#include <resolv.h> /* defines BYTE_ORDER on HPUX and Solaris */
#endif
-#if defined(__APPLE__) || defined (BSD)
+#if defined(__APPLE__) || defined (BSD) || defined(__FreeBSD_kernel__)
# if !(defined(MDB_USE_POSIX_MUTEX) || defined(MDB_USE_POSIX_SEM))
# define MDB_USE_SYSV_SEM 1
# endif
#ifndef _WIN32
#include <pthread.h>
+#include <signal.h>
#ifdef MDB_USE_POSIX_SEM
# define MDB_USE_HASH 1
#include <semaphore.h>
#define mdb_mutex_consistent(mutex) 0
#else /* MDB_USE_POSIX_MUTEX: */
- /** Shared mutex/semaphore as it is stored (mdb_mutex_t), and as
- * local variables keep it (mdb_mutexref_t).
+ /** Shared mutex/semaphore as the original is stored.
*
- * An mdb_mutex_t can be assigned to an mdb_mutexref_t. They can
- * be the same, or an array[size 1] and a pointer.
- * @{
+ * Not for copies. Instead it can be assigned to an #mdb_mutexref_t.
+ * When mdb_mutexref_t is a pointer and mdb_mutex_t is not, then it
+ * is array[size 1] so it can be assigned to the pointer.
*/
-typedef pthread_mutex_t mdb_mutex_t[1], *mdb_mutexref_t;
- /* @} */
+typedef pthread_mutex_t mdb_mutex_t[1];
+ /** Reference to an #mdb_mutex_t */
+typedef pthread_mutex_t *mdb_mutexref_t;
/** Lock the reader or writer mutex.
* Returns 0 or a code to give #mdb_mutex_failed(), as in #LOCK_MUTEX().
*/
#define Z MDB_FMT_Z /**< printf/scanf format modifier for size_t */
#define Yu MDB_PRIy(u) /**< printf format for #mdb_size_t */
-#define Yd MDB_PRIy(d) /**< printf format for "signed #mdb_size_t" */
+#define Yd MDB_PRIy(d) /**< printf format for 'signed #mdb_size_t' */
#if defined(_WIN32) || defined(MDB_USE_POSIX_SEM)
#define MNAME_LEN 32
/** Header for a single key/data pair within a page.
* Used in pages of type #P_BRANCH and #P_LEAF without #P_LEAF2.
* We guarantee 2-byte alignment for 'MDB_node's.
+ *
+ * #mn_lo and #mn_hi are used for data size on leaf nodes, and for child
+ * pgno on branch nodes. On 64 bit platforms, #mn_flags is also used
+ * for pgno. (Branch nodes have no flags). Lo and hi are in host byte
+ * order in case some accesses can be optimized to 32-bit word access.
+ *
+ * Leaf node flags describe node contents. #F_BIGDATA says the node's
+ * data part is the page number of an overflow page with actual data.
+ * #F_DUPDATA and #F_SUBDATA can be combined giving duplicate data in
+ * a sub-page/sub-database, and named databases (just #F_SUBDATA).
*/
typedef struct MDB_node {
- /** lo and hi are used for data size on leaf nodes and for
- * child pgno on branch nodes. On 64 bit platforms, flags
- * is also used for pgno. (Branch nodes have no flags).
- * They are in host byte order in case that lets some
- * accesses be optimized into a 32-bit word access.
- */
+ /** part of data size or pgno
+ * @{ */
#if BYTE_ORDER == LITTLE_ENDIAN
- unsigned short mn_lo, mn_hi; /**< part of data size or pgno */
+ unsigned short mn_lo, mn_hi;
#else
unsigned short mn_hi, mn_lo;
#endif
+ /** @} */
/** @defgroup mdb_node Node Flags
* @ingroup internal
* Flags for node headers.
