#define MDB_OWNERDEAD EOWNERDEAD /**< #LOCK_MUTEX0() result if dead owner */
#endif
-/* Android currently lacks Robust Mutex support */
-#if defined(ANDROID) && defined(MDB_USE_POSIX_MUTEX)
-#define MDB_NO_ROBUST 1
+#ifdef __GLIBC__
+#define GLIBC_VER ((__GLIBC__ << 16 )| __GLIBC_MINOR__)
#endif
-
/** Some platforms define the EOWNERDEAD error code
* even though they don't support Robust Mutexes.
- * Compile with -DMDB_NO_ROBUST, or use some other
+ * Compile with -DMDB_USE_ROBUST=0, or use some other
* mechanism like -DMDB_USE_SYSV_SEM instead of
* -DMDB_USE_POSIX_MUTEX. (SysV semaphores are
* also Robust, but some systems don't support them
* either.)
*/
+#ifndef MDB_USE_ROBUST
+/* Android currently lacks Robust Mutex support. So does glibc < 2.4. */
+# if defined(MDB_USE_POSIX_MUTEX) && (defined(ANDROID) || \
+ (defined(__GLIBC__) && GLIBC_VER < 0x020004))
+# define MDB_USE_ROBUST 0
+# else
+# define MDB_USE_ROBUST 1
+/* glibc < 2.10 only provided _np API */
+# if defined(__GLIBC__) && GLIBC_VER < 0x02000a
+# define PTHREAD_MUTEX_ROBUST PTHREAD_MUTEX_ROBUST_NP
+# define pthread_mutexattr_setrobust(attr, flag) pthread_mutexattr_setrobust_np(attr, flag)
+# define pthread_mutex_consistent(mutex) pthread_mutex_consistent_np(mutex)
+# endif
+# endif
+#endif /* MDB_USE_ROBUST */
-#ifndef MDB_NO_ROBUST
-#define MDB_NO_ROBUST 0
-#endif
-
-#if defined(MDB_OWNERDEAD) && !(MDB_NO_ROBUST)
+#if defined(MDB_OWNERDEAD) && MDB_USE_ROBUST
#define MDB_ROBUST_SUPPORTED 1
#endif
*
* @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
+# ifdef O_DSYNC
# define MDB_DSYNC O_DSYNC
+# else
+# define MDB_DSYNC O_SYNC
+# endif
#endif
#endif
static SECURITY_DESCRIPTOR mdb_null_sd;
static SECURITY_ATTRIBUTES mdb_all_sa;
static int mdb_sec_inited;
+
+static int utf8_to_utf16(const char *src, int srcsize, wchar_t **dst, int *dstsize);
#endif
/** Return the library version info. */
}
/* Append our loose page list to parent's */
- for (lp = &parent->mt_loose_pgs; *lp; lp = &NEXT_LOOSE_PAGE(lp))
+ for (lp = &parent->mt_loose_pgs; *lp; lp = &NEXT_LOOSE_PAGE(*lp))
;
*lp = txn->mt_loose_pgs;
parent->mt_loose_count += txn->mt_loose_count;
off_t size, rsize;
#ifdef _WIN32
- env->me_lfd = CreateFileA(lpath, GENERIC_READ|GENERIC_WRITE,
+ wchar_t *wlpath;
+ utf8_to_utf16(lpath, -1, &wlpath, NULL);
+ env->me_lfd = CreateFileW(wlpath, GENERIC_READ|GENERIC_WRITE,
FILE_SHARE_READ|FILE_SHARE_WRITE, NULL, OPEN_ALWAYS,
FILE_ATTRIBUTE_NORMAL, NULL);
+ free(wlpath);
#else
env->me_lfd = open(lpath, O_RDWR|O_CREAT|MDB_CLOEXEC, mode);
#endif
{
int oflags, rc, len, excl = -1;
char *lpath, *dpath;
+#ifdef _WIN32
+ wchar_t *wpath;
+#endif
if (env->me_fd!=INVALID_HANDLE_VALUE || (flags & ~(CHANGEABLE|CHANGELESS)))
return EINVAL;
len = OPEN_ALWAYS;
}
mode = FILE_ATTRIBUTE_NORMAL;
- env->me_fd = CreateFileA(dpath, oflags, FILE_SHARE_READ|FILE_SHARE_WRITE,
+ utf8_to_utf16(dpath, -1, &wpath, NULL);
+ env->me_fd = CreateFileW(wpath, oflags, FILE_SHARE_READ|FILE_SHARE_WRITE,
NULL, len, mode, NULL);
+ free(wpath);
#else
if (F_ISSET(flags, MDB_RDONLY))
oflags = O_RDONLY;
*/
#ifdef _WIN32
len = OPEN_EXISTING;
