummy.md_root = P_INVALID;
dummy.md_flags = flags & 0xffff;
- mdb_cursor_init(&mc, txn, MAIN_DBI, NULL);
rc = mdb_cursor_put(&mc, &key, &data, F_SUBDATA);
dbflag = DB_DIRTY;
}
rc = mdb_drop0(mc, mc->mc_db->md_flags & MDB_DUPSORT);
if (rc)
- mdb_cursor_close(mc);
- return rc;
+ goto leave;
/* Can't delete the main DB */
if (del && dbi > MAIN_DBI) {
txn->mt_dbs[dbi].md_entries = 0;
txn->mt_dbs[dbi].md_root = P_INVALID;
}
+leave:
mdb_cursor_close(mc);
return rc;
}