unsigned i, j;
int rc = MDB_SUCCESS, level;
- /* Mark pages seen by cursors */
- if (mc->mc_flags & C_UNTRACK)
- mc = NULL; /* will find mc in mt_cursors */
- for (i = txn->mt_numdbs;; mc = txn->mt_cursors[--i]) {
- for (; mc; mc=mc->mc_next) {
- if (!(mc->mc_flags & C_INITIALIZED))
- continue;
+ /* Mark pages seen by cursors: First m0, then tracked cursors */
+ for (i = txn->mt_numdbs;; ) {
+ if (mc->mc_flags & C_INITIALIZED) {
for (m3 = mc;; m3 = &mx->mx_cursor) {
mp = NULL;
for (j=0; j<m3->mc_snum; j++) {
break;
}
}
- if (i == 0)
- break;
+ mc = mc->mc_next;
+ for (; !mc || mc == m0; mc = txn->mt_cursors[--i])
+ if (i == 0)
+ goto mark_done;
}
+mark_done:
if (all) {
/* Mark dirty root pages */
for (i=0; i<txn->mt_numdbs; i++) {
if (flags & (MDB_PS_FIRST|MDB_PS_LAST)) {
i = 0;
- if (flags & MDB_PS_LAST)
+ if (flags & MDB_PS_LAST) {
i = NUMKEYS(mp) - 1;
+ /* if already init'd, see if we're already in right place */
+ if (mc->mc_flags & C_INITIALIZED) {
+ if (mc->mc_ki[mc->mc_top] == i) {
+ mc->mc_top = mc->mc_snum++;
+ mp = mc->mc_pg[mc->mc_top];
+ goto ready;
+ }
+ }
+ }
} else {
int exact;
node = mdb_node_search(mc, key, &exact);
if ((rc = mdb_cursor_push(mc, mp)))
return rc;
+ready:
if (flags & MDB_PS_MODIFY) {
if ((rc = mdb_page_touch(mc)) != 0)
return rc;
MDB_node *leaf;
int rc;
- if ((mc->mc_flags & C_EOF) ||
- ((mc->mc_flags & C_DEL) && op == MDB_NEXT_DUP)) {
+ if ((mc->mc_flags & C_DEL && op == MDB_NEXT_DUP))
return MDB_NOTFOUND;
- }
+
if (!(mc->mc_flags & C_INITIALIZED))
return mdb_cursor_first(mc, key, data);
mp = mc->mc_pg[mc->mc_top];
+ if (mc->mc_flags & C_EOF) {
+ if (mc->mc_ki[mc->mc_top] >= NUMKEYS(mp)-1)
+ return MDB_NOTFOUND;
+ mc->mc_flags ^= C_EOF;
+ }
+
if (mc->mc_db->md_flags & MDB_DUPSORT) {
leaf = NODEPTR(mp, mc->mc_ki[mc->mc_top]);
if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
mc->mc_xcursor->mx_cursor.mc_flags &= ~(C_INITIALIZED|C_EOF);
}
- if (!(mc->mc_flags & C_EOF)) {
-
- if (!(mc->mc_flags & C_INITIALIZED) || mc->mc_top) {
- rc = mdb_page_search(mc, NULL, MDB_PS_LAST);
- if (rc != MDB_SUCCESS)
- return rc;
- }
- mdb_cassert(mc, IS_LEAF(mc->mc_pg[mc->mc_top]));
-
+ if (!(mc->mc_flags & C_INITIALIZED) || mc->mc_top) {
+ rc = mdb_page_search(mc, NULL, MDB_PS_LAST);
+ if (rc != MDB_SUCCESS)
+ return rc;
}
+ mdb_cassert(mc, IS_LEAF(mc->mc_pg[mc->mc_top]));
+
mc->mc_ki[mc->mc_top] = NUMKEYS(mc->mc_pg[mc->mc_top]) - 1;
mc->mc_flags |= C_INITIALIZED|C_EOF;
leaf = NODEPTR(mc->mc_pg[mc->mc_top], mc->mc_ki[mc->mc_top]);
rc = MDB_INCOMPATIBLE;
break;
}
- rc = MDB_SUCCESS;