- env->me_mfd = CreateFileA(dpath, oflags,
+ utf8_to_utf16(dpath, -1, &wpath, NULL);
+ env->me_mfd = CreateFileW(wpath, oflags,
FILE_SHARE_READ|FILE_SHARE_WRITE, NULL, len,
mode | FILE_FLAG_WRITE_THROUGH, NULL);
+ free(wpath);
#else
oflags &= ~O_CREAT;
env->me_mfd = open(dpath, oflags | MDB_DSYNC, mode);
return MDB_CORRUPTED;
}
}
+ txn->mt_dirty_room++;
if (!(env->me_flags & MDB_WRITEMAP))
mdb_dpage_free(env, mp);
release:
* update branch key if there is a parent page
*/
if (mc->mc_top && !mc->mc_ki[mc->mc_top]) {
- unsigned short top = mc->mc_top;
+ unsigned short dtop = 1;
mc->mc_top--;
/* slot 0 is always an empty key, find real slot */
- while (mc->mc_top && !mc->mc_ki[mc->mc_top])
+ while (mc->mc_top && !mc->mc_ki[mc->mc_top]) {
mc->mc_top--;
+ dtop++;
+ }
if (mc->mc_ki[mc->mc_top])
rc2 = mdb_update_key(mc, key);
else
rc2 = MDB_SUCCESS;
- mc->mc_top = top;
+ mc->mc_top += dtop;
if (rc2)
return rc2;
}
return ENOMEM;
id2.mid = pg;
id2.mptr = np;
+ /* Note - this page is already counted in parent's dirty_room */
rc2 = mdb_mid2l_insert(mc->mc_txn->mt_u.dirty_list, &id2);
mdb_cassert(mc, rc2 == 0);
if (!(flags & MDB_RESERVE)) {
/* Adjust other cursors pointing to mp */
MDB_cursor *m2, *m3;
MDB_dbi dbi = csrc->mc_dbi;
- MDB_page *mp;
-
- mp = cdst->mc_pg[csrc->mc_top];
- for (m2 = csrc->mc_txn->mt_cursors[dbi]; m2; m2=m2->mc_next) {
- if (csrc->mc_flags & C_SUB)
- m3 = &m2->mc_xcursor->mx_cursor;
- else
- m3 = m2;
- if (m3 == cdst) continue;
- if (m3->mc_pg[csrc->mc_top] == mp && m3->mc_ki[csrc->mc_top] >=
- cdst->mc_ki[csrc->mc_top]) {
- m3->mc_ki[csrc->mc_top]++;
+ MDB_page *mpd, *mps;
+
+ mps = csrc->mc_pg[csrc->mc_top];
+ /* If we're adding on the left, bump others up */
+ if (!cdst->mc_ki[csrc->mc_top]) {
+ mpd = cdst->mc_pg[csrc->mc_top];
+ for (m2 = csrc->mc_txn->mt_cursors[dbi]; m2; m2=m2->mc_next) {
+ if (csrc->mc_flags & C_SUB)
+ m3 = &m2->mc_xcursor->mx_cursor;
+ else
+ m3 = m2;
+ if (m3 != cdst &&
+ m3->mc_pg[csrc->mc_top] == mpd &&
+ m3->mc_ki[csrc->mc_top] >= cdst->mc_ki[csrc->mc_top]) {
+ m3->mc_ki[csrc->mc_top]++;
+ }
+ if (m3 !=csrc &&
+ m3->mc_pg[csrc->mc_top] == mps &&
+ 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];
+ m3->mc_ki[csrc->mc_top-1]++;
+ }
}
- }
-
- mp = csrc->mc_pg[csrc->mc_top];
- for (m2 = csrc->mc_txn->mt_cursors[dbi]; m2; m2=m2->mc_next) {
- if (csrc->mc_flags & C_SUB)
- m3 = &m2->mc_xcursor->mx_cursor;
- else
- m3 = m2;
- if (m3 == csrc) continue;
- 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];
+ } else
+ /* Adding on the right, bump others down */
+ {
+ for (m2 = csrc->mc_txn->mt_cursors[dbi]; m2; m2=m2->mc_next) {
+ if (csrc->mc_flags & C_SUB)
+ m3 = &m2->mc_xcursor->mx_cursor;
+ else
+ m3 = m2;
+ if (m3 == csrc) continue;
+ if (m3->mc_pg[csrc->mc_top] == mps) {
+ if (!m3->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];
+ m3->mc_ki[csrc->mc_top-1]--;
+ } else {
+ m3->mc_ki[csrc->mc_top]--;
+ }
+ }
}
}
}
if ((rc = mdb_page_touch(cdst)))
return rc;
+ /* get dst page again now that we've touched it. */
+ pdst = cdst->mc_pg[cdst->mc_top];
+
/* Move all nodes from src to dst.
