* 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
/** 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.
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 (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;
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;
}
MDB_cursor *m2;
for (m2 = mc->mc_txn->mt_cursors[mc->mc_dbi]; m2; m2=m2->mc_next) {
- if (m2 == mc) continue;
+ 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;
}
{
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;
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);
}
}
}
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;
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();
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 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;
}
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;
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;
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;
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;
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) {
+ 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. */
m3 = &m2->mc_xcursor->mx_cursor;
else
m3 = m2;
- if (m3 == mc) continue;
+ 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]++;
}
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) continue;
+ 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);
}
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);
}
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;
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;
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|C_SUB;
mx->mx_cursor.mc_top = 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. */
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);
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;
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;
}
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;
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];
}
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;
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:
{
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--) {
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;
}