- if (!(mc->mc_xcursor->mx_cursor.mc_flags & C_INITIALIZED) ||
- (mc->mc_xcursor->mx_cursor.mc_flags & C_EOF))
+ if (!(mc->mc_xcursor->mx_cursor.mc_flags & C_INITIALIZED)) {
+ rc = EINVAL;
break;
+ }
+ if (mc->mc_xcursor->mx_cursor.mc_flags & C_EOF) {
+ MDB_cursor *mx = &mc->mc_xcursor->mx_cursor;
+ if (mx->mc_ki[mx->mc_top] >= NUMKEYS(mx->mc_pg[mx->mc_top])-1) {
+ rc = MDB_NOTFOUND;
+ break;
+ }
+ mx->mc_flags ^= C_EOF;
+ }
+ rc = MDB_SUCCESS;
goto fetchm;
case MDB_NEXT_MULTIPLE:
if (data == NULL) {
if (!(mc->mc_flags & C_INITIALIZED))
return EINVAL;
- if (!mc->mc_snum || (mc->mc_flags & C_EOF))
+ if (!mc->mc_snum)
return MDB_NOTFOUND;
+ if (mc->mc_flags & C_EOF) {
+ if (mc->mc_ki[mc->mc_top] >= NUMKEYS(mc->mc_pg[mc->mc_top]))
+ return MDB_NOTFOUND;
+ mc->mc_flags ^= C_EOF;
+ }
+
leaf = NODEPTR(mc->mc_pg[mc->mc_top], mc->mc_ki[mc->mc_top]);
if (!F_ISSET(leaf->mn_flags, F_DUPDATA)) {
*countp = 1;
MDB_CURSOR_UNREF(mc, 0);
}
if (mc && !mc->mc_backup) {
- /* remove from txn, if tracked */
+ /* Remove from txn, if tracked.
+ * A read-only txn (!C_UNTRACK) may have been freed already,
+ * so do not peek inside it. Only write txns track cursors.
+ */
if ((mc->mc_flags & C_UNTRACK) && mc->mc_txn->mt_cursors) {
MDB_cursor **prev = &mc->mc_txn->mt_cursors[mc->mc_dbi];
while (*prev && *prev != mc) prev = &(*prev)->mc_next;
* run out of space, triggering a split. We need this
* cursor to be consistent until the end of the rebalance.
*/
- mc.mc_flags |= C_UNTRACK;
mc.mc_next = txn->mt_cursors[dbi];
txn->mt_cursors[dbi] = &mc;
rc = mdb_cursor_del(&mc, flags);
} else {
/* find right page's left sibling */
mc->mc_ki[ptop] = mn.mc_ki[ptop];
- mdb_cursor_sibling(mc, 0);
+ rc = mdb_cursor_sibling(mc, 0);
}
}
} else {
mn.mc_top++;
}
if (rc != MDB_SUCCESS) {
+ if (rc == MDB_NOTFOUND) /* improper mdb_cursor_sibling() result */
+ rc = MDB_PROBLEM;
goto done;
}
if (nflags & MDB_APPEND) {
#else
int len;
#define DO_WRITE(rc, fd, ptr, w2, len) len = write(fd, ptr, w2); rc = (len >= 0)
+#ifdef SIGPIPE
+ sigset_t set;
+ sigemptyset(&set);
+ sigaddset(&set, SIGPIPE);
+ if ((rc = pthread_sigmask(SIG_BLOCK, &set, NULL)) != 0)
+ my->mc_error = rc;
+#endif
#endif
pthread_mutex_lock(&my->mc_mutex);
DO_WRITE(rc, my->mc_fd, ptr, wsize, len);
if (!rc) {
rc = ErrCode();
+#if defined(SIGPIPE) && !defined(_WIN32)
+ if (rc == EPIPE) {
+ /* Collect the pending SIGPIPE, otherwise at least OS X
+ * gives it to the process on thread-exit (ITS#8504).
+ */
+ int tmp;
+ sigwait(&set, &tmp);
+ }
+#endif
break;
} else if (len > 0) {
rc = MDB_SUCCESS;