*/
j = nkeys = NUMKEYS(pdst);
/* Adjust other cursors pointing to mp */
MDB_cursor *m2, *m3;
MDB_dbi dbi = csrc->mc_dbi;
+ unsigned int top = csrc->mc_top;
for (m2 = csrc->mc_txn->mt_cursors[dbi]; m2; m2=m2->mc_next) {
if (csrc->mc_flags & C_SUB)
m3 = m2;
if (m3 == csrc) continue;
if (m3->mc_snum < csrc->mc_snum) continue;
- if (m3->mc_pg[csrc->mc_top] == psrc) {
- m3->mc_pg[csrc->mc_top] = pdst;
- m3->mc_ki[csrc->mc_top] += nkeys;
+ if (m3->mc_pg[top] == psrc) {
+ m3->mc_pg[top] = pdst;
+ m3->mc_ki[top] += nkeys;
+ m3->mc_ki[top-1] = cdst->mc_ki[top-1];
+ } else if (m3->mc_pg[top-1] == csrc->mc_pg[top-1] &&
+ m3->mc_ki[top-1] > csrc->mc_ki[top-1]) {
+ m3->mc_ki[top-1]--;
}
}
}
m3 = &m2->mc_xcursor->mx_cursor;
else
m3 = m2;
- if (m3 == mc || m3->mc_snum < mc->mc_snum) continue;
+ if (m3 == mc) continue;
if (m3->mc_pg[0] == mp) {
- for (i=0; i<m3->mc_snum; i++) {
+ for (i=0; i<mc->mc_db->md_depth; i++) {
m3->mc_pg[i] = m3->mc_pg[i+1];
m3->mc_ki[i] = m3->mc_ki[i+1];
}
*/
if (PAGEFILL(mc->mc_txn->mt_env, mn.mc_pg[mn.mc_top]) >= thresh && NUMKEYS(mn.mc_pg[mn.mc_top]) > minkeys) {
rc = mdb_node_move(&mn, mc);
- if (mc->mc_ki[mc->mc_top-1]) {
+ if (!mc->mc_ki[mc->mc_top]) {
+ /* if we inserted on left, bump position up */
oldki++;
}
} else {
mn.mc_ki[mn.mc_top] += mc->mc_ki[mn.mc_top] + 1;
/* We want mdb_rebalance to find mn when doing fixups */
if (mc->mc_flags & C_SUB) {
+ dummy.mc_flags = C_INITIALIZED;
dummy.mc_next = mc->mc_txn->mt_cursors[mc->mc_dbi];
mc->mc_txn->mt_cursors[mc->mc_dbi] = &dummy;
dummy.mc_xcursor = (MDB_xcursor *)&mn;
rp->mp_pad = mp->mp_pad;
DPRINTF(("new right sibling: page %"Z"u", rp->mp_pgno));
- if (mc->mc_snum < 2) {
+ /* Usually when splitting the root page, the cursor
+ * height is 1. But when called from mdb_update_key,
+ * the cursor height may be greater because it walks
+ * up the stack while finding the branch slot to update.
+ */
+ if (mc->mc_top < 1) {
if ((rc = mdb_page_new(mc, P_BRANCH, 1, &pp)))
goto done;
/* shift current top to make room for new parent */
- mc->mc_pg[1] = mc->mc_pg[0];
- mc->mc_ki[1] = mc->mc_ki[0];
+ for (i=mc->mc_snum; i>0; i--) {
+ mc->mc_pg[i] = mc->mc_pg[i-1];
+ mc->mc_ki[i] = mc->mc_ki[i-1];
+ }
mc->mc_pg[0] = pp;
mc->mc_ki[0] = 0;
mc->mc_db->md_root = pp->mp_pgno;
mc->mc_db->md_depth--;
goto done;
}
- mc->mc_snum = 2;
- mc->mc_top = 1;
+ mc->mc_snum++;
+ mc->mc_top++;
ptop = 0;
} else {
ptop = mc->mc_top-1;
rp->mp_upper -= ksize - sizeof(indx_t);
mc->mc_ki[mc->mc_top] = x;
mc->mc_pg[mc->mc_top] = rp;
+ mc->mc_ki[ptop]++;
}
} else {
int psize, nsize, k;
/* Copy separator key to the parent.
*/
if (SIZELEFT(mn.mc_pg[ptop]) < mdb_branch_size(env, &sepkey)) {
+ int snum = mc->mc_snum;
mn.mc_snum--;
mn.mc_top--;
did_split = 1;
goto done;
/* root split? */
- if (mn.mc_snum == mc->mc_snum) {
- mc->mc_pg[mc->mc_snum] = mc->mc_pg[mc->mc_top];
- mc->mc_ki[mc->mc_snum] = mc->mc_ki[mc->mc_top];
- mc->mc_pg[mc->mc_top] = mc->mc_pg[ptop];
- mc->mc_ki[mc->mc_top] = mc->mc_ki[ptop];
- mc->mc_snum++;
- mc->mc_top++;
+ if (mc->mc_snum > snum) {
ptop++;
}
/* Right page might now have changed parent.
continue;
if (!(m2->mc_flags & m3->mc_flags & C_INITIALIZED))
continue;
- if (m3->mc_flags & C_SPLITTING)
- continue;
if (new_root) {
int k;
/* root split */
m3->mc_snum++;
m3->mc_top++;
}
+ if (m3->mc_flags & C_SPLITTING)
+ continue;
if (m3->mc_top >= mc->mc_top && m3->mc_pg[mc->mc_top] == mp) {
if (m3->mc_ki[mc->mc_top] >= newindx && !(nflags & MDB_SPLIT_REPLACE))
m3->mc_ki[mc->mc_top]++;
int rc, len;
char *lpath;
HANDLE newfd = INVALID_HANDLE_VALUE;
+#ifdef _WIN32
+ wchar_t *wpath;
+#endif
if (env->me_flags & MDB_NOSUBDIR) {
lpath = (char *)path;
* already in the OS cache.
*/
#ifdef _WIN32
- newfd = CreateFileA(lpath, GENERIC_WRITE, 0, NULL, CREATE_NEW,
+ utf8_to_utf16(lpath, -1, &wpath, NULL);
+ newfd = CreateFileW(wpath, GENERIC_WRITE, 0, NULL, CREATE_NEW,
FILE_FLAG_NO_BUFFERING|FILE_FLAG_WRITE_THROUGH, NULL);
+ free(wpath);
#else
newfd = open(lpath, O_WRONLY|O_CREAT|O_EXCL, 0666);
#endif
}
#endif /* MDB_ROBUST_SUPPORTED */
/** @} */
+
+#if defined(_WIN32)
+static int utf8_to_utf16(const char *src, int srcsize, wchar_t **dst, int *dstsize)
+{
+ int need;
+ wchar_t *result;
+ need = MultiByteToWideChar(CP_UTF8, 0, src, srcsize, NULL, 0);
+ if (need == 0xFFFD)
+ return EILSEQ;
+ if (need == 0)
+ return EINVAL;
+ result = malloc(sizeof(wchar_t) * need);
+ MultiByteToWideChar(CP_UTF8, 0, src, srcsize, result, need);
+ if (dstsize)
+ *dstsize = need;
+ *dst = result;
+ return 0;
+}
+#endif /* defined(_WIN32) */